aboutsummaryrefslogtreecommitdiffstats
path: root/arch/arc/mm
diff options
context:
space:
mode:
Diffstat (limited to 'arch/arc/mm')
-rw-r--r--arch/arc/mm/cache.c112
-rw-r--r--arch/arc/mm/fault.c20
-rw-r--r--arch/arc/mm/init.c5
-rw-r--r--arch/arc/mm/ioremap.c3
-rw-r--r--arch/arc/mm/tlb.c268
-rw-r--r--arch/arc/mm/tlbex.S84
6 files changed, 93 insertions, 399 deletions
diff --git a/arch/arc/mm/cache.c b/arch/arc/mm/cache.c
index a2fbea3ee07c..8aa1231865d1 100644
--- a/arch/arc/mm/cache.c
+++ b/arch/arc/mm/cache.c
@@ -205,93 +205,24 @@ slc_chk:
#define OP_INV_IC 0x4
/*
- * I-Cache Aliasing in ARC700 VIPT caches (MMU v1-v3)
+ * Cache Flush programming model
*
- * ARC VIPT I-cache uses vaddr to index into cache and paddr to match the tag.
- * The orig Cache Management Module "CDU" only required paddr to invalidate a
- * certain line since it sufficed as index in Non-Aliasing VIPT cache-geometry.
- * Infact for distinct V1,V2,P: all of {V1-P},{V2-P},{P-P} would end up fetching
- * the exact same line.
+ * ARC700 MMUv3 I$ and D$ are both VIPT and can potentially alias.
+ * Programming model requires both paddr and vaddr irrespecive of aliasing
+ * considerations:
+ * - vaddr in {I,D}C_IV?L
+ * - paddr in {I,D}C_PTAG
*
- * However for larger Caches (way-size > page-size) - i.e. in Aliasing config,
- * paddr alone could not be used to correctly index the cache.
+ * In HS38x (MMUv4), D$ is PIPT, I$ is VIPT and can still alias.
+ * Programming model is different for aliasing vs. non-aliasing I$
+ * - D$ / Non-aliasing I$: only paddr in {I,D}C_IV?L
+ * - Aliasing I$: same as ARC700 above (so MMUv3 routine used for MMUv4 I$)
*
- * ------------------
- * MMU v1/v2 (Fixed Page Size 8k)
- * ------------------
- * The solution was to provide CDU with these additonal vaddr bits. These
- * would be bits [x:13], x would depend on cache-geometry, 13 comes from
- * standard page size of 8k.
- * H/w folks chose [17:13] to be a future safe range, and moreso these 5 bits
- * of vaddr could easily be "stuffed" in the paddr as bits [4:0] since the
- * orig 5 bits of paddr were anyways ignored by CDU line ops, as they
- * represent the offset within cache-line. The adv of using this "clumsy"
- * interface for additional info was no new reg was needed in CDU programming
- * model.
- *
- * 17:13 represented the max num of bits passable, actual bits needed were
- * fewer, based on the num-of-aliases possible.
- * -for 2 alias possibility, only bit 13 needed (32K cache)
- * -for 4 alias possibility, bits 14:13 needed (64K cache)
- *
- * ------------------
- * MMU v3
- * ------------------
- * This ver of MMU supports variable page sizes (1k-16k): although Linux will
- * only support 8k (default), 16k and 4k.
- * However from hardware perspective, smaller page sizes aggravate aliasing
- * meaning more vaddr bits needed to disambiguate the cache-line-op ;
- * the existing scheme of piggybacking won't work for certain configurations.
- * Two new registers IC_PTAG and DC_PTAG inttoduced.
- * "tag" bits are provided in PTAG, index bits in existing IVIL/IVDL/FLDL regs
+ * - If PAE40 is enabled, independent of aliasing considerations, the higher
+ * bits needs to be written into PTAG_HI
*/
static inline
-void __cache_line_loop_v2(phys_addr_t paddr, unsigned long vaddr,
- unsigned long sz, const int op, const int full_page)
-{
- unsigned int aux_cmd;
- int num_lines;
-
- if (op == OP_INV_IC) {
- aux_cmd = ARC_REG_IC_IVIL;
- } else {
- /* d$ cmd: INV (discard or wback-n-discard) OR FLUSH (wback) */
- aux_cmd = op & OP_INV ? ARC_REG_DC_IVDL : ARC_REG_DC_FLDL;
- }
-
- /* Ensure we properly floor/ceil the non-line aligned/sized requests
- * and have @paddr - aligned to cache line and integral @num_lines.
- * This however can be avoided for page sized since:
- * -@paddr will be cache-line aligned already (being page aligned)
- * -@sz will be integral multiple of line size (being page sized).
- */
- if (!full_page) {
- sz += paddr & ~CACHE_LINE_MASK;
- paddr &= CACHE_LINE_MASK;
- vaddr &= CACHE_LINE_MASK;
- }
-
- num_lines = DIV_ROUND_UP(sz, L1_CACHE_BYTES);
-
- /* MMUv2 and before: paddr contains stuffed vaddrs bits */
- paddr |= (vaddr >> PAGE_SHIFT) & 0x1F;
-
- while (num_lines-- > 0) {
- write_aux_reg(aux_cmd, paddr);
- paddr += L1_CACHE_BYTES;
- }
-}
-
-/*
- * For ARC700 MMUv3 I-cache and D-cache flushes
- * - ARC700 programming model requires paddr and vaddr be passed in seperate
- * AUX registers (*_IV*L and *_PTAG respectively) irrespective of whether the
- * caches actually alias or not.
- * - For HS38, only the aliasing I-cache configuration uses the PTAG reg
- * (non aliasing I-cache version doesn't; while D-cache can't possibly alias)
- */
-static inline
void __cache_line_loop_v3(phys_addr_t paddr, unsigned long vaddr,
unsigned long sz, const int op, const int full_page)
{
@@ -350,17 +281,6 @@ void __cache_line_loop_v3(phys_addr_t paddr, unsigned long vaddr,
#ifndef USE_RGN_FLSH
/*
- * In HS38x (MMU v4), I-cache is VIPT (can alias), D-cache is PIPT
- * Here's how cache ops are implemented
- *
- * - D-cache: only paddr needed (in DC_IVDL/DC_FLDL)
- * - I-cache Non Aliasing: Despite VIPT, only paddr needed (in IC_IVIL)
- * - I-cache Aliasing: Both vaddr and paddr needed (in IC_IVIL, IC_PTAG
- * respectively, similar to MMU v3 programming model, hence
- * __cache_line_loop_v3() is used)
- *
- * If PAE40 is enabled, independent of aliasing considerations, the higher bits
- * needs to be written into PTAG_HI
*/
static inline
void __cache_line_loop_v4(phys_addr_t paddr, unsigned long vaddr,
@@ -460,11 +380,9 @@ void __cache_line_loop_v4(phys_addr_t paddr, unsigned long vaddr,
#endif
-#if (CONFIG_ARC_MMU_VER < 3)
-#define __cache_line_loop __cache_line_loop_v2
-#elif (CONFIG_ARC_MMU_VER == 3)
+#ifdef CONFIG_ARC_MMU_V3
#define __cache_line_loop __cache_line_loop_v3
-#elif (CONFIG_ARC_MMU_VER > 3)
+#else
#define __cache_line_loop __cache_line_loop_v4
#endif
@@ -1123,7 +1041,7 @@ void clear_user_page(void *to, unsigned long u_vaddr, struct page *page)
clear_page(to);
clear_bit(PG_dc_clean, &page->flags);
}
-
+EXPORT_SYMBOL(clear_user_page);
/**********************************************************************
* Explicit Cache flush request from user space via syscall
diff --git a/arch/arc/mm/fault.c b/arch/arc/mm/fault.c
index f5657cb68e4f..5787c261c9a4 100644
--- a/arch/arc/mm/fault.c
+++ b/arch/arc/mm/fault.c
@@ -33,28 +33,34 @@ noinline static int handle_kernel_vaddr_fault(unsigned long address)
pud_t *pud, *pud_k;
pmd_t *pmd, *pmd_k;
- pgd = pgd_offset_fast(current->active_mm, address);
+ pgd = pgd_offset(current->active_mm, address);
pgd_k = pgd_offset_k(address);
- if (!pgd_present(*pgd_k))
+ if (pgd_none (*pgd_k))
goto bad_area;
+ if (!pgd_present(*pgd))
+ set_pgd(pgd, *pgd_k);
p4d = p4d_offset(pgd, address);
p4d_k = p4d_offset(pgd_k, address);
- if (!p4d_present(*p4d_k))
+ if (p4d_none(*p4d_k))
goto bad_area;
+ if (!p4d_present(*p4d))
+ set_p4d(p4d, *p4d_k);
pud = pud_offset(p4d, address);
pud_k = pud_offset(p4d_k, address);
- if (!pud_present(*pud_k))
+ if (pud_none(*pud_k))
goto bad_area;
+ if (!pud_present(*pud))
+ set_pud(pud, *pud_k);
pmd = pmd_offset(pud, address);
pmd_k = pmd_offset(pud_k, address);
- if (!pmd_present(*pmd_k))
+ if (pmd_none(*pmd_k))
goto bad_area;
-
- set_pmd(pmd, *pmd_k);
+ if (!pmd_present(*pmd))
+ set_pmd(pmd, *pmd_k);
/* XXX: create the TLB entry here */
return 0;
diff --git a/arch/arc/mm/init.c b/arch/arc/mm/init.c
index c083bf660cec..699ecf119641 100644
--- a/arch/arc/mm/init.c
+++ b/arch/arc/mm/init.c
@@ -189,6 +189,11 @@ void __init mem_init(void)
{
memblock_free_all();
highmem_init();
+
+ BUILD_BUG_ON((PTRS_PER_PGD * sizeof(pgd_t)) > PAGE_SIZE);
+ BUILD_BUG_ON((PTRS_PER_PUD * sizeof(pud_t)) > PAGE_SIZE);
+ BUILD_BUG_ON((PTRS_PER_PMD * sizeof(pmd_t)) > PAGE_SIZE);
+ BUILD_BUG_ON((PTRS_PER_PTE * sizeof(pte_t)) > PAGE_SIZE);
}
#ifdef CONFIG_HIGHMEM
diff --git a/arch/arc/mm/ioremap.c b/arch/arc/mm/ioremap.c
index 95c649fbc95a..0ee75aca6e10 100644
--- a/arch/arc/mm/ioremap.c
+++ b/arch/arc/mm/ioremap.c
@@ -39,7 +39,8 @@ void __iomem *ioremap(phys_addr_t paddr, unsigned long size)
if (arc_uncached_addr_space(paddr))
return (void __iomem *)(u32)paddr;
- return ioremap_prot(paddr, size, PAGE_KERNEL_NO_CACHE);
+ return ioremap_prot(paddr, size,
+ pgprot_val(pgprot_noncached(PAGE_KERNEL)));
}
EXPORT_SYMBOL(ioremap);
diff --git a/arch/arc/mm/tlb.c b/arch/arc/mm/tlb.c
index 9c7c68247289..5f71445f26bd 100644
--- a/arch/arc/mm/tlb.c
+++ b/arch/arc/mm/tlb.c
@@ -1,51 +1,9 @@
// SPDX-License-Identifier: GPL-2.0-only
/*
- * TLB Management (flush/create/diagnostics) for ARC700
+ * TLB Management (flush/create/diagnostics) for MMUv3 and MMUv4
*
* Copyright (C) 2004, 2007-2010, 2011-2012 Synopsys, Inc. (www.synopsys.com)
*
- * vineetg: Aug 2011
- * -Reintroduce duplicate PD fixup - some customer chips still have the issue
- *
- * vineetg: May 2011
- * -No need to flush_cache_page( ) for each call to update_mmu_cache()
- * some of the LMBench tests improved amazingly
- * = page-fault thrice as fast (75 usec to 28 usec)
- * = mmap twice as fast (9.6 msec to 4.6 msec),
- * = fork (5.3 msec to 3.7 msec)
- *
- * vineetg: April 2011 :
- * -MMU v3: PD{0,1} bits layout changed: They don't overlap anymore,
- * helps avoid a shift when preparing PD0 from PTE
- *
- * vineetg: April 2011 : Preparing for MMU V3
- * -MMU v2/v3 BCRs decoded differently
- * -Remove TLB_SIZE hardcoding as it's variable now: 256 or 512
- * -tlb_entry_erase( ) can be void
- * -local_flush_tlb_range( ):
- * = need not "ceil" @end
- * = walks MMU only if range spans < 32 entries, as opposed to 256
- *
- * Vineetg: Sept 10th 2008
- * -Changes related to MMU v2 (Rel 4.8)
- *
- * Vineetg: Aug 29th 2008
- * -In TLB Flush operations (Metal Fix MMU) there is a explicit command to
- * flush Micro-TLBS. If TLB Index Reg is invalid prior to TLBIVUTLB cmd,
- * it fails. Thus need to load it with ANY valid value before invoking
- * TLBIVUTLB cmd
- *
- * Vineetg: Aug 21th 2008:
- * -Reduced the duration of IRQ lockouts in TLB Flush routines
- * -Multiple copies of TLB erase code separated into a "single" function
- * -In TLB Flush routines, interrupt disabling moved UP to retrieve ASID
- * in interrupt-safe region.
- *
- * Vineetg: April 23rd Bug #93131
- * Problem: tlb_flush_kernel_range() doesn't do anything if the range to
- * flush is more than the size of TLB itself.
- *
- * Rahul Trivedi : Codito Technologies 2004
*/
#include <linux/module.h>
@@ -57,47 +15,6 @@
#include <asm/mmu_context.h>
#include <asm/mmu.h>
-/* Need for ARC MMU v2
- *
- * ARC700 MMU-v1 had a Joint-TLB for Code and Data and is 2 way set-assoc.
- * For a memcpy operation with 3 players (src/dst/code) such that all 3 pages
- * map into same set, there would be contention for the 2 ways causing severe
- * Thrashing.
- *
- * Although J-TLB is 2 way set assoc, ARC700 caches J-TLB into uTLBS which has
- * much higher associativity. u-D-TLB is 8 ways, u-I-TLB is 4 ways.
- * Given this, the thrashing problem should never happen because once the 3
- * J-TLB entries are created (even though 3rd will knock out one of the prev
- * two), the u-D-TLB and u-I-TLB will have what is required to accomplish memcpy
- *
- * Yet we still see the Thrashing because a J-TLB Write cause flush of u-TLBs.
- * This is a simple design for keeping them in sync. So what do we do?
- * The solution which James came up was pretty neat. It utilised the assoc
- * of uTLBs by not invalidating always but only when absolutely necessary.
- *
- * - Existing TLB commands work as before
- * - New command (TLBWriteNI) for TLB write without clearing uTLBs
- * - New command (TLBIVUTLB) to invalidate uTLBs.
- *
- * The uTLBs need only be invalidated when pages are being removed from the
- * OS page table. If a 'victim' TLB entry is being overwritten in the main TLB
- * as a result of a miss, the removed entry is still allowed to exist in the
- * uTLBs as it is still valid and present in the OS page table. This allows the
- * full associativity of the uTLBs to hide the limited associativity of the main
- * TLB.
- *
- * During a miss handler, the new "TLBWriteNI" command is used to load
- * entries without clearing the uTLBs.
- *
- * When the OS page table is updated, TLB entries that may be associated with a
- * removed page are removed (flushed) from the TLB using TLBWrite. In this
- * circumstance, the uTLBs must also be cleared. This is done by using the
- * existing TLBWrite command. An explicit IVUTLB is also required for those
- * corner cases when TLBWrite was not executed at all because the corresp
- * J-TLB entry got evicted/replaced.
- */
-
-
/* A copy of the ASID from the PID reg is kept in asid_cache */
DEFINE_PER_CPU(unsigned int, asid_cache) = MM_CTXT_FIRST_CYCLE;
@@ -120,32 +37,10 @@ static inline void __tlb_entry_erase(void)
static void utlb_invalidate(void)
{
-#if (CONFIG_ARC_MMU_VER >= 2)
-
-#if (CONFIG_ARC_MMU_VER == 2)
- /* MMU v2 introduced the uTLB Flush command.
- * There was however an obscure hardware bug, where uTLB flush would
- * fail when a prior probe for J-TLB (both totally unrelated) would
- * return lkup err - because the entry didn't exist in MMU.
- * The Workaround was to set Index reg with some valid value, prior to
- * flush. This was fixed in MMU v3
- */
- unsigned int idx;
-
- /* make sure INDEX Reg is valid */
- idx = read_aux_reg(ARC_REG_TLBINDEX);
-
- /* If not write some dummy val */
- if (unlikely(idx & TLB_LKUP_ERR))
- write_aux_reg(ARC_REG_TLBINDEX, 0xa);
-#endif
-
write_aux_reg(ARC_REG_TLBCOMMAND, TLBIVUTLB);
-#endif
-
}
-#if (CONFIG_ARC_MMU_VER < 4)
+#ifdef CONFIG_ARC_MMU_V3
static inline unsigned int tlb_entry_lkup(unsigned long vaddr_n_asid)
{
@@ -176,7 +71,7 @@ static void tlb_entry_erase(unsigned int vaddr_n_asid)
}
}
-static void tlb_entry_insert(unsigned int pd0, pte_t pd1)
+static void tlb_entry_insert(unsigned int pd0, phys_addr_t pd1)
{
unsigned int idx;
@@ -206,7 +101,7 @@ static void tlb_entry_insert(unsigned int pd0, pte_t pd1)
write_aux_reg(ARC_REG_TLBCOMMAND, TLBWrite);
}
-#else /* CONFIG_ARC_MMU_VER >= 4) */
+#else /* MMUv4 */
static void tlb_entry_erase(unsigned int vaddr_n_asid)
{
@@ -214,13 +109,16 @@ static void tlb_entry_erase(unsigned int vaddr_n_asid)
write_aux_reg(ARC_REG_TLBCOMMAND, TLBDeleteEntry);
}
-static void tlb_entry_insert(unsigned int pd0, pte_t pd1)
+static void tlb_entry_insert(unsigned int pd0, phys_addr_t pd1)
{
write_aux_reg(ARC_REG_TLBPD0, pd0);
- write_aux_reg(ARC_REG_TLBPD1, pd1);
- if (is_pae40_enabled())
+ if (!is_pae40_enabled()) {
+ write_aux_reg(ARC_REG_TLBPD1, pd1);
+ } else {
+ write_aux_reg(ARC_REG_TLBPD1, pd1 & 0xFFFFFFFF);
write_aux_reg(ARC_REG_TLBPD1HI, (u64)pd1 >> 32);
+ }
write_aux_reg(ARC_REG_TLBCOMMAND, TLBInsertEntry);
}
@@ -496,7 +394,7 @@ void create_tlb(struct vm_area_struct *vma, unsigned long vaddr, pte_t *ptep)
unsigned long flags;
unsigned int asid_or_sasid, rwx;
unsigned long pd0;
- pte_t pd1;
+ phys_addr_t pd1;
/*
* create_tlb() assumes that current->mm == vma->mm, since
@@ -505,7 +403,6 @@ void create_tlb(struct vm_area_struct *vma, unsigned long vaddr, pte_t *ptep)
*
* Removing the assumption involves
* -Using vma->mm->context{ASID,SASID}, as opposed to MMU reg.
- * -Fix the TLB paranoid debug code to not trigger false negatives.
* -More importantly it makes this handler inconsistent with fast-path
* TLB Refill handler which always deals with "current"
*
@@ -528,8 +425,6 @@ void create_tlb(struct vm_area_struct *vma, unsigned long vaddr, pte_t *ptep)
local_irq_save(flags);
- tlb_paranoid_check(asid_mm(vma->vm_mm, smp_processor_id()), vaddr);
-
vaddr &= PAGE_MASK;
/* update this PTE credentials */
@@ -639,43 +534,6 @@ void update_mmu_cache_pmd(struct vm_area_struct *vma, unsigned long addr,
update_mmu_cache(vma, addr, &pte);
}
-void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
- pgtable_t pgtable)
-{
- struct list_head *lh = (struct list_head *) pgtable;
-
- assert_spin_locked(&mm->page_table_lock);
-
- /* FIFO */
- if (!pmd_huge_pte(mm, pmdp))
- INIT_LIST_HEAD(lh);
- else
- list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
- pmd_huge_pte(mm, pmdp) = pgtable;
-}
-
-pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
-{
- struct list_head *lh;
- pgtable_t pgtable;
-
- assert_spin_locked(&mm->page_table_lock);
-
- pgtable = pmd_huge_pte(mm, pmdp);
- lh = (struct list_head *) pgtable;
- if (list_empty(lh))
- pmd_huge_pte(mm, pmdp) = NULL;
- else {
- pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
- list_del(lh);
- }
-
- pte_val(pgtable[0]) = 0;
- pte_val(pgtable[1]) = 0;
-
- return pgtable;
-}
-
void local_flush_pmd_tlb_range(struct vm_area_struct *vma, unsigned long start,
unsigned long end)
{
@@ -706,14 +564,6 @@ void read_decode_mmu_bcr(void)
{
struct cpuinfo_arc_mmu *mmu = &cpuinfo_arc700[smp_processor_id()].mmu;
unsigned int tmp;
- struct bcr_mmu_1_2 {
-#ifdef CONFIG_CPU_BIG_ENDIAN
- unsigned int ver:8, ways:4, sets:4, u_itlb:8, u_dtlb:8;
-#else
- unsigned int u_dtlb:8, u_itlb:8, sets:4, ways:4, ver:8;
-#endif
- } *mmu2;
-
struct bcr_mmu_3 {
#ifdef CONFIG_CPU_BIG_ENDIAN
unsigned int ver:8, ways:4, sets:4, res:3, sasid:1, pg_sz:4,
@@ -738,23 +588,14 @@ void read_decode_mmu_bcr(void)
tmp = read_aux_reg(ARC_REG_MMU_BCR);
mmu->ver = (tmp >> 24);
- if (is_isa_arcompact()) {
- if (mmu->ver <= 2) {
- mmu2 = (struct bcr_mmu_1_2 *)&tmp;
- mmu->pg_sz_k = TO_KB(0x2000);
- mmu->sets = 1 << mmu2->sets;
- mmu->ways = 1 << mmu2->ways;
- mmu->u_dtlb = mmu2->u_dtlb;
- mmu->u_itlb = mmu2->u_itlb;
- } else {
- mmu3 = (struct bcr_mmu_3 *)&tmp;
- mmu->pg_sz_k = 1 << (mmu3->pg_sz - 1);
- mmu->sets = 1 << mmu3->sets;
- mmu->ways = 1 << mmu3->ways;
- mmu->u_dtlb = mmu3->u_dtlb;
- mmu->u_itlb = mmu3->u_itlb;
- mmu->sasid = mmu3->sasid;
- }
+ if (is_isa_arcompact() && mmu->ver == 3) {
+ mmu3 = (struct bcr_mmu_3 *)&tmp;
+ mmu->pg_sz_k = 1 << (mmu3->pg_sz - 1);
+ mmu->sets = 1 << mmu3->sets;
+ mmu->ways = 1 << mmu3->ways;
+ mmu->u_dtlb = mmu3->u_dtlb;
+ mmu->u_itlb = mmu3->u_itlb;
+ mmu->sasid = mmu3->sasid;
} else {
mmu4 = (struct bcr_mmu_4 *)&tmp;
mmu->pg_sz_k = 1 << (mmu4->sz0 - 1);
@@ -780,8 +621,8 @@ char *arc_mmu_mumbojumbo(int cpu_id, char *buf, int len)
IS_USED_CFG(CONFIG_TRANSPARENT_HUGEPAGE));
n += scnprintf(buf + n, len - n,
- "MMU [v%x]\t: %dk PAGE, %sJTLB %d (%dx%d), uDTLB %d, uITLB %d%s%s\n",
- p_mmu->ver, p_mmu->pg_sz_k, super_pg,
+ "MMU [v%x]\t: %dk PAGE, %s, swalk %d lvl, JTLB %d (%dx%d), uDTLB %d, uITLB %d%s%s\n",
+ p_mmu->ver, p_mmu->pg_sz_k, super_pg, CONFIG_PGTABLE_LEVELS,
p_mmu->sets * p_mmu->ways, p_mmu->sets, p_mmu->ways,
p_mmu->u_dtlb, p_mmu->u_itlb,
IS_AVAIL2(p_mmu->pae, ", PAE40 ", CONFIG_ARC_HAS_PAE40));
@@ -815,22 +656,17 @@ void arc_mmu_init(void)
/*
* Ensure that MMU features assumed by kernel exist in hardware.
- * For older ARC700 cpus, it has to be exact match, since the MMU
- * revisions were not backwards compatible (MMUv3 TLB layout changed
- * so even if kernel for v2 didn't use any new cmds of v3, it would
- * still not work.
- * For HS cpus, MMUv4 was baseline and v5 is backwards compatible
- * (will run older software).
+ * - For older ARC700 cpus, only v3 supported
+ * - For HS cpus, v4 was baseline and v5 is backwards compatible
+ * (will run older software).
*/
- if (is_isa_arcompact() && mmu->ver == CONFIG_ARC_MMU_VER)
+ if (is_isa_arcompact() && mmu->ver == 3)
compat = 1;
- else if (is_isa_arcv2() && mmu->ver >= CONFIG_ARC_MMU_VER)
+ else if (is_isa_arcv2() && mmu->ver >= 4)
compat = 1;
- if (!compat) {
- panic("MMU ver %d doesn't match kernel built for %d...\n",
- mmu->ver, CONFIG_ARC_MMU_VER);
- }
+ if (!compat)
+ panic("MMU ver %d doesn't match kernel built for\n", mmu->ver);
if (mmu->pg_sz_k != TO_KB(PAGE_SIZE))
panic("MMU pg size != PAGE_SIZE (%luk)\n", TO_KB(PAGE_SIZE));
@@ -843,14 +679,11 @@ void arc_mmu_init(void)
if (IS_ENABLED(CONFIG_ARC_HAS_PAE40) && !mmu->pae)
panic("Hardware doesn't support PAE40\n");
- /* Enable the MMU */
- write_aux_reg(ARC_REG_PID, MMU_ENABLE);
+ /* Enable the MMU with ASID 0 */
+ mmu_setup_asid(NULL, 0);
- /* In smp we use this reg for interrupt 1 scratch */
-#ifdef ARC_USE_SCRATCH_REG
- /* swapper_pg_dir is the pgd for the kernel, used by vmalloc */
- write_aux_reg(ARC_REG_SCRATCH_DATA0, swapper_pg_dir);
-#endif
+ /* cache the pgd pointer in MMU SCRATCH reg (ARCv2 only) */
+ mmu_setup_pgd(NULL, swapper_pg_dir);
if (pae40_exist_but_not_enab())
write_aux_reg(ARC_REG_TLBPD1HI, 0);
@@ -945,40 +778,3 @@ void do_tlb_overlap_fault(unsigned long cause, unsigned long address,
local_irq_restore(flags);
}
-
-/***********************************************************************
- * Diagnostic Routines
- * -Called from Low Level TLB Handlers if things don;t look good
- **********************************************************************/
-
-#ifdef CONFIG_ARC_DBG_TLB_PARANOIA
-
-/*
- * Low Level ASM TLB handler calls this if it finds that HW and SW ASIDS
- * don't match
- */
-void print_asid_mismatch(int mm_asid, int mmu_asid, int is_fast_path)
-{
- pr_emerg("ASID Mismatch in %s Path Handler: sw-pid=0x%x hw-pid=0x%x\n",
- is_fast_path ? "Fast" : "Slow", mm_asid, mmu_asid);
-
- __asm__ __volatile__("flag 1");
-}
-
-void tlb_paranoid_check(unsigned int mm_asid, unsigned long addr)
-{
- unsigned int mmu_asid;
-
- mmu_asid = read_aux_reg(ARC_REG_PID) & 0xff;
-
- /*
- * At the time of a TLB miss/installation
- * - HW version needs to match SW version
- * - SW needs to have a valid ASID
- */
- if (addr < 0x70000000 &&
- ((mm_asid == MM_CTXT_NO_ASID) ||
- (mmu_asid != (mm_asid & MM_CTXT_ASID_MASK))))
- print_asid_mismatch(mm_asid, mmu_asid, 0);
-}
-#endif
diff --git a/arch/arc/mm/tlbex.S b/arch/arc/mm/tlbex.S
index 062fae46c3f8..e054780a8fe0 100644
--- a/arch/arc/mm/tlbex.S
+++ b/arch/arc/mm/tlbex.S
@@ -39,7 +39,6 @@
#include <asm/arcregs.h>
#include <asm/cache.h>
#include <asm/processor.h>
-#include <asm/tlb-mmu1.h>
#ifdef CONFIG_ISA_ARCOMPACT
;-----------------------------------------------------------------
@@ -94,11 +93,6 @@ ex_saved_reg1:
st_s r1, [r0, 4]
st_s r2, [r0, 8]
st_s r3, [r0, 12]
-
- ; VERIFY if the ASID in MMU-PID Reg is same as
- ; one in Linux data structures
-
- tlb_paranoid_check_asm
.endm
.macro TLBMISS_RESTORE_REGS
@@ -148,53 +142,16 @@ ex_saved_reg1:
#endif
;============================================================================
-; Troubleshooting Stuff
+;TLB Miss handling Code
;============================================================================
-; Linux keeps ASID (Address Space ID) in task->active_mm->context.asid
-; When Creating TLB Entries, instead of doing 3 dependent loads from memory,
-; we use the MMU PID Reg to get current ASID.
-; In bizzare scenrios SW and HW ASID can get out-of-sync which is trouble.
-; So we try to detect this in TLB Mis shandler
-
-.macro tlb_paranoid_check_asm
-
-#ifdef CONFIG_ARC_DBG_TLB_PARANOIA
-
- GET_CURR_TASK_ON_CPU r3
- ld r0, [r3, TASK_ACT_MM]
- ld r0, [r0, MM_CTXT+MM_CTXT_ASID]
- breq r0, 0, 55f ; Error if no ASID allocated
-
- lr r1, [ARC_REG_PID]
- and r1, r1, 0xFF
-
- and r2, r0, 0xFF ; MMU PID bits only for comparison
- breq r1, r2, 5f
-
-55:
- ; Error if H/w and S/w ASID don't match, but NOT if in kernel mode
- lr r2, [erstatus]
- bbit0 r2, STATUS_U_BIT, 5f
-
- ; We sure are in troubled waters, Flag the error, but to do so
- ; need to switch to kernel mode stack to call error routine
- GET_TSK_STACK_BASE r3, sp
-
- ; Call printk to shoutout aloud
- mov r2, 1
- j print_asid_mismatch
-
-5: ; ASIDs match so proceed normally
- nop
-
+#ifndef PMD_SHIFT
+#define PMD_SHIFT PUD_SHIFT
#endif
-.endm
-
-;============================================================================
-;TLB Miss handling Code
-;============================================================================
+#ifndef PUD_SHIFT
+#define PUD_SHIFT PGDIR_SHIFT
+#endif
;-----------------------------------------------------------------------------
; This macro does the page-table lookup for the faulting address.
@@ -203,7 +160,7 @@ ex_saved_reg1:
lr r2, [efa]
-#ifdef ARC_USE_SCRATCH_REG
+#ifdef CONFIG_ISA_ARCV2
lr r1, [ARC_REG_SCRATCH_DATA0] ; current pgd
#else
GET_CURR_TASK_ON_CPU r1
@@ -216,6 +173,24 @@ ex_saved_reg1:
tst r3, r3
bz do_slow_path_pf ; if no Page Table, do page fault
+#if CONFIG_PGTABLE_LEVELS > 3
+ lsr r0, r2, PUD_SHIFT ; Bits for indexing into PUD
+ and r0, r0, (PTRS_PER_PUD - 1)
+ ld.as r1, [r3, r0] ; PMD entry
+ tst r1, r1
+ bz do_slow_path_pf
+ mov r3, r1
+#endif
+
+#if CONFIG_PGTABLE_LEVELS > 2
+ lsr r0, r2, PMD_SHIFT ; Bits for indexing into PMD
+ and r0, r0, (PTRS_PER_PMD - 1)
+ ld.as r1, [r3, r0] ; PMD entry
+ tst r1, r1
+ bz do_slow_path_pf
+ mov r3, r1
+#endif
+
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
and.f 0, r3, _PAGE_HW_SZ ; Is this Huge PMD (thp)
add2.nz r1, r1, r0
@@ -279,7 +254,7 @@ ex_saved_reg1:
; Commit the TLB entry into MMU
.macro COMMIT_ENTRY_TO_MMU
-#if (CONFIG_ARC_MMU_VER < 4)
+#ifdef CONFIG_ARC_MMU_V3
/* Get free TLB slot: Set = computed from vaddr, way = random */
sr TLBGetIndex, [ARC_REG_TLBCOMMAND]
@@ -375,13 +350,6 @@ ENTRY(EV_TLBMissD)
CONV_PTE_TO_TLB
-#if (CONFIG_ARC_MMU_VER == 1)
- ; MMU with 2 way set assoc J-TLB, needs some help in pathetic case of
- ; memcpy where 3 parties contend for 2 ways, ensuing a livelock.
- ; But only for old MMU or one with Metal Fix
- TLB_WRITE_HEURISTICS
-#endif
-
COMMIT_ENTRY_TO_MMU
TLBMISS_RESTORE_REGS
EV_TLBMissD_fast_ret: ; additional label for VDK OS-kit instrumentation
Copyright © 1996 – 2023 Jason A. Donenfeld. All Rights Reverse Engineered.