From patchwork Wed Oct 21 20:15:52 2015 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Matthew R. Ochs" X-Patchwork-Id: 534066 Return-Path: X-Original-To: patchwork-incoming@ozlabs.org Delivered-To: patchwork-incoming@ozlabs.org Received: from lists.ozlabs.org (lists.ozlabs.org [103.22.144.68]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 8B58E1412F9 for ; Thu, 22 Oct 2015 07:52:09 +1100 (AEDT) Received: from lists.ozlabs.org (lists.ozlabs.org [IPv6:2401:3900:2:1::3]) by lists.ozlabs.org (Postfix) with ESMTP id 716431A36FD for ; Thu, 22 Oct 2015 07:52:09 +1100 (AEDT) X-Original-To: linuxppc-dev@lists.ozlabs.org Delivered-To: linuxppc-dev@lists.ozlabs.org Received: from e18.ny.us.ibm.com (e18.ny.us.ibm.com [129.33.205.208]) (using TLSv1 with cipher CAMELLIA256-SHA (256/256 bits)) (No client certificate requested) by lists.ozlabs.org (Postfix) with ESMTPS id 60CDB1A0BBE for ; Thu, 22 Oct 2015 07:17:16 +1100 (AEDT) Received: from localhost by e18.ny.us.ibm.com with IBM ESMTP SMTP Gateway: Authorized Use Only! Violators will be prosecuted for from ; Wed, 21 Oct 2015 16:17:14 -0400 Received: from d01dlp01.pok.ibm.com (9.56.250.166) by e18.ny.us.ibm.com (146.89.104.205) with IBM ESMTP SMTP Gateway: Authorized Use Only! Violators will be prosecuted; Wed, 21 Oct 2015 16:17:12 -0400 X-IBM-Helo: d01dlp01.pok.ibm.com X-IBM-MailFrom: mrochs@linux.vnet.ibm.com X-IBM-RcptTo: linuxppc-dev@lists.ozlabs.org Received: from b01cxnp23034.gho.pok.ibm.com (b01cxnp23034.gho.pok.ibm.com [9.57.198.29]) by d01dlp01.pok.ibm.com (Postfix) with ESMTP id 5635F38C803B for ; Wed, 21 Oct 2015 16:17:11 -0400 (EDT) Received: from d01av02.pok.ibm.com (d01av02.pok.ibm.com [9.56.224.216]) by b01cxnp23034.gho.pok.ibm.com (8.14.9/8.14.9/NCO v10.0) with ESMTP id t9LKHBKr66584680 for ; Wed, 21 Oct 2015 20:17:11 GMT Received: from d01av02.pok.ibm.com (localhost [127.0.0.1]) by d01av02.pok.ibm.com (8.14.4/8.14.4/NCO v10.0 AVout) with ESMTP id t9LKHAm9001344 for ; Wed, 21 Oct 2015 16:17:11 -0400 Received: from p8tul1-build.aus.stglabs.ibm.com (als141206.austin.ibm.com [9.3.141.206]) by d01av02.pok.ibm.com (8.14.4/8.14.4/NCO v10.0 AVin) with ESMTP id t9LKH9JV001271; Wed, 21 Oct 2015 16:17:09 -0400 From: "Matthew R. Ochs" To: linux-scsi@vger.kernel.org, James Bottomley , "Nicholas A. Bellinger" , Brian King , Ian Munsie , Daniel Axtens , Andrew Donnellan , Tomas Henzl , David Laight Subject: [PATCH v6 32/37] cxlflash: Fix to avoid potential deadlock on EEH Date: Wed, 21 Oct 2015 15:15:52 -0500 Message-Id: <1445458552-61150-1-git-send-email-mrochs@linux.vnet.ibm.com> X-Mailer: git-send-email 2.1.0 In-Reply-To: <1445458134-63197-1-git-send-email-mrochs@linux.vnet.ibm.com> References: <1445458134-63197-1-git-send-email-mrochs@linux.vnet.ibm.com> X-TM-AS-MML: disable X-Content-Scanned: Fidelis XPS MAILER x-cbid: 15102120-0045-0000-0000-000001FC3EDB X-BeenThere: linuxppc-dev@lists.ozlabs.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: Michael Neuling , "Manoj N. Kumar" , linuxppc-dev@lists.ozlabs.org MIME-Version: 1.0 Errors-To: linuxppc-dev-bounces+patchwork-incoming=ozlabs.org@lists.ozlabs.org Sender: "Linuxppc-dev" Ioctl threads that use scsi_execute() can run for an excessive amount of time due to the fact that they have lengthy timeouts and retry logic built in. Under normal operation this is not an issue. However, once EEH enters the picture, a long execution time coupled with the possibility that a timeout can trigger entry to the driver via registered reset callbacks becomes a liability. In particular, a deadlock can occur when an EEH event is encountered while in running in scsi_execute(). As part of the recovery, the EEH handler drains all currently running ioctls, waiting until they have completed before proceeding with a reset. As the scsi_execute()'s are situated on the ioctl path, the EEH handler will wait until they (and the remainder of the ioctl handler they're associated with) have completed. Normally this would not be much of an issue aside from the longer recovery period. Unfortunately, the scsi_execute() triggers a reset when it times out. The reset handler will see that the device is already being reset and wait until that reset completed. This creates a condition where the EEH handler becomes stuck, infinitely waiting for the ioctl thread to complete. To avoid this behavior, temporarily unmark the scsi_execute() threads as an ioctl thread by releasing the ioctl read semaphore. This allows the EEH handler to proceed with a recovery while the thread is still running. Once the scsi_execute() returns, the ioctl read semaphore is reacquired and the adapter state is rechecked in case it changed while inside of scsi_execute(). The state check will wait if the adapter is still being recovered or returns a failure if the recovery failed. In the event that the adapter reset failed, the failure is simply returned as the ioctl would be unable to continue. Reported-by: Brian King Signed-off-by: Matthew R. Ochs Signed-off-by: Manoj N. Kumar Reviewed-by: Brian King Reviewed-by: Daniel Axtens Reviewed-by: Tomas Henzl --- drivers/scsi/cxlflash/superpipe.c | 30 +++++++++++++++++++++++++++++- drivers/scsi/cxlflash/superpipe.h | 2 ++ drivers/scsi/cxlflash/vlun.c | 29 +++++++++++++++++++++++++++++ 3 files changed, 60 insertions(+), 1 deletion(-) diff --git a/drivers/scsi/cxlflash/superpipe.c b/drivers/scsi/cxlflash/superpipe.c index f625e07..8af7cdc 100644 --- a/drivers/scsi/cxlflash/superpipe.c +++ b/drivers/scsi/cxlflash/superpipe.c @@ -283,6 +283,24 @@ out: * @sdev: SCSI device associated with LUN. * @lli: LUN destined for capacity request. * + * The READ_CAP16 can take quite a while to complete. Should an EEH occur while + * in scsi_execute(), the EEH handler will attempt to recover. As part of the + * recovery, the handler drains all currently running ioctls, waiting until they + * have completed before proceeding with a reset. As this routine is used on the + * ioctl path, this can create a condition where the EEH handler becomes stuck, + * infinitely waiting for this ioctl thread. To avoid this behavior, temporarily + * unmark this thread as an ioctl thread by releasing the ioctl read semaphore. + * This will allow the EEH handler to proceed with a recovery while this thread + * is still running. Once the scsi_execute() returns, reacquire the ioctl read + * semaphore and check the adapter state in case it changed while inside of + * scsi_execute(). The state check will wait if the adapter is still being + * recovered or return a failure if the recovery failed. In the event that the + * adapter reset failed, simply return the failure as the ioctl would be unable + * to continue. + * + * Note that the above puts a requirement on this routine to only be called on + * an ioctl thread. + * * Return: 0 on success, -errno on failure */ static int read_cap16(struct scsi_device *sdev, struct llun_info *lli) @@ -314,8 +332,18 @@ retry: dev_dbg(dev, "%s: %ssending cmd(0x%x)\n", __func__, retry_cnt ? "re" : "", scsi_cmd[0]); + /* Drop the ioctl read semahpore across lengthy call */ + up_read(&cfg->ioctl_rwsem); result = scsi_execute(sdev, scsi_cmd, DMA_FROM_DEVICE, cmd_buf, CMD_BUFSIZE, sense_buf, to, CMD_RETRIES, 0, NULL); + down_read(&cfg->ioctl_rwsem); + rc = check_state(cfg); + if (rc) { + dev_err(dev, "%s: Failed state! result=0x08%X\n", + __func__, result); + rc = -ENODEV; + goto out; + } if (driver_byte(result) == DRIVER_SENSE) { result &= ~(0xFF<<24); /* DRIVER_SENSE is not an error */ @@ -1221,7 +1249,7 @@ static const struct file_operations null_fops = { * * Return: 0 on success, -errno on failure */ -static int check_state(struct cxlflash_cfg *cfg) +int check_state(struct cxlflash_cfg *cfg) { struct device *dev = &cfg->dev->dev; int rc = 0; diff --git a/drivers/scsi/cxlflash/superpipe.h b/drivers/scsi/cxlflash/superpipe.h index 7df88ee..06a805a 100644 --- a/drivers/scsi/cxlflash/superpipe.h +++ b/drivers/scsi/cxlflash/superpipe.h @@ -147,4 +147,6 @@ void cxlflash_ba_terminate(struct ba_lun *); int cxlflash_manage_lun(struct scsi_device *, struct dk_cxlflash_manage_lun *); +int check_state(struct cxlflash_cfg *); + #endif /* ifndef _CXLFLASH_SUPERPIPE_H */ diff --git a/drivers/scsi/cxlflash/vlun.c b/drivers/scsi/cxlflash/vlun.c index b0eaf55..a53f583 100644 --- a/drivers/scsi/cxlflash/vlun.c +++ b/drivers/scsi/cxlflash/vlun.c @@ -400,6 +400,24 @@ static int init_vlun(struct llun_info *lli) * @lba: Logical block address to start write same. * @nblks: Number of logical blocks to write same. * + * The SCSI WRITE_SAME16 can take quite a while to complete. Should an EEH occur + * while in scsi_execute(), the EEH handler will attempt to recover. As part of + * the recovery, the handler drains all currently running ioctls, waiting until + * they have completed before proceeding with a reset. As this routine is used + * on the ioctl path, this can create a condition where the EEH handler becomes + * stuck, infinitely waiting for this ioctl thread. To avoid this behavior, + * temporarily unmark this thread as an ioctl thread by releasing the ioctl read + * semaphore. This will allow the EEH handler to proceed with a recovery while + * this thread is still running. Once the scsi_execute() returns, reacquire the + * ioctl read semaphore and check the adapter state in case it changed while + * inside of scsi_execute(). The state check will wait if the adapter is still + * being recovered or return a failure if the recovery failed. In the event that + * the adapter reset failed, simply return the failure as the ioctl would be + * unable to continue. + * + * Note that the above puts a requirement on this routine to only be called on + * an ioctl thread. + * * Return: 0 on success, -errno on failure */ static int write_same16(struct scsi_device *sdev, @@ -433,9 +451,20 @@ static int write_same16(struct scsi_device *sdev, put_unaligned_be32(ws_limit < left ? ws_limit : left, &scsi_cmd[10]); + /* Drop the ioctl read semahpore across lengthy call */ + up_read(&cfg->ioctl_rwsem); result = scsi_execute(sdev, scsi_cmd, DMA_TO_DEVICE, cmd_buf, CMD_BUFSIZE, sense_buf, to, CMD_RETRIES, 0, NULL); + down_read(&cfg->ioctl_rwsem); + rc = check_state(cfg); + if (rc) { + dev_err(dev, "%s: Failed state! result=0x08%X\n", + __func__, result); + rc = -ENODEV; + goto out; + } + if (result) { dev_err_ratelimited(dev, "%s: command failed for " "offset %lld result=0x%x\n",