From af512ed0f8a7e6a3c6fd93b2b5882c8e837a6939 Mon Sep 17 00:00:00 2001 From: Ralph Wuerthner Date: Tue, 10 Jul 2007 11:24:19 +0200 Subject: [S390] zcrypt: fix request timeout handling Under very high load zcrypt requests may timeout while waiting on the request queue. Modify zcrypt that timeouts are based on crypto adapter responses. A timeout occurs only if a crypto adapter does not respond within a given time frame to sumitted requests. Signed-off-by: Ralph Wuerthner Signed-off-by: Martin Schwidefsky --- drivers/s390/crypto/ap_bus.c | 98 +++++++++++++++++++++++++++++++++---- drivers/s390/crypto/ap_bus.h | 11 +++++ drivers/s390/crypto/zcrypt_cex2a.c | 27 ++++------ drivers/s390/crypto/zcrypt_pcica.c | 27 ++++------ drivers/s390/crypto/zcrypt_pcicc.c | 27 ++++------ drivers/s390/crypto/zcrypt_pcixcc.c | 40 +++++---------- 6 files changed, 140 insertions(+), 90 deletions(-) (limited to 'drivers/s390') diff --git a/drivers/s390/crypto/ap_bus.c b/drivers/s390/crypto/ap_bus.c index 5aac0ec36368..90bd22014513 100644 --- a/drivers/s390/crypto/ap_bus.c +++ b/drivers/s390/crypto/ap_bus.c @@ -43,6 +43,7 @@ static void ap_poll_all(unsigned long); static void ap_poll_timeout(unsigned long); static int ap_poll_thread_start(void); static void ap_poll_thread_stop(void); +static void ap_request_timeout(unsigned long); /** * Module description. @@ -189,6 +190,7 @@ int ap_send(ap_qid_t qid, unsigned long long psmid, void *msg, size_t length) case AP_RESPONSE_NORMAL: return 0; case AP_RESPONSE_Q_FULL: + case AP_RESPONSE_RESET_IN_PROGRESS: return -EBUSY; default: /* Device is gone. */ return -ENODEV; @@ -252,6 +254,8 @@ int ap_recv(ap_qid_t qid, unsigned long long *psmid, void *msg, size_t length) if (status.queue_empty) return -ENOENT; return -EBUSY; + case AP_RESPONSE_RESET_IN_PROGRESS: + return -EBUSY; default: return -ENODEV; } @@ -326,11 +330,12 @@ static int ap_init_queue(ap_qid_t qid) i = AP_MAX_RESET; /* return with -ENODEV */ break; case AP_RESPONSE_RESET_IN_PROGRESS: + rc = -EBUSY; case AP_RESPONSE_BUSY: default: break; } - if (rc != -ENODEV) + if (rc != -ENODEV && rc != -EBUSY) break; if (i < AP_MAX_RESET - 1) { udelay(5); @@ -340,6 +345,40 @@ static int ap_init_queue(ap_qid_t qid) return rc; } +/** + * Arm request timeout if a AP device was idle and a new request is submitted. + */ +static void ap_increase_queue_count(struct ap_device *ap_dev) +{ + int timeout = ap_dev->drv->request_timeout; + + ap_dev->queue_count++; + if (ap_dev->queue_count == 1) { + mod_timer(&ap_dev->timeout, jiffies + timeout); + ap_dev->reset = AP_RESET_ARMED; + } +} + +/** + * AP device is still alive, re-schedule request timeout if there are still + * pending requests. + */ +static void ap_decrease_queue_count(struct ap_device *ap_dev) +{ + int timeout = ap_dev->drv->request_timeout; + + ap_dev->queue_count--; + if (ap_dev->queue_count > 0) + mod_timer(&ap_dev->timeout, jiffies + timeout); + else + /** + * The timeout timer should to be disabled now - since + * del_timer_sync() is very expensive, we just tell via the + * reset flag to ignore the pending timeout timer. + */ + ap_dev->reset = AP_RESET_IGNORE; +} + /** * AP device related attributes. */ @@ -498,6 +537,7 @@ static int ap_device_remove(struct device *dev) struct ap_driver *ap_drv = ap_dev->drv; ap_flush_queue(ap_dev); + del_timer_sync(&ap_dev->timeout); if (ap_drv->remove) ap_drv->remove(ap_dev); spin_lock_bh(&ap_device_lock); @@ -759,17 +799,21 @@ static void ap_scan_bus(struct work_struct *unused) __ap_scan_bus); rc = ap_query_queue(qid, &queue_depth, &device_type); if (dev) { + if (rc == -EBUSY) { + set_current_state(TASK_UNINTERRUPTIBLE); + schedule_timeout(AP_RESET_TIMEOUT); + rc = ap_query_queue(qid, &queue_depth, + &device_type); + } ap_dev = to_ap_dev(dev); spin_lock_bh(&ap_dev->lock); if (rc || ap_dev->unregistered) { spin_unlock_bh(&ap_dev->lock); - put_device(dev); device_unregister(dev); + put_device(dev); continue; - } else - spin_unlock_bh(&ap_dev->lock); - } - if (dev) { + } + spin_unlock_bh(&ap_dev->lock); put_device(dev); continue; } @@ -788,6 +832,8 @@ static void ap_scan_bus(struct work_struct *unused) INIT_LIST_HEAD(&ap_dev->pendingq); INIT_LIST_HEAD(&ap_dev->requestq); INIT_LIST_HEAD(&ap_dev->list); + setup_timer(&ap_dev->timeout, ap_request_timeout, + (unsigned long) ap_dev); if (device_type == 0) ap_probe_device_type(ap_dev); else @@ -853,7 +899,7 @@ static int ap_poll_read(struct ap_device *ap_dev, unsigned long *flags) switch (status.response_code) { case AP_RESPONSE_NORMAL: atomic_dec(&ap_poll_requests); - ap_dev->queue_count--; + ap_decrease_queue_count(ap_dev); list_for_each_entry(ap_msg, &ap_dev->pendingq, list) { if (ap_msg->psmid != ap_dev->reply->psmid) continue; @@ -904,7 +950,7 @@ static int ap_poll_write(struct ap_device *ap_dev, unsigned long *flags) switch (status.response_code) { case AP_RESPONSE_NORMAL: atomic_inc(&ap_poll_requests); - ap_dev->queue_count++; + ap_increase_queue_count(ap_dev); list_move_tail(&ap_msg->list, &ap_dev->pendingq); ap_dev->requestq_count--; ap_dev->pendingq_count++; @@ -914,6 +960,7 @@ static int ap_poll_write(struct ap_device *ap_dev, unsigned long *flags) *flags |= 2; break; case AP_RESPONSE_Q_FULL: + case AP_RESPONSE_RESET_IN_PROGRESS: *flags |= 2; break; case AP_RESPONSE_MESSAGE_TOO_BIG: @@ -960,10 +1007,11 @@ static int __ap_queue_message(struct ap_device *ap_dev, struct ap_message *ap_ms list_add_tail(&ap_msg->list, &ap_dev->pendingq); atomic_inc(&ap_poll_requests); ap_dev->pendingq_count++; - ap_dev->queue_count++; + ap_increase_queue_count(ap_dev); ap_dev->total_request_count++; break; case AP_RESPONSE_Q_FULL: + case AP_RESPONSE_RESET_IN_PROGRESS: list_add_tail(&ap_msg->list, &ap_dev->requestq); ap_dev->requestq_count++; ap_dev->total_request_count++; @@ -1045,6 +1093,25 @@ static void ap_poll_timeout(unsigned long unused) tasklet_schedule(&ap_tasklet); } +/** + * Reset a not responding AP device and move all requests from the + * pending queue to the request queue. + */ +static void ap_reset(struct ap_device *ap_dev) +{ + int rc; + + ap_dev->reset = AP_RESET_IGNORE; + atomic_sub(ap_dev->queue_count, &ap_poll_requests); + ap_dev->queue_count = 0; + list_splice_init(&ap_dev->pendingq, &ap_dev->requestq); + ap_dev->requestq_count += ap_dev->pendingq_count; + ap_dev->pendingq_count = 0; + rc = ap_init_queue(ap_dev->qid); + if (rc == -ENODEV) + ap_dev->unregistered = 1; +} + /** * Poll all AP devices on the bus in a round robin fashion. Continue * polling until bit 2^0 of the control flags is not set. If bit 2^1 @@ -1056,6 +1123,8 @@ static int __ap_poll_all(struct ap_device *ap_dev, unsigned long *flags) if (!ap_dev->unregistered) { if (ap_poll_queue(ap_dev, flags)) ap_dev->unregistered = 1; + if (ap_dev->reset == AP_RESET_DO) + ap_reset(ap_dev); } spin_unlock(&ap_dev->lock); return 0; @@ -1147,6 +1216,17 @@ static void ap_poll_thread_stop(void) mutex_unlock(&ap_poll_thread_mutex); } +/** + * Handling of request timeouts + */ +static void ap_request_timeout(unsigned long data) +{ + struct ap_device *ap_dev = (struct ap_device *) data; + + if (ap_dev->reset == AP_RESET_ARMED) + ap_dev->reset = AP_RESET_DO; +} + static void ap_reset_domain(void) { int i; diff --git a/drivers/s390/crypto/ap_bus.h b/drivers/s390/crypto/ap_bus.h index 008559ea742b..87c2d6442875 100644 --- a/drivers/s390/crypto/ap_bus.h +++ b/drivers/s390/crypto/ap_bus.h @@ -33,6 +33,7 @@ #define AP_DEVICES 64 /* Number of AP devices. */ #define AP_DOMAINS 16 /* Number of AP domains. */ #define AP_MAX_RESET 90 /* Maximum number of resets. */ +#define AP_RESET_TIMEOUT (HZ/2) /* Time in ticks for reset timeouts. */ #define AP_CONFIG_TIME 30 /* Time in seconds between AP bus rescans. */ #define AP_POLL_TIME 1 /* Time in ticks between receive polls. */ @@ -83,6 +84,13 @@ struct ap_queue_status { #define AP_DEVICE_TYPE_CEX2A 6 #define AP_DEVICE_TYPE_CEX2C 7 +/** + * AP reset flag states + */ +#define AP_RESET_IGNORE 0 /* request timeout will be ignored */ +#define AP_RESET_ARMED 1 /* request timeout timer is active */ +#define AP_RESET_DO 2 /* AP reset required */ + struct ap_device; struct ap_message; @@ -95,6 +103,7 @@ struct ap_driver { /* receive is called from tasklet context */ void (*receive)(struct ap_device *, struct ap_message *, struct ap_message *); + int request_timeout; /* request timeout in jiffies */ }; #define to_ap_drv(x) container_of((x), struct ap_driver, driver) @@ -112,6 +121,8 @@ struct ap_device { int queue_depth; /* AP queue depth.*/ int device_type; /* AP device type. */ int unregistered; /* marks AP device as unregistered */ + struct timer_list timeout; /* Timer for request timeouts. */ + int reset; /* Reset required after req. timeout. */ int queue_count; /* # messages currently on AP queue. */ diff --git a/drivers/s390/crypto/zcrypt_cex2a.c b/drivers/s390/crypto/zcrypt_cex2a.c index 5bb13a9d0898..08657f604b8c 100644 --- a/drivers/s390/crypto/zcrypt_cex2a.c +++ b/drivers/s390/crypto/zcrypt_cex2a.c @@ -70,6 +70,7 @@ static struct ap_driver zcrypt_cex2a_driver = { .remove = zcrypt_cex2a_remove, .receive = zcrypt_cex2a_receive, .ids = zcrypt_cex2a_ids, + .request_timeout = CEX2A_CLEANUP_TIME, }; /** @@ -306,18 +307,13 @@ static long zcrypt_cex2a_modexpo(struct zcrypt_device *zdev, goto out_free; init_completion(&work); ap_queue_message(zdev->ap_dev, &ap_msg); - rc = wait_for_completion_interruptible_timeout( - &work, CEX2A_CLEANUP_TIME); - if (rc > 0) + rc = wait_for_completion_interruptible(&work); + if (rc == 0) rc = convert_response(zdev, &ap_msg, mex->outputdata, mex->outputdatalength); - else { - /* Signal pending or message timed out. */ + else + /* Signal pending. */ ap_cancel_message(zdev->ap_dev, &ap_msg); - if (rc == 0) - /* Message timed out. */ - rc = -ETIME; - } out_free: kfree(ap_msg.message); return rc; @@ -348,18 +344,13 @@ static long zcrypt_cex2a_modexpo_crt(struct zcrypt_device *zdev, goto out_free; init_completion(&work); ap_queue_message(zdev->ap_dev, &ap_msg); - rc = wait_for_completion_interruptible_timeout( - &work, CEX2A_CLEANUP_TIME); - if (rc > 0) + rc = wait_for_completion_interruptible(&work); + if (rc == 0) rc = convert_response(zdev, &ap_msg, crt->outputdata, crt->outputdatalength); - else { - /* Signal pending or message timed out. */ + else + /* Signal pending. */ ap_cancel_message(zdev->ap_dev, &ap_msg); - if (rc == 0) - /* Message timed out. */ - rc = -ETIME; - } out_free: kfree(ap_msg.message); return rc; diff --git a/drivers/s390/crypto/zcrypt_pcica.c b/drivers/s390/crypto/zcrypt_pcica.c index 818ffe05ac00..6e93b4751782 100644 --- a/drivers/s390/crypto/zcrypt_pcica.c +++ b/drivers/s390/crypto/zcrypt_pcica.c @@ -70,6 +70,7 @@ static struct ap_driver zcrypt_pcica_driver = { .remove = zcrypt_pcica_remove, .receive = zcrypt_pcica_receive, .ids = zcrypt_pcica_ids, + .request_timeout = PCICA_CLEANUP_TIME, }; /** @@ -290,18 +291,13 @@ static long zcrypt_pcica_modexpo(struct zcrypt_device *zdev, goto out_free; init_completion(&work); ap_queue_message(zdev->ap_dev, &ap_msg); - rc = wait_for_completion_interruptible_timeout( - &work, PCICA_CLEANUP_TIME); - if (rc > 0) + rc = wait_for_completion_interruptible(&work); + if (rc == 0) rc = convert_response(zdev, &ap_msg, mex->outputdata, mex->outputdatalength); - else { - /* Signal pending or message timed out. */ + else + /* Signal pending. */ ap_cancel_message(zdev->ap_dev, &ap_msg); - if (rc == 0) - /* Message timed out. */ - rc = -ETIME; - } out_free: kfree(ap_msg.message); return rc; @@ -332,18 +328,13 @@ static long zcrypt_pcica_modexpo_crt(struct zcrypt_device *zdev, goto out_free; init_completion(&work); ap_queue_message(zdev->ap_dev, &ap_msg); - rc = wait_for_completion_interruptible_timeout( - &work, PCICA_CLEANUP_TIME); - if (rc > 0) + rc = wait_for_completion_interruptible(&work); + if (rc == 0) rc = convert_response(zdev, &ap_msg, crt->outputdata, crt->outputdatalength); - else { - /* Signal pending or message timed out. */ + else + /* Signal pending. */ ap_cancel_message(zdev->ap_dev, &ap_msg); - if (rc == 0) - /* Message timed out. */ - rc = -ETIME; - } out_free: kfree(ap_msg.message); return rc; diff --git a/drivers/s390/crypto/zcrypt_pcicc.c b/drivers/s390/crypto/zcrypt_pcicc.c index f295a403b29a..d6d59bf9ac38 100644 --- a/drivers/s390/crypto/zcrypt_pcicc.c +++ b/drivers/s390/crypto/zcrypt_pcicc.c @@ -82,6 +82,7 @@ static struct ap_driver zcrypt_pcicc_driver = { .remove = zcrypt_pcicc_remove, .receive = zcrypt_pcicc_receive, .ids = zcrypt_pcicc_ids, + .request_timeout = PCICC_CLEANUP_TIME, }; /** @@ -501,18 +502,13 @@ static long zcrypt_pcicc_modexpo(struct zcrypt_device *zdev, goto out_free; init_completion(&work); ap_queue_message(zdev->ap_dev, &ap_msg); - rc = wait_for_completion_interruptible_timeout( - &work, PCICC_CLEANUP_TIME); - if (rc > 0) + rc = wait_for_completion_interruptible(&work); + if (rc == 0) rc = convert_response(zdev, &ap_msg, mex->outputdata, mex->outputdatalength); - else { - /* Signal pending or message timed out. */ + else + /* Signal pending. */ ap_cancel_message(zdev->ap_dev, &ap_msg); - if (rc == 0) - /* Message timed out. */ - rc = -ETIME; - } out_free: free_page((unsigned long) ap_msg.message); return rc; @@ -544,18 +540,13 @@ static long zcrypt_pcicc_modexpo_crt(struct zcrypt_device *zdev, goto out_free; init_completion(&work); ap_queue_message(zdev->ap_dev, &ap_msg); - rc = wait_for_completion_interruptible_timeout( - &work, PCICC_CLEANUP_TIME); - if (rc > 0) + rc = wait_for_completion_interruptible(&work); + if (rc == 0) rc = convert_response(zdev, &ap_msg, crt->outputdata, crt->outputdatalength); - else { - /* Signal pending or message timed out. */ + else + /* Signal pending. */ ap_cancel_message(zdev->ap_dev, &ap_msg); - if (rc == 0) - /* Message timed out. */ - rc = -ETIME; - } out_free: free_page((unsigned long) ap_msg.message); return rc; diff --git a/drivers/s390/crypto/zcrypt_pcixcc.c b/drivers/s390/crypto/zcrypt_pcixcc.c index 252443b6bd1b..64948788d301 100644 --- a/drivers/s390/crypto/zcrypt_pcixcc.c +++ b/drivers/s390/crypto/zcrypt_pcixcc.c @@ -93,6 +93,7 @@ static struct ap_driver zcrypt_pcixcc_driver = { .remove = zcrypt_pcixcc_remove, .receive = zcrypt_pcixcc_receive, .ids = zcrypt_pcixcc_ids, + .request_timeout = PCIXCC_CLEANUP_TIME, }; /** @@ -641,18 +642,13 @@ static long zcrypt_pcixcc_modexpo(struct zcrypt_device *zdev, goto out_free; init_completion(&resp_type.work); ap_queue_message(zdev->ap_dev, &ap_msg); - rc = wait_for_completion_interruptible_timeout( - &resp_type.work, PCIXCC_CLEANUP_TIME); - if (rc > 0) + rc = wait_for_completion_interruptible(&resp_type.work); + if (rc == 0) rc = convert_response_ica(zdev, &ap_msg, mex->outputdata, mex->outputdatalength); - else { - /* Signal pending or message timed out. */ + else + /* Signal pending. */ ap_cancel_message(zdev->ap_dev, &ap_msg); - if (rc == 0) - /* Message timed out. */ - rc = -ETIME; - } out_free: free_page((unsigned long) ap_msg.message); return rc; @@ -685,18 +681,13 @@ static long zcrypt_pcixcc_modexpo_crt(struct zcrypt_device *zdev, goto out_free; init_completion(&resp_type.work); ap_queue_message(zdev->ap_dev, &ap_msg); - rc = wait_for_completion_interruptible_timeout( - &resp_type.work, PCIXCC_CLEANUP_TIME); - if (rc > 0) + rc = wait_for_completion_interruptible(&resp_type.work); + if (rc == 0) rc = convert_response_ica(zdev, &ap_msg, crt->outputdata, crt->outputdatalength); - else { - /* Signal pending or message timed out. */ + else + /* Signal pending. */ ap_cancel_message(zdev->ap_dev, &ap_msg); - if (rc == 0) - /* Message timed out. */ - rc = -ETIME; - } out_free: free_page((unsigned long) ap_msg.message); return rc; @@ -729,17 +720,12 @@ static long zcrypt_pcixcc_send_cprb(struct zcrypt_device *zdev, goto out_free; init_completion(&resp_type.work); ap_queue_message(zdev->ap_dev, &ap_msg); - rc = wait_for_completion_interruptible_timeout( - &resp_type.work, PCIXCC_CLEANUP_TIME); - if (rc > 0) + rc = wait_for_completion_interruptible(&resp_type.work); + if (rc == 0) rc = convert_response_xcrb(zdev, &ap_msg, xcRB); - else { - /* Signal pending or message timed out. */ + else + /* Signal pending. */ ap_cancel_message(zdev->ap_dev, &ap_msg); - if (rc == 0) - /* Message timed out. */ - rc = -ETIME; - } out_free: memset(ap_msg.message, 0x0, ap_msg.length); kfree(ap_msg.message); -- cgit v1.2.3-59-g8ed1b