2014-12-29 17:29:55

by Nicholas Mc Guire

Subject: [SCSI RFC] mpt2sas: wait_for_completion_timeout timeout not reported

wait_for_completion_timeout reaching timeout was being ignored,
this probably also should fail if timeout condition occurs ?

this was only compile tested with
x86_64_defconfig + CONFIG_SCSI_LOWLEVEL=y + CONFIG_SCSI_MPT2SAS=m

patch is against linux-next 3.19.0-rc1 -next-20141226

Signed-off-by: Nicholas Mc Guire <[email protected]>
---
drivers/scsi/mpt2sas/mpt2sas_config.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/scsi/mpt2sas/mpt2sas_config.c b/drivers/scsi/mpt2sas/mpt2sas_config.c
index c72a2ff..02dc2d8 100644
--- a/drivers/scsi/mpt2sas/mpt2sas_config.c
+++ b/drivers/scsi/mpt2sas/mpt2sas_config.c
@@ -391,7 +391,7 @@ _config_request(struct MPT2SAS_ADAPTER *ioc, Mpi2ConfigRequest_t
mpt2sas_base_put_smid_default(ioc, smid);
timeleft = wait_for_completion_timeout(&ioc->config_cmds.done,
timeout*HZ);
- if (!(ioc->config_cmds.status & MPT2_CMD_COMPLETE)) {
+ if (timeleft == 0 || !(ioc->config_cmds.status & MPT2_CMD_COMPLETE)) {
printk(MPT2SAS_ERR_FMT "%s: timeout\n",
ioc->name, __func__);
_debug_dump_mf(mpi_request,
--
1.7.10.4

2014-12-30 13:33:27

by Joe Lawrence

Subject: Re: [SCSI RFC] mpt2sas: wait_for_completion_timeout timeout not reported

On 12/29/2014 12:25 PM, Nicholas Mc Guire wrote:
> wait_for_completion_timeout reaching timeout was being ignored,
> this probably also should fail if timeout condition occurs ?
>
> this was only compile tested with
> x86_64_defconfig + CONFIG_SCSI_LOWLEVEL=y + CONFIG_SCSI_MPT2SAS=m
>
> patch is against linux-next 3.19.0-rc1 -next-20141226
>
> Signed-off-by: Nicholas Mc Guire <[email protected]>
> ---
> drivers/scsi/mpt2sas/mpt2sas_config.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/scsi/mpt2sas/mpt2sas_config.c b/drivers/scsi/mpt2sas/mpt2sas_config.c
> index c72a2ff..02dc2d8 100644
> --- a/drivers/scsi/mpt2sas/mpt2sas_config.c
> +++ b/drivers/scsi/mpt2sas/mpt2sas_config.c
> @@ -391,7 +391,7 @@ _config_request(struct MPT2SAS_ADAPTER *ioc, Mpi2ConfigRequest_t
> mpt2sas_base_put_smid_default(ioc, smid);
> timeleft = wait_for_completion_timeout(&ioc->config_cmds.done,
> timeout*HZ);
> - if (!(ioc->config_cmds.status & MPT2_CMD_COMPLETE)) {
> + if (timeleft == 0 || !(ioc->config_cmds.status & MPT2_CMD_COMPLETE)) {
> printk(MPT2SAS_ERR_FMT "%s: timeout\n",
> ioc->name, __func__);
> _debug_dump_mf(mpi_request,
>

Hi Nicholas,

This would look like the correct thing to do here, for it eliminates a
small window where we timeout when mpt2sas_config_done had set
MPT2_CMD_COMPLETE and MPT3_CMD_REPLY_VALID, but has not finished copying
the reply data into config_cmds.reply.

It looks like the same patch should be applied to the mpt3sas driver as
well.

-- Joe