Re: SG does not ignore dxferp (direct io + mmap)

[Laurence Oberman <loberman@redhat.com>]

----- Original Message -----
> From: "Ewan Milne" <emilne@redhat.com>
> To: "Johannes Thumshirn" <jthumshirn@suse.de>
> Cc: "Laurence Oberman" <loberman@redhat.com>, "Eyal Ben David" <bdeyal@gmail.com>, dgilbert@interlog.com,
> linux-scsi@vger.kernel.org
> Sent: Friday, November 25, 2016 12:56:16 PM
> Subject: Re: SG does not ignore dxferp (direct io + mmap)
> 
> >> ---
> >> 
> >> In other words, this commit made the bad behavior go away in 4.8.
> >> Need to look at this in more detail, it doesn't appear as if this patch
> >> was intended to fix such a problem.
> >> 
> >> -Ewan
> >
> >Are you sure it did? I can repropduce copy_to_user() errors with 4.8 as
> >well.
> >Using the very same reproducer. On 4.8 it's just harder to trigger and
> >doesn't trigger on AHCI as fas as I can telli (maybe I just haven't hit
> >it hard enough). I can trigger it on QEMUs SCSI CDROM emulation and hpsa
> >though. I cannot however trigger this with a minimal config inside an
> >initrd.
> 
> It did for Eyal's supplied test case on my machine, but that was not an
> exhaustive test, and I am a little suspicious that the behavior change was
> due to a side-effect of the patch rather than actually fixing something.
> 
> I think what we need to understand is what caused the regression in the
> first place, I probably should have been bisecting the original failure
> rather than trying to find where it started working.
> 
> I was running against an internal (physical) drive.
> 
> -Ewan
> 

My 100000 loop was against an HPSA target and passed all tests.
Again, all I did was patch the 4.7.9 with the 2 line changes, the rest of the patch was line breaks.

I guess we need to understand when it first broke and what caused that, versus what seems to correct it.
Thanks
Laurence