[SCSI] ipr: Fix oops while resetting an ipr adapter

When resetting an ipr adapter, we use scsi_block_requests to
block any new commands from scsi core, and then unblock after
the reset. When hotplug removing an adapter, we shut it down
and go through this same code, but we've seen issues with
scsi_unblock_requests running after the adapter's memory has
been freed. There is really no need to block/unblock when
the adapter is being removed, so this patch skips the
block/unblock and will immediately fail any commands that
happen to make it to queuecommand while the adapter is
being shutdown.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: Wen Xiong <wenxiong@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>
This commit is contained in:
Brian King 2013-01-30 23:45:08 -06:00 committed by James Bottomley
parent 4d7007b49d
commit bfae7820b8
2 changed files with 24 additions and 10 deletions

View file

@ -6112,7 +6112,7 @@ static int ipr_queuecommand(struct Scsi_Host *shost,
* We have told the host to stop giving us new requests, but
* ERP ops don't count. FIXME
*/
if (unlikely(!hrrq->allow_cmds && !hrrq->ioa_is_dead)) {
if (unlikely(!hrrq->allow_cmds && !hrrq->ioa_is_dead && !hrrq->removing_ioa)) {
spin_unlock_irqrestore(hrrq->lock, hrrq_flags);
return SCSI_MLQUEUE_HOST_BUSY;
}
@ -6121,7 +6121,7 @@ static int ipr_queuecommand(struct Scsi_Host *shost,
* FIXME - Create scsi_set_host_offline interface
* and the ioa_is_dead check can be removed
*/
if (unlikely(hrrq->ioa_is_dead || !res)) {
if (unlikely(hrrq->ioa_is_dead || hrrq->removing_ioa || !res)) {
spin_unlock_irqrestore(hrrq->lock, hrrq_flags);
goto err_nodev;
}
@ -6741,14 +6741,17 @@ static int ipr_ioa_bringdown_done(struct ipr_cmnd *ipr_cmd)
struct ipr_ioa_cfg *ioa_cfg = ipr_cmd->ioa_cfg;
ENTER;
if (!ioa_cfg->hrrq[IPR_INIT_HRRQ].removing_ioa) {
ipr_trace;
spin_unlock_irq(ioa_cfg->host->host_lock);
scsi_unblock_requests(ioa_cfg->host);
spin_lock_irq(ioa_cfg->host->host_lock);
}
ioa_cfg->in_reset_reload = 0;
ioa_cfg->reset_retries = 0;
list_add_tail(&ipr_cmd->queue, &ipr_cmd->hrrq->hrrq_free_q);
wake_up_all(&ioa_cfg->reset_wait_q);
spin_unlock_irq(ioa_cfg->host->host_lock);
scsi_unblock_requests(ioa_cfg->host);
spin_lock_irq(ioa_cfg->host->host_lock);
LEAVE;
return IPR_RC_JOB_RETURN;
@ -8494,7 +8497,8 @@ static void _ipr_initiate_ioa_reset(struct ipr_ioa_cfg *ioa_cfg,
spin_unlock(&ioa_cfg->hrrq[i]._lock);
}
wmb();
scsi_block_requests(ioa_cfg->host);
if (!ioa_cfg->hrrq[IPR_INIT_HRRQ].removing_ioa)
scsi_block_requests(ioa_cfg->host);
ipr_cmd = ipr_get_free_ipr_cmnd(ioa_cfg);
ioa_cfg->reset_cmd = ipr_cmd;
@ -8549,9 +8553,11 @@ static void ipr_initiate_ioa_reset(struct ipr_ioa_cfg *ioa_cfg,
ipr_fail_all_ops(ioa_cfg);
wake_up_all(&ioa_cfg->reset_wait_q);
spin_unlock_irq(ioa_cfg->host->host_lock);
scsi_unblock_requests(ioa_cfg->host);
spin_lock_irq(ioa_cfg->host->host_lock);
if (!ioa_cfg->hrrq[IPR_INIT_HRRQ].removing_ioa) {
spin_unlock_irq(ioa_cfg->host->host_lock);
scsi_unblock_requests(ioa_cfg->host);
spin_lock_irq(ioa_cfg->host->host_lock);
}
return;
} else {
ioa_cfg->in_ioa_bringdown = 1;
@ -9695,6 +9701,7 @@ static void __ipr_remove(struct pci_dev *pdev)
{
unsigned long host_lock_flags = 0;
struct ipr_ioa_cfg *ioa_cfg = pci_get_drvdata(pdev);
int i;
ENTER;
spin_lock_irqsave(ioa_cfg->host->host_lock, host_lock_flags);
@ -9704,6 +9711,12 @@ static void __ipr_remove(struct pci_dev *pdev)
spin_lock_irqsave(ioa_cfg->host->host_lock, host_lock_flags);
}
for (i = 0; i < ioa_cfg->hrrq_num; i++) {
spin_lock(&ioa_cfg->hrrq[i]._lock);
ioa_cfg->hrrq[i].removing_ioa = 1;
spin_unlock(&ioa_cfg->hrrq[i]._lock);
}
wmb();
ipr_initiate_ioa_bringdown(ioa_cfg, IPR_SHUTDOWN_NORMAL);
spin_unlock_irqrestore(ioa_cfg->host->host_lock, host_lock_flags);

View file

@ -493,6 +493,7 @@ struct ipr_hrr_queue {
u8 allow_interrupts:1;
u8 ioa_is_dead:1;
u8 allow_cmds:1;
u8 removing_ioa:1;
struct blk_iopoll iopoll;
};