On 08/01/2018 05:05 PM, Stephen Rothwell wrote:
Hi all,
On Wed, 01 Aug 2018 16:00:54 -0700 James Bottomley <James.Bottomley@xxxxxxxxxxxxxxxxxxxxx> wrote:
So what seems to be happening to cause this is that there's a patch
somewhere between the merge base of my scsi-next series and the next
tree and the patch just before scsi-next was actually merged that
actually causes a boot failure with blk-mq enabled. Could you try to
find this patch? I think the way to do it is to try to bisect this
range of linux-next using the command line
scsi_mod.use_blk_mq=1
Which forces block mq to be the default and seeing where the first boot
failure is (you don't need my scsi-next tree merged to do this because
all the offending patch does is flip the default state of the above
flag).
So this means using v4.8-rc1 as the first good commit and 453f1d821165
("Merge remote-tracking branch 'cgroup/for-next'") as the first bad
(assuming that this latter fails to boot with "scsi_mod.use_blk_mq=1").
Puzzled. Same results. 453f1d821165 works with both scsi_mod.use_blk_mq=0
and scsi_mod.use_blk_mq=1. next-20180801 works with scsi_mod.use_blk_mq=0
and fails with scsi_mod.use_blk_mq=1. Bisect still points to the same commit
(which just changes the default) as culprit. I know that doesn't make sense.
I'll need to think about it.
Guenter