Re: Thinkpad E595 system deadlock on resume from S3

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On 10/4/23 05:07, Petr Tesařík wrote:
> I just want to confirm that my understanding of the issue is correct
> now. After applying the patch below, my laptop has just survived half a
> dozen suspend/resume cycles with autosuspend enabled for the SATA SSD
> disk. Without the patch, it usually freezes on first attempt.
> 
> diff --git a/drivers/ata/libata-scsi.c b/drivers/ata/libata-scsi.c
> index a371b497035e..87000f89319e 100644
> --- a/drivers/ata/libata-scsi.c
> +++ b/drivers/ata/libata-scsi.c
> @@ -4768,6 +4768,14 @@ void ata_scsi_dev_rescan(struct work_struct *work)
>  		ata_for_each_dev(dev, link, ENABLED) {
>  			struct scsi_device *sdev = dev->sdev;
>  
> +			/*
> +			 * If the queue accepts only PM, bail out.
> +			 */
> +			if (blk_queue_pm_only(sdev->request_queue)) {
> +				ret = -EAGAIN;
> +				goto unlock;
> +			}
> +
>  			/*
>  			 * If the port was suspended before this was scheduled,
>  			 * bail out.

This seems sensible to me: scsi_rescan_device() only checks that the device is
running, without checking that the device queue is also resumed. So the right
place for this check is scsi_rescan_device():

diff --git a/drivers/scsi/scsi_scan.c b/drivers/scsi/scsi_scan.c
index 3db4d31a03a1..b05a55f498a2 100644
--- a/drivers/scsi/scsi_scan.c
+++ b/drivers/scsi/scsi_scan.c
@@ -1627,12 +1627,13 @@ int scsi_rescan_device(struct scsi_device *sdev)
        device_lock(dev);

        /*
-        * Bail out if the device is not running. Otherwise, the rescan may
-        * block waiting for commands to be executed, with us holding the
-        * device lock. This can result in a potential deadlock in the power
-        * management core code when system resume is on-going.
+        * Bail out if the device or its queue are not running. Otherwise,
+        * the rescan may block waiting for commands to be executed, with us
+        * holding the device lock. This can result in a potential deadlock
+        * in the power management core code when system resume is on-going.
         */
-       if (sdev->sdev_state != SDEV_RUNNING) {
+       if (sdev->sdev_state != SDEV_RUNNING ||
+           blk_queue_pm_only(sdev->request_queue)) {
                ret = -EWOULDBLOCK;
                goto unlock;
        }

Can you try the above to confirm it works for you ? It should, as that is
essentially the same as you did.

-- 
Damien Le Moal
Western Digital Research




[Index of Archives]     [Linux Filesystems]     [Linux SCSI]     [Linux RAID]     [Git]     [Kernel Newbies]     [Linux Newbie]     [Security]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Samba]     [Device Mapper]

  Powered by Linux