Re: [PATCH 07/19] scsi: sd: Do not issue commands to suspended disks on remove

[Damien Le Moal <dlemoal@kernel.org>]

On 9/14/23 05:50, Bart Van Assche wrote:
> On 9/10/23 21:02, Damien Le Moal wrote:
>> If an error occurs when resuming a host adapter before the devices
>> attached to the adapter are resumed, the adapter low level driver may
>> remove the scsi host, resulting in a call to sd_remove() for the
>> disks of the host. However, since this function calls sd_shutdown(),
>> a synchronize cache command and a start stop unit may be issued with the
>> drive still sleeping and the HBA non-functional. This causes PM resume
>> to hang, forcing a reset of the machine to recover.
>>
>> Fix this by checking a device host state in sd_shutdown() and by
>> returning early doing nothing if the host state is not SHOST_RUNNING.
>>
>> Cc: stable@vger.kernel.org
>> Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
>> ---
>>   drivers/scsi/sd.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c
>> index c92a317ba547..a415abb721d3 100644
>> --- a/drivers/scsi/sd.c
>> +++ b/drivers/scsi/sd.c
>> @@ -3763,7 +3763,8 @@ static void sd_shutdown(struct device *dev)
>>   	if (!sdkp)
>>   		return;         /* this can happen */
>>   
>> -	if (pm_runtime_suspended(dev))
>> +	if (pm_runtime_suspended(dev) ||
>> +	    sdkp->device->host->shost_state != SHOST_RUNNING)
>>   		return;
>>   
>>   	if (sdkp->WCE && sdkp->media_present) {
> 
> Why to test the host state instead of dev->power.runtime_status? I don't
> think that it is safe to skip shutdown if the error handler is active.
> If the error handler can recover the device a SYNCHRONIZE CACHE command
> should be submitted.

But there is no synchronization with EH that I can see anyway. At least for
sd_remove(), I would assume that this is called only once the device references
were all dropped, so presumably EH is not doing anything with the drive when
that happen, no ?

In any case, looking at dev->power.runtime_status is not correct as this is set
to RPM_ACTIVE when the device is suspended through system suspend. We could
replace the test "sdkp->device->host->shost_state != SHOST_RUNNING" with
"dev->power.is_suspended", as that indicates true (1) for a suspended device.
However, I really do not like that as that is a PM internal field and should not
be accessing it directly. The PM code comments say as much. Any better idea ?


-- 
Damien Le Moal
Western Digital Research