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

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

[Alex Deucher]

Initialize nsamples to 1 in case userspace does not emit
PA_SC_AA_CONFIG.

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Cc: Andre Maasikas <amaasikas@gmail.com>
---
 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)
-- 
1.7.1.1

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

[Jerome Glisse]

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@gmail.com>
> Cc: Andre Maasikas <amaasikas@gmail.com>
> ---
>  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

Cheers,
Jerome

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

[Alex Deucher]

On Mon, Aug 16, 2010 at 9:19 AM, Jerome Glisse <glisse@freedesktop.org> 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@gmail.com>
>> Cc: Andre Maasikas <amaasikas@gmail.com>
>> ---
>>  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

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

[Jerome Glisse]

On 08/16/2010 09:22 AM, Alex Deucher wrote:
> On Mon, Aug 16, 2010 at 9:19 AM, Jerome Glisse <glisse@freedesktop.org> 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@gmail.com>
>>> Cc: Andre Maasikas <amaasikas@gmail.com>
>>> ---
>>>  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

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

[Alex Deucher]

On Mon, Aug 16, 2010 at 9:54 AM, Jerome Glisse <glisse@freedesktop.org> wrote:
> On 08/16/2010 09:22 AM, Alex Deucher wrote:
>> On Mon, Aug 16, 2010 at 9:19 AM, Jerome Glisse <glisse@freedesktop.org> 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@gmail.com>
>>>> Cc: Andre Maasikas <amaasikas@gmail.com>
>>>> ---
>>>>  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.

I *think* the only thing that doesn't emit this reg is the blit code
in mesa git master, so we could just fix it there as it's not
officially released, but there's quite a bit of mesa git master out in
the wild.

Alex

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

[Dave Airlie]

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@freedesktop.org> 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@gmail.com>
> >>> Cc: Andre Maasikas <amaasikas@gmail.com>
> >>> ---
> >>>  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?

Dave.

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

[Alex Deucher]

On Mon, Aug 16, 2010 at 4:39 PM, Dave Airlie <airlied@redhat.com> 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@freedesktop.org> 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@gmail.com>
>> >>> Cc: Andre Maasikas <amaasikas@gmail.com>
>> >>> ---
>> >>>  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.

Alex

> Dave.
>
>
>

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

[Alex Deucher]

On Mon, Aug 16, 2010 at 4:53 PM, Alex Deucher <alexdeucher@gmail.com> wrote:
> On Mon, Aug 16, 2010 at 4:39 PM, Dave Airlie <airlied@redhat.com> 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@freedesktop.org> 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@gmail.com>
>>> >>> Cc: Andre Maasikas <amaasikas@gmail.com>
>>> >>> ---
>>> >>>  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.
>>
>>
>>
>