Re: [PATCH 3/5] drm/xe/vram: report FLAT_CCS base misalignment

[Matthew Brost <matthew.brost@intel.com>]

On Tue, Sep 01, 2026 at 09:23:22AM +0100, Matthew Auld wrote:
> On 28/08/2026 21:10, Matthew Brost wrote:
> > On Fri, Aug 28, 2026 at 04:14:09PM +0100, Matthew Auld wrote:
> > > So we can easily check if a machine had the CCS bug, when looking back
> > > over bug reports where we have the same machine with newer kernel.
> > > 
> > 
> > I'm a little confused about how this helps, so I'm thinking out loud here:
> > - On old kernels without Linus's CCS fix or this patch, there is no
> >    message and we get silent corruption if the affected memory belongs to
> >    user space. Of course, things fall apart completely if the corruption
> >    affects kernel memory, as in the case Linus hit.
> > - On new kernels with Linus's CCS fix plus this patch, does this message
> >    appear under load on some configurations, or on all configurations?
> 
> Yeah, this is logged before the round_down(offset, 4K). On my b570, for
> example, I get a nice dmesg print to indicate that this machine had the bug
> (fix from Linus also applied). In our CI farm, there is no print, since
> nothing seems to have this bug, also confirms the memtest is not just
> missing the issue somehow. But not sure about wider RIL, I assume something
> in there maybe has this bug.
> 
> > 
> > Is the idea that the first thing we check for is whether this message is
> > present?
> 
> Yeah, just a record in dmesg of whether the device had this CCS bug, when
> looking at bug reports. The memtest is also gated behind debug builds, for
> now. For example, some older bug report with impossible corruption or other
> weird stuff, update to latest stable or tip and issue seems to go away.
> Would be good to confirm if this machine had the CCS bug. Users might also
> want to check this.
> 
> > Would it be better to simply print the FLAT_CCS base address along with
> > a field indicating whether it is misaligned? That would let us quickly
> > determine, across all devices, whether the CCS base is aligned or
> > misaligned and if the CCS fix is present.
> 
> Could print the address also. Happy to do either approach.
> 

I think just delete the 'if (!IS_ALIGNED(offset, SZ_128K))' gate and
print the CCS info + flag a misalignment.

Matt

> > 
> > Matt
> > 
> > > Fixes: 37173392741c ("drm/xe/vram: fix ccs offset calculation")
> > > Signed-off-by: Matthew Auld <matthew.auld@intel.com>
> > > Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> > > Cc: Matthew Brost <matthew.brost@intel.com>
> > > Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
> > > Cc: stable@kernel.org
> > > ---
> > >   drivers/gpu/drm/xe/xe_vram.c | 3 +++
> > >   1 file changed, 3 insertions(+)
> > > 
> > > diff --git a/drivers/gpu/drm/xe/xe_vram.c b/drivers/gpu/drm/xe/xe_vram.c
> > > index 7b4478fb1697..0a2f1ab8416e 100644
> > > --- a/drivers/gpu/drm/xe/xe_vram.c
> > > +++ b/drivers/gpu/drm/xe/xe_vram.c
> > > @@ -90,6 +90,9 @@ static int get_flat_ccs_offset(struct xe_gt *gt, u64 tile_size, u64 *poffset)
> > >   		offset |= offset_lo << 6; /* HW view bits 31:6 */
> > >   		offset *= num_enabled; /* convert to SW view */
> > > +		if (!IS_ALIGNED(offset, SZ_128K))
> > > +			drm_info(&xe->drm, "FLAT_CCS base misaligned: %llx\n", offset);
> > > +
> > >   		/*
> > >   		 * Everything below this offset is handed to the VRAM
> > >   		 * allocator, so it has to be the *first* address the
> > > -- 
> > > 2.55.0
> > > 
>