summaryrefslogtreecommitdiff
path: root/drivers/scsi/ipr.c
diff options
context:
space:
mode:
authorBrian King <brking@linux.vnet.ibm.com>2017-03-15 16:58:41 -0500
committerMartin K. Petersen <martin.petersen@oracle.com>2017-03-23 12:04:05 -0400
commitef97d8ae12eb20d81bea5d16a0c1e33dcbe44492 (patch)
tree76f20e6152227c9680ad9127b99bd6b5246cc222 /drivers/scsi/ipr.c
parentf646f325a829d73abc708088d2b67cd11b775fdf (diff)
downloadlwn-ef97d8ae12eb20d81bea5d16a0c1e33dcbe44492.tar.gz
lwn-ef97d8ae12eb20d81bea5d16a0c1e33dcbe44492.zip
scsi: ipr: Fix SATA EH hang
This patch fixes a hang that can occur in ATA EH with ipr. With ipr's usage of libata, commands should never end up on ap->eh_done_q. The timeout function we use for ipr, even for SATA devices, is scsi_times_out, so ATA_QCFLAG_EH_SCHEDULED never gets set for ipr and EH is driven completely by ipr and SCSI. The SCSI EH thread ends up calling ipr's eh_device_reset_handler, which then calls ata_std_error_handler. This ends up calling ipr_sata_reset, which issues a reset to the device. This should result in all pending commands getting failed back and having ata_qc_complete called for them, which should end up clearing ATA_QCFLAG_FAILED as qc->flags gets zeroed in ata_qc_free. This ensures that when we end up in ata_eh_finish, we don't do anything more with the command. On adapters that only support a single interrupt and when running with two MSI-X vectors or less, the adapter firmware guarantees that responses to all outstanding commands are sent back prior to sending the response to the SATA reset command. On newer adapters supporting multiple HRRQs, however, this can no longer be guaranteed, since the command responses and reset response may be processed on different HRRQs. If ipr returns from ipr_sata_reset before the outstanding command was returned, this sends us down the path of __ata_eh_qc_complete which then moves the associated scsi_cmd from the work_q in scsi_eh_bus_device_reset to ap->eh_done_q, which then will sit there forever and we will be wedged. This patch fixes this up by ensuring that any outstanding commands are flushed before returning from eh_device_reset_handler for a SATA device. Reported-by: David Jeffery <djeffery@redhat.com> Signed-off-by: Brian King <brking@linux.vnet.ibm.com> Reviewed-by: Wendy Xiong <wenxiong@linux.vnet.ibm.com> Tested-by: Wendy Xiong <wenxiong@linux.vnet.ibm.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Diffstat (limited to 'drivers/scsi/ipr.c')
-rw-r--r--drivers/scsi/ipr.c62
1 files changed, 41 insertions, 21 deletions
diff --git a/drivers/scsi/ipr.c b/drivers/scsi/ipr.c
index 95f1e6457664..b3d1e4426ecc 100644
--- a/drivers/scsi/ipr.c
+++ b/drivers/scsi/ipr.c
@@ -5067,6 +5067,23 @@ static bool ipr_cmnd_is_free(struct ipr_cmnd *ipr_cmd)
}
/**
+ * ipr_match_res - Match function for specified resource entry
+ * @ipr_cmd: ipr command struct
+ * @resource: resource entry to match
+ *
+ * Returns:
+ * 1 if command matches sdev / 0 if command does not match sdev
+ **/
+static int ipr_match_res(struct ipr_cmnd *ipr_cmd, void *resource)
+{
+ struct ipr_resource_entry *res = resource;
+
+ if (res && ipr_cmd->ioarcb.res_handle == res->res_handle)
+ return 1;
+ return 0;
+}
+
+/**
* ipr_wait_for_ops - Wait for matching commands to complete
* @ipr_cmd: ipr command struct
* @device: device to match (sdev)
@@ -5246,7 +5263,7 @@ static int ipr_sata_reset(struct ata_link *link, unsigned int *classes,
struct ipr_ioa_cfg *ioa_cfg = sata_port->ioa_cfg;
struct ipr_resource_entry *res;
unsigned long lock_flags = 0;
- int rc = -ENXIO;
+ int rc = -ENXIO, ret;
ENTER;
spin_lock_irqsave(ioa_cfg->host->host_lock, lock_flags);
@@ -5260,9 +5277,19 @@ static int ipr_sata_reset(struct ata_link *link, unsigned int *classes,
if (res) {
rc = ipr_device_reset(ioa_cfg, res);
*classes = res->ata_class;
- }
+ spin_unlock_irqrestore(ioa_cfg->host->host_lock, lock_flags);
+
+ ret = ipr_wait_for_ops(ioa_cfg, res, ipr_match_res);
+ if (ret != SUCCESS) {
+ spin_lock_irqsave(ioa_cfg->host->host_lock, lock_flags);
+ ipr_initiate_ioa_reset(ioa_cfg, IPR_SHUTDOWN_ABBREV);
+ spin_unlock_irqrestore(ioa_cfg->host->host_lock, lock_flags);
+
+ wait_event(ioa_cfg->reset_wait_q, !ioa_cfg->in_reset_reload);
+ }
+ } else
+ spin_unlock_irqrestore(ioa_cfg->host->host_lock, lock_flags);
- spin_unlock_irqrestore(ioa_cfg->host->host_lock, lock_flags);
LEAVE;
return rc;
}
@@ -5291,9 +5318,6 @@ static int __ipr_eh_dev_reset(struct scsi_cmnd *scsi_cmd)
ioa_cfg = (struct ipr_ioa_cfg *) scsi_cmd->device->host->hostdata;
res = scsi_cmd->device->hostdata;
- if (!res)
- return FAILED;
-
/*
* If we are currently going through reset/reload, return failed. This will force the
* mid-layer to call ipr_eh_host_reset, which will then go to sleep and wait for the
@@ -5332,19 +5356,6 @@ static int __ipr_eh_dev_reset(struct scsi_cmnd *scsi_cmd)
spin_unlock_irq(scsi_cmd->device->host->host_lock);
ata_std_error_handler(ap);
spin_lock_irq(scsi_cmd->device->host->host_lock);
-
- for_each_hrrq(hrrq, ioa_cfg) {
- spin_lock(&hrrq->_lock);
- list_for_each_entry(ipr_cmd,
- &hrrq->hrrq_pending_q, queue) {
- if (ipr_cmd->ioarcb.res_handle ==
- res->res_handle) {
- rc = -EIO;
- break;
- }
- }
- spin_unlock(&hrrq->_lock);
- }
} else
rc = ipr_device_reset(ioa_cfg, res);
res->resetting_device = 0;
@@ -5358,15 +5369,24 @@ static int ipr_eh_dev_reset(struct scsi_cmnd *cmd)
{
int rc;
struct ipr_ioa_cfg *ioa_cfg;
+ struct ipr_resource_entry *res;
ioa_cfg = (struct ipr_ioa_cfg *) cmd->device->host->hostdata;
+ res = cmd->device->hostdata;
+
+ if (!res)
+ return FAILED;
spin_lock_irq(cmd->device->host->host_lock);
rc = __ipr_eh_dev_reset(cmd);
spin_unlock_irq(cmd->device->host->host_lock);
- if (rc == SUCCESS)
- rc = ipr_wait_for_ops(ioa_cfg, cmd->device, ipr_match_lun);
+ if (rc == SUCCESS) {
+ if (ipr_is_gata(res) && res->sata_port)
+ rc = ipr_wait_for_ops(ioa_cfg, res, ipr_match_res);
+ else
+ rc = ipr_wait_for_ops(ioa_cfg, cmd->device, ipr_match_lun);
+ }
return rc;
}