Re: [PATCH] drm/radeon/kms: fix nsample tracking in r6xx/r7xx CS tracker

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

 



On Mon, Aug 16, 2010 at 4:53 PM, Alex Deucher <alexdeucher@xxxxxxxxx> wrote:
> On Mon, Aug 16, 2010 at 4:39 PM, Dave Airlie <airlied@xxxxxxxxxx> wrote:
>> On Mon, 2010-08-16 at 09:54 -0400, Jerome Glisse wrote:
>>> On 08/16/2010 09:22 AM, Alex Deucher wrote:
>>> > On Mon, Aug 16, 2010 at 9:19 AM, Jerome Glisse <glisse@xxxxxxxxxxxxxxx> wrote:
>>> >> On 08/13/2010 12:48 AM, Alex Deucher wrote:
>>> >>> Initialize nsamples to 1 in case userspace does not emit
>>> >>> PA_SC_AA_CONFIG.
>>> >>>
>>> >>> Signed-off-by: Alex Deucher <alexdeucher@xxxxxxxxx>
>>> >>> Cc: Andre Maasikas <amaasikas@xxxxxxxxx>
>>> >>> ---
>>> >>>  drivers/gpu/drm/radeon/r600_cs.c |    1 +
>>> >>>  1 files changed, 1 insertions(+), 0 deletions(-)
>>> >>>
>>> >>> diff --git a/drivers/gpu/drm/radeon/r600_cs.c b/drivers/gpu/drm/radeon/r600_cs.c
>>> >>> index 8e165b2..b03f93b 100644
>>> >>> --- a/drivers/gpu/drm/radeon/r600_cs.c
>>> >>> +++ b/drivers/gpu/drm/radeon/r600_cs.c
>>> >>> @@ -163,6 +163,7 @@ static void r600_cs_track_init(struct r600_cs_track *track)
>>> >>>       track->db_depth_size = 0xFFFFFFFF;
>>> >>>       track->db_depth_size_idx = 0;
>>> >>>       track->db_depth_control = 0xFFFFFFFF;
>>> >>> +     track->nsamples = 1;
>>> >>>  }
>>> >>>
>>> >>>  static inline int r600_cs_track_validate_cb(struct radeon_cs_parser *p, int i)
>>> >>
>>> >> I would NAK this one, accepting such things pretty much defeat the idea
>>> >> of checker, i think userspace should always set nsamples as otherwise it
>>> >> might trick kernel and overwritte others memory. ie:
>>> >> cs1 set nsample 8 & all others properly so cs1 is accepted
>>> >> cs2 don't set nsample but bind a smaller bo and so can overwritte others
>>> >> buffer
>>> >
>>> > Unfortunately, existing userspace already sends buffers without this
>>> > reg set. It's just that prior to tiling we didn't care what it was set
>>> > to.
>>> >
>>> > Alex
>>>
>>> I guess once again we hit by our own fault of prematurely going out of
>>> staging.
>>
>> Jerome, staying in staging forever until every feature of the hw is
>> implemented isn't possible, at some point you have to do real engineer
>> work and this sounds like its workaroundable
>>
>> How about if we get a tiling CB in the checker, we make sure nsamples is
>> set? will that work?
>>
>
> I've fixed the issue in mesa.  The problematic mesa code has not been
> officially part of a release yet.  That said, I think 1 is a
> reasonable default for nsamples regardless as unless MSAA is enabled,
> you can pretty safely assume 1.
>

If we don't set a default value, we would need to make sure IBs always
contain PA_SC_AA_CONFIG which is nor currently enforced.  Thoughts?

Alex

> Alex
>
>> Dave.
>>
>>
>>
>
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
http://lists.freedesktop.org/mailman/listinfo/dri-devel



[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux