From 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 Mon Sep 17 00:00:00 2001 From: Linus Torvalds Date: Sat, 16 Apr 2005 15:20:36 -0700 Subject: Linux-2.6.12-rc2 Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip! --- drivers/scsi/53c7xx.c | 6102 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 6102 insertions(+) create mode 100644 drivers/scsi/53c7xx.c (limited to 'drivers/scsi/53c7xx.c') diff --git a/drivers/scsi/53c7xx.c b/drivers/scsi/53c7xx.c new file mode 100644 index 000000000000..8ead55f75d02 --- /dev/null +++ b/drivers/scsi/53c7xx.c @@ -0,0 +1,6102 @@ +/* + * 53c710 driver. Modified from Drew Eckhardts driver + * for 53c810 by Richard Hirst [richard@sleepie.demon.co.uk] + * Check out PERM_OPTIONS and EXPECTED_CLOCK, which may be defined in the + * relevant machine specific file (eg. mvme16x.[ch], amiga7xx.[ch]). + * There are also currently some defines at the top of 53c7xx.scr. + * The chip type is #defined in script_asm.pl, as well as the Makefile. + * Host scsi ID expected to be 7 - see NCR53c7x0_init(). + * + * I have removed the PCI code and some of the 53c8xx specific code - + * simply to make this file smaller and easier to manage. + * + * MVME16x issues: + * Problems trying to read any chip registers in NCR53c7x0_init(), as they + * may never have been set by 16xBug (eg. If kernel has come in over tftp). + */ + +/* + * Adapted for Linux/m68k Amiga platforms for the A4000T/A4091 and + * WarpEngine SCSI controllers. + * By Alan Hourihane + * Thanks to Richard Hirst for making it possible with the MVME additions + */ + +/* + * 53c710 rev 0 doesn't support add with carry. Rev 1 and 2 does. To + * overcome this problem you can define FORCE_DSA_ALIGNMENT, which ensures + * that the DSA address is always xxxxxx00. If disconnection is not allowed, + * then the script only ever tries to add small (< 256) positive offsets to + * DSA, so lack of carry isn't a problem. FORCE_DSA_ALIGNMENT can, of course, + * be defined for all chip revisions at a small cost in memory usage. + */ + +#define FORCE_DSA_ALIGNMENT + +/* + * Selection timer does not always work on the 53c710, depending on the + * timing at the last disconnect, if this is a problem for you, try + * using validids as detailed below. + * + * Options for the NCR7xx driver + * + * noasync:0 - disables sync and asynchronous negotiation + * nosync:0 - disables synchronous negotiation (does async) + * nodisconnect:0 - disables disconnection + * validids:0x?? - Bitmask field that disallows certain ID's. + * - e.g. 0x03 allows ID 0,1 + * - 0x1F allows ID 0,1,2,3,4 + * opthi:n - replace top word of options with 'n' + * optlo:n - replace bottom word of options with 'n' + * - ALWAYS SPECIFY opthi THEN optlo <<<<<<<<<< + */ + +/* + * PERM_OPTIONS are driver options which will be enabled for all NCR boards + * in the system at driver initialization time. + * + * Don't THINK about touching these in PERM_OPTIONS : + * OPTION_MEMORY_MAPPED + * 680x0 doesn't have an IO map! + * + * OPTION_DEBUG_TEST1 + * Test 1 does bus mastering and interrupt tests, which will help weed + * out brain damaged main boards. + * + * Other PERM_OPTIONS settings are listed below. Note the actual options + * required are set in the relevant file (mvme16x.c, amiga7xx.c, etc): + * + * OPTION_NO_ASYNC + * Don't negotiate for asynchronous transfers on the first command + * when OPTION_ALWAYS_SYNCHRONOUS is set. Useful for dain bramaged + * devices which do something bad rather than sending a MESSAGE + * REJECT back to us like they should if they can't cope. + * + * OPTION_SYNCHRONOUS + * Enable support for synchronous transfers. Target negotiated + * synchronous transfers will be responded to. To initiate + * a synchronous transfer request, call + * + * request_synchronous (hostno, target) + * + * from within KGDB. + * + * OPTION_ALWAYS_SYNCHRONOUS + * Negotiate for synchronous transfers with every target after + * driver initialization or a SCSI bus reset. This is a bit dangerous, + * since there are some dain bramaged SCSI devices which will accept + * SDTR messages but keep talking asynchronously. + * + * OPTION_DISCONNECT + * Enable support for disconnect/reconnect. To change the + * default setting on a given host adapter, call + * + * request_disconnect (hostno, allow) + * + * where allow is non-zero to allow, 0 to disallow. + * + * If you really want to run 10MHz FAST SCSI-II transfers, you should + * know that the NCR driver currently ignores parity information. Most + * systems do 5MHz SCSI fine. I've seen a lot that have problems faster + * than 8MHz. To play it safe, we only request 5MHz transfers. + * + * If you'd rather get 10MHz transfers, edit sdtr_message and change + * the fourth byte from 50 to 25. + */ + +/* + * Sponsored by + * iX Multiuser Multitasking Magazine + * Hannover, Germany + * hm@ix.de + * + * Copyright 1993, 1994, 1995 Drew Eckhardt + * Visionary Computing + * (Unix and Linux consulting and custom programming) + * drew@PoohSticks.ORG + * +1 (303) 786-7975 + * + * TolerANT and SCSI SCRIPTS are registered trademarks of NCR Corporation. + * + * For more information, please consult + * + * NCR53C810 + * SCSI I/O Processor + * Programmer's Guide + * + * NCR 53C810 + * PCI-SCSI I/O Processor + * Data Manual + * + * NCR 53C810/53C820 + * PCI-SCSI I/O Processor Design In Guide + * + * For literature on Symbios Logic Inc. formerly NCR, SCSI, + * and Communication products please call (800) 334-5454 or + * (719) 536-3300. + * + * PCI BIOS Specification Revision + * PCI Local Bus Specification + * PCI System Design Guide + * + * PCI Special Interest Group + * M/S HF3-15A + * 5200 N.E. Elam Young Parkway + * Hillsboro, Oregon 97124-6497 + * +1 (503) 696-2000 + * +1 (800) 433-5177 + */ + +/* + * Design issues : + * The cumulative latency needed to propagate a read/write request + * through the file system, buffer cache, driver stacks, SCSI host, and + * SCSI device is ultimately the limiting factor in throughput once we + * have a sufficiently fast host adapter. + * + * So, to maximize performance we want to keep the ratio of latency to data + * transfer time to a minimum by + * 1. Minimizing the total number of commands sent (typical command latency + * including drive and bus mastering host overhead is as high as 4.5ms) + * to transfer a given amount of data. + * + * This is accomplished by placing no arbitrary limit on the number + * of scatter/gather buffers supported, since we can transfer 1K + * per scatter/gather buffer without Eric's cluster patches, + * 4K with. + * + * 2. Minimizing the number of fatal interrupts serviced, since + * fatal interrupts halt the SCSI I/O processor. Basically, + * this means offloading the practical maximum amount of processing + * to the SCSI chip. + * + * On the NCR53c810/820/720, this is accomplished by using + * interrupt-on-the-fly signals when commands complete, + * and only handling fatal errors and SDTR / WDTR messages + * in the host code. + * + * On the NCR53c710, interrupts are generated as on the NCR53c8x0, + * only the lack of a interrupt-on-the-fly facility complicates + * things. Also, SCSI ID registers and commands are + * bit fielded rather than binary encoded. + * + * On the NCR53c700 and NCR53c700-66, operations that are done via + * indirect, table mode on the more advanced chips must be + * replaced by calls through a jump table which + * acts as a surrogate for the DSA. Unfortunately, this + * will mean that we must service an interrupt for each + * disconnect/reconnect. + * + * 3. Eliminating latency by pipelining operations at the different levels. + * + * This driver allows a configurable number of commands to be enqueued + * for each target/lun combination (experimentally, I have discovered + * that two seems to work best) and will ultimately allow for + * SCSI-II tagged queuing. + * + * + * Architecture : + * This driver is built around a Linux queue of commands waiting to + * be executed, and a shared Linux/NCR array of commands to start. Commands + * are transferred to the array by the run_process_issue_queue() function + * which is called whenever a command completes. + * + * As commands are completed, the interrupt routine is triggered, + * looks for commands in the linked list of completed commands with + * valid status, removes these commands from a list of running commands, + * calls the done routine, and flags their target/luns as not busy. + * + * Due to limitations in the intelligence of the NCR chips, certain + * concessions are made. In many cases, it is easier to dynamically + * generate/fix-up code rather than calculate on the NCR at run time. + * So, code is generated or fixed up for + * + * - Handling data transfers, using a variable number of MOVE instructions + * interspersed with CALL MSG_IN, WHEN MSGIN instructions. + * + * The DATAIN and DATAOUT routines are separate, so that an incorrect + * direction can be trapped, and space isn't wasted. + * + * It may turn out that we're better off using some sort + * of table indirect instruction in a loop with a variable + * sized table on the NCR53c710 and newer chips. + * + * - Checking for reselection (NCR53c710 and better) + * + * - Handling the details of SCSI context switches (NCR53c710 and better), + * such as reprogramming appropriate synchronous parameters, + * removing the dsa structure from the NCR's queue of outstanding + * commands, etc. + * + */ + +#include + +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#ifdef CONFIG_AMIGA +#include +#include +#include + +#define BIG_ENDIAN +#define NO_IO_SPACE +#endif + +#ifdef CONFIG_MVME16x +#include + +#define BIG_ENDIAN +#define NO_IO_SPACE +#define VALID_IDS +#endif + +#ifdef CONFIG_BVME6000 +#include + +#define BIG_ENDIAN +#define NO_IO_SPACE +#define VALID_IDS +#endif + +#include "scsi.h" +#include +#include "53c7xx.h" +#include +#include + +#ifdef NO_IO_SPACE +/* + * The following make the definitions in 53c7xx.h (write8, etc) smaller, + * we don't have separate i/o space anyway. + */ +#undef inb +#undef outb +#undef inw +#undef outw +#undef inl +#undef outl +#define inb(x) 1 +#define inw(x) 1 +#define inl(x) 1 +#define outb(x,y) 1 +#define outw(x,y) 1 +#define outl(x,y) 1 +#endif + +static int check_address (unsigned long addr, int size); +static void dump_events (struct Scsi_Host *host, int count); +static Scsi_Cmnd * return_outstanding_commands (struct Scsi_Host *host, + int free, int issue); +static void hard_reset (struct Scsi_Host *host); +static void ncr_scsi_reset (struct Scsi_Host *host); +static void print_lots (struct Scsi_Host *host); +static void set_synchronous (struct Scsi_Host *host, int target, int sxfer, + int scntl3, int now_connected); +static int datapath_residual (struct Scsi_Host *host); +static const char * sbcl_to_phase (int sbcl); +static void print_progress (Scsi_Cmnd *cmd); +static void print_queues (struct Scsi_Host *host); +static void process_issue_queue (unsigned long flags); +static int shutdown (struct Scsi_Host *host); +static void abnormal_finished (struct NCR53c7x0_cmd *cmd, int result); +static int disable (struct Scsi_Host *host); +static int NCR53c7xx_run_tests (struct Scsi_Host *host); +static irqreturn_t NCR53c7x0_intr(int irq, void *dev_id, struct pt_regs * regs); +static void NCR53c7x0_intfly (struct Scsi_Host *host); +static int ncr_halt (struct Scsi_Host *host); +static void intr_phase_mismatch (struct Scsi_Host *host, struct NCR53c7x0_cmd + *cmd); +static void intr_dma (struct Scsi_Host *host, struct NCR53c7x0_cmd *cmd); +static void print_dsa (struct Scsi_Host *host, u32 *dsa, + const char *prefix); +static int print_insn (struct Scsi_Host *host, const u32 *insn, + const char *prefix, int kernel); + +static void NCR53c7xx_dsa_fixup (struct NCR53c7x0_cmd *cmd); +static void NCR53c7x0_init_fixup (struct Scsi_Host *host); +static int NCR53c7x0_dstat_sir_intr (struct Scsi_Host *host, struct + NCR53c7x0_cmd *cmd); +static void NCR53c7x0_soft_reset (struct Scsi_Host *host); + +/* Size of event list (per host adapter) */ +static int track_events = 0; +static struct Scsi_Host *first_host = NULL; /* Head of list of NCR boards */ +static Scsi_Host_Template *the_template = NULL; + +/* NCR53c710 script handling code */ + +#include "53c7xx_d.h" +#ifdef A_int_debug_sync +#define DEBUG_SYNC_INTR A_int_debug_sync +#endif +int NCR53c7xx_script_len = sizeof (SCRIPT); +int NCR53c7xx_dsa_len = A_dsa_end + Ent_dsa_zero - Ent_dsa_code_template; +#ifdef FORCE_DSA_ALIGNMENT +int CmdPageStart = (0 - Ent_dsa_zero - sizeof(struct NCR53c7x0_cmd)) & 0xff; +#endif + +static char *setup_strings[] = + {"","","","","","","",""}; + +#define MAX_SETUP_STRINGS (sizeof(setup_strings) / sizeof(char *)) +#define SETUP_BUFFER_SIZE 200 +static char setup_buffer[SETUP_BUFFER_SIZE]; +static char setup_used[MAX_SETUP_STRINGS]; + +void ncr53c7xx_setup (char *str, int *ints) +{ + int i; + char *p1, *p2; + + p1 = setup_buffer; + *p1 = '\0'; + if (str) + strncpy(p1, str, SETUP_BUFFER_SIZE - strlen(setup_buffer)); + setup_buffer[SETUP_BUFFER_SIZE - 1] = '\0'; + p1 = setup_buffer; + i = 0; + while (*p1 && (i < MAX_SETUP_STRINGS)) { + p2 = strchr(p1, ','); + if (p2) { + *p2 = '\0'; + if (p1 != p2) + setup_strings[i] = p1; + p1 = p2 + 1; + i++; + } + else { + setup_strings[i] = p1; + break; + } + } + for (i=0; i= '0') && (*cp <= '9')) { + *val = simple_strtoul(cp,NULL,0); + } + return ++x; +} + + + +/* + * KNOWN BUGS : + * - There is some sort of conflict when the PPP driver is compiled with + * support for 16 channels? + * + * - On systems which predate the 1.3.x initialization order change, + * the NCR driver will cause Cannot get free page messages to appear. + * These are harmless, but I don't know of an easy way to avoid them. + * + * - With OPTION_DISCONNECT, on two systems under unknown circumstances, + * we get a PHASE MISMATCH with DSA set to zero (suggests that we + * are occurring somewhere in the reselection code) where + * DSP=some value DCMD|DBC=same value. + * + * Closer inspection suggests that we may be trying to execute + * some portion of the DSA? + * scsi0 : handling residual transfer (+ 0 bytes from DMA FIFO) + * scsi0 : handling residual transfer (+ 0 bytes from DMA FIFO) + * scsi0 : no current command : unexpected phase MSGIN. + * DSP=0x1c46cc, DCMD|DBC=0x1c46ac, DSA=0x0 + * DSPS=0x0, TEMP=0x1c3e70, DMODE=0x80 + * scsi0 : DSP-> + * 001c46cc : 0x001c46cc 0x00000000 + * 001c46d4 : 0x001c5ea0 0x000011f8 + * + * Changed the print code in the phase_mismatch handler so + * that we call print_lots to try to diagnose this. + * + */ + +/* + * Possible future direction of architecture for max performance : + * + * We're using a single start array for the NCR chip. This is + * sub-optimal, because we cannot add a command which would conflict with + * an executing command to this start queue, and therefore must insert the + * next command for a given I/T/L combination after the first has completed; + * incurring our interrupt latency between SCSI commands. + * + * To allow further pipelining of the NCR and host CPU operation, we want + * to set things up so that immediately on termination of a command destined + * for a given LUN, we get that LUN busy again. + * + * To do this, we need to add a 32 bit pointer to which is jumped to + * on completion of a command. If no new command is available, this + * would point to the usual DSA issue queue select routine. + * + * If one were, it would point to a per-NCR53c7x0_cmd select routine + * which starts execution immediately, inserting the command at the head + * of the start queue if the NCR chip is selected or reselected. + * + * We would change so that we keep a list of outstanding commands + * for each unit, rather than a single running_list. We'd insert + * a new command into the right running list; if the NCR didn't + * have something running for that yet, we'd put it in the + * start queue as well. Some magic needs to happen to handle the + * race condition between the first command terminating before the + * new one is written. + * + * Potential for profiling : + * Call do_gettimeofday(struct timeval *tv) to get 800ns resolution. + */ + + +/* + * TODO : + * 1. To support WIDE transfers, not much needs to happen. We + * should do CHMOVE instructions instead of MOVEs when + * we have scatter/gather segments of uneven length. When + * we do this, we need to handle the case where we disconnect + * between segments. + * + * 2. Currently, when Icky things happen we do a FATAL(). Instead, + * we want to do an integrity check on the parts of the NCR hostdata + * structure which were initialized at boot time; FATAL() if that + * fails, and otherwise try to recover. Keep track of how many + * times this has happened within a single SCSI command; if it + * gets excessive, then FATAL(). + * + * 3. Parity checking is currently disabled, and a few things should + * happen here now that we support synchronous SCSI transfers : + * 1. On soft-reset, we shoould set the EPC (Enable Parity Checking) + * and AAP (Assert SATN/ on parity error) bits in SCNTL0. + * + * 2. We should enable the parity interrupt in the SIEN0 register. + * + * 3. intr_phase_mismatch() needs to believe that message out is + * always an "acceptable" phase to have a mismatch in. If + * the old phase was MSG_IN, we should send a MESSAGE PARITY + * error. If the old phase was something else, we should send + * a INITIATOR_DETECTED_ERROR message. Note that this could + * cause a RESTORE POINTERS message; so we should handle that + * correctly first. Instead, we should probably do an + * initiator_abort. + * + * 4. MPEE bit of CTEST4 should be set so we get interrupted if + * we detect an error. + * + * + * 5. The initial code has been tested on the NCR53c810. I don't + * have access to NCR53c700, 700-66 (Forex boards), NCR53c710 + * (NCR Pentium systems), NCR53c720, NCR53c820, or NCR53c825 boards to + * finish development on those platforms. + * + * NCR53c820/825/720 - need to add wide transfer support, including WDTR + * negotiation, programming of wide transfer capabilities + * on reselection and table indirect selection. + * + * NCR53c710 - need to add fatal interrupt or GEN code for + * command completion signaling. Need to modify all + * SDID, SCID, etc. registers, and table indirect select code + * since these use bit fielded (ie 1<NOP_insn) ? + /* + * If the IF TRUE bit is set, it's a JUMP instruction. The + * operand is a bus pointer to the dsa_begin routine for this DSA. The + * dsa field of the NCR53c7x0_cmd structure starts with the + * DSA code template. By converting to a virtual address, + * subtracting the code template size, and offset of the + * dsa field, we end up with a pointer to the start of the + * structure (alternatively, we could use the + * dsa_cmnd field, an anachronism from when we weren't + * sure what the relationship between the NCR structures + * and host structures were going to be. + */ + (struct NCR53c7x0_cmd *) ((char *) bus_to_virt (issue[1]) - + (hostdata->E_dsa_code_begin - hostdata->E_dsa_code_template) - + offsetof(struct NCR53c7x0_cmd, dsa)) + /* If the IF TRUE bit is not set, it's a NOP */ + : NULL; +} + + +/* + * FIXME: we should junk these, in favor of synchronous_want and + * wide_want in the NCR53c7x0_hostdata structure. + */ + +/* Template for "preferred" synchronous transfer parameters. */ + +static const unsigned char sdtr_message[] = { +#ifdef CONFIG_SCSI_NCR53C7xx_FAST + EXTENDED_MESSAGE, 3 /* length */, EXTENDED_SDTR, 25 /* *4ns */, 8 /* off */ +#else + EXTENDED_MESSAGE, 3 /* length */, EXTENDED_SDTR, 50 /* *4ns */, 8 /* off */ +#endif +}; + +/* Template to request asynchronous transfers */ + +static const unsigned char async_message[] = { + EXTENDED_MESSAGE, 3 /* length */, EXTENDED_SDTR, 0, 0 /* asynchronous */ +}; + +/* Template for "preferred" WIDE transfer parameters */ + +static const unsigned char wdtr_message[] = { + EXTENDED_MESSAGE, 2 /* length */, EXTENDED_WDTR, 1 /* 2^1 bytes */ +}; + +#if 0 +/* + * Function : struct Scsi_Host *find_host (int host) + * + * Purpose : KGDB support function which translates a host number + * to a host structure. + * + * Inputs : host - number of SCSI host + * + * Returns : NULL on failure, pointer to host structure on success. + */ + +static struct Scsi_Host * +find_host (int host) { + struct Scsi_Host *h; + for (h = first_host; h && h->host_no != host; h = h->next); + if (!h) { + printk (KERN_ALERT "scsi%d not found\n", host); + return NULL; + } else if (h->hostt != the_template) { + printk (KERN_ALERT "scsi%d is not a NCR board\n", host); + return NULL; + } + return h; +} + +#if 0 +/* + * Function : request_synchronous (int host, int target) + * + * Purpose : KGDB interface which will allow us to negotiate for + * synchronous transfers. This ill be replaced with a more + * integrated function; perhaps a new entry in the scsi_host + * structure, accessible via an ioctl() or perhaps /proc/scsi. + * + * Inputs : host - number of SCSI host; target - number of target. + * + * Returns : 0 when negotiation has been setup for next SCSI command, + * -1 on failure. + */ + +static int +request_synchronous (int host, int target) { + struct Scsi_Host *h; + struct NCR53c7x0_hostdata *hostdata; + unsigned long flags; + if (target < 0) { + printk (KERN_ALERT "target %d is bogus\n", target); + return -1; + } + if (!(h = find_host (host))) + return -1; + else if (h->this_id == target) { + printk (KERN_ALERT "target %d is host ID\n", target); + return -1; + } + else if (target > h->max_id) { + printk (KERN_ALERT "target %d exceeds maximum of %d\n", target, + h->max_id); + return -1; + } + hostdata = (struct NCR53c7x0_hostdata *)h->hostdata[0]; + + local_irq_save(flags); + if (hostdata->initiate_sdtr & (1 << target)) { + local_irq_restore(flags); + printk (KERN_ALERT "target %d already doing SDTR\n", target); + return -1; + } + hostdata->initiate_sdtr |= (1 << target); + local_irq_restore(flags); + return 0; +} +#endif + +/* + * Function : request_disconnect (int host, int on_or_off) + * + * Purpose : KGDB support function, tells us to allow or disallow + * disconnections. + * + * Inputs : host - number of SCSI host; on_or_off - non-zero to allow, + * zero to disallow. + * + * Returns : 0 on success, * -1 on failure. + */ + +static int +request_disconnect (int host, int on_or_off) { + struct Scsi_Host *h; + struct NCR53c7x0_hostdata *hostdata; + if (!(h = find_host (host))) + return -1; + hostdata = (struct NCR53c7x0_hostdata *) h->hostdata[0]; + if (on_or_off) + hostdata->options |= OPTION_DISCONNECT; + else + hostdata->options &= ~OPTION_DISCONNECT; + return 0; +} +#endif + +/* + * Function : static void NCR53c7x0_driver_init (struct Scsi_Host *host) + * + * Purpose : Initialize internal structures, as required on startup, or + * after a SCSI bus reset. + * + * Inputs : host - pointer to this host adapter's structure + */ + +static void +NCR53c7x0_driver_init (struct Scsi_Host *host) { + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + int i, j; + u32 *ncrcurrent; + + for (i = 0; i < 16; ++i) { + hostdata->request_sense[i] = 0; + for (j = 0; j < 8; ++j) + hostdata->busy[i][j] = 0; + set_synchronous (host, i, /* sxfer */ 0, hostdata->saved_scntl3, 0); + } + hostdata->issue_queue = NULL; + hostdata->running_list = hostdata->finished_queue = + hostdata->ncrcurrent = NULL; + for (i = 0, ncrcurrent = (u32 *) hostdata->schedule; + i < host->can_queue; ++i, ncrcurrent += 2) { + ncrcurrent[0] = hostdata->NOP_insn; + ncrcurrent[1] = 0xdeadbeef; + } + ncrcurrent[0] = ((DCMD_TYPE_TCI|DCMD_TCI_OP_JUMP) << 24) | DBC_TCI_TRUE; + ncrcurrent[1] = (u32) virt_to_bus (hostdata->script) + + hostdata->E_wait_reselect; + hostdata->reconnect_dsa_head = 0; + hostdata->addr_reconnect_dsa_head = (u32) + virt_to_bus((void *) &(hostdata->reconnect_dsa_head)); + hostdata->expecting_iid = 0; + hostdata->expecting_sto = 0; + if (hostdata->options & OPTION_ALWAYS_SYNCHRONOUS) + hostdata->initiate_sdtr = 0xffff; + else + hostdata->initiate_sdtr = 0; + hostdata->talked_to = 0; + hostdata->idle = 1; +} + +/* + * Function : static int clock_to_ccf_710 (int clock) + * + * Purpose : Return the clock conversion factor for a given SCSI clock. + * + * Inputs : clock - SCSI clock expressed in Hz. + * + * Returns : ccf on success, -1 on failure. + */ + +static int +clock_to_ccf_710 (int clock) { + if (clock <= 16666666) + return -1; + if (clock <= 25000000) + return 2; /* Divide by 1.0 */ + else if (clock <= 37500000) + return 1; /* Divide by 1.5 */ + else if (clock <= 50000000) + return 0; /* Divide by 2.0 */ + else if (clock <= 66000000) + return 3; /* Divide by 3.0 */ + else + return -1; +} + +/* + * Function : static int NCR53c7x0_init (struct Scsi_Host *host) + * + * Purpose : initialize the internal structures for a given SCSI host + * + * Inputs : host - pointer to this host adapter's structure + * + * Preconditions : when this function is called, the chip_type + * field of the hostdata structure MUST have been set. + * + * Returns : 0 on success, -1 on failure. + */ + +int +NCR53c7x0_init (struct Scsi_Host *host) { + NCR53c7x0_local_declare(); + int i, ccf; + unsigned char revision; + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + /* + * There are some things which we need to know about in order to provide + * a semblance of support. Print 'em if they aren't what we expect, + * otherwise don't add to the noise. + * + * -1 means we don't know what to expect. + */ + int val, flags; + char buf[32]; + int expected_id = -1; + int expected_clock = -1; + int uninitialized = 0; +#ifdef NO_IO_SPACE + int expected_mapping = OPTION_MEMORY_MAPPED; +#else + int expected_mapping = OPTION_IO_MAPPED; +#endif + for (i=0;i<7;i++) + hostdata->valid_ids[i] = 1; /* Default all ID's to scan */ + + /* Parse commandline flags */ + if (check_setup_strings("noasync",&flags,&val,buf)) + { + hostdata->options |= OPTION_NO_ASYNC; + hostdata->options &= ~(OPTION_SYNCHRONOUS | OPTION_ALWAYS_SYNCHRONOUS); + } + + if (check_setup_strings("nosync",&flags,&val,buf)) + { + hostdata->options &= ~(OPTION_SYNCHRONOUS | OPTION_ALWAYS_SYNCHRONOUS); + } + + if (check_setup_strings("nodisconnect",&flags,&val,buf)) + hostdata->options &= ~OPTION_DISCONNECT; + + if (check_setup_strings("validids",&flags,&val,buf)) + { + for (i=0;i<7;i++) + hostdata->valid_ids[i] = val & (1<options = (long long)val << 32; + if (check_setup_strings("optlo",&flags,&val,buf)) + hostdata->options |= val; + + NCR53c7x0_local_setup(host); + switch (hostdata->chip) { + case 710: + case 770: + hostdata->dstat_sir_intr = NCR53c7x0_dstat_sir_intr; + hostdata->init_save_regs = NULL; + hostdata->dsa_fixup = NCR53c7xx_dsa_fixup; + hostdata->init_fixup = NCR53c7x0_init_fixup; + hostdata->soft_reset = NCR53c7x0_soft_reset; + hostdata->run_tests = NCR53c7xx_run_tests; + expected_clock = hostdata->scsi_clock; + expected_id = 7; + break; + default: + printk ("scsi%d : chip type of %d is not supported yet, detaching.\n", + host->host_no, hostdata->chip); + scsi_unregister (host); + return -1; + } + + /* Assign constants accessed by NCR */ + hostdata->NCR53c7xx_zero = 0; + hostdata->NCR53c7xx_msg_reject = MESSAGE_REJECT; + hostdata->NCR53c7xx_msg_abort = ABORT; + hostdata->NCR53c7xx_msg_nop = NOP; + hostdata->NOP_insn = (DCMD_TYPE_TCI|DCMD_TCI_OP_JUMP) << 24; + if (expected_mapping == -1 || + (hostdata->options & (OPTION_MEMORY_MAPPED)) != + (expected_mapping & OPTION_MEMORY_MAPPED)) + printk ("scsi%d : using %s mapped access\n", host->host_no, + (hostdata->options & OPTION_MEMORY_MAPPED) ? "memory" : + "io"); + + hostdata->dmode = (hostdata->chip == 700 || hostdata->chip == 70066) ? + DMODE_REG_00 : DMODE_REG_10; + hostdata->istat = ((hostdata->chip / 100) == 8) ? + ISTAT_REG_800 : ISTAT_REG_700; + +/* We have to assume that this may be the first access to the chip, so + * we must set EA in DCNTL. */ + + NCR53c7x0_write8 (DCNTL_REG, DCNTL_10_EA|DCNTL_10_COM); + + +/* Only the ISTAT register is readable when the NCR is running, so make + sure it's halted. */ + ncr_halt(host); + +/* + * XXX - the NCR53c700 uses bitfielded registers for SCID, SDID, etc, + * as does the 710 with one bit per SCSI ID. Conversely, the NCR + * uses a normal, 3 bit binary representation of these values. + * + * Get the rest of the NCR documentation, and FIND OUT where the change + * was. + */ + +#if 0 + /* May not be able to do this - chip my not have been set up yet */ + tmp = hostdata->this_id_mask = NCR53c7x0_read8(SCID_REG); + for (host->this_id = 0; tmp != 1; tmp >>=1, ++host->this_id); +#else + host->this_id = 7; +#endif + +/* + * Note : we should never encounter a board setup for ID0. So, + * if we see ID0, assume that it was uninitialized and set it + * to the industry standard 7. + */ + if (!host->this_id) { + printk("scsi%d : initiator ID was %d, changing to 7\n", + host->host_no, host->this_id); + host->this_id = 7; + hostdata->this_id_mask = 1 << 7; + uninitialized = 1; + }; + + if (expected_id == -1 || host->this_id != expected_id) + printk("scsi%d : using initiator ID %d\n", host->host_no, + host->this_id); + + /* + * Save important registers to allow a soft reset. + */ + + /* + * CTEST7 controls cache snooping, burst mode, and support for + * external differential drivers. This isn't currently used - the + * default value may not be optimal anyway. + * Even worse, it may never have been set up since reset. + */ + hostdata->saved_ctest7 = NCR53c7x0_read8(CTEST7_REG) & CTEST7_SAVE; + revision = (NCR53c7x0_read8(CTEST8_REG) & 0xF0) >> 4; + switch (revision) { + case 1: revision = 0; break; + case 2: revision = 1; break; + case 4: revision = 2; break; + case 8: revision = 3; break; + default: revision = 255; break; + } + printk("scsi%d: Revision 0x%x\n",host->host_no,revision); + + if ((revision == 0 || revision == 255) && (hostdata->options & (OPTION_SYNCHRONOUS|OPTION_DISCONNECT|OPTION_ALWAYS_SYNCHRONOUS))) + { + printk ("scsi%d: Disabling sync working and disconnect/reselect\n", + host->host_no); + hostdata->options &= ~(OPTION_SYNCHRONOUS|OPTION_DISCONNECT|OPTION_ALWAYS_SYNCHRONOUS); + } + + /* + * On NCR53c700 series chips, DCNTL controls the SCSI clock divisor, + * on 800 series chips, it allows for a totem-pole IRQ driver. + * NOTE saved_dcntl currently overwritten in init function. + * The value read here may be garbage anyway, MVME16x board at least + * does not initialise chip if kernel arrived via tftp. + */ + + hostdata->saved_dcntl = NCR53c7x0_read8(DCNTL_REG); + + /* + * DMODE controls DMA burst length, and on 700 series chips, + * 286 mode and bus width + * NOTE: On MVME16x, chip may have been reset, so this could be a + * power-on/reset default value. + */ + hostdata->saved_dmode = NCR53c7x0_read8(hostdata->dmode); + + /* + * Now that burst length and enabled/disabled status is known, + * clue the user in on it. + */ + + ccf = clock_to_ccf_710 (expected_clock); + + for (i = 0; i < 16; ++i) + hostdata->cmd_allocated[i] = 0; + + if (hostdata->init_save_regs) + hostdata->init_save_regs (host); + if (hostdata->init_fixup) + hostdata->init_fixup (host); + + if (!the_template) { + the_template = host->hostt; + first_host = host; + } + + /* + * Linux SCSI drivers have always been plagued with initialization + * problems - some didn't work with the BIOS disabled since they expected + * initialization from it, some didn't work when the networking code + * was enabled and registers got scrambled, etc. + * + * To avoid problems like this, in the future, we will do a soft + * reset on the SCSI chip, taking it back to a sane state. + */ + + hostdata->soft_reset (host); + +#if 1 + hostdata->debug_count_limit = -1; +#else + hostdata->debug_count_limit = 1; +#endif + hostdata->intrs = -1; + hostdata->resets = -1; + memcpy ((void *) hostdata->synchronous_want, (void *) sdtr_message, + sizeof (hostdata->synchronous_want)); + + NCR53c7x0_driver_init (host); + + if (request_irq(host->irq, NCR53c7x0_intr, SA_SHIRQ, "53c7xx", host)) + { + printk("scsi%d : IRQ%d not free, detaching\n", + host->host_no, host->irq); + goto err_unregister; + } + + if ((hostdata->run_tests && hostdata->run_tests(host) == -1) || + (hostdata->options & OPTION_DEBUG_TESTS_ONLY)) { + /* XXX Should disable interrupts, etc. here */ + goto err_free_irq; + } else { + if (host->io_port) { + host->n_io_port = 128; + if (!request_region (host->io_port, host->n_io_port, "ncr53c7xx")) + goto err_free_irq; + } + } + + if (NCR53c7x0_read8 (SBCL_REG) & SBCL_BSY) { + printk ("scsi%d : bus wedge, doing SCSI reset\n", host->host_no); + hard_reset (host); + } + return 0; + + err_free_irq: + free_irq(host->irq, NCR53c7x0_intr); + err_unregister: + scsi_unregister(host); + return -1; +} + +/* + * Function : int ncr53c7xx_init(Scsi_Host_Template *tpnt, int board, int chip, + * unsigned long base, int io_port, int irq, int dma, long long options, + * int clock); + * + * Purpose : initializes a NCR53c7,8x0 based on base addresses, + * IRQ, and DMA channel. + * + * Inputs : tpnt - Template for this SCSI adapter, board - board level + * product, chip - 710 + * + * Returns : 0 on success, -1 on failure. + * + */ + +int +ncr53c7xx_init (Scsi_Host_Template *tpnt, int board, int chip, + unsigned long base, int io_port, int irq, int dma, + long long options, int clock) +{ + struct Scsi_Host *instance; + struct NCR53c7x0_hostdata *hostdata; + char chip_str[80]; + int script_len = 0, dsa_len = 0, size = 0, max_cmd_size = 0, + schedule_size = 0, ok = 0; + void *tmp; + unsigned long page; + + switch (chip) { + case 710: + case 770: + schedule_size = (tpnt->can_queue + 1) * 8 /* JUMP instruction size */; + script_len = NCR53c7xx_script_len; + dsa_len = NCR53c7xx_dsa_len; + options |= OPTION_INTFLY; + sprintf (chip_str, "NCR53c%d", chip); + break; + default: + printk("scsi-ncr53c7xx : unsupported SCSI chip %d\n", chip); + return -1; + } + + printk("scsi-ncr53c7xx : %s at memory 0x%lx, io 0x%x, irq %d", + chip_str, base, io_port, irq); + if (dma == DMA_NONE) + printk("\n"); + else + printk(", dma %d\n", dma); + + if (options & OPTION_DEBUG_PROBE_ONLY) { + printk ("scsi-ncr53c7xx : probe only enabled, aborting initialization\n"); + return -1; + } + + max_cmd_size = sizeof(struct NCR53c7x0_cmd) + dsa_len + + /* Size of dynamic part of command structure : */ + 2 * /* Worst case : we don't know if we need DATA IN or DATA out */ + ( 2 * /* Current instructions per scatter/gather segment */ + tpnt->sg_tablesize + + 3 /* Current startup / termination required per phase */ + ) * + 8 /* Each instruction is eight bytes */; + + /* Allocate fixed part of hostdata, dynamic part to hold appropriate + SCSI SCRIPT(tm) plus a single, maximum-sized NCR53c7x0_cmd structure. + + We need a NCR53c7x0_cmd structure for scan_scsis() when we are + not loaded as a module, and when we're loaded as a module, we + can't use a non-dynamically allocated structure because modules + are vmalloc()'d, which can allow structures to cross page + boundaries and breaks our physical/virtual address assumptions + for DMA. + + So, we stick it past the end of our hostdata structure. + + ASSUMPTION : + Regardless of how many simultaneous SCSI commands we allow, + the probe code only executes a _single_ instruction at a time, + so we only need one here, and don't need to allocate NCR53c7x0_cmd + structures for each target until we are no longer in scan_scsis + and kmalloc() has become functional (memory_init() happens + after all device driver initialization). + */ + + size = sizeof(struct NCR53c7x0_hostdata) + script_len + + /* Note that alignment will be guaranteed, since we put the command + allocated at probe time after the fixed-up SCSI script, which + consists of 32 bit words, aligned on a 32 bit boundary. But + on a 64bit machine we need 8 byte alignment for hostdata->free, so + we add in another 4 bytes to take care of potential misalignment + */ + (sizeof(void *) - sizeof(u32)) + max_cmd_size + schedule_size; + + page = __get_free_pages(GFP_ATOMIC,1); + if(page==0) + { + printk(KERN_ERR "53c7xx: out of memory.\n"); + return -ENOMEM; + } +#ifdef FORCE_DSA_ALIGNMENT + /* + * 53c710 rev.0 doesn't have an add-with-carry instruction. + * Ensure we allocate enough memory to force DSA alignment. + */ + size += 256; +#endif + /* Size should be < 8K, so we can fit it in two pages. */ + if (size > 8192) { + printk(KERN_ERR "53c7xx: hostdata > 8K\n"); + return -1; + } + + instance = scsi_register (tpnt, 4); + if (!instance) + { + free_page(page); + return -1; + } + instance->hostdata[0] = page; + memset((void *)instance->hostdata[0], 0, 8192); + cache_push(virt_to_phys((void *)(instance->hostdata[0])), 8192); + cache_clear(virt_to_phys((void *)(instance->hostdata[0])), 8192); + kernel_set_cachemode((void *)instance->hostdata[0], 8192, IOMAP_NOCACHE_SER); + + /* FIXME : if we ever support an ISA NCR53c7xx based board, we + need to check if the chip is running in a 16 bit mode, and if so + unregister it if it is past the 16M (0x1000000) mark */ + + hostdata = (struct NCR53c7x0_hostdata *)instance->hostdata[0]; + hostdata->size = size; + hostdata->script_count = script_len / sizeof(u32); + hostdata->board = board; + hostdata->chip = chip; + + /* + * Being memory mapped is more desirable, since + * + * - Memory accesses may be faster. + * + * - The destination and source address spaces are the same for + * all instructions, meaning we don't have to twiddle dmode or + * any other registers. + * + * So, we try for memory mapped, and if we don't get it, + * we go for port mapped, and that failing we tell the user + * it can't work. + */ + + if (base) { + instance->base = base; + /* Check for forced I/O mapping */ + if (!(options & OPTION_IO_MAPPED)) { + options |= OPTION_MEMORY_MAPPED; + ok = 1; + } + } else { + options &= ~OPTION_MEMORY_MAPPED; + } + + if (io_port) { + instance->io_port = io_port; + options |= OPTION_IO_MAPPED; + ok = 1; + } else { + options &= ~OPTION_IO_MAPPED; + } + + if (!ok) { + printk ("scsi%d : not initializing, no I/O or memory mapping known \n", + instance->host_no); + scsi_unregister (instance); + return -1; + } + instance->irq = irq; + instance->dma_channel = dma; + + hostdata->options = options; + hostdata->dsa_len = dsa_len; + hostdata->max_cmd_size = max_cmd_size; + hostdata->num_cmds = 1; + hostdata->scsi_clock = clock; + /* Initialize single command */ + tmp = (hostdata->script + hostdata->script_count); +#ifdef FORCE_DSA_ALIGNMENT + { + void *t = ROUNDUP(tmp, void *); + if (((u32)t & 0xff) > CmdPageStart) + t = (void *)((u32)t + 255); + t = (void *)(((u32)t & ~0xff) + CmdPageStart); + hostdata->free = t; +#if 0 + printk ("scsi: Registered size increased by 256 to %d\n", size); + printk ("scsi: CmdPageStart = 0x%02x\n", CmdPageStart); + printk ("scsi: tmp = 0x%08x, hostdata->free set to 0x%08x\n", + (u32)tmp, (u32)t); +#endif + } +#else + hostdata->free = ROUNDUP(tmp, void *); +#endif + hostdata->free->real = tmp; + hostdata->free->size = max_cmd_size; + hostdata->free->free = NULL; + hostdata->free->next = NULL; + hostdata->extra_allocate = 0; + + /* Allocate command start code space */ + hostdata->schedule = (chip == 700 || chip == 70066) ? + NULL : (u32 *) ((char *)hostdata->free + max_cmd_size); + +/* + * For diagnostic purposes, we don't really care how fast things blaze. + * For profiling, we want to access the 800ns resolution system clock, + * using a 'C' call on the host processor. + * + * Therefore, there's no need for the NCR chip to directly manipulate + * this data, and we should put it wherever is most convenient for + * Linux. + */ + if (track_events) + hostdata->events = (struct NCR53c7x0_event *) (track_events ? + vmalloc (sizeof (struct NCR53c7x0_event) * track_events) : NULL); + else + hostdata->events = NULL; + + if (hostdata->events) { + memset ((void *) hostdata->events, 0, sizeof(struct NCR53c7x0_event) * + track_events); + hostdata->event_size = track_events; + hostdata->event_index = 0; + } else + hostdata->event_size = 0; + + return NCR53c7x0_init(instance); +} + + +/* + * Function : static void NCR53c7x0_init_fixup (struct Scsi_Host *host) + * + * Purpose : copy and fixup the SCSI SCRIPTS(tm) code for this device. + * + * Inputs : host - pointer to this host adapter's structure + * + */ + +static void +NCR53c7x0_init_fixup (struct Scsi_Host *host) { + NCR53c7x0_local_declare(); + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + unsigned char tmp; + int i, ncr_to_memory, memory_to_ncr; + u32 base; + NCR53c7x0_local_setup(host); + + + /* XXX - NOTE : this code MUST be made endian aware */ + /* Copy code into buffer that was allocated at detection time. */ + memcpy ((void *) hostdata->script, (void *) SCRIPT, + sizeof(SCRIPT)); + /* Fixup labels */ + for (i = 0; i < PATCHES; ++i) + hostdata->script[LABELPATCHES[i]] += + virt_to_bus(hostdata->script); + /* Fixup addresses of constants that used to be EXTERNAL */ + + patch_abs_32 (hostdata->script, 0, NCR53c7xx_msg_abort, + virt_to_bus(&(hostdata->NCR53c7xx_msg_abort))); + patch_abs_32 (hostdata->script, 0, NCR53c7xx_msg_reject, + virt_to_bus(&(hostdata->NCR53c7xx_msg_reject))); + patch_abs_32 (hostdata->script, 0, NCR53c7xx_zero, + virt_to_bus(&(hostdata->NCR53c7xx_zero))); + patch_abs_32 (hostdata->script, 0, NCR53c7xx_sink, + virt_to_bus(&(hostdata->NCR53c7xx_sink))); + patch_abs_32 (hostdata->script, 0, NOP_insn, + virt_to_bus(&(hostdata->NOP_insn))); + patch_abs_32 (hostdata->script, 0, schedule, + virt_to_bus((void *) hostdata->schedule)); + + /* Fixup references to external variables: */ + for (i = 0; i < EXTERNAL_PATCHES_LEN; ++i) + hostdata->script[EXTERNAL_PATCHES[i].offset] += + virt_to_bus(EXTERNAL_PATCHES[i].address); + + /* + * Fixup absolutes set at boot-time. + * + * All non-code absolute variables suffixed with "dsa_" and "int_" + * are constants, and need no fixup provided the assembler has done + * it for us (I don't know what the "real" NCR assembler does in + * this case, my assembler does the right magic). + */ + + patch_abs_rwri_data (hostdata->script, 0, dsa_save_data_pointer, + Ent_dsa_code_save_data_pointer - Ent_dsa_zero); + patch_abs_rwri_data (hostdata->script, 0, dsa_restore_pointers, + Ent_dsa_code_restore_pointers - Ent_dsa_zero); + patch_abs_rwri_data (hostdata->script, 0, dsa_check_reselect, + Ent_dsa_code_check_reselect - Ent_dsa_zero); + + /* + * Just for the hell of it, preserve the settings of + * Burst Length and Enable Read Line bits from the DMODE + * register. Make sure SCRIPTS start automagically. + */ + +#if defined(CONFIG_MVME16x) || defined(CONFIG_BVME6000) + /* We know better what we want than 16xBug does! */ + tmp = DMODE_10_BL_8 | DMODE_10_FC2; +#else + tmp = NCR53c7x0_read8(DMODE_REG_10); + tmp &= (DMODE_BL_MASK | DMODE_10_FC2 | DMODE_10_FC1 | DMODE_710_PD | + DMODE_710_UO); +#endif + + if (!(hostdata->options & OPTION_MEMORY_MAPPED)) { + base = (u32) host->io_port; + memory_to_ncr = tmp|DMODE_800_DIOM; + ncr_to_memory = tmp|DMODE_800_SIOM; + } else { + base = virt_to_bus((void *)host->base); + memory_to_ncr = ncr_to_memory = tmp; + } + + /* SCRATCHB_REG_10 == SCRATCHA_REG_800, as it happens */ + patch_abs_32 (hostdata->script, 0, addr_scratch, base + SCRATCHA_REG_800); + patch_abs_32 (hostdata->script, 0, addr_temp, base + TEMP_REG); + patch_abs_32 (hostdata->script, 0, addr_dsa, base + DSA_REG); + + /* + * I needed some variables in the script to be accessible to + * both the NCR chip and the host processor. For these variables, + * I made the arbitrary decision to store them directly in the + * hostdata structure rather than in the RELATIVE area of the + * SCRIPTS. + */ + + + patch_abs_rwri_data (hostdata->script, 0, dmode_memory_to_memory, tmp); + patch_abs_rwri_data (hostdata->script, 0, dmode_memory_to_ncr, memory_to_ncr); + patch_abs_rwri_data (hostdata->script, 0, dmode_ncr_to_memory, ncr_to_memory); + + patch_abs_32 (hostdata->script, 0, msg_buf, + virt_to_bus((void *)&(hostdata->msg_buf))); + patch_abs_32 (hostdata->script, 0, reconnect_dsa_head, + virt_to_bus((void *)&(hostdata->reconnect_dsa_head))); + patch_abs_32 (hostdata->script, 0, addr_reconnect_dsa_head, + virt_to_bus((void *)&(hostdata->addr_reconnect_dsa_head))); + patch_abs_32 (hostdata->script, 0, reselected_identify, + virt_to_bus((void *)&(hostdata->reselected_identify))); +/* reselected_tag is currently unused */ +#if 0 + patch_abs_32 (hostdata->script, 0, reselected_tag, + virt_to_bus((void *)&(hostdata->reselected_tag))); +#endif + + patch_abs_32 (hostdata->script, 0, test_dest, + virt_to_bus((void*)&hostdata->test_dest)); + patch_abs_32 (hostdata->script, 0, test_src, + virt_to_bus(&hostdata->test_source)); + patch_abs_32 (hostdata->script, 0, saved_dsa, + virt_to_bus((void *)&hostdata->saved2_dsa)); + patch_abs_32 (hostdata->script, 0, emulfly, + virt_to_bus((void *)&hostdata->emulated_intfly)); + + patch_abs_rwri_data (hostdata->script, 0, dsa_check_reselect, + (unsigned char)(Ent_dsa_code_check_reselect - Ent_dsa_zero)); + +/* These are for event logging; the ncr_event enum contains the + actual interrupt numbers. */ +#ifdef A_int_EVENT_SELECT + patch_abs_32 (hostdata->script, 0, int_EVENT_SELECT, (u32) EVENT_SELECT); +#endif +#ifdef A_int_EVENT_DISCONNECT + patch_abs_32 (hostdata->script, 0, int_EVENT_DISCONNECT, (u32) EVENT_DISCONNECT); +#endif +#ifdef A_int_EVENT_RESELECT + patch_abs_32 (hostdata->script, 0, int_EVENT_RESELECT, (u32) EVENT_RESELECT); +#endif +#ifdef A_int_EVENT_COMPLETE + patch_abs_32 (hostdata->script, 0, int_EVENT_COMPLETE, (u32) EVENT_COMPLETE); +#endif +#ifdef A_int_EVENT_IDLE + patch_abs_32 (hostdata->script, 0, int_EVENT_IDLE, (u32) EVENT_IDLE); +#endif +#ifdef A_int_EVENT_SELECT_FAILED + patch_abs_32 (hostdata->script, 0, int_EVENT_SELECT_FAILED, + (u32) EVENT_SELECT_FAILED); +#endif +#ifdef A_int_EVENT_BEFORE_SELECT + patch_abs_32 (hostdata->script, 0, int_EVENT_BEFORE_SELECT, + (u32) EVENT_BEFORE_SELECT); +#endif +#ifdef A_int_EVENT_RESELECT_FAILED + patch_abs_32 (hostdata->script, 0, int_EVENT_RESELECT_FAILED, + (u32) EVENT_RESELECT_FAILED); +#endif + + /* + * Make sure the NCR and Linux code agree on the location of + * certain fields. + */ + + hostdata->E_accept_message = Ent_accept_message; + hostdata->E_command_complete = Ent_command_complete; + hostdata->E_cmdout_cmdout = Ent_cmdout_cmdout; + hostdata->E_data_transfer = Ent_data_transfer; + hostdata->E_debug_break = Ent_debug_break; + hostdata->E_dsa_code_template = Ent_dsa_code_template; + hostdata->E_dsa_code_template_end = Ent_dsa_code_template_end; + hostdata->E_end_data_transfer = Ent_end_data_transfer; + hostdata->E_initiator_abort = Ent_initiator_abort; + hostdata->E_msg_in = Ent_msg_in; + hostdata->E_other_transfer = Ent_other_transfer; + hostdata->E_other_in = Ent_other_in; + hostdata->E_other_out = Ent_other_out; + hostdata->E_reject_message = Ent_reject_message; + hostdata->E_respond_message = Ent_respond_message; + hostdata->E_select = Ent_select; + hostdata->E_select_msgout = Ent_select_msgout; + hostdata->E_target_abort = Ent_target_abort; +#ifdef Ent_test_0 + hostdata->E_test_0 = Ent_test_0; +#endif + hostdata->E_test_1 = Ent_test_1; + hostdata->E_test_2 = Ent_test_2; +#ifdef Ent_test_3 + hostdata->E_test_3 = Ent_test_3; +#endif + hostdata->E_wait_reselect = Ent_wait_reselect; + hostdata->E_dsa_code_begin = Ent_dsa_code_begin; + + hostdata->dsa_cmdout = A_dsa_cmdout; + hostdata->dsa_cmnd = A_dsa_cmnd; + hostdata->dsa_datain = A_dsa_datain; + hostdata->dsa_dataout = A_dsa_dataout; + hostdata->dsa_end = A_dsa_end; + hostdata->dsa_msgin = A_dsa_msgin; + hostdata->dsa_msgout = A_dsa_msgout; + hostdata->dsa_msgout_other = A_dsa_msgout_other; + hostdata->dsa_next = A_dsa_next; + hostdata->dsa_select = A_dsa_select; + hostdata->dsa_start = Ent_dsa_code_template - Ent_dsa_zero; + hostdata->dsa_status = A_dsa_status; + hostdata->dsa_jump_dest = Ent_dsa_code_fix_jump - Ent_dsa_zero + + 8 /* destination operand */; + + /* sanity check */ + if (A_dsa_fields_start != Ent_dsa_code_template_end - + Ent_dsa_zero) + printk("scsi%d : NCR dsa_fields start is %d not %d\n", + host->host_no, A_dsa_fields_start, Ent_dsa_code_template_end - + Ent_dsa_zero); + + printk("scsi%d : NCR code relocated to 0x%lx (virt 0x%p)\n", host->host_no, + virt_to_bus(hostdata->script), hostdata->script); +} + +/* + * Function : static int NCR53c7xx_run_tests (struct Scsi_Host *host) + * + * Purpose : run various verification tests on the NCR chip, + * including interrupt generation, and proper bus mastering + * operation. + * + * Inputs : host - a properly initialized Scsi_Host structure + * + * Preconditions : the NCR chip must be in a halted state. + * + * Returns : 0 if all tests were successful, -1 on error. + * + */ + +static int +NCR53c7xx_run_tests (struct Scsi_Host *host) { + NCR53c7x0_local_declare(); + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + unsigned long timeout; + u32 start; + int failed, i; + unsigned long flags; + NCR53c7x0_local_setup(host); + + /* The NCR chip _must_ be idle to run the test scripts */ + + local_irq_save(flags); + if (!hostdata->idle) { + printk ("scsi%d : chip not idle, aborting tests\n", host->host_no); + local_irq_restore(flags); + return -1; + } + + /* + * Check for functional interrupts, this could work as an + * autoprobe routine. + */ + + if ((hostdata->options & OPTION_DEBUG_TEST1) && + hostdata->state != STATE_DISABLED) { + hostdata->idle = 0; + hostdata->test_running = 1; + hostdata->test_completed = -1; + hostdata->test_dest = 0; + hostdata->test_source = 0xdeadbeef; + start = virt_to_bus (hostdata->script) + hostdata->E_test_1; + hostdata->state = STATE_RUNNING; + printk ("scsi%d : test 1", host->host_no); + NCR53c7x0_write32 (DSP_REG, start); + if (hostdata->options & OPTION_DEBUG_TRACE) + NCR53c7x0_write8 (DCNTL_REG, hostdata->saved_dcntl | DCNTL_SSM | + DCNTL_STD); + printk (" started\n"); + local_irq_restore(flags); + + /* + * This is currently a .5 second timeout, since (in theory) no slow + * board will take that long. In practice, we've seen one + * pentium which occassionally fails with this, but works with + * 10 times as much? + */ + + timeout = jiffies + 5 * HZ / 10; + while ((hostdata->test_completed == -1) && time_before(jiffies, timeout)) + barrier(); + + failed = 1; + if (hostdata->test_completed == -1) + printk ("scsi%d : driver test 1 timed out%s\n",host->host_no , + (hostdata->test_dest == 0xdeadbeef) ? + " due to lost interrupt.\n" + " Please verify that the correct IRQ is being used for your board,\n" + : ""); + else if (hostdata->test_completed != 1) + printk ("scsi%d : test 1 bad interrupt value (%d)\n", + host->host_no, hostdata->test_completed); + else + failed = (hostdata->test_dest != 0xdeadbeef); + + if (hostdata->test_dest != 0xdeadbeef) { + printk ("scsi%d : driver test 1 read 0x%x instead of 0xdeadbeef indicating a\n" + " probable cache invalidation problem. Please configure caching\n" + " as write-through or disabled\n", + host->host_no, hostdata->test_dest); + } + + if (failed) { + printk ("scsi%d : DSP = 0x%p (script at 0x%p, start at 0x%x)\n", + host->host_no, bus_to_virt(NCR53c7x0_read32(DSP_REG)), + hostdata->script, start); + printk ("scsi%d : DSPS = 0x%x\n", host->host_no, + NCR53c7x0_read32(DSPS_REG)); + local_irq_restore(flags); + return -1; + } + hostdata->test_running = 0; + } + + if ((hostdata->options & OPTION_DEBUG_TEST2) && + hostdata->state != STATE_DISABLED) { + u32 dsa[48]; + unsigned char identify = IDENTIFY(0, 0); + unsigned char cmd[6]; + unsigned char data[36]; + unsigned char status = 0xff; + unsigned char msg = 0xff; + + cmd[0] = INQUIRY; + cmd[1] = cmd[2] = cmd[3] = cmd[5] = 0; + cmd[4] = sizeof(data); + + dsa[2] = 1; + dsa[3] = virt_to_bus(&identify); + dsa[4] = 6; + dsa[5] = virt_to_bus(&cmd); + dsa[6] = sizeof(data); + dsa[7] = virt_to_bus(&data); + dsa[8] = 1; + dsa[9] = virt_to_bus(&status); + dsa[10] = 1; + dsa[11] = virt_to_bus(&msg); + + for (i = 0; i < 6; ++i) { +#ifdef VALID_IDS + if (!hostdata->valid_ids[i]) + continue; +#endif + local_irq_disable(); + if (!hostdata->idle) { + printk ("scsi%d : chip not idle, aborting tests\n", host->host_no); + local_irq_restore(flags); + return -1; + } + + /* 710: bit mapped scsi ID, async */ + dsa[0] = (1 << i) << 16; + hostdata->idle = 0; + hostdata->test_running = 2; + hostdata->test_completed = -1; + start = virt_to_bus(hostdata->script) + hostdata->E_test_2; + hostdata->state = STATE_RUNNING; + NCR53c7x0_write32 (DSA_REG, virt_to_bus(dsa)); + NCR53c7x0_write32 (DSP_REG, start); + if (hostdata->options & OPTION_DEBUG_TRACE) + NCR53c7x0_write8 (DCNTL_REG, hostdata->saved_dcntl | + DCNTL_SSM | DCNTL_STD); + local_irq_restore(flags); + + timeout = jiffies + 5 * HZ; /* arbitrary */ + while ((hostdata->test_completed == -1) && time_before(jiffies, timeout)) + barrier(); + + NCR53c7x0_write32 (DSA_REG, 0); + + if (hostdata->test_completed == 2) { + data[35] = 0; + printk ("scsi%d : test 2 INQUIRY to target %d, lun 0 : %s\n", + host->host_no, i, data + 8); + printk ("scsi%d : status ", host->host_no); + print_status (status); + printk ("\nscsi%d : message ", host->host_no); + print_msg (&msg); + printk ("\n"); + } else if (hostdata->test_completed == 3) { + printk("scsi%d : test 2 no connection with target %d\n", + host->host_no, i); + if (!hostdata->idle) { + printk("scsi%d : not idle\n", host->host_no); + local_irq_restore(flags); + return -1; + } + } else if (hostdata->test_completed == -1) { + printk ("scsi%d : test 2 timed out\n", host->host_no); + local_irq_restore(flags); + return -1; + } + hostdata->test_running = 0; + } + } + + local_irq_restore(flags); + return 0; +} + +/* + * Function : static void NCR53c7xx_dsa_fixup (struct NCR53c7x0_cmd *cmd) + * + * Purpose : copy the NCR53c8xx dsa structure into cmd's dsa buffer, + * performing all necessary relocation. + * + * Inputs : cmd, a NCR53c7x0_cmd structure with a dsa area large + * enough to hold the NCR53c8xx dsa. + */ + +static void +NCR53c7xx_dsa_fixup (struct NCR53c7x0_cmd *cmd) { + Scsi_Cmnd *c = cmd->cmd; + struct Scsi_Host *host = c->device->host; + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + int i; + + memcpy (cmd->dsa, hostdata->script + (hostdata->E_dsa_code_template / 4), + hostdata->E_dsa_code_template_end - hostdata->E_dsa_code_template); + + /* + * Note : within the NCR 'C' code, dsa points to the _start_ + * of the DSA structure, and _not_ the offset of dsa_zero within + * that structure used to facilitate shorter signed offsets + * for the 8 bit ALU. + * + * The implications of this are that + * + * - 32 bit A_dsa_* absolute values require an additional + * dsa_zero added to their value to be correct, since they are + * relative to dsa_zero which is in essentially a separate + * space from the code symbols. + * + * - All other symbols require no special treatment. + */ + + patch_abs_tci_data (cmd->dsa, Ent_dsa_code_template / sizeof(u32), + dsa_temp_lun, c->device->lun); + patch_abs_32 (cmd->dsa, Ent_dsa_code_template / sizeof(u32), + dsa_temp_addr_next, virt_to_bus(&cmd->dsa_next_addr)); + patch_abs_32 (cmd->dsa, Ent_dsa_code_template / sizeof(u32), + dsa_temp_next, virt_to_bus(cmd->dsa) + Ent_dsa_zero - + Ent_dsa_code_template + A_dsa_next); + patch_abs_32 (cmd->dsa, Ent_dsa_code_template / sizeof(u32), + dsa_temp_sync, virt_to_bus((void *)hostdata->sync[c->device->id].script)); + patch_abs_32 (cmd->dsa, Ent_dsa_code_template / sizeof(u32), + dsa_sscf_710, virt_to_bus((void *)&hostdata->sync[c->device->id].sscf_710)); + patch_abs_tci_data (cmd->dsa, Ent_dsa_code_template / sizeof(u32), + dsa_temp_target, 1 << c->device->id); + /* XXX - new pointer stuff */ + patch_abs_32 (cmd->dsa, Ent_dsa_code_template / sizeof(u32), + dsa_temp_addr_saved_pointer, virt_to_bus(&cmd->saved_data_pointer)); + patch_abs_32 (cmd->dsa, Ent_dsa_code_template / sizeof(u32), + dsa_temp_addr_saved_residual, virt_to_bus(&cmd->saved_residual)); + patch_abs_32 (cmd->dsa, Ent_dsa_code_template / sizeof(u32), + dsa_temp_addr_residual, virt_to_bus(&cmd->residual)); + + /* XXX - new start stuff */ + + patch_abs_32 (cmd->dsa, Ent_dsa_code_template / sizeof(u32), + dsa_temp_addr_dsa_value, virt_to_bus(&cmd->dsa_addr)); +} + +/* + * Function : run_process_issue_queue (void) + * + * Purpose : insure that the coroutine is running and will process our + * request. process_issue_queue_running is checked/set here (in an + * inline function) rather than in process_issue_queue itself to reduce + * the chances of stack overflow. + * + */ + +static volatile int process_issue_queue_running = 0; + +static __inline__ void +run_process_issue_queue(void) { + unsigned long flags; + local_irq_save(flags); + if (!process_issue_queue_running) { + process_issue_queue_running = 1; + process_issue_queue(flags); + /* + * process_issue_queue_running is cleared in process_issue_queue + * once it can't do more work, and process_issue_queue exits with + * interrupts disabled. + */ + } + local_irq_restore(flags); +} + +/* + * Function : static void abnormal_finished (struct NCR53c7x0_cmd *cmd, int + * result) + * + * Purpose : mark SCSI command as finished, OR'ing the host portion + * of the result word into the result field of the corresponding + * Scsi_Cmnd structure, and removing it from the internal queues. + * + * Inputs : cmd - command, result - entire result field + * + * Preconditions : the NCR chip should be in a halted state when + * abnormal_finished is run, since it modifies structures which + * the NCR expects to have exclusive access to. + */ + +static void +abnormal_finished (struct NCR53c7x0_cmd *cmd, int result) { + Scsi_Cmnd *c = cmd->cmd; + struct Scsi_Host *host = c->device->host; + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + unsigned long flags; + int left, found; + volatile struct NCR53c7x0_cmd * linux_search; + volatile struct NCR53c7x0_cmd * volatile *linux_prev; + volatile u32 *ncr_prev, *ncrcurrent, ncr_search; + +#if 0 + printk ("scsi%d: abnormal finished\n", host->host_no); +#endif + + local_irq_save(flags); + found = 0; + /* + * Traverse the NCR issue array until we find a match or run out + * of instructions. Instructions in the NCR issue array are + * either JUMP or NOP instructions, which are 2 words in length. + */ + + + for (found = 0, left = host->can_queue, ncrcurrent = hostdata->schedule; + left > 0; --left, ncrcurrent += 2) + { + if (issue_to_cmd (host, hostdata, (u32 *) ncrcurrent) == cmd) + { + ncrcurrent[0] = hostdata->NOP_insn; + ncrcurrent[1] = 0xdeadbeef; + ++found; + break; + } + } + + /* + * Traverse the NCR reconnect list of DSA structures until we find + * a pointer to this dsa or have found too many command structures. + * We let prev point at the next field of the previous element or + * head of the list, so we don't do anything different for removing + * the head element. + */ + + for (left = host->can_queue, + ncr_search = hostdata->reconnect_dsa_head, + ncr_prev = &hostdata->reconnect_dsa_head; + left >= 0 && ncr_search && + ((char*)bus_to_virt(ncr_search) + hostdata->dsa_start) + != (char *) cmd->dsa; + ncr_prev = (u32*) ((char*)bus_to_virt(ncr_search) + + hostdata->dsa_next), ncr_search = *ncr_prev, --left); + + if (left < 0) + printk("scsi%d: loop detected in ncr reconncect list\n", + host->host_no); + else if (ncr_search) { + if (found) + printk("scsi%d: scsi %ld in ncr issue array and reconnect lists\n", + host->host_no, c->pid); + else { + volatile u32 * next = (u32 *) + ((char *)bus_to_virt(ncr_search) + hostdata->dsa_next); + *ncr_prev = *next; +/* If we're at the tail end of the issue queue, update that pointer too. */ + found = 1; + } + } + + /* + * Traverse the host running list until we find this command or discover + * we have too many elements, pointing linux_prev at the next field of the + * linux_previous element or head of the list, search at this element. + */ + + for (left = host->can_queue, linux_search = hostdata->running_list, + linux_prev = &hostdata->running_list; + left >= 0 && linux_search && linux_search != cmd; + linux_prev = &(linux_search->next), + linux_search = linux_search->next, --left); + + if (left < 0) + printk ("scsi%d: loop detected in host running list for scsi pid %ld\n", + host->host_no, c->pid); + else if (linux_search) { + *linux_prev = linux_search->next; + --hostdata->busy[c->device->id][c->device->lun]; + } + + /* Return the NCR command structure to the free list */ + cmd->next = hostdata->free; + hostdata->free = cmd; + c->host_scribble = NULL; + + /* And return */ + c->result = result; + c->scsi_done(c); + + local_irq_restore(flags); + run_process_issue_queue(); +} + +/* + * Function : static void intr_break (struct Scsi_Host *host, + * struct NCR53c7x0_cmd *cmd) + * + * Purpose : Handler for breakpoint interrupts from a SCSI script + * + * Inputs : host - pointer to this host adapter's structure, + * cmd - pointer to the command (if any) dsa was pointing + * to. + * + */ + +static void +intr_break (struct Scsi_Host *host, struct + NCR53c7x0_cmd *cmd) { + NCR53c7x0_local_declare(); + struct NCR53c7x0_break *bp; +#if 0 + Scsi_Cmnd *c = cmd ? cmd->cmd : NULL; +#endif + u32 *dsp; + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + unsigned long flags; + NCR53c7x0_local_setup(host); + + /* + * Find the break point corresponding to this address, and + * dump the appropriate debugging information to standard + * output. + */ + local_irq_save(flags); + dsp = (u32 *) bus_to_virt(NCR53c7x0_read32(DSP_REG)); + for (bp = hostdata->breakpoints; bp && bp->address != dsp; + bp = bp->next); + if (!bp) + panic("scsi%d : break point interrupt from %p with no breakpoint!", + host->host_no, dsp); + + /* + * Configure the NCR chip for manual start mode, so that we can + * point the DSP register at the instruction that follows the + * INT int_debug_break instruction. + */ + + NCR53c7x0_write8 (hostdata->dmode, + NCR53c7x0_read8(hostdata->dmode)|DMODE_MAN); + + /* + * And update the DSP register, using the size of the old + * instruction in bytes. + */ + + local_irq_restore(flags); +} +/* + * Function : static void print_synchronous (const char *prefix, + * const unsigned char *msg) + * + * Purpose : print a pretty, user and machine parsable representation + * of a SDTR message, including the "real" parameters, data + * clock so we can tell transfer rate at a glance. + * + * Inputs ; prefix - text to prepend, msg - SDTR message (5 bytes) + */ + +static void +print_synchronous (const char *prefix, const unsigned char *msg) { + if (msg[4]) { + int Hz = 1000000000 / (msg[3] * 4); + int integer = Hz / 1000000; + int fraction = (Hz - (integer * 1000000)) / 10000; + printk ("%speriod %dns offset %d %d.%02dMHz %s SCSI%s\n", + prefix, (int) msg[3] * 4, (int) msg[4], integer, fraction, + (((msg[3] * 4) < 200) ? "FAST" : "synchronous"), + (((msg[3] * 4) < 200) ? "-II" : "")); + } else + printk ("%sasynchronous SCSI\n", prefix); +} + +/* + * Function : static void set_synchronous (struct Scsi_Host *host, + * int target, int sxfer, int scntl3, int now_connected) + * + * Purpose : reprogram transfers between the selected SCSI initiator and + * target with the given register values; in the indirect + * select operand, reselection script, and chip registers. + * + * Inputs : host - NCR53c7,8xx SCSI host, target - number SCSI target id, + * sxfer and scntl3 - NCR registers. now_connected - if non-zero, + * we should reprogram the registers now too. + * + * NOTE: For 53c710, scntl3 is actually used for SCF bits from + * SBCL, as we don't have a SCNTL3. + */ + +static void +set_synchronous (struct Scsi_Host *host, int target, int sxfer, int scntl3, + int now_connected) { + NCR53c7x0_local_declare(); + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + u32 *script; + NCR53c7x0_local_setup(host); + + /* These are eight bit registers */ + sxfer &= 0xff; + scntl3 &= 0xff; + + hostdata->sync[target].sxfer_sanity = sxfer; + hostdata->sync[target].scntl3_sanity = scntl3; + +/* + * HARD CODED : synchronous script is EIGHT words long. This + * must agree with 53c7.8xx.h + */ + + if ((hostdata->chip != 700) && (hostdata->chip != 70066)) { + hostdata->sync[target].select_indirect = (1 << target) << 16 | + (sxfer << 8); + hostdata->sync[target].sscf_710 = scntl3; + + script = (u32 *) hostdata->sync[target].script; + + /* XXX - add NCR53c7x0 code to reprogram SCF bits if we want to */ + script[0] = ((DCMD_TYPE_RWRI | DCMD_RWRI_OPC_MODIFY | + DCMD_RWRI_OP_MOVE) << 24) | + (SBCL_REG << 16) | (scntl3 << 8); + script[1] = 0; + script += 2; + + script[0] = ((DCMD_TYPE_RWRI | DCMD_RWRI_OPC_MODIFY | + DCMD_RWRI_OP_MOVE) << 24) | + (SXFER_REG << 16) | (sxfer << 8); + script[1] = 0; + script += 2; + +#ifdef DEBUG_SYNC_INTR + if (hostdata->options & OPTION_DEBUG_DISCONNECT) { + script[0] = ((DCMD_TYPE_TCI|DCMD_TCI_OP_INT) << 24) | DBC_TCI_TRUE; + script[1] = DEBUG_SYNC_INTR; + script += 2; + } +#endif + + script[0] = ((DCMD_TYPE_TCI|DCMD_TCI_OP_RETURN) << 24) | DBC_TCI_TRUE; + script[1] = 0; + script += 2; + } + + if (hostdata->options & OPTION_DEBUG_SYNCHRONOUS) + printk ("scsi%d : target %d sync parameters are sxfer=0x%x, scntl3=0x%x\n", + host->host_no, target, sxfer, scntl3); + + if (now_connected) { + NCR53c7x0_write8(SBCL_REG, scntl3); + NCR53c7x0_write8(SXFER_REG, sxfer); + } +} + + +/* + * Function : static int asynchronous (struct Scsi_Host *host, int target) + * + * Purpose : reprogram between the selected SCSI Host adapter and target + * (assumed to be currently connected) for asynchronous transfers. + * + * Inputs : host - SCSI host structure, target - numeric target ID. + * + * Preconditions : the NCR chip should be in one of the halted states + */ + +static void +asynchronous (struct Scsi_Host *host, int target) { + NCR53c7x0_local_declare(); + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + NCR53c7x0_local_setup(host); + set_synchronous (host, target, /* no offset */ 0, hostdata->saved_scntl3, + 1); + printk ("scsi%d : setting target %d to asynchronous SCSI\n", + host->host_no, target); +} + +/* + * XXX - do we want to go out of our way (ie, add extra code to selection + * in the NCR53c710/NCR53c720 script) to reprogram the synchronous + * conversion bits, or can we be content in just setting the + * sxfer bits? I chose to do so [richard@sleepie.demon.co.uk] + */ + +/* Table for NCR53c8xx synchronous values */ + +/* This table is also correct for 710, allowing that scf=4 is equivalent + * of SSCF=0 (ie use DCNTL, divide by 3) for a 50.01-66.00MHz clock. + * For any other clock values, we cannot use entries with SCF values of + * 4. I guess that for a 66MHz clock, the slowest it will set is 2MHz, + * and for a 50MHz clock, the slowest will be 2.27Mhz. Should check + * that a device doesn't try and negotiate sync below these limits! + */ + +static const struct { + int div; /* Total clock divisor * 10 */ + unsigned char scf; /* */ + unsigned char tp; /* 4 + tp = xferp divisor */ +} syncs[] = { +/* div scf tp div scf tp div scf tp */ + { 40, 1, 0}, { 50, 1, 1}, { 60, 1, 2}, + { 70, 1, 3}, { 75, 2, 1}, { 80, 1, 4}, + { 90, 1, 5}, { 100, 1, 6}, { 105, 2, 3}, + { 110, 1, 7}, { 120, 2, 4}, { 135, 2, 5}, + { 140, 3, 3}, { 150, 2, 6}, { 160, 3, 4}, + { 165, 2, 7}, { 180, 3, 5}, { 200, 3, 6}, + { 210, 4, 3}, { 220, 3, 7}, { 240, 4, 4}, + { 270, 4, 5}, { 300, 4, 6}, { 330, 4, 7} +}; + +/* + * Function : static void synchronous (struct Scsi_Host *host, int target, + * char *msg) + * + * Purpose : reprogram transfers between the selected SCSI initiator and + * target for synchronous SCSI transfers such that the synchronous + * offset is less than that requested and period at least as long + * as that requested. Also modify *msg such that it contains + * an appropriate response. + * + * Inputs : host - NCR53c7,8xx SCSI host, target - number SCSI target id, + * msg - synchronous transfer request. + */ + + +static void +synchronous (struct Scsi_Host *host, int target, char *msg) { + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + int desire, divisor, i, limit; + unsigned char scntl3, sxfer; +/* The diagnostic message fits on one line, even with max. width integers */ + char buf[80]; + +/* Desired transfer clock in Hz */ + desire = 1000000000L / (msg[3] * 4); +/* Scale the available SCSI clock by 10 so we get tenths */ + divisor = (hostdata->scsi_clock * 10) / desire; + +/* NCR chips can handle at most an offset of 8 */ + if (msg[4] > 8) + msg[4] = 8; + + if (hostdata->options & OPTION_DEBUG_SDTR) + printk("scsi%d : optimal synchronous divisor of %d.%01d\n", + host->host_no, divisor / 10, divisor % 10); + + limit = (sizeof(syncs) / sizeof(syncs[0]) -1); + for (i = 0; (i < limit) && (divisor > syncs[i].div); ++i); + + if (hostdata->options & OPTION_DEBUG_SDTR) + printk("scsi%d : selected synchronous divisor of %d.%01d\n", + host->host_no, syncs[i].div / 10, syncs[i].div % 10); + + msg[3] = ((1000000000L / hostdata->scsi_clock) * syncs[i].div / 10 / 4); + + if (hostdata->options & OPTION_DEBUG_SDTR) + printk("scsi%d : selected synchronous period of %dns\n", host->host_no, + msg[3] * 4); + + scntl3 = syncs[i].scf; + sxfer = (msg[4] << SXFER_MO_SHIFT) | (syncs[i].tp << 4); + if (hostdata->options & OPTION_DEBUG_SDTR) + printk ("scsi%d : sxfer=0x%x scntl3=0x%x\n", + host->host_no, (int) sxfer, (int) scntl3); + set_synchronous (host, target, sxfer, scntl3, 1); + sprintf (buf, "scsi%d : setting target %d to ", host->host_no, target); + print_synchronous (buf, msg); +} + +/* + * Function : static int NCR53c7x0_dstat_sir_intr (struct Scsi_Host *host, + * struct NCR53c7x0_cmd *cmd) + * + * Purpose : Handler for INT generated instructions for the + * NCR53c810/820 SCSI SCRIPT + * + * Inputs : host - pointer to this host adapter's structure, + * cmd - pointer to the command (if any) dsa was pointing + * to. + * + */ + +static int +NCR53c7x0_dstat_sir_intr (struct Scsi_Host *host, struct + NCR53c7x0_cmd *cmd) { + NCR53c7x0_local_declare(); + int print; + Scsi_Cmnd *c = cmd ? cmd->cmd : NULL; + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + u32 dsps,*dsp; /* Argument of the INT instruction */ + + NCR53c7x0_local_setup(host); + dsps = NCR53c7x0_read32(DSPS_REG); + dsp = (u32 *) bus_to_virt(NCR53c7x0_read32(DSP_REG)); + + /* RGH 150597: Frig. Commands which fail with Check Condition are + * Flagged as successful - hack dsps to indicate check condition */ +#if 0 + /* RGH 200597: Need to disable for BVME6000, as it gets Check Conditions + * and then dies. Seems to handle Check Condition at startup, but + * not mid kernel build. */ + if (dsps == A_int_norm_emulateintfly && cmd && cmd->result == 2) + dsps = A_int_err_check_condition; +#endif + + if (hostdata->options & OPTION_DEBUG_INTR) + printk ("scsi%d : DSPS = 0x%x\n", host->host_no, dsps); + + switch (dsps) { + case A_int_msg_1: + print = 1; + switch (hostdata->msg_buf[0]) { + /* + * Unless we've initiated synchronous negotiation, I don't + * think that this should happen. + */ + case MESSAGE_REJECT: + hostdata->dsp = hostdata->script + hostdata->E_accept_message / + sizeof(u32); + hostdata->dsp_changed = 1; + if (cmd && (cmd->flags & CMD_FLAG_SDTR)) { + printk ("scsi%d : target %d rejected SDTR\n", host->host_no, + c->device->id); + cmd->flags &= ~CMD_FLAG_SDTR; + asynchronous (host, c->device->id); + print = 0; + } + break; + case INITIATE_RECOVERY: + printk ("scsi%d : extended contingent allegiance not supported yet, rejecting\n", + host->host_no); + /* Fall through to default */ + hostdata->dsp = hostdata->script + hostdata->E_reject_message / + sizeof(u32); + hostdata->dsp_changed = 1; + break; + default: + printk ("scsi%d : unsupported message, rejecting\n", + host->host_no); + hostdata->dsp = hostdata->script + hostdata->E_reject_message / + sizeof(u32); + hostdata->dsp_changed = 1; + } + if (print) { + printk ("scsi%d : received message", host->host_no); + if (c) + printk (" from target %d lun %d ", c->device->id, c->device->lun); + print_msg ((unsigned char *) hostdata->msg_buf); + printk("\n"); + } + + return SPECIFIC_INT_NOTHING; + + + case A_int_msg_sdtr: +/* + * At this point, hostdata->msg_buf contains + * 0 EXTENDED MESSAGE + * 1 length + * 2 SDTR + * 3 period * 4ns + * 4 offset + */ + + if (cmd) { + char buf[80]; + sprintf (buf, "scsi%d : target %d %s ", host->host_no, c->device->id, + (cmd->flags & CMD_FLAG_SDTR) ? "accepting" : "requesting"); + print_synchronous (buf, (unsigned char *) hostdata->msg_buf); + + /* + * Initiator initiated, won't happen unless synchronous + * transfers are enabled. If we get a SDTR message in + * response to our SDTR, we should program our parameters + * such that + * offset <= requested offset + * period >= requested period + */ + if (cmd->flags & CMD_FLAG_SDTR) { + cmd->flags &= ~CMD_FLAG_SDTR; + if (hostdata->msg_buf[4]) + synchronous (host, c->device->id, (unsigned char *) + hostdata->msg_buf); + else + asynchronous (host, c->device->id); + hostdata->dsp = hostdata->script + hostdata->E_accept_message / + sizeof(u32); + hostdata->dsp_changed = 1; + return SPECIFIC_INT_NOTHING; + } else { + if (hostdata->options & OPTION_SYNCHRONOUS) { + cmd->flags |= CMD_FLAG_DID_SDTR; + synchronous (host, c->device->id, (unsigned char *) + hostdata->msg_buf); + } else { + hostdata->msg_buf[4] = 0; /* 0 offset = async */ + asynchronous (host, c->device->id); + } + patch_dsa_32 (cmd->dsa, dsa_msgout_other, 0, 5); + patch_dsa_32 (cmd->dsa, dsa_msgout_other, 1, (u32) + virt_to_bus ((void *)&hostdata->msg_buf)); + hostdata->dsp = hostdata->script + + hostdata->E_respond_message / sizeof(u32); + hostdata->dsp_changed = 1; + } + return SPECIFIC_INT_NOTHING; + } + /* Fall through to abort if we couldn't find a cmd, and + therefore a dsa structure to twiddle */ + case A_int_msg_wdtr: + hostdata->dsp = hostdata->script + hostdata->E_reject_message / + sizeof(u32); + hostdata->dsp_changed = 1; + return SPECIFIC_INT_NOTHING; + case A_int_err_unexpected_phase: + if (hostdata->options & OPTION_DEBUG_INTR) + printk ("scsi%d : unexpected phase\n", host->host_no); + return SPECIFIC_INT_ABORT; + case A_int_err_selected: + if ((hostdata->chip / 100) == 8) + printk ("scsi%d : selected by target %d\n", host->host_no, + (int) NCR53c7x0_read8(SDID_REG_800) &7); + else + printk ("scsi%d : selected by target LCRC=0x%02x\n", host->host_no, + (int) NCR53c7x0_read8(LCRC_REG_10)); + hostdata->dsp = hostdata->script + hostdata->E_target_abort / + sizeof(u32); + hostdata->dsp_changed = 1; + return SPECIFIC_INT_NOTHING; + case A_int_err_unexpected_reselect: + if ((hostdata->chip / 100) == 8) + printk ("scsi%d : unexpected reselect by target %d lun %d\n", + host->host_no, (int) NCR53c7x0_read8(SDID_REG_800) & 7, + hostdata->reselected_identify & 7); + else + printk ("scsi%d : unexpected reselect LCRC=0x%02x\n", host->host_no, + (int) NCR53c7x0_read8(LCRC_REG_10)); + hostdata->dsp = hostdata->script + hostdata->E_initiator_abort / + sizeof(u32); + hostdata->dsp_changed = 1; + return SPECIFIC_INT_NOTHING; +/* + * Since contingent allegiance conditions are cleared by the next + * command issued to a target, we must issue a REQUEST SENSE + * command after receiving a CHECK CONDITION status, before + * another command is issued. + * + * Since this NCR53c7x0_cmd will be freed after use, we don't + * care if we step on the various fields, so modify a few things. + */ + case A_int_err_check_condition: +#if 0 + if (hostdata->options & OPTION_DEBUG_INTR) +#endif + printk ("scsi%d : CHECK CONDITION\n", host->host_no); + if (!c) { + printk("scsi%d : CHECK CONDITION with no SCSI command\n", + host->host_no); + return SPECIFIC_INT_PANIC; + } + + /* + * FIXME : this uses the normal one-byte selection message. + * We may want to renegotiate for synchronous & WIDE transfers + * since these could be the crux of our problem. + * + hostdata->NOP_insn* FIXME : once SCSI-II tagged queuing is implemented, we'll + * have to set this up so that the rest of the DSA + * agrees with this being an untagged queue'd command. + */ + + patch_dsa_32 (cmd->dsa, dsa_msgout, 0, 1); + + /* + * Modify the table indirect for COMMAND OUT phase, since + * Request Sense is a six byte command. + */ + + patch_dsa_32 (cmd->dsa, dsa_cmdout, 0, 6); + + /* + * The CDB is now mirrored in our local non-cached + * structure, but keep the old structure up to date as well, + * just in case anyone looks at it. + */ + + /* + * XXX Need to worry about data buffer alignment/cache state + * XXX here, but currently never get A_int_err_check_condition, + * XXX so ignore problem for now. + */ + cmd->cmnd[0] = c->cmnd[0] = REQUEST_SENSE; + cmd->cmnd[0] = c->cmnd[1] &= 0xe0; /* Zero all but LUN */ + cmd->cmnd[0] = c->cmnd[2] = 0; + cmd->cmnd[0] = c->cmnd[3] = 0; + cmd->cmnd[0] = c->cmnd[4] = sizeof(c->sense_buffer); + cmd->cmnd[0] = c->cmnd[5] = 0; + + /* + * Disable dataout phase, and program datain to transfer to the + * sense buffer, and add a jump to other_transfer after the + * command so overflow/underrun conditions are detected. + */ + + patch_dsa_32 (cmd->dsa, dsa_dataout, 0, + virt_to_bus(hostdata->script) + hostdata->E_other_transfer); + patch_dsa_32 (cmd->dsa, dsa_datain, 0, + virt_to_bus(cmd->data_transfer_start)); + cmd->data_transfer_start[0] = (((DCMD_TYPE_BMI | DCMD_BMI_OP_MOVE_I | + DCMD_BMI_IO)) << 24) | sizeof(c->sense_buffer); + cmd->data_transfer_start[1] = (u32) virt_to_bus(c->sense_buffer); + + cmd->data_transfer_start[2] = ((DCMD_TYPE_TCI | DCMD_TCI_OP_JUMP) + << 24) | DBC_TCI_TRUE; + cmd->data_transfer_start[3] = (u32) virt_to_bus(hostdata->script) + + hostdata->E_other_transfer; + + /* + * Currently, this command is flagged as completed, ie + * it has valid status and message data. Reflag it as + * incomplete. Q - need to do something so that original + * status, etc are used. + */ + + cmd->result = cmd->cmd->result = 0xffff; + + /* + * Restart command as a REQUEST SENSE. + */ + hostdata->dsp = (u32 *) hostdata->script + hostdata->E_select / + sizeof(u32); + hostdata->dsp_changed = 1; + return SPECIFIC_INT_NOTHING; + case A_int_debug_break: + return SPECIFIC_INT_BREAK; + case A_int_norm_aborted: + hostdata->dsp = (u32 *) hostdata->schedule; + hostdata->dsp_changed = 1; + if (cmd) + abnormal_finished (cmd, DID_ERROR << 16); + return SPECIFIC_INT_NOTHING; + case A_int_norm_emulateintfly: + NCR53c7x0_intfly(host); + return SPECIFIC_INT_NOTHING; + case A_int_test_1: + case A_int_test_2: + hostdata->idle = 1; + hostdata->test_completed = (dsps - A_int_test_1) / 0x00010000 + 1; + if (hostdata->options & OPTION_DEBUG_INTR) + printk("scsi%d : test%d complete\n", host->host_no, + hostdata->test_completed); + return SPECIFIC_INT_NOTHING; +#ifdef A_int_debug_reselected_ok + case A_int_debug_reselected_ok: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR| + OPTION_DEBUG_DISCONNECT)) { + /* + * Note - this dsa is not based on location relative to + * the command structure, but to location relative to the + * DSA register + */ + u32 *dsa; + dsa = (u32 *) bus_to_virt (NCR53c7x0_read32(DSA_REG)); + + printk("scsi%d : reselected_ok (DSA = 0x%x (virt 0x%p)\n", + host->host_no, NCR53c7x0_read32(DSA_REG), dsa); + printk("scsi%d : resume address is 0x%x (virt 0x%p)\n", + host->host_no, cmd->saved_data_pointer, + bus_to_virt(cmd->saved_data_pointer)); + print_insn (host, hostdata->script + Ent_reselected_ok / + sizeof(u32), "", 1); + if ((hostdata->chip / 100) == 8) + printk ("scsi%d : sxfer=0x%x, scntl3=0x%x\n", + host->host_no, NCR53c7x0_read8(SXFER_REG), + NCR53c7x0_read8(SCNTL3_REG_800)); + else + printk ("scsi%d : sxfer=0x%x, cannot read SBCL\n", + host->host_no, NCR53c7x0_read8(SXFER_REG)); + if (c) { + print_insn (host, (u32 *) + hostdata->sync[c->device->id].script, "", 1); + print_insn (host, (u32 *) + hostdata->sync[c->device->id].script + 2, "", 1); + } + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_reselect_check + case A_int_debug_reselect_check: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR)) { + u32 *dsa; +#if 0 + u32 *code; +#endif + /* + * Note - this dsa is not based on location relative to + * the command structure, but to location relative to the + * DSA register + */ + dsa = bus_to_virt (NCR53c7x0_read32(DSA_REG)); + printk("scsi%d : reselected_check_next (DSA = 0x%lx (virt 0x%p))\n", + host->host_no, virt_to_bus(dsa), dsa); + if (dsa) { + printk("scsi%d : resume address is 0x%x (virt 0x%p)\n", + host->host_no, cmd->saved_data_pointer, + bus_to_virt (cmd->saved_data_pointer)); +#if 0 + printk("scsi%d : template code :\n", host->host_no); + for (code = dsa + (Ent_dsa_code_check_reselect - Ent_dsa_zero) + / sizeof(u32); code < (dsa + Ent_dsa_zero / sizeof(u32)); + code += print_insn (host, code, "", 1)); +#endif + } + print_insn (host, hostdata->script + Ent_reselected_ok / + sizeof(u32), "", 1); + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_dsa_schedule + case A_int_debug_dsa_schedule: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR)) { + u32 *dsa; + /* + * Note - this dsa is not based on location relative to + * the command structure, but to location relative to the + * DSA register + */ + dsa = (u32 *) bus_to_virt (NCR53c7x0_read32(DSA_REG)); + printk("scsi%d : dsa_schedule (old DSA = 0x%lx (virt 0x%p))\n", + host->host_no, virt_to_bus(dsa), dsa); + if (dsa) + printk("scsi%d : resume address is 0x%x (virt 0x%p)\n" + " (temp was 0x%x (virt 0x%p))\n", + host->host_no, cmd->saved_data_pointer, + bus_to_virt (cmd->saved_data_pointer), + NCR53c7x0_read32 (TEMP_REG), + bus_to_virt (NCR53c7x0_read32(TEMP_REG))); + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_scheduled + case A_int_debug_scheduled: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR)) { + printk("scsi%d : new I/O 0x%x (virt 0x%p) scheduled\n", + host->host_no, NCR53c7x0_read32(DSA_REG), + bus_to_virt(NCR53c7x0_read32(DSA_REG))); + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_idle + case A_int_debug_idle: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR)) { + printk("scsi%d : idle\n", host->host_no); + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_cmd + case A_int_debug_cmd: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR)) { + printk("scsi%d : command sent\n"); + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_dsa_loaded + case A_int_debug_dsa_loaded: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR)) { + printk("scsi%d : DSA loaded with 0x%x (virt 0x%p)\n", host->host_no, + NCR53c7x0_read32(DSA_REG), + bus_to_virt(NCR53c7x0_read32(DSA_REG))); + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_reselected + case A_int_debug_reselected: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR| + OPTION_DEBUG_DISCONNECT)) { + if ((hostdata->chip / 100) == 8) + printk("scsi%d : reselected by target %d lun %d\n", + host->host_no, (int) NCR53c7x0_read8(SDID_REG_800) & ~0x80, + (int) hostdata->reselected_identify & 7); + else + printk("scsi%d : reselected by LCRC=0x%02x lun %d\n", + host->host_no, (int) NCR53c7x0_read8(LCRC_REG_10), + (int) hostdata->reselected_identify & 7); + print_queues(host); + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_disconnect_msg + case A_int_debug_disconnect_msg: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR)) { + if (c) + printk("scsi%d : target %d lun %d disconnecting\n", + host->host_no, c->device->id, c->device->lun); + else + printk("scsi%d : unknown target disconnecting\n", + host->host_no); + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_disconnected + case A_int_debug_disconnected: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR| + OPTION_DEBUG_DISCONNECT)) { + printk ("scsi%d : disconnected, new queues are\n", + host->host_no); + print_queues(host); +#if 0 + /* Not valid on ncr53c710! */ + printk ("scsi%d : sxfer=0x%x, scntl3=0x%x\n", + host->host_no, NCR53c7x0_read8(SXFER_REG), + NCR53c7x0_read8(SCNTL3_REG_800)); +#endif + if (c) { + print_insn (host, (u32 *) + hostdata->sync[c->device->id].script, "", 1); + print_insn (host, (u32 *) + hostdata->sync[c->device->id].script + 2, "", 1); + } + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_panic + case A_int_debug_panic: + printk("scsi%d : int_debug_panic received\n", host->host_no); + print_lots (host); + return SPECIFIC_INT_PANIC; +#endif +#ifdef A_int_debug_saved + case A_int_debug_saved: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR| + OPTION_DEBUG_DISCONNECT)) { + printk ("scsi%d : saved data pointer 0x%x (virt 0x%p)\n", + host->host_no, cmd->saved_data_pointer, + bus_to_virt (cmd->saved_data_pointer)); + print_progress (c); + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_restored + case A_int_debug_restored: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR| + OPTION_DEBUG_DISCONNECT)) { + if (cmd) { + int size; + printk ("scsi%d : restored data pointer 0x%x (virt 0x%p)\n", + host->host_no, cmd->saved_data_pointer, bus_to_virt ( + cmd->saved_data_pointer)); + size = print_insn (host, (u32 *) + bus_to_virt(cmd->saved_data_pointer), "", 1); + size = print_insn (host, (u32 *) + bus_to_virt(cmd->saved_data_pointer) + size, "", 1); + print_progress (c); + } +#if 0 + printk ("scsi%d : datapath residual %d\n", + host->host_no, datapath_residual (host)) ; +#endif + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_sync + case A_int_debug_sync: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR| + OPTION_DEBUG_DISCONNECT|OPTION_DEBUG_SDTR)) { + unsigned char sxfer = NCR53c7x0_read8 (SXFER_REG), scntl3; + if ((hostdata->chip / 100) == 8) { + scntl3 = NCR53c7x0_read8 (SCNTL3_REG_800); + if (c) { + if (sxfer != hostdata->sync[c->device->id].sxfer_sanity || + scntl3 != hostdata->sync[c->device->id].scntl3_sanity) { + printk ("scsi%d : sync sanity check failed sxfer=0x%x, scntl3=0x%x", + host->host_no, sxfer, scntl3); + NCR53c7x0_write8 (SXFER_REG, sxfer); + NCR53c7x0_write8 (SCNTL3_REG_800, scntl3); + } + } else + printk ("scsi%d : unknown command sxfer=0x%x, scntl3=0x%x\n", + host->host_no, (int) sxfer, (int) scntl3); + } else { + if (c) { + if (sxfer != hostdata->sync[c->device->id].sxfer_sanity) { + printk ("scsi%d : sync sanity check failed sxfer=0x%x", + host->host_no, sxfer); + NCR53c7x0_write8 (SXFER_REG, sxfer); + NCR53c7x0_write8 (SBCL_REG, + hostdata->sync[c->device->id].sscf_710); + } + } else + printk ("scsi%d : unknown command sxfer=0x%x\n", + host->host_no, (int) sxfer); + } + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_datain + case A_int_debug_datain: + if (hostdata->options & (OPTION_DEBUG_SCRIPT|OPTION_DEBUG_INTR| + OPTION_DEBUG_DISCONNECT|OPTION_DEBUG_SDTR)) { + int size; + if ((hostdata->chip / 100) == 8) + printk ("scsi%d : In do_datain (%s) sxfer=0x%x, scntl3=0x%x\n" + " datapath residual=%d\n", + host->host_no, sbcl_to_phase (NCR53c7x0_read8 (SBCL_REG)), + (int) NCR53c7x0_read8(SXFER_REG), + (int) NCR53c7x0_read8(SCNTL3_REG_800), + datapath_residual (host)) ; + else + printk ("scsi%d : In do_datain (%s) sxfer=0x%x\n" + " datapath residual=%d\n", + host->host_no, sbcl_to_phase (NCR53c7x0_read8 (SBCL_REG)), + (int) NCR53c7x0_read8(SXFER_REG), + datapath_residual (host)) ; + print_insn (host, dsp, "", 1); + size = print_insn (host, (u32 *) bus_to_virt(dsp[1]), "", 1); + print_insn (host, (u32 *) bus_to_virt(dsp[1]) + size, "", 1); + } + return SPECIFIC_INT_RESTART; +#endif +#ifdef A_int_debug_check_dsa + case A_int_debug_check_dsa: + if (NCR53c7x0_read8 (SCNTL1_REG) & SCNTL1_CON) { + int sdid; + int tmp; + char *where; + if (hostdata->chip / 100 == 8) + sdid = NCR53c7x0_read8 (SDID_REG_800) & 15; + else { + tmp = NCR53c7x0_read8 (SDID_REG_700); + if (!tmp) + panic ("SDID_REG_700 = 0"); + tmp >>= 1; + sdid = 0; + while (tmp) { + tmp >>= 1; + sdid++; + } + } + where = dsp - NCR53c7x0_insn_size(NCR53c7x0_read8 + (DCMD_REG)) == hostdata->script + + Ent_select_check_dsa / sizeof(u32) ? + "selection" : "reselection"; + if (c && sdid != c->device->id) { + printk ("scsi%d : SDID target %d != DSA target %d at %s\n", + host->host_no, sdid, c->device->id, where); + print_lots(host); + dump_events (host, 20); + return SPECIFIC_INT_PANIC; + } + } + return SPECIFIC_INT_RESTART; +#endif + default: + if ((dsps & 0xff000000) == 0x03000000) { + printk ("scsi%d : misc debug interrupt 0x%x\n", + host->host_no, dsps); + return SPECIFIC_INT_RESTART; + } else if ((dsps & 0xff000000) == 0x05000000) { + if (hostdata->events) { + struct NCR53c7x0_event *event; + ++hostdata->event_index; + if (hostdata->event_index >= hostdata->event_size) + hostdata->event_index = 0; + event = (struct NCR53c7x0_event *) hostdata->events + + hostdata->event_index; + event->event = (enum ncr_event) dsps; + event->dsa = bus_to_virt(NCR53c7x0_read32(DSA_REG)); + if (NCR53c7x0_read8 (SCNTL1_REG) & SCNTL1_CON) { + if (hostdata->chip / 100 == 8) + event->target = NCR53c7x0_read8(SSID_REG_800); + else { + unsigned char tmp, sdid; + tmp = NCR53c7x0_read8 (SDID_REG_700); + if (!tmp) + panic ("SDID_REG_700 = 0"); + tmp >>= 1; + sdid = 0; + while (tmp) { + tmp >>= 1; + sdid++; + } + event->target = sdid; + } + } + else + event->target = 255; + + if (event->event == EVENT_RESELECT) + event->lun = hostdata->reselected_identify & 0xf; + else if (c) + event->lun = c->device->lun; + else + event->lun = 255; + do_gettimeofday(&(event->time)); + if (c) { + event->pid = c->pid; + memcpy ((void *) event->cmnd, (void *) c->cmnd, + sizeof (event->cmnd)); + } else { + event->pid = -1; + } + } + return SPECIFIC_INT_RESTART; + } + + printk ("scsi%d : unknown user interrupt 0x%x\n", + host->host_no, (unsigned) dsps); + return SPECIFIC_INT_PANIC; + } +} + +/* + * XXX - the stock NCR assembler won't output the scriptu.h file, + * which undefine's all #define'd CPP symbols from the script.h + * file, which will create problems if you use multiple scripts + * with the same symbol names. + * + * If you insist on using NCR's assembler, you could generate + * scriptu.h from script.h using something like + * + * grep #define script.h | \ + * sed 's/#define[ ][ ]*\([_a-zA-Z][_a-zA-Z0-9]*\).*$/#undefine \1/' \ + * > scriptu.h + */ + +#include "53c7xx_u.h" + +/* XXX - add alternate script handling code here */ + + +/* + * Function : static void NCR537xx_soft_reset (struct Scsi_Host *host) + * + * Purpose : perform a soft reset of the NCR53c7xx chip + * + * Inputs : host - pointer to this host adapter's structure + * + * Preconditions : NCR53c7x0_init must have been called for this + * host. + * + */ + +static void +NCR53c7x0_soft_reset (struct Scsi_Host *host) { + NCR53c7x0_local_declare(); + unsigned long flags; + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + NCR53c7x0_local_setup(host); + + local_irq_save(flags); + + /* Disable scsi chip and s/w level 7 ints */ + +#ifdef CONFIG_MVME16x + if (MACH_IS_MVME16x) + { + volatile unsigned long v; + + v = *(volatile unsigned long *)0xfff4006c; + v &= ~0x8000; + *(volatile unsigned long *)0xfff4006c = v; + v = *(volatile unsigned long *)0xfff4202c; + v &= ~0x10; + *(volatile unsigned long *)0xfff4202c = v; + } +#endif + /* Anything specific for your hardware? */ + + /* + * Do a soft reset of the chip so that everything is + * reinitialized to the power-on state. + * + * Basically follow the procedure outlined in the NCR53c700 + * data manual under Chapter Six, How to Use, Steps Necessary to + * Start SCRIPTS, with the exception of actually starting the + * script and setting up the synchronous transfer gunk. + */ + + /* Should we reset the scsi bus here??????????????????? */ + + NCR53c7x0_write8(ISTAT_REG_700, ISTAT_10_SRST); + NCR53c7x0_write8(ISTAT_REG_700, 0); + + /* + * saved_dcntl is set up in NCR53c7x0_init() before it is overwritten + * here. We should have some better way of working out the CF bit + * setting.. + */ + + hostdata->saved_dcntl = DCNTL_10_EA|DCNTL_10_COM; + if (hostdata->scsi_clock > 50000000) + hostdata->saved_dcntl |= DCNTL_700_CF_3; + else + if (hostdata->scsi_clock > 37500000) + hostdata->saved_dcntl |= DCNTL_700_CF_2; +#if 0 + else + /* Any clocks less than 37.5MHz? */ +#endif + + if (hostdata->options & OPTION_DEBUG_TRACE) + NCR53c7x0_write8(DCNTL_REG, hostdata->saved_dcntl | DCNTL_SSM); + else + NCR53c7x0_write8(DCNTL_REG, hostdata->saved_dcntl); + /* Following disables snooping - snooping is not required, as non- + * cached pages are used for shared data, and appropriate use is + * made of cache_push/cache_clear. Indeed, for 68060 + * enabling snooping causes disk corruption of ext2fs free block + * bitmaps and the like. If you have a 68060 with snooping hardwared + * on, then you need to enable CONFIG_060_WRITETHROUGH. + */ + NCR53c7x0_write8(CTEST7_REG, CTEST7_10_TT1|CTEST7_STD); + /* Actually burst of eight, according to my 53c710 databook */ + NCR53c7x0_write8(hostdata->dmode, DMODE_10_BL_8 | DMODE_10_FC2); + NCR53c7x0_write8(SCID_REG, 1 << host->this_id); + NCR53c7x0_write8(SBCL_REG, 0); + NCR53c7x0_write8(SCNTL1_REG, SCNTL1_ESR_700); + NCR53c7x0_write8(SCNTL0_REG, ((hostdata->options & OPTION_PARITY) ? + SCNTL0_EPC : 0) | SCNTL0_EPG_700 | SCNTL0_ARB1 | SCNTL0_ARB2); + + /* + * Enable all interrupts, except parity which we only want when + * the user requests it. + */ + + NCR53c7x0_write8(DIEN_REG, DIEN_700_BF | + DIEN_ABRT | DIEN_SSI | DIEN_SIR | DIEN_700_OPC); + + NCR53c7x0_write8(SIEN_REG_700, ((hostdata->options & OPTION_PARITY) ? + SIEN_PAR : 0) | SIEN_700_STO | SIEN_RST | SIEN_UDC | + SIEN_SGE | SIEN_MA); + +#ifdef CONFIG_MVME16x + if (MACH_IS_MVME16x) + { + volatile unsigned long v; + + /* Enable scsi chip and s/w level 7 ints */ + v = *(volatile unsigned long *)0xfff40080; + v = (v & ~(0xf << 28)) | (4 << 28); + *(volatile unsigned long *)0xfff40080 = v; + v = *(volatile unsigned long *)0xfff4006c; + v |= 0x8000; + *(volatile unsigned long *)0xfff4006c = v; + v = *(volatile unsigned long *)0xfff4202c; + v = (v & ~0xff) | 0x10 | 4; + *(volatile unsigned long *)0xfff4202c = v; + } +#endif + /* Anything needed for your hardware? */ + local_irq_restore(flags); +} + + +/* + * Function static struct NCR53c7x0_cmd *allocate_cmd (Scsi_Cmnd *cmd) + * + * Purpose : Return the first free NCR53c7x0_cmd structure (which are + * reused in a LIFO manner to minimize cache thrashing). + * + * Side effects : If we haven't yet scheduled allocation of NCR53c7x0_cmd + * structures for this device, do so. Attempt to complete all scheduled + * allocations using get_zeroed_page(), putting NCR53c7x0_cmd structures on + * the free list. Teach programmers not to drink and hack. + * + * Inputs : cmd - SCSI command + * + * Returns : NCR53c7x0_cmd structure allocated on behalf of cmd; + * NULL on failure. + */ + +static void +my_free_page (void *addr, int dummy) +{ + /* XXX This assumes default cache mode to be IOMAP_FULL_CACHING, which + * XXX may be invalid (CONFIG_060_WRITETHROUGH) + */ + kernel_set_cachemode((void *)addr, 4096, IOMAP_FULL_CACHING); + free_page ((u32)addr); +} + +static struct NCR53c7x0_cmd * +allocate_cmd (Scsi_Cmnd *cmd) { + struct Scsi_Host *host = cmd->device->host; + struct NCR53c7x0_hostdata *hostdata = + (struct NCR53c7x0_hostdata *) host->hostdata[0]; + u32 real; /* Real address */ + int size; /* Size of *tmp */ + struct NCR53c7x0_cmd *tmp; + unsigned long flags; + + if (hostdata->options & OPTION_DEBUG_ALLOCATION) + printk ("scsi%d : num_cmds = %d, can_queue = %d\n" + " target = %d, lun = %d, %s\n", + host->host_no, hostdata->num_cmds, host->can_queue, + cmd->device->id, cmd->device->lun, (hostdata->cmd_allocated[cmd->device->id] & + (1 << cmd->device->lun)) ? "already allocated" : "not allocated"); + +/* + * If we have not yet reserved commands for this I_T_L nexus, and + * the device exists (as indicated by permanent Scsi_Cmnd structures + * being allocated under 1.3.x, or being outside of scan_scsis in + * 1.2.x), do so now. + */ + if (!(hostdata->cmd_allocated[cmd->device->id] & (1 << cmd->device->lun)) && + cmd->device && cmd->device->has_cmdblocks) { + if ((hostdata->extra_allocate + hostdata->num_cmds) < host->can_queue) + hostdata->extra_allocate += host->cmd_per_lun; + hostdata->cmd_allocated[cmd->device->id] |= (1 << cmd->device->lun); + } + + for (; hostdata->extra_allocate > 0 ; --hostdata->extra_allocate, + ++hostdata->num_cmds) { + /* historically, kmalloc has returned unaligned addresses; pad so we + have enough room to ROUNDUP */ + size = hostdata->max_cmd_size + sizeof (void *); +#ifdef FORCE_DSA_ALIGNMENT + /* + * 53c710 rev.0 doesn't have an add-with-carry instruction. + * Ensure we allocate enough memory to force alignment. + */ + size += 256; +#endif +/* FIXME: for ISA bus '7xx chips, we need to or GFP_DMA in here */ + + if (size > 4096) { + printk (KERN_ERR "53c7xx: allocate_cmd size > 4K\n"); + return NULL; + } + real = get_zeroed_page(GFP_ATOMIC); + if (real == 0) + return NULL; + memset((void *)real, 0, 4096); + cache_push(virt_to_phys((void *)real), 4096); + cache_clear(virt_to_phys((void *)real), 4096); + kernel_set_cachemode((void *)real, 4096, IOMAP_NOCACHE_SER); + tmp = ROUNDUP(real, void *); +#ifdef FORCE_DSA_ALIGNMENT + { + if (((u32)tmp & 0xff) > CmdPageStart) + tmp = (struct NCR53c7x0_cmd *)((u32)tmp + 255); + tmp = (struct NCR53c7x0_cmd *)(((u32)tmp & ~0xff) + CmdPageStart); +#if 0 + printk ("scsi: size = %d, real = 0x%08x, tmp set to 0x%08x\n", + size, real, (u32)tmp); +#endif + } +#endif + tmp->real = (void *)real; + tmp->size = size; + tmp->free = ((void (*)(void *, int)) my_free_page); + local_irq_save(flags); + tmp->next = hostdata->free; + hostdata->free = tmp; + local_irq_restore(flags); + } + local_irq_save(flags); + tmp = (struct NCR53c7x0_cmd *) hostdata->free; + if (tmp) { + hostdata->free = tmp->next; + } + local_irq_restore(flags); + if (!tmp) + printk ("scsi%d : can't allocate command for target %d lun %d\n", + host->host_no, cmd->device->id, cmd->device->lun); + return tmp; +} + +/* + * Function static struct NCR53c7x0_cmd *create_cmd (Scsi_Cmnd *cmd) + * + * + * Purpose : allocate a NCR53c7x0_cmd structure, initialize it based on the + * Scsi_Cmnd structure passed in cmd, including dsa and Linux field + * initialization, and dsa code relocation. + * + * Inputs : cmd - SCSI command + * + * Returns : NCR53c7x0_cmd structure corresponding to cmd, + * NULL on failure. + */ +static struct NCR53c7x0_cmd * +create_cmd (Scsi_Cmnd *cmd) { + NCR53c7x0_local_declare(); + struct Scsi_Host *host = cmd->device->host; + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + struct NCR53c7x0_cmd *tmp; /* NCR53c7x0_cmd structure for this command */ + int datain, /* Number of instructions per phase */ + dataout; + int data_transfer_instructions, /* Count of dynamic instructions */ + i; /* Counter */ + u32 *cmd_datain, /* Address of datain/dataout code */ + *cmd_dataout; /* Incremented as we assemble */ +#ifdef notyet + unsigned char *msgptr; /* Current byte in select message */ + int msglen; /* Length of whole select message */ +#endif + unsigned long flags; + u32 exp_select_indirect; /* Used in sanity check */ + NCR53c7x0_local_setup(cmd->device->host); + + if (!(tmp = allocate_cmd (cmd))) + return NULL; + + /* + * Copy CDB and initialised result fields from Scsi_Cmnd to NCR53c7x0_cmd. + * We do this because NCR53c7x0_cmd may have a special cache mode + * selected to cope with lack of bus snooping, etc. + */ + + memcpy(tmp->cmnd, cmd->cmnd, 12); + tmp->result = cmd->result; + + /* + * Decide whether we need to generate commands for DATA IN, + * DATA OUT, neither, or both based on the SCSI command + */ + + switch (cmd->cmnd[0]) { + /* These commands do DATA IN */ + case INQUIRY: + case MODE_SENSE: + case READ_6: + case READ_10: + case READ_CAPACITY: + case REQUEST_SENSE: + case READ_BLOCK_LIMITS: + case READ_TOC: + datain = 2 * (cmd->use_sg ? cmd->use_sg : 1) + 3; + dataout = 0; + break; + /* These commands do DATA OUT */ + case MODE_SELECT: + case WRITE_6: + case WRITE_10: +#if 0 + printk("scsi%d : command is ", host->host_no); + print_command(cmd->cmnd); +#endif +#if 0 + printk ("scsi%d : %d scatter/gather segments\n", host->host_no, + cmd->use_sg); +#endif + datain = 0; + dataout = 2 * (cmd->use_sg ? cmd->use_sg : 1) + 3; +#if 0 + hostdata->options |= OPTION_DEBUG_INTR; +#endif + break; + /* + * These commands do no data transfer, we should force an + * interrupt if a data phase is attempted on them. + */ + case TEST_UNIT_READY: + case ALLOW_MEDIUM_REMOVAL: + case START_STOP: + datain = dataout = 0; + break; + /* + * We don't know about these commands, so generate code to handle + * both DATA IN and DATA OUT phases. More efficient to identify them + * and add them to the above cases. + */ + default: + printk("scsi%d : datain+dataout for command ", host->host_no); + print_command(cmd->cmnd); + datain = dataout = 2 * (cmd->use_sg ? cmd->use_sg : 1) + 3; + } + + /* + * New code : so that active pointers work correctly regardless + * of where the saved data pointer is at, we want to immediately + * enter the dynamic code after selection, and on a non-data + * phase perform a CALL to the non-data phase handler, with + * returns back to this address. + * + * If a phase mismatch is encountered in the middle of a + * Block MOVE instruction, we want to _leave_ that instruction + * unchanged as the current case is, modify a temporary buffer, + * and point the active pointer (TEMP) at that. + * + * Furthermore, we want to implement a saved data pointer, + * set by the SAVE_DATA_POINTERs message. + * + * So, the data transfer segments will change to + * CALL data_transfer, WHEN NOT data phase + * MOVE x, x, WHEN data phase + * ( repeat ) + * JUMP other_transfer + */ + + data_transfer_instructions = datain + dataout; + + /* + * When we perform a request sense, we overwrite various things, + * including the data transfer code. Make sure we have enough + * space to do that. + */ + + if (data_transfer_instructions < 2) + data_transfer_instructions = 2; + + + /* + * The saved data pointer is set up so that a RESTORE POINTERS message + * will start the data transfer over at the beginning. + */ + + tmp->saved_data_pointer = virt_to_bus (hostdata->script) + + hostdata->E_data_transfer; + + /* + * Initialize Linux specific fields. + */ + + tmp->cmd = cmd; + tmp->next = NULL; + tmp->flags = 0; + tmp->dsa_next_addr = virt_to_bus(tmp->dsa) + hostdata->dsa_next - + hostdata->dsa_start; + tmp->dsa_addr = virt_to_bus(tmp->dsa) - hostdata->dsa_start; + + /* + * Calculate addresses of dynamic code to fill in DSA + */ + + tmp->data_transfer_start = tmp->dsa + (hostdata->dsa_end - + hostdata->dsa_start) / sizeof(u32); + tmp->data_transfer_end = tmp->data_transfer_start + + 2 * data_transfer_instructions; + + cmd_datain = datain ? tmp->data_transfer_start : NULL; + cmd_dataout = dataout ? (datain ? cmd_datain + 2 * datain : tmp-> + data_transfer_start) : NULL; + + /* + * Fill in the NCR53c7x0_cmd structure as follows + * dsa, with fixed up DSA code + * datain code + * dataout code + */ + + /* Copy template code into dsa and perform all necessary fixups */ + if (hostdata->dsa_fixup) + hostdata->dsa_fixup(tmp); + + patch_dsa_32(tmp->dsa, dsa_next, 0, 0); + /* + * XXX is this giving 53c710 access to the Scsi_Cmnd in some way? + * Do we need to change it for caching reasons? + */ + patch_dsa_32(tmp->dsa, dsa_cmnd, 0, virt_to_bus(cmd)); + + if (hostdata->options & OPTION_DEBUG_SYNCHRONOUS) { + + exp_select_indirect = ((1 << cmd->device->id) << 16) | + (hostdata->sync[cmd->device->id].sxfer_sanity << 8); + + if (hostdata->sync[cmd->device->id].select_indirect != + exp_select_indirect) { + printk ("scsi%d : sanity check failed select_indirect=0x%x\n", + host->host_no, hostdata->sync[cmd->device->id].select_indirect); + FATAL(host); + + } + } + + patch_dsa_32(tmp->dsa, dsa_select, 0, + hostdata->sync[cmd->device->id].select_indirect); + + /* + * Right now, we'll do the WIDE and SYNCHRONOUS negotiations on + * different commands; although it should be trivial to do them + * both at the same time. + */ + if (hostdata->initiate_wdtr & (1 << cmd->device->id)) { + memcpy ((void *) (tmp->select + 1), (void *) wdtr_message, + sizeof(wdtr_message)); + patch_dsa_32(tmp->dsa, dsa_msgout, 0, 1 + sizeof(wdtr_message)); + local_irq_save(flags); + hostdata->initiate_wdtr &= ~(1 << cmd->device->id); + local_irq_restore(flags); + } else if (hostdata->initiate_sdtr & (1 << cmd->device->id)) { + memcpy ((void *) (tmp->select + 1), (void *) sdtr_message, + sizeof(sdtr_message)); + patch_dsa_32(tmp->dsa, dsa_msgout, 0, 1 + sizeof(sdtr_message)); + tmp->flags |= CMD_FLAG_SDTR; + local_irq_save(flags); + hostdata->initiate_sdtr &= ~(1 << cmd->device->id); + local_irq_restore(flags); + + } +#if 1 + else if (!(hostdata->talked_to & (1 << cmd->device->id)) && + !(hostdata->options & OPTION_NO_ASYNC)) { + + memcpy ((void *) (tmp->select + 1), (void *) async_message, + sizeof(async_message)); + patch_dsa_32(tmp->dsa, dsa_msgout, 0, 1 + sizeof(async_message)); + tmp->flags |= CMD_FLAG_SDTR; + } +#endif + else + patch_dsa_32(tmp->dsa, dsa_msgout, 0, 1); + + hostdata->talked_to |= (1 << cmd->device->id); + tmp->select[0] = (hostdata->options & OPTION_DISCONNECT) ? + IDENTIFY (1, cmd->device->lun) : IDENTIFY (0, cmd->device->lun); + patch_dsa_32(tmp->dsa, dsa_msgout, 1, virt_to_bus(tmp->select)); + patch_dsa_32(tmp->dsa, dsa_cmdout, 0, cmd->cmd_len); + patch_dsa_32(tmp->dsa, dsa_cmdout, 1, virt_to_bus(tmp->cmnd)); + patch_dsa_32(tmp->dsa, dsa_dataout, 0, cmd_dataout ? + virt_to_bus (cmd_dataout) + : virt_to_bus (hostdata->script) + hostdata->E_other_transfer); + patch_dsa_32(tmp->dsa, dsa_datain, 0, cmd_datain ? + virt_to_bus (cmd_datain) + : virt_to_bus (hostdata->script) + hostdata->E_other_transfer); + /* + * XXX - need to make endian aware, should use separate variables + * for both status and message bytes. + */ + patch_dsa_32(tmp->dsa, dsa_msgin, 0, 1); +/* + * FIXME : these only works for little endian. We probably want to + * provide message and status fields in the NCR53c7x0_cmd + * structure, and assign them to cmd->result when we're done. + */ +#ifdef BIG_ENDIAN + patch_dsa_32(tmp->dsa, dsa_msgin, 1, virt_to_bus(&tmp->result) + 2); + patch_dsa_32(tmp->dsa, dsa_status, 0, 1); + patch_dsa_32(tmp->dsa, dsa_status, 1, virt_to_bus(&tmp->result) + 3); +#else + patch_dsa_32(tmp->dsa, dsa_msgin, 1, virt_to_bus(&tmp->result) + 1); + patch_dsa_32(tmp->dsa, dsa_status, 0, 1); + patch_dsa_32(tmp->dsa, dsa_status, 1, virt_to_bus(&tmp->result)); +#endif + patch_dsa_32(tmp->dsa, dsa_msgout_other, 0, 1); + patch_dsa_32(tmp->dsa, dsa_msgout_other, 1, + virt_to_bus(&(hostdata->NCR53c7xx_msg_nop))); + + /* + * Generate code for zero or more of the DATA IN, DATA OUT phases + * in the format + * + * CALL data_transfer, WHEN NOT phase + * MOVE first buffer length, first buffer address, WHEN phase + * ... + * MOVE last buffer length, last buffer address, WHEN phase + * JUMP other_transfer + */ + +/* + * See if we're getting to data transfer by generating an unconditional + * interrupt. + */ +#if 0 + if (datain) { + cmd_datain[0] = 0x98080000; + cmd_datain[1] = 0x03ffd00d; + cmd_datain += 2; + } +#endif + +/* + * XXX - I'm undecided whether all of this nonsense is faster + * in the long run, or whether I should just go and implement a loop + * on the NCR chip using table indirect mode? + * + * In any case, this is how it _must_ be done for 53c700/700-66 chips, + * so this stays even when we come up with something better. + * + * When we're limited to 1 simultaneous command, no overlapping processing, + * we're seeing 630K/sec, with 7% CPU usage on a slow Syquest 45M + * drive. + * + * Not bad, not good. We'll see. + */ + + tmp->bounce.len = 0; /* Assume aligned buffer */ + + for (i = 0; cmd->use_sg ? (i < cmd->use_sg) : !i; cmd_datain += 4, + cmd_dataout += 4, ++i) { + u32 vbuf = cmd->use_sg + ? (u32)page_address(((struct scatterlist *)cmd->buffer)[i].page)+ + ((struct scatterlist *)cmd->buffer)[i].offset + : (u32)(cmd->request_buffer); + u32 bbuf = virt_to_bus((void *)vbuf); + u32 count = cmd->use_sg ? + ((struct scatterlist *)cmd->buffer)[i].length : + cmd->request_bufflen; + + /* + * If we have buffers which are not aligned with 16 byte cache + * lines, then we just hope nothing accesses the other parts of + * those cache lines while the transfer is in progress. That would + * fill the cache, and subsequent reads of the dma data would pick + * up the wrong thing. + * XXX We need a bounce buffer to handle that correctly. + */ + + if (((bbuf & 15) || (count & 15)) && (datain || dataout)) + { + /* Bounce buffer needed */ + if (cmd->use_sg) + printk ("53c7xx: Non-aligned buffer with use_sg\n"); + else if (datain && dataout) + printk ("53c7xx: Non-aligned buffer with datain && dataout\n"); + else if (count > 256) + printk ("53c7xx: Non-aligned transfer > 256 bytes\n"); + else + { + if (datain) + { + tmp->bounce.len = count; + tmp->bounce.addr = vbuf; + bbuf = virt_to_bus(tmp->bounce.buf); + tmp->bounce.buf[0] = 0xff; + tmp->bounce.buf[1] = 0xfe; + tmp->bounce.buf[2] = 0xfd; + tmp->bounce.buf[3] = 0xfc; + } + if (dataout) + { + memcpy ((void *)tmp->bounce.buf, (void *)vbuf, count); + bbuf = virt_to_bus(tmp->bounce.buf); + } + } + } + + if (datain) { + cache_clear(virt_to_phys((void *)vbuf), count); + /* CALL other_in, WHEN NOT DATA_IN */ + cmd_datain[0] = ((DCMD_TYPE_TCI | DCMD_TCI_OP_CALL | + DCMD_TCI_IO) << 24) | + DBC_TCI_WAIT_FOR_VALID | DBC_TCI_COMPARE_PHASE; + cmd_datain[1] = virt_to_bus (hostdata->script) + + hostdata->E_other_in; + /* MOVE count, buf, WHEN DATA_IN */ + cmd_datain[2] = ((DCMD_TYPE_BMI | DCMD_BMI_OP_MOVE_I | DCMD_BMI_IO) + << 24) | count; + cmd_datain[3] = bbuf; +#if 0 + print_insn (host, cmd_datain, "dynamic ", 1); + print_insn (host, cmd_datain + 2, "dynamic ", 1); +#endif + } + if (dataout) { + cache_push(virt_to_phys((void *)vbuf), count); + /* CALL other_out, WHEN NOT DATA_OUT */ + cmd_dataout[0] = ((DCMD_TYPE_TCI | DCMD_TCI_OP_CALL) << 24) | + DBC_TCI_WAIT_FOR_VALID | DBC_TCI_COMPARE_PHASE; + cmd_dataout[1] = virt_to_bus(hostdata->script) + + hostdata->E_other_out; + /* MOVE count, buf, WHEN DATA+OUT */ + cmd_dataout[2] = ((DCMD_TYPE_BMI | DCMD_BMI_OP_MOVE_I) << 24) + | count; + cmd_dataout[3] = bbuf; +#if 0 + print_insn (host, cmd_dataout, "dynamic ", 1); + print_insn (host, cmd_dataout + 2, "dynamic ", 1); +#endif + } + } + + /* + * Install JUMP instructions after the data transfer routines to return + * control to the do_other_transfer routines. + */ + + + if (datain) { + cmd_datain[0] = ((DCMD_TYPE_TCI | DCMD_TCI_OP_JUMP) << 24) | + DBC_TCI_TRUE; + cmd_datain[1] = virt_to_bus(hostdata->script) + + hostdata->E_other_transfer; +#if 0 + print_insn (host, cmd_datain, "dynamic jump ", 1); +#endif + cmd_datain += 2; + } +#if 0 + if (datain) { + cmd_datain[0] = 0x98080000; + cmd_datain[1] = 0x03ffdeed; + cmd_datain += 2; + } +#endif + if (dataout) { + cmd_dataout[0] = ((DCMD_TYPE_TCI | DCMD_TCI_OP_JUMP) << 24) | + DBC_TCI_TRUE; + cmd_dataout[1] = virt_to_bus(hostdata->script) + + hostdata->E_other_transfer; +#if 0 + print_insn (host, cmd_dataout, "dynamic jump ", 1); +#endif + cmd_dataout += 2; + } + + return tmp; +} + +/* + * Function : int NCR53c7xx_queue_command (Scsi_Cmnd *cmd, + * void (*done)(Scsi_Cmnd *)) + * + * Purpose : enqueues a SCSI command + * + * Inputs : cmd - SCSI command, done - function called on completion, with + * a pointer to the command descriptor. + * + * Returns : 0 + * + * Side effects : + * cmd is added to the per instance driver issue_queue, with major + * twiddling done to the host specific fields of cmd. If the + * process_issue_queue coroutine isn't running, it is restarted. + * + * NOTE : we use the host_scribble field of the Scsi_Cmnd structure to + * hold our own data, and pervert the ptr field of the SCp field + * to create a linked list. + */ + +int +NCR53c7xx_queue_command (Scsi_Cmnd *cmd, void (* done)(Scsi_Cmnd *)) { + struct Scsi_Host *host = cmd->device->host; + struct NCR53c7x0_hostdata *hostdata = + (struct NCR53c7x0_hostdata *) host->hostdata[0]; + unsigned long flags; + Scsi_Cmnd *tmp; + + cmd->scsi_done = done; + cmd->host_scribble = NULL; + cmd->SCp.ptr = NULL; + cmd->SCp.buffer = NULL; + +#ifdef VALID_IDS + /* Ignore commands on invalid IDs */ + if (!hostdata->valid_ids[cmd->device->id]) { + printk("scsi%d : ignoring target %d lun %d\n", host->host_no, + cmd->device->id, cmd->device->lun); + cmd->result = (DID_BAD_TARGET << 16); + done(cmd); + return 0; + } +#endif + + local_irq_save(flags); + if ((hostdata->options & (OPTION_DEBUG_INIT_ONLY|OPTION_DEBUG_PROBE_ONLY)) + || ((hostdata->options & OPTION_DEBUG_TARGET_LIMIT) && + !(hostdata->debug_lun_limit[cmd->device->id] & (1 << cmd->device->lun))) +#ifdef LINUX_1_2 + || cmd->device->id > 7 +#else + || cmd->device->id > host->max_id +#endif + || cmd->device->id == host->this_id + || hostdata->state == STATE_DISABLED) { + printk("scsi%d : disabled or bad target %d lun %d\n", host->host_no, + cmd->device->id, cmd->device->lun); + cmd->result = (DID_BAD_TARGET << 16); + done(cmd); + local_irq_restore(flags); + return 0; + } + + if ((hostdata->options & OPTION_DEBUG_NCOMMANDS_LIMIT) && + (hostdata->debug_count_limit == 0)) { + printk("scsi%d : maximum commands exceeded\n", host->host_no); + cmd->result = (DID_BAD_TARGET << 16); + done(cmd); + local_irq_restore(flags); + return 0; + } + + if (hostdata->options & OPTION_DEBUG_READ_ONLY) { + switch (cmd->cmnd[0]) { + case WRITE_6: + case WRITE_10: + printk("scsi%d : WRITE attempted with NO_WRITE debugging flag set\n", + host->host_no); + cmd->result = (DID_BAD_TARGET << 16); + done(cmd); + local_irq_restore(flags); + return 0; + } + } + + if ((hostdata->options & OPTION_DEBUG_TARGET_LIMIT) && + hostdata->debug_count_limit != -1) + --hostdata->debug_count_limit; + + cmd->result = 0xffff; /* The NCR will overwrite message + and status with valid data */ + cmd->host_scribble = (unsigned char *) tmp = create_cmd (cmd); + + /* + * REQUEST SENSE commands are inserted at the head of the queue + * so that we do not clear the contingent allegiance condition + * they may be looking at. + */ + + if (!(hostdata->issue_queue) || (cmd->cmnd[0] == REQUEST_SENSE)) { + cmd->SCp.ptr = (unsigned char *) hostdata->issue_queue; + hostdata->issue_queue = cmd; + } else { + for (tmp = (Scsi_Cmnd *) hostdata->issue_queue; tmp->SCp.ptr; + tmp = (Scsi_Cmnd *) tmp->SCp.ptr); + tmp->SCp.ptr = (unsigned char *) cmd; + } + local_irq_restore(flags); + run_process_issue_queue(); + return 0; +} + +/* + * Function : void to_schedule_list (struct Scsi_Host *host, + * struct NCR53c7x0_hostdata * hostdata, Scsi_Cmnd *cmd) + * + * Purpose : takes a SCSI command which was just removed from the + * issue queue, and deals with it by inserting it in the first + * free slot in the schedule list or by terminating it immediately. + * + * Inputs : + * host - SCSI host adapter; hostdata - hostdata structure for + * this adapter; cmd - a pointer to the command; should have + * the host_scribble field initialized to point to a valid + * + * Side effects : + * cmd is added to the per instance schedule list, with minor + * twiddling done to the host specific fields of cmd. + * + */ + +static __inline__ void +to_schedule_list (struct Scsi_Host *host, struct NCR53c7x0_hostdata *hostdata, + struct NCR53c7x0_cmd *cmd) { + NCR53c7x0_local_declare(); + Scsi_Cmnd *tmp = cmd->cmd; + unsigned long flags; + /* dsa start is negative, so subtraction is used */ + volatile u32 *ncrcurrent; + + int i; + NCR53c7x0_local_setup(host); +#if 0 + printk("scsi%d : new dsa is 0x%lx (virt 0x%p)\n", host->host_no, + virt_to_bus(hostdata->dsa), hostdata->dsa); +#endif + + local_irq_save(flags); + + /* + * Work around race condition : if an interrupt fired and we + * got disabled forget about this command. + */ + + if (hostdata->state == STATE_DISABLED) { + printk("scsi%d : driver disabled\n", host->host_no); + tmp->result = (DID_BAD_TARGET << 16); + cmd->next = (struct NCR53c7x0_cmd *) hostdata->free; + hostdata->free = cmd; + tmp->scsi_done(tmp); + local_irq_restore(flags); + return; + } + + for (i = host->can_queue, ncrcurrent = hostdata->schedule; + i > 0 && ncrcurrent[0] != hostdata->NOP_insn; + --i, ncrcurrent += 2 /* JUMP instructions are two words */); + + if (i > 0) { + ++hostdata->busy[tmp->device->id][tmp->device->lun]; + cmd->next = hostdata->running_list; + hostdata->running_list = cmd; + + /* Restore this instruction to a NOP once the command starts */ + cmd->dsa [(hostdata->dsa_jump_dest - hostdata->dsa_start) / + sizeof(u32)] = (u32) virt_to_bus ((void *)ncrcurrent); + /* Replace the current jump operand. */ + ncrcurrent[1] = + virt_to_bus ((void *) cmd->dsa) + hostdata->E_dsa_code_begin - + hostdata->E_dsa_code_template; + /* Replace the NOP instruction with a JUMP */ + ncrcurrent[0] = ((DCMD_TYPE_TCI|DCMD_TCI_OP_JUMP) << 24) | + DBC_TCI_TRUE; + } else { + printk ("scsi%d: no free slot\n", host->host_no); + disable(host); + tmp->result = (DID_ERROR << 16); + cmd->next = (struct NCR53c7x0_cmd *) hostdata->free; + hostdata->free = cmd; + tmp->scsi_done(tmp); + local_irq_restore(flags); + return; + } + + /* + * If the NCR chip is in an idle state, start it running the scheduler + * immediately. Otherwise, signal the chip to jump to schedule as + * soon as it is idle. + */ + + if (hostdata->idle) { + hostdata->idle = 0; + hostdata->state = STATE_RUNNING; + NCR53c7x0_write32 (DSP_REG, virt_to_bus ((void *)hostdata->schedule)); + if (hostdata->options & OPTION_DEBUG_TRACE) + NCR53c7x0_write8 (DCNTL_REG, hostdata->saved_dcntl | + DCNTL_SSM | DCNTL_STD); + } else { + NCR53c7x0_write8(hostdata->istat, ISTAT_10_SIGP); + } + + local_irq_restore(flags); +} + +/* + * Function : busyp (struct Scsi_Host *host, struct NCR53c7x0_hostdata + * *hostdata, Scsi_Cmnd *cmd) + * + * Purpose : decide if we can pass the given SCSI command on to the + * device in question or not. + * + * Returns : non-zero when we're busy, 0 when we aren't. + */ + +static __inline__ int +busyp (struct Scsi_Host *host, struct NCR53c7x0_hostdata *hostdata, + Scsi_Cmnd *cmd) { + /* FIXME : in the future, this needs to accommodate SCSI-II tagged + queuing, and we may be able to play with fairness here a bit. + */ + return hostdata->busy[cmd->device->id][cmd->device->lun]; +} + +/* + * Function : process_issue_queue (void) + * + * Purpose : transfer commands from the issue queue to NCR start queue + * of each NCR53c7/8xx in the system, avoiding kernel stack + * overflows when the scsi_done() function is invoked recursively. + * + * NOTE : process_issue_queue exits with interrupts *disabled*, so the + * caller must reenable them if it desires. + * + * NOTE : process_issue_queue should be called from both + * NCR53c7x0_queue_command() and from the interrupt handler + * after command completion in case NCR53c7x0_queue_command() + * isn't invoked again but we've freed up resources that are + * needed. + */ + +static void +process_issue_queue (unsigned long flags) { + Scsi_Cmnd *tmp, *prev; + struct Scsi_Host *host; + struct NCR53c7x0_hostdata *hostdata; + int done; + + /* + * We run (with interrupts disabled) until we're sure that none of + * the host adapters have anything that can be done, at which point + * we set process_issue_queue_running to 0 and exit. + * + * Interrupts are enabled before doing various other internal + * instructions, after we've decided that we need to run through + * the loop again. + * + */ + + do { + local_irq_disable(); /* Freeze request queues */ + done = 1; + for (host = first_host; host && host->hostt == the_template; + host = host->next) { + hostdata = (struct NCR53c7x0_hostdata *) host->hostdata[0]; + local_irq_disable(); + if (hostdata->issue_queue) { + if (hostdata->state == STATE_DISABLED) { + tmp = (Scsi_Cmnd *) hostdata->issue_queue; + hostdata->issue_queue = (Scsi_Cmnd *) tmp->SCp.ptr; + tmp->result = (DID_BAD_TARGET << 16); + if (tmp->host_scribble) { + ((struct NCR53c7x0_cmd *)tmp->host_scribble)->next = + hostdata->free; + hostdata->free = + (struct NCR53c7x0_cmd *)tmp->host_scribble; + tmp->host_scribble = NULL; + } + tmp->scsi_done (tmp); + done = 0; + } else + for (tmp = (Scsi_Cmnd *) hostdata->issue_queue, + prev = NULL; tmp; prev = tmp, tmp = (Scsi_Cmnd *) + tmp->SCp.ptr) + if (!tmp->host_scribble || + !busyp (host, hostdata, tmp)) { + if (prev) + prev->SCp.ptr = tmp->SCp.ptr; + else + hostdata->issue_queue = (Scsi_Cmnd *) + tmp->SCp.ptr; + tmp->SCp.ptr = NULL; + if (tmp->host_scribble) { + if (hostdata->options & OPTION_DEBUG_QUEUES) + printk ("scsi%d : moving command for target %d lun %d to start list\n", + host->host_no, tmp->device->id, tmp->device->lun); + + + to_schedule_list (host, hostdata, + (struct NCR53c7x0_cmd *) + tmp->host_scribble); + } else { + if (((tmp->result & 0xff) == 0xff) || + ((tmp->result & 0xff00) == 0xff00)) { + printk ("scsi%d : danger Will Robinson!\n", + host->host_no); + tmp->result = DID_ERROR << 16; + disable (host); + } + tmp->scsi_done(tmp); + } + done = 0; + } /* if target/lun is not busy */ + } /* if hostdata->issue_queue */ + if (!done) + local_irq_restore(flags); + } /* for host */ + } while (!done); + process_issue_queue_running = 0; +} + +/* + * Function : static void intr_scsi (struct Scsi_Host *host, + * struct NCR53c7x0_cmd *cmd) + * + * Purpose : handle all SCSI interrupts, indicated by the setting + * of the SIP bit in the ISTAT register. + * + * Inputs : host, cmd - host and NCR command causing the interrupt, cmd + * may be NULL. + */ + +static void +intr_scsi (struct Scsi_Host *host, struct NCR53c7x0_cmd *cmd) { + NCR53c7x0_local_declare(); + struct NCR53c7x0_hostdata *hostdata = + (struct NCR53c7x0_hostdata *) host->hostdata[0]; + unsigned char sstat0_sist0, sist1, /* Registers */ + fatal; /* Did a fatal interrupt + occur ? */ + + NCR53c7x0_local_setup(host); + + fatal = 0; + + sstat0_sist0 = NCR53c7x0_read8(SSTAT0_REG); + sist1 = 0; + + if (hostdata->options & OPTION_DEBUG_INTR) + printk ("scsi%d : SIST0 0x%0x, SIST1 0x%0x\n", host->host_no, + sstat0_sist0, sist1); + + /* 250ms selection timeout */ + if (sstat0_sist0 & SSTAT0_700_STO) { + fatal = 1; + if (hostdata->options & OPTION_DEBUG_INTR) { + printk ("scsi%d : Selection Timeout\n", host->host_no); + if (cmd) { + printk("scsi%d : target %d, lun %d, command ", + host->host_no, cmd->cmd->device->id, cmd->cmd->device->lun); + print_command (cmd->cmd->cmnd); + printk("scsi%d : dsp = 0x%x (virt 0x%p)\n", host->host_no, + NCR53c7x0_read32(DSP_REG), + bus_to_virt(NCR53c7x0_read32(DSP_REG))); + } else { + printk("scsi%d : no command\n", host->host_no); + } + } +/* + * XXX - question : how do we want to handle the Illegal Instruction + * interrupt, which may occur before or after the Selection Timeout + * interrupt? + */ + + if (1) { + hostdata->idle = 1; + hostdata->expecting_sto = 0; + + if (hostdata->test_running) { + hostdata->test_running = 0; + hostdata->test_completed = 3; + } else if (cmd) { + abnormal_finished(cmd, DID_BAD_TARGET << 16); + } +#if 0 + hostdata->intrs = 0; +#endif + } + } + +/* + * FIXME : in theory, we can also get a UDC when a STO occurs. + */ + if (sstat0_sist0 & SSTAT0_UDC) { + fatal = 1; + if (cmd) { + printk("scsi%d : target %d lun %d unexpected disconnect\n", + host->host_no, cmd->cmd->device->id, cmd->cmd->device->lun); + print_lots (host); + abnormal_finished(cmd, DID_ERROR << 16); + } else + printk("scsi%d : unexpected disconnect (no command)\n", + host->host_no); + + hostdata->dsp = (u32 *) hostdata->schedule; + hostdata->dsp_changed = 1; + } + + /* SCSI PARITY error */ + if (sstat0_sist0 & SSTAT0_PAR) { + fatal = 1; + if (cmd && cmd->cmd) { + printk("scsi%d : target %d lun %d parity error.\n", + host->host_no, cmd->cmd->device->id, cmd->cmd->device->lun); + abnormal_finished (cmd, DID_PARITY << 16); + } else + printk("scsi%d : parity error\n", host->host_no); + /* Should send message out, parity error */ + + /* XXX - Reduce synchronous transfer rate! */ + hostdata->dsp = hostdata->script + hostdata->E_initiator_abort / + sizeof(u32); + hostdata->dsp_changed = 1; + /* SCSI GROSS error */ + } + + if (sstat0_sist0 & SSTAT0_SGE) { + fatal = 1; + printk("scsi%d : gross error, saved2_dsa = 0x%x\n", host->host_no, + (unsigned int)hostdata->saved2_dsa); + print_lots (host); + + /* + * A SCSI gross error may occur when we have + * + * - A synchronous offset which causes the SCSI FIFO to be overwritten. + * + * - A REQ which causes the maximum synchronous offset programmed in + * the SXFER register to be exceeded. + * + * - A phase change with an outstanding synchronous offset. + * + * - Residual data in the synchronous data FIFO, with a transfer + * other than a synchronous receive is started.$# + */ + + + /* XXX Should deduce synchronous transfer rate! */ + hostdata->dsp = hostdata->script + hostdata->E_initiator_abort / + sizeof(u32); + hostdata->dsp_changed = 1; + /* Phase mismatch */ + } + + if (sstat0_sist0 & SSTAT0_MA) { + fatal = 1; + if (hostdata->options & OPTION_DEBUG_INTR) + printk ("scsi%d : SSTAT0_MA\n", host->host_no); + intr_phase_mismatch (host, cmd); + } + +#if 0 + if (sstat0_sist0 & SIST0_800_RSL) + printk ("scsi%d : Oh no Mr. Bill!\n", host->host_no); +#endif + +/* + * If a fatal SCSI interrupt occurs, we must insure that the DMA and + * SCSI FIFOs were flushed. + */ + + if (fatal) { + if (!hostdata->dstat_valid) { + hostdata->dstat = NCR53c7x0_read8(DSTAT_REG); + hostdata->dstat_valid = 1; + } + + if (!(hostdata->dstat & DSTAT_DFE)) { + printk ("scsi%d : DMA FIFO not empty\n", host->host_no); + /* + * Really need to check this code for 710 RGH. + * Havn't seen any problems, but maybe we should FLUSH before + * clearing sometimes. + */ + NCR53c7x0_write8 (CTEST8_REG, CTEST8_10_CLF); + while (NCR53c7x0_read8 (CTEST8_REG) & CTEST8_10_CLF) + ; + hostdata->dstat |= DSTAT_DFE; + } + } +} + +#ifdef CYCLIC_TRACE + +/* + * The following implements a cyclic log of instructions executed, if you turn + * TRACE on. It will also print the log for you. Very useful when debugging + * 53c710 support, possibly not really needed any more. + */ + +u32 insn_log[4096]; +u32 insn_log_index = 0; + +void log1 (u32 i) +{ + insn_log[insn_log_index++] = i; + if (insn_log_index == 4096) + insn_log_index = 0; +} + +void log_insn (u32 *ip) +{ + log1 ((u32)ip); + log1 (*ip); + log1 (*(ip+1)); + if (((*ip >> 24) & DCMD_TYPE_MASK) == DCMD_TYPE_MMI) + log1 (*(ip+2)); +} + +void dump_log(void) +{ + int cnt = 0; + int i = insn_log_index; + int size; + struct Scsi_Host *host = first_host; + + while (cnt < 4096) { + printk ("%08x (+%6x): ", insn_log[i], (insn_log[i] - (u32)&(((struct NCR53c7x0_hostdata *)host->hostdata[0])->script))/4); + if (++i == 4096) + i = 0; + cnt++; + if (((insn_log[i] >> 24) & DCMD_TYPE_MASK) == DCMD_TYPE_MMI) + size = 3; + else + size = 2; + while (size--) { + printk ("%08x ", insn_log[i]); + if (++i == 4096) + i = 0; + cnt++; + } + printk ("\n"); + } +} +#endif + + +/* + * Function : static void NCR53c7x0_intfly (struct Scsi_Host *host) + * + * Purpose : Scan command queue for specified host, looking for completed + * commands. + * + * Inputs : Scsi_Host pointer. + * + * This is called from the interrupt handler, when a simulated INTFLY + * interrupt occurs. + */ + +static void +NCR53c7x0_intfly (struct Scsi_Host *host) +{ + NCR53c7x0_local_declare(); + struct NCR53c7x0_hostdata *hostdata; /* host->hostdata[0] */ + struct NCR53c7x0_cmd *cmd, /* command which halted */ + **cmd_prev_ptr; + unsigned long flags; + char search_found = 0; /* Got at least one ? */ + + hostdata = (struct NCR53c7x0_hostdata *) host->hostdata[0]; + NCR53c7x0_local_setup(host); + + if (hostdata->options & OPTION_DEBUG_INTR) + printk ("scsi%d : INTFLY\n", host->host_no); + + /* + * Traverse our list of running commands, and look + * for those with valid (non-0xff ff) status and message + * bytes encoded in the result which signify command + * completion. + */ + + local_irq_save(flags); +restart: + for (cmd_prev_ptr = (struct NCR53c7x0_cmd **)&(hostdata->running_list), + cmd = (struct NCR53c7x0_cmd *) hostdata->running_list; cmd ; + cmd_prev_ptr = (struct NCR53c7x0_cmd **) &(cmd->next), + cmd = (struct NCR53c7x0_cmd *) cmd->next) + { + Scsi_Cmnd *tmp; + + if (!cmd) { + printk("scsi%d : very weird.\n", host->host_no); + break; + } + + if (!(tmp = cmd->cmd)) { + printk("scsi%d : weird. NCR53c7x0_cmd has no Scsi_Cmnd\n", + host->host_no); + continue; + } + /* Copy the result over now; may not be complete, + * but subsequent tests may as well be done on + * cached memory. + */ + tmp->result = cmd->result; + + if (((tmp->result & 0xff) == 0xff) || + ((tmp->result & 0xff00) == 0xff00)) + continue; + + search_found = 1; + + if (cmd->bounce.len) + memcpy ((void *)cmd->bounce.addr, + (void *)cmd->bounce.buf, cmd->bounce.len); + + /* Important - remove from list _before_ done is called */ + if (cmd_prev_ptr) + *cmd_prev_ptr = (struct NCR53c7x0_cmd *) cmd->next; + + --hostdata->busy[tmp->device->id][tmp->device->lun]; + cmd->next = hostdata->free; + hostdata->free = cmd; + + tmp->host_scribble = NULL; + + if (hostdata->options & OPTION_DEBUG_INTR) { + printk ("scsi%d : command complete : pid %lu, id %d,lun %d result 0x%x ", + host->host_no, tmp->pid, tmp->device->id, tmp->device->lun, tmp->result); + print_command (tmp->cmnd); + } + + tmp->scsi_done(tmp); + goto restart; + } + local_irq_restore(flags); + + if (!search_found) { + printk ("scsi%d : WARNING : INTFLY with no completed commands.\n", + host->host_no); + } else { + run_process_issue_queue(); + } + return; +} + +/* + * Function : static irqreturn_t NCR53c7x0_intr (int irq, void *dev_id, struct pt_regs * regs) + * + * Purpose : handle NCR53c7x0 interrupts for all NCR devices sharing + * the same IRQ line. + * + * Inputs : Since we're using the SA_INTERRUPT interrupt handler + * semantics, irq indicates the interrupt which invoked + * this handler. + * + * On the 710 we simualte an INTFLY with a script interrupt, and the + * script interrupt handler will call back to this function. + */ + +static irqreturn_t +NCR53c7x0_intr (int irq, void *dev_id, struct pt_regs * regs) +{ + NCR53c7x0_local_declare(); + struct Scsi_Host *host; /* Host we are looking at */ + unsigned char istat; /* Values of interrupt regs */ + struct NCR53c7x0_hostdata *hostdata; /* host->hostdata[0] */ + struct NCR53c7x0_cmd *cmd; /* command which halted */ + u32 *dsa; /* DSA */ + int handled = 0; + +#ifdef NCR_DEBUG + char buf[80]; /* Debugging sprintf buffer */ + size_t buflen; /* Length of same */ +#endif + + host = (struct Scsi_Host *)dev_id; + hostdata = (struct NCR53c7x0_hostdata *) host->hostdata[0]; + NCR53c7x0_local_setup(host); + + /* + * Only read istat once per loop, since reading it again will unstack + * interrupts + */ + + while ((istat = NCR53c7x0_read8(hostdata->istat)) & (ISTAT_SIP|ISTAT_DIP)) { + handled = 1; + hostdata->dsp_changed = 0; + hostdata->dstat_valid = 0; + hostdata->state = STATE_HALTED; + + if (NCR53c7x0_read8 (SSTAT2_REG) & SSTAT2_FF_MASK) + printk ("scsi%d : SCSI FIFO not empty\n", host->host_no); + + /* + * NCR53c700 and NCR53c700-66 change the current SCSI + * process, hostdata->ncrcurrent, in the Linux driver so + * cmd = hostdata->ncrcurrent. + * + * With other chips, we must look through the commands + * executing and find the command structure which + * corresponds to the DSA register. + */ + + if (hostdata->options & OPTION_700) { + cmd = (struct NCR53c7x0_cmd *) hostdata->ncrcurrent; + } else { + dsa = bus_to_virt(NCR53c7x0_read32(DSA_REG)); + for (cmd = (struct NCR53c7x0_cmd *) hostdata->running_list; + cmd && (dsa + (hostdata->dsa_start / sizeof(u32))) != cmd->dsa; + cmd = (struct NCR53c7x0_cmd *)(cmd->next)) + ; + } + if (hostdata->options & OPTION_DEBUG_INTR) { + if (cmd) { + printk("scsi%d : interrupt for pid %lu, id %d, lun %d ", + host->host_no, cmd->cmd->pid, (int) cmd->cmd->device->id, + (int) cmd->cmd->device->lun); + print_command (cmd->cmd->cmnd); + } else { + printk("scsi%d : no active command\n", host->host_no); + } + } + + if (istat & ISTAT_SIP) { + if (hostdata->options & OPTION_DEBUG_INTR) + printk ("scsi%d : ISTAT_SIP\n", host->host_no); + intr_scsi (host, cmd); + } + + if (istat & ISTAT_DIP) { + if (hostdata->options & OPTION_DEBUG_INTR) + printk ("scsi%d : ISTAT_DIP\n", host->host_no); + intr_dma (host, cmd); + } + + if (!hostdata->dstat_valid) { + hostdata->dstat = NCR53c7x0_read8(DSTAT_REG); + hostdata->dstat_valid = 1; + } + + if (!(hostdata->dstat & DSTAT_DFE)) { + printk ("scsi%d : DMA FIFO not empty\n", host->host_no); + /* Really need to check this out for 710 RGH */ + NCR53c7x0_write8 (CTEST8_REG, CTEST8_10_CLF); + while (NCR53c7x0_read8 (CTEST8_REG) & CTEST8_10_CLF) + ; + hostdata->dstat |= DSTAT_DFE; + } + + if (!hostdata->idle && hostdata->state == STATE_HALTED) { + if (!hostdata->dsp_changed) + hostdata->dsp = (u32 *)bus_to_virt(NCR53c7x0_read32(DSP_REG)); +#if 0 + printk("scsi%d : new dsp is 0x%lx (virt 0x%p)\n", + host->host_no, virt_to_bus(hostdata->dsp), hostdata->dsp); +#endif + + hostdata->state = STATE_RUNNING; + NCR53c7x0_write32 (DSP_REG, virt_to_bus(hostdata->dsp)); + if (hostdata->options & OPTION_DEBUG_TRACE) { +#ifdef CYCLIC_TRACE + log_insn (hostdata->dsp); +#else + print_insn (host, hostdata->dsp, "t ", 1); +#endif + NCR53c7x0_write8 (DCNTL_REG, + hostdata->saved_dcntl | DCNTL_SSM | DCNTL_STD); + } + } + } + return IRQ_HANDLED; +} + + +/* + * Function : static int abort_connected (struct Scsi_Host *host) + * + * Purpose : Assuming that the NCR SCSI processor is currently + * halted, break the currently established nexus. Clean + * up of the NCR53c7x0_cmd and Scsi_Cmnd structures should + * be done on receipt of the abort interrupt. + * + * Inputs : host - SCSI host + * + */ + +static int +abort_connected (struct Scsi_Host *host) { +#ifdef NEW_ABORT + NCR53c7x0_local_declare(); +#endif + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; +/* FIXME : this probably should change for production kernels; at the + least, counter should move to a per-host structure. */ + static int counter = 5; +#ifdef NEW_ABORT + int sstat, phase, offset; + u32 *script; + NCR53c7x0_local_setup(host); +#endif + + if (--counter <= 0) { + disable(host); + return 0; + } + + printk ("scsi%d : DANGER : abort_connected() called \n", + host->host_no); + +#ifdef NEW_ABORT + +/* + * New strategy : Rather than using a generic abort routine, + * we'll specifically try to source or sink the appropriate + * amount of data for the phase we're currently in (taking into + * account the current synchronous offset) + */ + + sstat = (NCR53c8x0_read8 (SSTAT2_REG); + offset = OFFSET (sstat & SSTAT2_FF_MASK) >> SSTAT2_FF_SHIFT; + phase = sstat & SSTAT2_PHASE_MASK; + +/* + * SET ATN + * MOVE source_or_sink, WHEN CURRENT PHASE + * < repeat for each outstanding byte > + * JUMP send_abort_message + */ + + script = hostdata->abort_script = kmalloc ( + 8 /* instruction size */ * ( + 1 /* set ATN */ + + (!offset ? 1 : offset) /* One transfer per outstanding byte */ + + 1 /* send abort message */), + GFP_ATOMIC); + + +#else /* def NEW_ABORT */ + hostdata->dsp = hostdata->script + hostdata->E_initiator_abort / + sizeof(u32); +#endif /* def NEW_ABORT */ + hostdata->dsp_changed = 1; + +/* XXX - need to flag the command as aborted after the abort_connected + code runs + */ + return 0; +} + +/* + * Function : static int datapath_residual (Scsi_Host *host) + * + * Purpose : return residual data count of what's in the chip. + * + * Inputs : host - SCSI host + */ + +static int +datapath_residual (struct Scsi_Host *host) { + NCR53c7x0_local_declare(); + int count, synchronous, sstat; + unsigned int ddir; + + NCR53c7x0_local_setup(host); + /* COMPAT : the 700 and 700-66 need to use DFIFO_00_BO_MASK */ + count = ((NCR53c7x0_read8 (DFIFO_REG) & DFIFO_10_BO_MASK) - + (NCR53c7x0_read32 (DBC_REG) & DFIFO_10_BO_MASK)) & DFIFO_10_BO_MASK; + synchronous = NCR53c7x0_read8 (SXFER_REG) & SXFER_MO_MASK; + /* COMPAT : DDIR is elsewhere on non-'8xx chips. */ + ddir = NCR53c7x0_read8 (CTEST0_REG_700) & CTEST0_700_DDIR; + + if (ddir) { + /* Receive */ + if (synchronous) + count += (NCR53c7x0_read8 (SSTAT2_REG) & SSTAT2_FF_MASK) >> SSTAT2_FF_SHIFT; + else + if (NCR53c7x0_read8 (SSTAT1_REG) & SSTAT1_ILF) + ++count; + } else { + /* Send */ + sstat = NCR53c7x0_read8 (SSTAT1_REG); + if (sstat & SSTAT1_OLF) + ++count; + if (synchronous && (sstat & SSTAT1_ORF)) + ++count; + } + return count; +} + +/* + * Function : static const char * sbcl_to_phase (int sbcl)_ + * + * Purpose : Convert SBCL register to user-parsable phase representation + * + * Inputs : sbcl - value of sbcl register + */ + + +static const char * +sbcl_to_phase (int sbcl) { + switch (sbcl & SBCL_PHASE_MASK) { + case SBCL_PHASE_DATAIN: + return "DATAIN"; + case SBCL_PHASE_DATAOUT: + return "DATAOUT"; + case SBCL_PHASE_MSGIN: + return "MSGIN"; + case SBCL_PHASE_MSGOUT: + return "MSGOUT"; + case SBCL_PHASE_CMDOUT: + return "CMDOUT"; + case SBCL_PHASE_STATIN: + return "STATUSIN"; + default: + return "unknown"; + } +} + +/* + * Function : static const char * sstat2_to_phase (int sstat)_ + * + * Purpose : Convert SSTAT2 register to user-parsable phase representation + * + * Inputs : sstat - value of sstat register + */ + + +static const char * +sstat2_to_phase (int sstat) { + switch (sstat & SSTAT2_PHASE_MASK) { + case SSTAT2_PHASE_DATAIN: + return "DATAIN"; + case SSTAT2_PHASE_DATAOUT: + return "DATAOUT"; + case SSTAT2_PHASE_MSGIN: + return "MSGIN"; + case SSTAT2_PHASE_MSGOUT: + return "MSGOUT"; + case SSTAT2_PHASE_CMDOUT: + return "CMDOUT"; + case SSTAT2_PHASE_STATIN: + return "STATUSIN"; + default: + return "unknown"; + } +} + +/* + * Function : static void intr_phase_mismatch (struct Scsi_Host *host, + * struct NCR53c7x0_cmd *cmd) + * + * Purpose : Handle phase mismatch interrupts + * + * Inputs : host, cmd - host and NCR command causing the interrupt, cmd + * may be NULL. + * + * Side effects : The abort_connected() routine is called or the NCR chip + * is restarted, jumping to the command_complete entry point, or + * patching the address and transfer count of the current instruction + * and calling the msg_in entry point as appropriate. + */ + +static void +intr_phase_mismatch (struct Scsi_Host *host, struct NCR53c7x0_cmd *cmd) { + NCR53c7x0_local_declare(); + u32 dbc_dcmd, *dsp, *dsp_next; + unsigned char dcmd, sbcl; + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + int residual; + enum {ACTION_ABORT, ACTION_ABORT_PRINT, ACTION_CONTINUE} action = + ACTION_ABORT_PRINT; + const char *where = NULL; + + NCR53c7x0_local_setup(host); + + /* + * Corrective action is based on where in the SCSI SCRIPT(tm) the error + * occurred, as well as which SCSI phase we are currently in. + */ + dsp_next = bus_to_virt(NCR53c7x0_read32(DSP_REG)); + + /* + * Fetch the current instruction, and remove the operands for easier + * interpretation. + */ + dbc_dcmd = NCR53c7x0_read32(DBC_REG); + dcmd = (dbc_dcmd & 0xff000000) >> 24; + /* + * Like other processors, the NCR adjusts the instruction pointer before + * instruction decode. Set the DSP address back to what it should + * be for this instruction based on its size (2 or 3 32 bit words). + */ + dsp = dsp_next - NCR53c7x0_insn_size(dcmd); + + + /* + * Read new SCSI phase from the SBCL lines. Since all of our code uses + * a WHEN conditional instead of an IF conditional, we don't need to + * wait for a new REQ. + */ + sbcl = NCR53c7x0_read8(SBCL_REG) & SBCL_PHASE_MASK; + + if (!cmd) { + action = ACTION_ABORT_PRINT; + where = "no current command"; + /* + * The way my SCSI SCRIPTS(tm) are architected, recoverable phase + * mismatches should only occur where we're doing a multi-byte + * BMI instruction. Specifically, this means + * + * - select messages (a SCSI-I target may ignore additional messages + * after the IDENTIFY; any target may reject a SDTR or WDTR) + * + * - command out (targets may send a message to signal an error + * condition, or go into STATUSIN after they've decided + * they don't like the command. + * + * - reply_message (targets may reject a multi-byte message in the + * middle) + * + * - data transfer routines (command completion with buffer space + * left, disconnect message, or error message) + */ + } else if (((dsp >= cmd->data_transfer_start && + dsp < cmd->data_transfer_end)) || dsp == (cmd->residual + 2)) { + if ((dcmd & (DCMD_TYPE_MASK|DCMD_BMI_OP_MASK|DCMD_BMI_INDIRECT| + DCMD_BMI_MSG|DCMD_BMI_CD)) == (DCMD_TYPE_BMI| + DCMD_BMI_OP_MOVE_I)) { + residual = datapath_residual (host); + if (hostdata->options & OPTION_DEBUG_DISCONNECT) + printk ("scsi%d : handling residual transfer (+ %d bytes from DMA FIFO)\n", + host->host_no, residual); + + /* + * The first instruction is a CALL to the alternate handler for + * this data transfer phase, so we can do calls to + * munge_msg_restart as we would if control were passed + * from normal dynamic code. + */ + if (dsp != cmd->residual + 2) { + cmd->residual[0] = ((DCMD_TYPE_TCI | DCMD_TCI_OP_CALL | + ((dcmd & DCMD_BMI_IO) ? DCMD_TCI_IO : 0)) << 24) | + DBC_TCI_WAIT_FOR_VALID | DBC_TCI_COMPARE_PHASE; + cmd->residual[1] = virt_to_bus(hostdata->script) + + ((dcmd & DCMD_BMI_IO) + ? hostdata->E_other_in : hostdata->E_other_out); + } + + /* + * The second instruction is the a data transfer block + * move instruction, reflecting the pointer and count at the + * time of the phase mismatch. + */ + cmd->residual[2] = dbc_dcmd + residual; + cmd->residual[3] = NCR53c7x0_read32(DNAD_REG) - residual; + + /* + * The third and final instruction is a jump to the instruction + * which follows the instruction which had to be 'split' + */ + if (dsp != cmd->residual + 2) { + cmd->residual[4] = ((DCMD_TYPE_TCI|DCMD_TCI_OP_JUMP) + << 24) | DBC_TCI_TRUE; + cmd->residual[5] = virt_to_bus(dsp_next); + } + + /* + * For the sake of simplicity, transfer control to the + * conditional CALL at the start of the residual buffer. + */ + hostdata->dsp = cmd->residual; + hostdata->dsp_changed = 1; + action = ACTION_CONTINUE; + } else { + where = "non-BMI dynamic DSA code"; + action = ACTION_ABORT_PRINT; + } + } else if (dsp == (hostdata->script + hostdata->E_select_msgout / 4 + 2)) { + /* RGH 290697: Added +2 above, to compensate for the script + * instruction which disables the selection timer. */ + /* Release ATN */ + NCR53c7x0_write8 (SOCL_REG, 0); + switch (sbcl) { + /* + * Some devices (SQ555 come to mind) grab the IDENTIFY message + * sent on selection, and decide to go into COMMAND OUT phase + * rather than accepting the rest of the messages or rejecting + * them. Handle these devices gracefully. + */ + case SBCL_PHASE_CMDOUT: + hostdata->dsp = dsp + 2 /* two _words_ */; + hostdata->dsp_changed = 1; + printk ("scsi%d : target %d ignored SDTR and went into COMMAND OUT\n", + host->host_no, cmd->cmd->device->id); + cmd->flags &= ~CMD_FLAG_SDTR; + action = ACTION_CONTINUE; + break; + case SBCL_PHASE_MSGIN: + hostdata->dsp = hostdata->script + hostdata->E_msg_in / + sizeof(u32); + hostdata->dsp_changed = 1; + action = ACTION_CONTINUE; + break; + default: + where="select message out"; + action = ACTION_ABORT_PRINT; + } + /* + * Some SCSI devices will interpret a command as they read the bytes + * off the SCSI bus, and may decide that the command is Bogus before + * they've read the entire command off the bus. + */ + } else if (dsp == hostdata->script + hostdata->E_cmdout_cmdout / sizeof + (u32)) { + hostdata->dsp = hostdata->script + hostdata->E_data_transfer / + sizeof (u32); + hostdata->dsp_changed = 1; + action = ACTION_CONTINUE; + /* FIXME : we need to handle message reject, etc. within msg_respond. */ +#ifdef notyet + } else if (dsp == hostdata->script + hostdata->E_reply_message) { + switch (sbcl) { + /* Any other phase mismatches abort the currently executing command. */ +#endif + } else { + where = "unknown location"; + action = ACTION_ABORT_PRINT; + } + + /* Flush DMA FIFO */ + if (!hostdata->dstat_valid) { + hostdata->dstat = NCR53c7x0_read8(DSTAT_REG); + hostdata->dstat_valid = 1; + } + if (!(hostdata->dstat & DSTAT_DFE)) { + /* Really need to check this out for 710 RGH */ + NCR53c7x0_write8 (CTEST8_REG, CTEST8_10_CLF); + while (NCR53c7x0_read8 (CTEST8_REG) & CTEST8_10_CLF); + hostdata->dstat |= DSTAT_DFE; + } + + switch (action) { + case ACTION_ABORT_PRINT: + printk("scsi%d : %s : unexpected phase %s.\n", + host->host_no, where ? where : "unknown location", + sbcl_to_phase(sbcl)); + print_lots (host); + /* Fall through to ACTION_ABORT */ + case ACTION_ABORT: + abort_connected (host); + break; + case ACTION_CONTINUE: + break; + } + +#if 0 + if (hostdata->dsp_changed) { + printk("scsi%d: new dsp 0x%p\n", host->host_no, hostdata->dsp); + print_insn (host, hostdata->dsp, "", 1); + } +#endif +} + +/* + * Function : static void intr_bf (struct Scsi_Host *host, + * struct NCR53c7x0_cmd *cmd) + * + * Purpose : handle BUS FAULT interrupts + * + * Inputs : host, cmd - host and NCR command causing the interrupt, cmd + * may be NULL. + */ + +static void +intr_bf (struct Scsi_Host *host, struct NCR53c7x0_cmd *cmd) { + NCR53c7x0_local_declare(); + u32 *dsp, + *next_dsp, /* Current dsp */ + *dsa, + dbc_dcmd; /* DCMD (high eight bits) + DBC */ + char *reason = NULL; + /* Default behavior is for a silent error, with a retry until we've + exhausted retries. */ + enum {MAYBE, ALWAYS, NEVER} retry = MAYBE; + int report = 0; + NCR53c7x0_local_setup(host); + + dbc_dcmd = NCR53c7x0_read32 (DBC_REG); + next_dsp = bus_to_virt (NCR53c7x0_read32(DSP_REG)); + dsp = next_dsp - NCR53c7x0_insn_size ((dbc_dcmd >> 24) & 0xff); +/* FIXME - check chip type */ + dsa = bus_to_virt (NCR53c7x0_read32(DSA_REG)); + + /* + * Bus faults can be caused by either a Bad Address or + * Target Abort. We should check the Received Target Abort + * bit of the PCI status register and Master Abort Bit. + * + * - Master Abort bit indicates that no device claimed + * the address with DEVSEL within five clocks + * + * - Target Abort bit indicates that a target claimed it, + * but changed its mind once it saw the byte enables. + * + */ + + /* 53c710, not PCI system */ + report = 1; + reason = "Unknown"; + +#ifndef notyet + report = 1; +#endif + if (report && reason) + { + printk(KERN_ALERT "scsi%d : BUS FAULT reason = %s\n", + host->host_no, reason ? reason : "unknown"); + print_lots (host); + } + +#ifndef notyet + retry = NEVER; +#endif + + /* + * TODO : we should attempt to recover from any spurious bus + * faults. After X retries, we should figure that things are + * sufficiently wedged, and call NCR53c7xx_reset. + * + * This code should only get executed once we've decided that we + * cannot retry. + */ + + if (retry == NEVER) { + printk(KERN_ALERT " mail richard@sleepie.demon.co.uk\n"); + FATAL (host); + } +} + +/* + * Function : static void intr_dma (struct Scsi_Host *host, + * struct NCR53c7x0_cmd *cmd) + * + * Purpose : handle all DMA interrupts, indicated by the setting + * of the DIP bit in the ISTAT register. + * + * Inputs : host, cmd - host and NCR command causing the interrupt, cmd + * may be NULL. + */ + +static void +intr_dma (struct Scsi_Host *host, struct NCR53c7x0_cmd *cmd) { + NCR53c7x0_local_declare(); + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + unsigned char dstat; /* DSTAT */ + u32 *dsp, + *next_dsp, /* Current dsp */ + *dsa, + dbc_dcmd; /* DCMD (high eight bits) + DBC */ + int tmp; + unsigned long flags; + NCR53c7x0_local_setup(host); + + if (!hostdata->dstat_valid) { + hostdata->dstat = NCR53c7x0_read8(DSTAT_REG); + hostdata->dstat_valid = 1; + } + + dstat = hostdata->dstat; + + if (hostdata->options & OPTION_DEBUG_INTR) + printk("scsi%d : DSTAT=0x%x\n", host->host_no, (int) dstat); + + dbc_dcmd = NCR53c7x0_read32 (DBC_REG); + next_dsp = bus_to_virt(NCR53c7x0_read32(DSP_REG)); + dsp = next_dsp - NCR53c7x0_insn_size ((dbc_dcmd >> 24) & 0xff); +/* XXX - check chip type */ + dsa = bus_to_virt(NCR53c7x0_read32(DSA_REG)); + + /* + * DSTAT_ABRT is the aborted interrupt. This is set whenever the + * SCSI chip is aborted. + * + * With NCR53c700 and NCR53c700-66 style chips, we should only + * get this when the chip is currently running the accept + * reselect/select code and we have set the abort bit in the + * ISTAT register. + * + */ + + if (dstat & DSTAT_ABRT) { +#if 0 + /* XXX - add code here to deal with normal abort */ + if ((hostdata->options & OPTION_700) && (hostdata->state == + STATE_ABORTING)) { + } else +#endif + { + printk(KERN_ALERT "scsi%d : unexpected abort interrupt at\n" + " ", host->host_no); + print_insn (host, dsp, KERN_ALERT "s ", 1); + FATAL (host); + } + } + + /* + * DSTAT_SSI is the single step interrupt. Should be generated + * whenever we have single stepped or are tracing. + */ + + if (dstat & DSTAT_SSI) { + if (hostdata->options & OPTION_DEBUG_TRACE) { + /* Don't print instr. until we write DSP at end of intr function */ + } else if (hostdata->options & OPTION_DEBUG_SINGLE) { + print_insn (host, dsp, "s ", 0); + local_irq_save(flags); +/* XXX - should we do this, or can we get away with writing dsp? */ + + NCR53c7x0_write8 (DCNTL_REG, (NCR53c7x0_read8(DCNTL_REG) & + ~DCNTL_SSM) | DCNTL_STD); + local_irq_restore(flags); + } else { + printk(KERN_ALERT "scsi%d : unexpected single step interrupt at\n" + " ", host->host_no); + print_insn (host, dsp, KERN_ALERT "", 1); + printk(KERN_ALERT " mail drew@PoohSticks.ORG\n"); + FATAL (host); + } + } + + /* + * DSTAT_IID / DSTAT_OPC (same bit, same meaning, only the name + * is different) is generated whenever an illegal instruction is + * encountered. + * + * XXX - we may want to emulate INTFLY here, so we can use + * the same SCSI SCRIPT (tm) for NCR53c710 through NCR53c810 + * chips. + */ + + if (dstat & DSTAT_OPC) { + /* + * Ascertain if this IID interrupts occurred before or after a STO + * interrupt. Since the interrupt handling code now leaves + * DSP unmodified until _after_ all stacked interrupts have been + * processed, reading the DSP returns the original DSP register. + * This means that if dsp lies between the select code, and + * message out following the selection code (where the IID interrupt + * would have to have occurred by due to the implicit wait for REQ), + * we have an IID interrupt resulting from a STO condition and + * can ignore it. + */ + + if (((dsp >= (hostdata->script + hostdata->E_select / sizeof(u32))) && + (dsp <= (hostdata->script + hostdata->E_select_msgout / + sizeof(u32) + 8))) || (hostdata->test_running == 2)) { + if (hostdata->options & OPTION_DEBUG_INTR) + printk ("scsi%d : ignoring DSTAT_IID for SSTAT_STO\n", + host->host_no); + if (hostdata->expecting_iid) { + hostdata->expecting_iid = 0; + hostdata->idle = 1; + if (hostdata->test_running == 2) { + hostdata->test_running = 0; + hostdata->test_completed = 3; + } else if (cmd) + abnormal_finished (cmd, DID_BAD_TARGET << 16); + } else { + hostdata->expecting_sto = 1; + } + /* + * We can't guarantee we'll be able to execute the WAIT DISCONNECT + * instruction within the 3.4us of bus free and arbitration delay + * that a target can RESELECT in and assert REQ after we've dropped + * ACK. If this happens, we'll get an illegal instruction interrupt. + * Doing away with the WAIT DISCONNECT instructions broke everything, + * so instead I'll settle for moving one WAIT DISCONNECT a few + * instructions closer to the CLEAR ACK before it to minimize the + * chances of this happening, and handle it if it occurs anyway. + * + * Simply continue with what we were doing, and control should + * be transferred to the schedule routine which will ultimately + * pass control onto the reselection or selection (not yet) + * code. + */ + } else if (dbc_dcmd == 0x48000000 && (NCR53c7x0_read8 (SBCL_REG) & + SBCL_REQ)) { + if (!(hostdata->options & OPTION_NO_PRINT_RACE)) + { + printk("scsi%d: REQ before WAIT DISCONNECT IID\n", + host->host_no); + hostdata->options |= OPTION_NO_PRINT_RACE; + } + } else { + printk(KERN_ALERT "scsi%d : invalid instruction\n", host->host_no); + print_lots (host); + printk(KERN_ALERT " mail Richard@sleepie.demon.co.uk with ALL\n" + " boot messages and diagnostic output\n"); + FATAL (host); + } + } + + /* + * DSTAT_BF are bus fault errors. DSTAT_800_BF is valid for 710 also. + */ + + if (dstat & DSTAT_800_BF) { + intr_bf (host, cmd); + } + + + /* + * DSTAT_SIR interrupts are generated by the execution of + * the INT instruction. Since the exact values available + * are determined entirely by the SCSI script running, + * and are local to a particular script, a unique handler + * is called for each script. + */ + + if (dstat & DSTAT_SIR) { + if (hostdata->options & OPTION_DEBUG_INTR) + printk ("scsi%d : DSTAT_SIR\n", host->host_no); + switch ((tmp = hostdata->dstat_sir_intr (host, cmd))) { + case SPECIFIC_INT_NOTHING: + case SPECIFIC_INT_RESTART: + break; + case SPECIFIC_INT_ABORT: + abort_connected(host); + break; + case SPECIFIC_INT_PANIC: + printk(KERN_ALERT "scsi%d : failure at ", host->host_no); + print_insn (host, dsp, KERN_ALERT "", 1); + printk(KERN_ALERT " dstat_sir_intr() returned SPECIFIC_INT_PANIC\n"); + FATAL (host); + break; + case SPECIFIC_INT_BREAK: + intr_break (host, cmd); + break; + default: + printk(KERN_ALERT "scsi%d : failure at ", host->host_no); + print_insn (host, dsp, KERN_ALERT "", 1); + printk(KERN_ALERT" dstat_sir_intr() returned unknown value %d\n", + tmp); + FATAL (host); + } + } +} + +/* + * Function : static int print_insn (struct Scsi_Host *host, + * u32 *insn, int kernel) + * + * Purpose : print numeric representation of the instruction pointed + * to by insn to the debugging or kernel message buffer + * as appropriate. + * + * If desired, a user level program can interpret this + * information. + * + * Inputs : host, insn - host, pointer to instruction, prefix - + * string to prepend, kernel - use printk instead of debugging buffer. + * + * Returns : size, in u32s, of instruction printed. + */ + +/* + * FIXME: should change kernel parameter so that it takes an ENUM + * specifying severity - either KERN_ALERT or KERN_PANIC so + * all panic messages are output with the same severity. + */ + +static int +print_insn (struct Scsi_Host *host, const u32 *insn, + const char *prefix, int kernel) { + char buf[160], /* Temporary buffer and pointer. ICKY + arbitrary length. */ + + + *tmp; + unsigned char dcmd; /* dcmd register for *insn */ + int size; + + /* + * Check to see if the instruction pointer is not bogus before + * indirecting through it; avoiding red-zone at start of + * memory. + * + * FIXME: icky magic needs to happen here on non-intel boxes which + * don't have kernel memory mapped in like this. Might be reasonable + * to use vverify()? + */ + + if (virt_to_phys((void *)insn) < PAGE_SIZE || + virt_to_phys((void *)(insn + 8)) > virt_to_phys(high_memory) || + ((((dcmd = (insn[0] >> 24) & 0xff) & DCMD_TYPE_MMI) == DCMD_TYPE_MMI) && + virt_to_phys((void *)(insn + 12)) > virt_to_phys(high_memory))) { + size = 0; + sprintf (buf, "%s%p: address out of range\n", + prefix, insn); + } else { +/* + * FIXME : (void *) cast in virt_to_bus should be unnecessary, because + * it should take const void * as argument. + */ +#if !defined(CONFIG_MVME16x) && !defined(CONFIG_BVME6000) + sprintf(buf, "%s0x%lx (virt 0x%p) : 0x%08x 0x%08x (virt 0x%p)", + (prefix ? prefix : ""), virt_to_bus((void *) insn), insn, + insn[0], insn[1], bus_to_virt (insn[1])); +#else + /* Remove virtual addresses to reduce output, as they are the same */ + sprintf(buf, "%s0x%x (+%x) : 0x%08x 0x%08x", + (prefix ? prefix : ""), (u32)insn, ((u32)insn - + (u32)&(((struct NCR53c7x0_hostdata *)host->hostdata[0])->script))/4, + insn[0], insn[1]); +#endif + tmp = buf + strlen(buf); + if ((dcmd & DCMD_TYPE_MASK) == DCMD_TYPE_MMI) { +#if !defined(CONFIG_MVME16x) && !defined(CONFIG_BVME6000) + sprintf (tmp, " 0x%08x (virt 0x%p)\n", insn[2], + bus_to_virt(insn[2])); +#else + /* Remove virtual addr to reduce output, as it is the same */ + sprintf (tmp, " 0x%08x\n", insn[2]); +#endif + size = 3; + } else { + sprintf (tmp, "\n"); + size = 2; + } + } + + if (kernel) + printk ("%s", buf); +#ifdef NCR_DEBUG + else { + size_t len = strlen(buf); + debugger_kernel_write(host, buf, len); + } +#endif + return size; +} + +/* + * Function : int NCR53c7xx_abort (Scsi_Cmnd *cmd) + * + * Purpose : Abort an errant SCSI command, doing all necessary + * cleanup of the issue_queue, running_list, shared Linux/NCR + * dsa issue and reconnect queues. + * + * Inputs : cmd - command to abort, code - entire result field + * + * Returns : 0 on success, -1 on failure. + */ + +int +NCR53c7xx_abort (Scsi_Cmnd *cmd) { + NCR53c7x0_local_declare(); + struct Scsi_Host *host = cmd->device->host; + struct NCR53c7x0_hostdata *hostdata = host ? (struct NCR53c7x0_hostdata *) + host->hostdata[0] : NULL; + unsigned long flags; + struct NCR53c7x0_cmd *curr, **prev; + Scsi_Cmnd *me, **last; +#if 0 + static long cache_pid = -1; +#endif + + + if (!host) { + printk ("Bogus SCSI command pid %ld; no host structure\n", + cmd->pid); + return SCSI_ABORT_ERROR; + } else if (!hostdata) { + printk ("Bogus SCSI host %d; no hostdata\n", host->host_no); + return SCSI_ABORT_ERROR; + } + NCR53c7x0_local_setup(host); + +/* + * CHECK : I don't think that reading ISTAT will unstack any interrupts, + * since we need to write the INTF bit to clear it, and SCSI/DMA + * interrupts don't clear until we read SSTAT/SIST and DSTAT registers. + * + * See that this is the case. Appears to be correct on the 710, at least. + * + * I suspect that several of our failures may be coming from a new fatal + * interrupt (possibly due to a phase mismatch) happening after we've left + * the interrupt handler, but before the PIC has had the interrupt condition + * cleared. + */ + + if (NCR53c7x0_read8(hostdata->istat) & (ISTAT_DIP|ISTAT_SIP)) { + printk ("scsi%d : dropped interrupt for command %ld\n", host->host_no, + cmd->pid); + NCR53c7x0_intr (host->irq, NULL, NULL); + return SCSI_ABORT_BUSY; + } + + local_irq_save(flags); +#if 0 + if (cache_pid == cmd->pid) + panic ("scsi%d : bloody fetus %d\n", host->host_no, cmd->pid); + else + cache_pid = cmd->pid; +#endif + + +/* + * The command could be hiding in the issue_queue. This would be very + * nice, as commands can't be moved from the high level driver's issue queue + * into the shared queue until an interrupt routine is serviced, and this + * moving is atomic. + * + * If this is the case, we don't have to worry about anything - we simply + * pull the command out of the old queue, and call it aborted. + */ + + for (me = (Scsi_Cmnd *) hostdata->issue_queue, + last = (Scsi_Cmnd **) &(hostdata->issue_queue); + me && me != cmd; last = (Scsi_Cmnd **)&(me->SCp.ptr), + me = (Scsi_Cmnd *)me->SCp.ptr); + + if (me) { + *last = (Scsi_Cmnd *) me->SCp.ptr; + if (me->host_scribble) { + ((struct NCR53c7x0_cmd *)me->host_scribble)->next = hostdata->free; + hostdata->free = (struct NCR53c7x0_cmd *) me->host_scribble; + me->host_scribble = NULL; + } + cmd->result = DID_ABORT << 16; + cmd->scsi_done(cmd); + printk ("scsi%d : found command %ld in Linux issue queue\n", + host->host_no, me->pid); + local_irq_restore(flags); + run_process_issue_queue(); + return SCSI_ABORT_SUCCESS; + } + +/* + * That failing, the command could be in our list of already executing + * commands. If this is the case, drastic measures are called for. + */ + + for (curr = (struct NCR53c7x0_cmd *) hostdata->running_list, + prev = (struct NCR53c7x0_cmd **) &(hostdata->running_list); + curr && curr->cmd != cmd; prev = (struct NCR53c7x0_cmd **) + &(curr->next), curr = (struct NCR53c7x0_cmd *) curr->next); + + if (curr) { + if ((curr->result & 0xff) != 0xff && (curr->result & 0xff00) != 0xff00) { + cmd->result = curr->result; + if (prev) + *prev = (struct NCR53c7x0_cmd *) curr->next; + curr->next = (struct NCR53c7x0_cmd *) hostdata->free; + cmd->host_scribble = NULL; + hostdata->free = curr; + cmd->scsi_done(cmd); + printk ("scsi%d : found finished command %ld in running list\n", + host->host_no, cmd->pid); + local_irq_restore(flags); + return SCSI_ABORT_NOT_RUNNING; + } else { + printk ("scsi%d : DANGER : command running, can not abort.\n", + cmd->device->host->host_no); + local_irq_restore(flags); + return SCSI_ABORT_BUSY; + } + } + +/* + * And if we couldn't find it in any of our queues, it must have been + * a dropped interrupt. + */ + + curr = (struct NCR53c7x0_cmd *) cmd->host_scribble; + if (curr) { + curr->next = hostdata->free; + hostdata->free = curr; + cmd->host_scribble = NULL; + } + + if (curr == NULL || ((curr->result & 0xff00) == 0xff00) || + ((curr->result & 0xff) == 0xff)) { + printk ("scsi%d : did this command ever run?\n", host->host_no); + cmd->result = DID_ABORT << 16; + } else { + printk ("scsi%d : probably lost INTFLY, normal completion\n", + host->host_no); + cmd->result = curr->result; +/* + * FIXME : We need to add an additional flag which indicates if a + * command was ever counted as BUSY, so if we end up here we can + * decrement the busy count if and only if it is necessary. + */ + --hostdata->busy[cmd->device->id][cmd->device->lun]; + } + local_irq_restore(flags); + cmd->scsi_done(cmd); + +/* + * We need to run process_issue_queue since termination of this command + * may allow another queued command to execute first? + */ + return SCSI_ABORT_NOT_RUNNING; +} + +/* + * Function : int NCR53c7xx_reset (Scsi_Cmnd *cmd) + * + * Purpose : perform a hard reset of the SCSI bus and NCR + * chip. + * + * Inputs : cmd - command which caused the SCSI RESET + * + * Returns : 0 on success. + */ + +int +NCR53c7xx_reset (Scsi_Cmnd *cmd, unsigned int reset_flags) { + NCR53c7x0_local_declare(); + unsigned long flags; + int found = 0; + struct NCR53c7x0_cmd * c; + Scsi_Cmnd *tmp; + /* + * When we call scsi_done(), it's going to wake up anything sleeping on the + * resources which were in use by the aborted commands, and we'll start to + * get new commands. + * + * We can't let this happen until after we've re-initialized the driver + * structures, and can't reinitialize those structures until after we've + * dealt with their contents. + * + * So, we need to find all of the commands which were running, stick + * them on a linked list of completed commands (we'll use the host_scribble + * pointer), do our reinitialization, and then call the done function for + * each command. + */ + Scsi_Cmnd *nuke_list = NULL; + struct Scsi_Host *host = cmd->device->host; + struct NCR53c7x0_hostdata *hostdata = + (struct NCR53c7x0_hostdata *) host->hostdata[0]; + + NCR53c7x0_local_setup(host); + local_irq_save(flags); + ncr_halt (host); + print_lots (host); + dump_events (host, 30); + ncr_scsi_reset (host); + for (tmp = nuke_list = return_outstanding_commands (host, 1 /* free */, + 0 /* issue */ ); tmp; tmp = (Scsi_Cmnd *) tmp->SCp.buffer) + if (tmp == cmd) { + found = 1; + break; + } + + /* + * If we didn't find the command which caused this reset in our running + * list, then we've lost it. See that it terminates normally anyway. + */ + if (!found) { + c = (struct NCR53c7x0_cmd *) cmd->host_scribble; + if (c) { + cmd->host_scribble = NULL; + c->next = hostdata->free; + hostdata->free = c; + } else + printk ("scsi%d: lost command %ld\n", host->host_no, cmd->pid); + cmd->SCp.buffer = (struct scatterlist *) nuke_list; + nuke_list = cmd; + } + + NCR53c7x0_driver_init (host); + hostdata->soft_reset (host); + if (hostdata->resets == 0) + disable(host); + else if (hostdata->resets != -1) + --hostdata->resets; + local_irq_restore(flags); + for (; nuke_list; nuke_list = tmp) { + tmp = (Scsi_Cmnd *) nuke_list->SCp.buffer; + nuke_list->result = DID_RESET << 16; + nuke_list->scsi_done (nuke_list); + } + local_irq_restore(flags); + return SCSI_RESET_SUCCESS; +} + +/* + * The NCR SDMS bios follows Annex A of the SCSI-CAM draft, and + * therefore shares the scsicam_bios_param function. + */ + +/* + * Function : int insn_to_offset (Scsi_Cmnd *cmd, u32 *insn) + * + * Purpose : convert instructions stored at NCR pointer into data + * pointer offset. + * + * Inputs : cmd - SCSI command; insn - pointer to instruction. Either current + * DSP, or saved data pointer. + * + * Returns : offset on success, -1 on failure. + */ + + +static int +insn_to_offset (Scsi_Cmnd *cmd, u32 *insn) { + struct NCR53c7x0_hostdata *hostdata = + (struct NCR53c7x0_hostdata *) cmd->device->host->hostdata[0]; + struct NCR53c7x0_cmd *ncmd = + (struct NCR53c7x0_cmd *) cmd->host_scribble; + int offset = 0, buffers; + struct scatterlist *segment; + char *ptr; + int found = 0; + +/* + * With the current code implementation, if the insn is inside dynamically + * generated code, the data pointer will be the instruction preceding + * the next transfer segment. + */ + + if (!check_address ((unsigned long) ncmd, sizeof (struct NCR53c7x0_cmd)) && + ((insn >= ncmd->data_transfer_start && + insn < ncmd->data_transfer_end) || + (insn >= ncmd->residual && + insn < (ncmd->residual + + sizeof(ncmd->residual))))) { + ptr = bus_to_virt(insn[3]); + + if ((buffers = cmd->use_sg)) { + for (offset = 0, + segment = (struct scatterlist *) cmd->buffer; + buffers && !((found = ((ptr >= (char *)page_address(segment->page)+segment->offset) && + (ptr < ((char *)page_address(segment->page)+segment->offset+segment->length))))); + --buffers, offset += segment->length, ++segment) +#if 0 + printk("scsi%d: comparing 0x%p to 0x%p\n", + cmd->device->host->host_no, saved, page_address(segment->page+segment->offset); +#else + ; +#endif + offset += ptr - ((char *)page_address(segment->page)+segment->offset); + } else { + found = 1; + offset = ptr - (char *) (cmd->request_buffer); + } + } else if ((insn >= hostdata->script + + hostdata->E_data_transfer / sizeof(u32)) && + (insn <= hostdata->script + + hostdata->E_end_data_transfer / sizeof(u32))) { + found = 1; + offset = 0; + } + return found ? offset : -1; +} + + + +/* + * Function : void print_progress (Scsi_Cmnd *cmd) + * + * Purpose : print the current location of the saved data pointer + * + * Inputs : cmd - command we are interested in + * + */ + +static void +print_progress (Scsi_Cmnd *cmd) { + NCR53c7x0_local_declare(); + struct NCR53c7x0_cmd *ncmd = + (struct NCR53c7x0_cmd *) cmd->host_scribble; + int offset, i; + char *where; + u32 *ptr; + NCR53c7x0_local_setup (cmd->device->host); + + if (check_address ((unsigned long) ncmd,sizeof (struct NCR53c7x0_cmd)) == 0) + { + printk("\nNCR53c7x0_cmd fields:\n"); + printk(" bounce.len=0x%x, addr=0x%0x, buf[]=0x%02x %02x %02x %02x\n", + ncmd->bounce.len, ncmd->bounce.addr, ncmd->bounce.buf[0], + ncmd->bounce.buf[1], ncmd->bounce.buf[2], ncmd->bounce.buf[3]); + printk(" result=%04x, cdb[0]=0x%02x\n", ncmd->result, ncmd->cmnd[0]); + } + + for (i = 0; i < 2; ++i) { + if (check_address ((unsigned long) ncmd, + sizeof (struct NCR53c7x0_cmd)) == -1) + continue; + if (!i) { + where = "saved"; + ptr = bus_to_virt(ncmd->saved_data_pointer); + } else { + where = "active"; + ptr = bus_to_virt (NCR53c7x0_read32 (DSP_REG) - + NCR53c7x0_insn_size (NCR53c7x0_read8 (DCMD_REG)) * + sizeof(u32)); + } + offset = insn_to_offset (cmd, ptr); + + if (offset != -1) + printk ("scsi%d : %s data pointer at offset %d\n", + cmd->device->host->host_no, where, offset); + else { + int size; + printk ("scsi%d : can't determine %s data pointer offset\n", + cmd->device->host->host_no, where); + if (ncmd) { + size = print_insn (cmd->device->host, + bus_to_virt(ncmd->saved_data_pointer), "", 1); + print_insn (cmd->device->host, + bus_to_virt(ncmd->saved_data_pointer) + size * sizeof(u32), + "", 1); + } + } + } +} + + +static void +print_dsa (struct Scsi_Host *host, u32 *dsa, const char *prefix) { + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + int i, len; + char *ptr; + Scsi_Cmnd *cmd; + + if (check_address ((unsigned long) dsa, hostdata->dsa_end - + hostdata->dsa_start) == -1) { + printk("scsi%d : bad dsa virt 0x%p\n", host->host_no, dsa); + return; + } + printk("%sscsi%d : dsa at phys 0x%lx (virt 0x%p)\n" + " + %d : dsa_msgout length = %u, data = 0x%x (virt 0x%p)\n" , + prefix ? prefix : "", + host->host_no, virt_to_bus (dsa), dsa, hostdata->dsa_msgout, + dsa[hostdata->dsa_msgout / sizeof(u32)], + dsa[hostdata->dsa_msgout / sizeof(u32) + 1], + bus_to_virt (dsa[hostdata->dsa_msgout / sizeof(u32) + 1])); + + /* + * Only print messages if they're sane in length so we don't + * blow the kernel printk buffer on something which won't buy us + * anything. + */ + + if (dsa[hostdata->dsa_msgout / sizeof(u32)] < + sizeof (hostdata->free->select)) + for (i = dsa[hostdata->dsa_msgout / sizeof(u32)], + ptr = bus_to_virt (dsa[hostdata->dsa_msgout / sizeof(u32) + 1]); + i > 0 && !check_address ((unsigned long) ptr, 1); + ptr += len, i -= len) { + printk(" "); + len = print_msg (ptr); + printk("\n"); + if (!len) + break; + } + + printk(" + %d : select_indirect = 0x%x\n", + hostdata->dsa_select, dsa[hostdata->dsa_select / sizeof(u32)]); + cmd = (Scsi_Cmnd *) bus_to_virt(dsa[hostdata->dsa_cmnd / sizeof(u32)]); + printk(" + %d : dsa_cmnd = 0x%x ", hostdata->dsa_cmnd, + (u32) virt_to_bus(cmd)); + /* XXX Maybe we should access cmd->host_scribble->result here. RGH */ + if (cmd) { + printk(" result = 0x%x, target = %d, lun = %d, cmd = ", + cmd->result, cmd->device->id, cmd->device->lun); + print_command(cmd->cmnd); + } else + printk("\n"); + printk(" + %d : dsa_next = 0x%x\n", hostdata->dsa_next, + dsa[hostdata->dsa_next / sizeof(u32)]); + if (cmd) { + printk("scsi%d target %d : sxfer_sanity = 0x%x, scntl3_sanity = 0x%x\n" + " script : ", + host->host_no, cmd->device->id, + hostdata->sync[cmd->device->id].sxfer_sanity, + hostdata->sync[cmd->device->id].scntl3_sanity); + for (i = 0; i < (sizeof(hostdata->sync[cmd->device->id].script) / 4); ++i) + printk ("0x%x ", hostdata->sync[cmd->device->id].script[i]); + printk ("\n"); + print_progress (cmd); + } +} +/* + * Function : void print_queues (Scsi_Host *host) + * + * Purpose : print the contents of the NCR issue and reconnect queues + * + * Inputs : host - SCSI host we are interested in + * + */ + +static void +print_queues (struct Scsi_Host *host) { + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + u32 *dsa, *next_dsa; + volatile u32 *ncrcurrent; + int left; + Scsi_Cmnd *cmd, *next_cmd; + unsigned long flags; + + printk ("scsi%d : issue queue\n", host->host_no); + + for (left = host->can_queue, cmd = (Scsi_Cmnd *) hostdata->issue_queue; + left >= 0 && cmd; + cmd = next_cmd) { + next_cmd = (Scsi_Cmnd *) cmd->SCp.ptr; + local_irq_save(flags); + if (cmd->host_scribble) { + if (check_address ((unsigned long) (cmd->host_scribble), + sizeof (cmd->host_scribble)) == -1) + printk ("scsi%d: scsi pid %ld bad pointer to NCR53c7x0_cmd\n", + host->host_no, cmd->pid); + /* print_dsa does sanity check on address, no need to check */ + else + print_dsa (host, ((struct NCR53c7x0_cmd *) cmd->host_scribble) + -> dsa, ""); + } else + printk ("scsi%d : scsi pid %ld for target %d lun %d has no NCR53c7x0_cmd\n", + host->host_no, cmd->pid, cmd->device->id, cmd->device->lun); + local_irq_restore(flags); + } + + if (left <= 0) { + printk ("scsi%d : loop detected in issue queue\n", + host->host_no); + } + + /* + * Traverse the NCR reconnect and start DSA structures, printing out + * each element until we hit the end or detect a loop. Currently, + * the reconnect structure is a linked list; and the start structure + * is an array. Eventually, the reconnect structure will become a + * list as well, since this simplifies the code. + */ + + printk ("scsi%d : schedule dsa array :\n", host->host_no); + for (left = host->can_queue, ncrcurrent = hostdata->schedule; + left > 0; ncrcurrent += 2, --left) + if (ncrcurrent[0] != hostdata->NOP_insn) +/* FIXME : convert pointer to dsa_begin to pointer to dsa. */ + print_dsa (host, bus_to_virt (ncrcurrent[1] - + (hostdata->E_dsa_code_begin - + hostdata->E_dsa_code_template)), ""); + printk ("scsi%d : end schedule dsa array\n", host->host_no); + + printk ("scsi%d : reconnect_dsa_head :\n", host->host_no); + + for (left = host->can_queue, + dsa = bus_to_virt (hostdata->reconnect_dsa_head); + left >= 0 && dsa; + dsa = next_dsa) { + local_irq_save(flags); + if (check_address ((unsigned long) dsa, sizeof(dsa)) == -1) { + printk ("scsi%d: bad DSA pointer 0x%p", host->host_no, + dsa); + next_dsa = NULL; + } + else + { + next_dsa = bus_to_virt(dsa[hostdata->dsa_next / sizeof(u32)]); + print_dsa (host, dsa, ""); + } + local_irq_restore(flags); + } + printk ("scsi%d : end reconnect_dsa_head\n", host->host_no); + if (left < 0) + printk("scsi%d: possible loop in ncr reconnect list\n", + host->host_no); +} + +static void +print_lots (struct Scsi_Host *host) { + NCR53c7x0_local_declare(); + struct NCR53c7x0_hostdata *hostdata = + (struct NCR53c7x0_hostdata *) host->hostdata[0]; + u32 *dsp_next, *dsp, *dsa, dbc_dcmd; + unsigned char dcmd, sbcl; + int i, size; + NCR53c7x0_local_setup(host); + + if ((dsp_next = bus_to_virt(NCR53c7x0_read32 (DSP_REG)))) { + dbc_dcmd = NCR53c7x0_read32(DBC_REG); + dcmd = (dbc_dcmd & 0xff000000) >> 24; + dsp = dsp_next - NCR53c7x0_insn_size(dcmd); + dsa = bus_to_virt(NCR53c7x0_read32(DSA_REG)); + sbcl = NCR53c7x0_read8 (SBCL_REG); + + /* + * For the 53c710, the following will report value 0 for SCNTL3 + * and STEST0 - we don't have these registers. + */ + printk ("scsi%d : DCMD|DBC=0x%x, DNAD=0x%x (virt 0x%p)\n" + " DSA=0x%lx (virt 0x%p)\n" + " DSPS=0x%x, TEMP=0x%x (virt 0x%p), DMODE=0x%x\n" + " SXFER=0x%x, SCNTL3=0x%x\n" + " %s%s%sphase=%s, %d bytes in SCSI FIFO\n" + " SCRATCH=0x%x, saved2_dsa=0x%0lx\n", + host->host_no, dbc_dcmd, NCR53c7x0_read32(DNAD_REG), + bus_to_virt(NCR53c7x0_read32(DNAD_REG)), + virt_to_bus(dsa), dsa, + NCR53c7x0_read32(DSPS_REG), NCR53c7x0_read32(TEMP_REG), + bus_to_virt (NCR53c7x0_read32(TEMP_REG)), + (int) NCR53c7x0_read8(hostdata->dmode), + (int) NCR53c7x0_read8(SXFER_REG), + ((hostdata->chip / 100) == 8) ? + (int) NCR53c7x0_read8(SCNTL3_REG_800) : 0, + (sbcl & SBCL_BSY) ? "BSY " : "", + (sbcl & SBCL_SEL) ? "SEL " : "", + (sbcl & SBCL_REQ) ? "REQ " : "", + sstat2_to_phase(NCR53c7x0_read8 (((hostdata->chip / 100) == 8) ? + SSTAT1_REG : SSTAT2_REG)), + (NCR53c7x0_read8 ((hostdata->chip / 100) == 8 ? + SSTAT1_REG : SSTAT2_REG) & SSTAT2_FF_MASK) >> SSTAT2_FF_SHIFT, + ((hostdata->chip / 100) == 8) ? NCR53c7x0_read8 (STEST0_REG_800) : + NCR53c7x0_read32(SCRATCHA_REG_800), + hostdata->saved2_dsa); + printk ("scsi%d : DSP 0x%lx (virt 0x%p) ->\n", host->host_no, + virt_to_bus(dsp), dsp); + for (i = 6; i > 0; --i, dsp += size) + size = print_insn (host, dsp, "", 1); + if (NCR53c7x0_read8 (SCNTL1_REG) & SCNTL1_CON) { + if ((hostdata->chip / 100) == 8) + printk ("scsi%d : connected (SDID=0x%x, SSID=0x%x)\n", + host->host_no, NCR53c7x0_read8 (SDID_REG_800), + NCR53c7x0_read8 (SSID_REG_800)); + else + printk ("scsi%d : connected (SDID=0x%x)\n", + host->host_no, NCR53c7x0_read8 (SDID_REG_700)); + print_dsa (host, dsa, ""); + } + +#if 1 + print_queues (host); +#endif + } +} + +/* + * Function : static int shutdown (struct Scsi_Host *host) + * + * Purpose : does a clean (we hope) shutdown of the NCR SCSI + * chip. Use prior to dumping core, unloading the NCR driver, + * + * Returns : 0 on success + */ +static int +shutdown (struct Scsi_Host *host) { + NCR53c7x0_local_declare(); + unsigned long flags; + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + NCR53c7x0_local_setup(host); + local_irq_save(flags); +/* Get in a state where we can reset the SCSI bus */ + ncr_halt (host); + ncr_scsi_reset (host); + hostdata->soft_reset(host); + + disable (host); + local_irq_restore(flags); + return 0; +} + +/* + * Function : void ncr_scsi_reset (struct Scsi_Host *host) + * + * Purpose : reset the SCSI bus. + */ + +static void +ncr_scsi_reset (struct Scsi_Host *host) { + NCR53c7x0_local_declare(); + unsigned long flags; + NCR53c7x0_local_setup(host); + local_irq_save(flags); + NCR53c7x0_write8(SCNTL1_REG, SCNTL1_RST); + udelay(25); /* Minimum amount of time to assert RST */ + NCR53c7x0_write8(SCNTL1_REG, 0); + local_irq_restore(flags); +} + +/* + * Function : void hard_reset (struct Scsi_Host *host) + * + */ + +static void +hard_reset (struct Scsi_Host *host) { + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + unsigned long flags; + local_irq_save(flags); + ncr_scsi_reset(host); + NCR53c7x0_driver_init (host); + if (hostdata->soft_reset) + hostdata->soft_reset (host); + local_irq_restore(flags); +} + + +/* + * Function : Scsi_Cmnd *return_outstanding_commands (struct Scsi_Host *host, + * int free, int issue) + * + * Purpose : return a linked list (using the SCp.buffer field as next, + * so we don't perturb hostdata. We don't use a field of the + * NCR53c7x0_cmd structure since we may not have allocated one + * for the command causing the reset.) of Scsi_Cmnd structures that + * had propagated below the Linux issue queue level. If free is set, + * free the NCR53c7x0_cmd structures which are associated with + * the Scsi_Cmnd structures, and clean up any internal + * NCR lists that the commands were on. If issue is set, + * also return commands in the issue queue. + * + * Returns : linked list of commands + * + * NOTE : the caller should insure that the NCR chip is halted + * if the free flag is set. + */ + +static Scsi_Cmnd * +return_outstanding_commands (struct Scsi_Host *host, int free, int issue) { + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + struct NCR53c7x0_cmd *c; + int i; + u32 *ncrcurrent; + Scsi_Cmnd *list = NULL, *tmp; + for (c = (struct NCR53c7x0_cmd *) hostdata->running_list; c; + c = (struct NCR53c7x0_cmd *) c->next) { + if (c->cmd->SCp.buffer) { + printk ("scsi%d : loop detected in running list!\n", host->host_no); + break; + } else { + printk ("Duh? Bad things happening in the NCR driver\n"); + break; + } + + c->cmd->SCp.buffer = (struct scatterlist *) list; + list = c->cmd; + if (free) { + c->next = hostdata->free; + hostdata->free = c; + } + } + + if (free) { + for (i = 0, ncrcurrent = (u32 *) hostdata->schedule; + i < host->can_queue; ++i, ncrcurrent += 2) { + ncrcurrent[0] = hostdata->NOP_insn; + ncrcurrent[1] = 0xdeadbeef; + } + hostdata->ncrcurrent = NULL; + } + + if (issue) { + for (tmp = (Scsi_Cmnd *) hostdata->issue_queue; tmp; tmp = tmp->next) { + if (tmp->SCp.buffer) { + printk ("scsi%d : loop detected in issue queue!\n", + host->host_no); + break; + } + tmp->SCp.buffer = (struct scatterlist *) list; + list = tmp; + } + if (free) + hostdata->issue_queue = NULL; + + } + return list; +} + +/* + * Function : static int disable (struct Scsi_Host *host) + * + * Purpose : disables the given NCR host, causing all commands + * to return a driver error. Call this so we can unload the + * module during development and try again. Eventually, + * we should be able to find clean workarounds for these + * problems. + * + * Inputs : host - hostadapter to twiddle + * + * Returns : 0 on success. + */ + +static int +disable (struct Scsi_Host *host) { + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + unsigned long flags; + Scsi_Cmnd *nuke_list, *tmp; + local_irq_save(flags); + if (hostdata->state != STATE_HALTED) + ncr_halt (host); + nuke_list = return_outstanding_commands (host, 1 /* free */, 1 /* issue */); + hard_reset (host); + hostdata->state = STATE_DISABLED; + local_irq_restore(flags); + printk ("scsi%d : nuking commands\n", host->host_no); + for (; nuke_list; nuke_list = tmp) { + tmp = (Scsi_Cmnd *) nuke_list->SCp.buffer; + nuke_list->result = DID_ERROR << 16; + nuke_list->scsi_done(nuke_list); + } + printk ("scsi%d : done. \n", host->host_no); + printk (KERN_ALERT "scsi%d : disabled. Unload and reload\n", + host->host_no); + return 0; +} + +/* + * Function : static int ncr_halt (struct Scsi_Host *host) + * + * Purpose : halts the SCSI SCRIPTS(tm) processor on the NCR chip + * + * Inputs : host - SCSI chip to halt + * + * Returns : 0 on success + */ + +static int +ncr_halt (struct Scsi_Host *host) { + NCR53c7x0_local_declare(); + unsigned long flags; + unsigned char istat, tmp; + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + int stage; + NCR53c7x0_local_setup(host); + + local_irq_save(flags); + /* Stage 0 : eat all interrupts + Stage 1 : set ABORT + Stage 2 : eat all but abort interrupts + Stage 3 : eat all interrupts + */ + for (stage = 0;;) { + if (stage == 1) { + NCR53c7x0_write8(hostdata->istat, ISTAT_ABRT); + ++stage; + } + istat = NCR53c7x0_read8 (hostdata->istat); + if (istat & ISTAT_SIP) { + tmp = NCR53c7x0_read8(SSTAT0_REG); + } else if (istat & ISTAT_DIP) { + tmp = NCR53c7x0_read8(DSTAT_REG); + if (stage == 2) { + if (tmp & DSTAT_ABRT) { + NCR53c7x0_write8(hostdata->istat, 0); + ++stage; + } else { + printk(KERN_ALERT "scsi%d : could not halt NCR chip\n", + host->host_no); + disable (host); + } + } + } + if (!(istat & (ISTAT_SIP|ISTAT_DIP))) { + if (stage == 0) + ++stage; + else if (stage == 3) + break; + } + } + hostdata->state = STATE_HALTED; + local_irq_restore(flags); +#if 0 + print_lots (host); +#endif + return 0; +} + +/* + * Function: event_name (int event) + * + * Purpose: map event enum into user-readable strings. + */ + +static const char * +event_name (int event) { + switch (event) { + case EVENT_NONE: return "none"; + case EVENT_ISSUE_QUEUE: return "to issue queue"; + case EVENT_START_QUEUE: return "to start queue"; + case EVENT_SELECT: return "selected"; + case EVENT_DISCONNECT: return "disconnected"; + case EVENT_RESELECT: return "reselected"; + case EVENT_COMPLETE: return "completed"; + case EVENT_IDLE: return "idle"; + case EVENT_SELECT_FAILED: return "select failed"; + case EVENT_BEFORE_SELECT: return "before select"; + case EVENT_RESELECT_FAILED: return "reselect failed"; + default: return "unknown"; + } +} + +/* + * Function : void dump_events (struct Scsi_Host *host, count) + * + * Purpose : print last count events which have occurred. + */ +static void +dump_events (struct Scsi_Host *host, int count) { + struct NCR53c7x0_hostdata *hostdata = (struct NCR53c7x0_hostdata *) + host->hostdata[0]; + struct NCR53c7x0_event event; + int i; + unsigned long flags; + if (hostdata->events) { + if (count > hostdata->event_size) + count = hostdata->event_size; + for (i = hostdata->event_index; count > 0; + i = (i ? i - 1 : hostdata->event_size -1), --count) { +/* + * By copying the event we're currently examining with interrupts + * disabled, we can do multiple printk(), etc. operations and + * still be guaranteed that they're happening on the same + * event structure. + */ + local_irq_save(flags); +#if 0 + event = hostdata->events[i]; +#else + memcpy ((void *) &event, (void *) &(hostdata->events[i]), + sizeof(event)); +#endif + + local_irq_restore(flags); + printk ("scsi%d : %s event %d at %ld secs %ld usecs target %d lun %d\n", + host->host_no, event_name (event.event), count, + (long) event.time.tv_sec, (long) event.time.tv_usec, + event.target, event.lun); + if (event.dsa) + printk (" event for dsa 0x%lx (virt 0x%p)\n", + virt_to_bus(event.dsa), event.dsa); + if (event.pid != -1) { + printk (" event for pid %ld ", event.pid); + print_command (event.cmnd); + } + } + } +} + +/* + * Function: check_address + * + * Purpose: Check to see if a possibly corrupt pointer will fault the + * kernel. + * + * Inputs: addr - address; size - size of area + * + * Returns: 0 if area is OK, -1 on error. + * + * NOTES: should be implemented in terms of vverify on kernels + * that have it. + */ + +static int +check_address (unsigned long addr, int size) { + return (virt_to_phys((void *)addr) < PAGE_SIZE || virt_to_phys((void *)(addr + size)) > virt_to_phys(high_memory) ? -1 : 0); +} + +#ifdef MODULE +int +NCR53c7x0_release(struct Scsi_Host *host) { + struct NCR53c7x0_hostdata *hostdata = + (struct NCR53c7x0_hostdata *) host->hostdata[0]; + struct NCR53c7x0_cmd *cmd, *tmp; + shutdown (host); + if (host->irq != SCSI_IRQ_NONE) + { + int irq_count; + struct Scsi_Host *tmp; + for (irq_count = 0, tmp = first_host; tmp; tmp = tmp->next) + if (tmp->hostt == the_template && tmp->irq == host->irq) + ++irq_count; + if (irq_count == 1) + free_irq(host->irq, NULL); + } + if (host->dma_channel != DMA_NONE) + free_dma(host->dma_channel); + if (host->io_port) + release_region(host->io_port, host->n_io_port); + + for (cmd = (struct NCR53c7x0_cmd *) hostdata->free; cmd; cmd = tmp, + --hostdata->num_cmds) { + tmp = (struct NCR53c7x0_cmd *) cmd->next; + /* + * If we're going to loop, try to stop it to get a more accurate + * count of the leaked commands. + */ + cmd->next = NULL; + if (cmd->free) + cmd->free ((void *) cmd->real, cmd->size); + } + if (hostdata->num_cmds) + printk ("scsi%d : leaked %d NCR53c7x0_cmd structures\n", + host->host_no, hostdata->num_cmds); + if (hostdata->events) + vfree ((void *)hostdata->events); + + /* XXX This assumes default cache mode to be IOMAP_FULL_CACHING, which + * XXX may be invalid (CONFIG_060_WRITETHROUGH) + */ + kernel_set_cachemode((void *)hostdata, 8192, IOMAP_FULL_CACHING); + free_pages ((u32)hostdata, 1); + return 1; +} +#endif /* def MODULE */ -- cgit v1.2.3-59-g8ed1b