More small FBC fixes
Subsystems (2)
cover More small FBC fixes
From: Paulo Zanoni <[email protected]>
Serve these while the others are still baking.
Paulo Zanoni (6):
drm/i915: make sure we're not changing the FBC CFB with FBC enabled drm/i915: fix the FBC work allocation failure path drm/i915: fix FBC for cases where crtc->base.y is non-zero drm/i915: set ILK_DPFC_FENCE_YOFF to 0 on SNB
drm/i915: check for the supported strides on HSW+ FBC drm/i915: try a little harder to find an FBC CRTC
drivers/gpu/drm/i915/i915_drv.h | 1 + drivers/gpu/drm/i915/intel_fbc.c | 76 ++++++++++++++++++++++++++++++---------- 2 files changed, 58 insertions(+), 19 deletions(-)
Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 1/6 drm/i915: make sure we're not changing the FBC CFB with FBC enabled
From: Paulo Zanoni <[email protected]>
We used to have this bug in the past, but now that we properly track the size of the CFB, we don't have it anymore. Still, add the WARN just to make sure we don't go back to the bad state.
--- drivers/gpu/drm/i915/intel_fbc.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/gpu/drm/i915/intel_fbc.c +2 -0
@@ -589,6 +589,8 @@ static int intel_fbc_alloc_cfb(struct drm_i915_private *dev_priv, int size, dev_priv->fbc.threshold = ret; +WARN_ON (dev_priv->fbc.enabled); +if (INTEL_INFO (dev_priv)->gen >=5 )I915_WRITE (ILK_DPFC_CB_BASE , dev_priv->fbc.compressed_fb.start);else if (IS_GM45 (dev_priv)) {
Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 2/6 drm/i915: fix the FBC work allocation failure path
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
We were forgetting to set fbc.crtc, fbc.fb_id and fbc.y when the kzalloc call failed. Luckily, this is not a very common case, I hope.
--- drivers/gpu/drm/i915/intel_fbc.c | 27 +++++++++++++++++---------- 1 file changed, 17 insertions(+), 10 deletions(-)
drivers/gpu/drm/i915/intel_fbc.c +17 -10
@@ -308,6 +308,18 @@ bool intel_fbc_enabled(struct drm_i915_private *dev_priv)return dev_priv->fbc.enabled; } +static void intel_fbc_enable (struct intel_crtc *crtc, +struct drm_framebuffer *fb) +{ +struct drm_i915_private *dev_priv = crtc->base.dev->dev_private; + + dev_priv->fbc.enable_fbc (crtc); + + dev_priv->fbc.crtc = crtc; + dev_priv->fbc.fb_id = fb->base.id; + dev_priv->fbc.y = crtc->base.y; +} +static void intel_fbc_work_fn (struct work_struct *__work) { struct intel_fbc_work *work =
@@ -321,13 +333,8 @@ static void intel_fbc_work_fn(struct work_struct *__work)/* Double check that we haven't switched fb without cancelling * the prior work. */ -if (crtc_fb == work->fb) { - dev_priv->fbc.enable_fbc (work->crtc); - - dev_priv->fbc.crtc = work->crtc; - dev_priv->fbc.fb_id = crtc_fb->base.id; - dev_priv->fbc.y = work->crtc->base.y; - } +if (crtc_fb == work->fb) +intel_fbc_enable (work->crtc, work->fb); dev_priv->fbc.fbc_work =NULL ; }
@@ -361,7 +368,7 @@ static void intel_fbc_cancel_work(struct drm_i915_private *dev_priv) dev_priv->fbc.fbc_work =NULL ; } -static void intel_fbc_enable (struct intel_crtc *crtc) +static void intel_fbc_schedule_enable (struct intel_crtc *crtc) {struct intel_fbc_work *work;struct drm_i915_private *dev_priv = crtc->base.dev->dev_private;
@@ -373,7 +380,7 @@ static void intel_fbc_enable(struct intel_crtc *crtc) work =kzalloc (sizeof (*work),GFP_KERNEL );if (work ==NULL ) {DRM_ERROR ("Failed to allocate FBC work structure\n" ); - dev_priv->fbc.enable_fbc (crtc); +intel_fbc_enable (crtc, crtc->base.primary->fb);return ; }
@@ -826,7 +833,7 @@ static void __intel_fbc_update(struct drm_i915_private *dev_priv)__intel_fbc_disable (dev_priv); } -intel_fbc_enable (intel_crtc); +intel_fbc_schedule_enable (intel_crtc); dev_priv->fbc.no_fbc_reason =FBC_OK ;return ;
Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 3/6 drm/i915: fix FBC for cases where crtc->base.y is non-zero
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
I only tested this on BDW, but since the register description is the same ever since gen4, let's assume that all gens take the same register format. If that's not true, then hopefully someone will bisect a bug to this patch and we'll fix it.
Notice that the wrong fence offset register just means that the hardware tracking will be wrong.
Testcases:
- igt/kms_frontbuffer_tracking/fbc-1p-primscrn-pri-mfb-draw-mmap-gtt
- igt/kms_frontbuffer_tracking/fbc-2p-primscrn-pri-mfb-draw-mmap-gtt
--- drivers/gpu/drm/i915/intel_fbc.c | 26 +++++++++++++++++++++----- 1 file changed, 21 insertions(+), 5 deletions(-)
drivers/gpu/drm/i915/intel_fbc.c +21 -5
@@ -64,6 +64,20 @@ static void i8xx_fbc_disable(struct drm_i915_private *dev_priv)DRM_DEBUG_KMS ("disabled FBC\n" ); } +static unsigned int get_crtc_fence_y_offset (struct intel_crtc *crtc) +{ +struct drm_i915_private *dev_priv = crtc->base.dev->dev_private; +struct drm_framebuffer *fb = crtc->base.primary->fb; +unsigned int tile_height; + + tile_height =intel_tile_height (dev_priv->dev, fb->pixel_format, + fb->modifier[0 ]); + +/* The value we want is the line number of the first tile that contains +* the first vertical line of the CRTC. */ +return crtc->base.y - (crtc->base.y % tile_height); +} +static void i8xx_fbc_enable (struct intel_crtc *crtc) {struct drm_i915_private *dev_priv = crtc->base.dev->dev_private;
@@ -97,7 +111,7 @@ static void i8xx_fbc_enable(struct intel_crtc *crtc) fbc_ctl2 =FBC_CTL_FENCE_DBL |FBC_CTL_IDLE_IMM |FBC_CTL_CPU_FENCE ; fbc_ctl2 |=FBC_CTL_PLANE (crtc->plane);I915_WRITE (FBC_CONTROL2 , fbc_ctl2); -I915_WRITE (FBC_FENCE_OFF , crtc->base.y); +I915_WRITE (FBC_FENCE_OFF ,get_crtc_fence_y_offset (crtc)); }/* enable it... */
@@ -135,7 +149,7 @@ static void g4x_fbc_enable(struct intel_crtc *crtc) dpfc_ctl |=DPFC_CTL_LIMIT_1X ; dpfc_ctl |=DPFC_CTL_FENCE_EN | obj->fence_reg; -I915_WRITE (DPFC_FENCE_YOFF , crtc->base.y); +I915_WRITE (DPFC_FENCE_YOFF ,get_crtc_fence_y_offset (crtc));/* enable it... */ I915_WRITE (DPFC_CONTROL , dpfc_ctl |DPFC_CTL_EN );
@@ -177,6 +191,7 @@ static void ilk_fbc_enable(struct intel_crtc *crtc)struct drm_i915_gem_object *obj =intel_fb_obj (fb);u32 dpfc_ctl;int threshold = dev_priv->fbc.threshold; +unsigned int y_offset; dev_priv->fbc.enabled = true;
@@ -200,7 +215,8 @@ static void ilk_fbc_enable(struct intel_crtc *crtc)if (IS_GEN5 (dev_priv)) dpfc_ctl |= obj->fence_reg; -I915_WRITE (ILK_DPFC_FENCE_YOFF , crtc->base.y); + y_offset =get_crtc_fence_y_offset (crtc); +I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset);I915_WRITE (ILK_FBC_RT_BASE ,i915_gem_obj_ggtt_offset (obj) |ILK_FBC_RT_VALID );/* enable it... */ I915_WRITE (ILK_DPFC_CONTROL , dpfc_ctl |DPFC_CTL_EN );
@@ -208,7 +224,7 @@ static void ilk_fbc_enable(struct intel_crtc *crtc)if (IS_GEN6 (dev_priv)) {I915_WRITE (SNB_DPFC_CTL_SA ,SNB_CPU_FENCE_ENABLE | obj->fence_reg); -I915_WRITE (DPFC_CPU_FENCE_OFFSET , crtc->base.y); +I915_WRITE (DPFC_CPU_FENCE_OFFSET , y_offset); }intel_fbc_nuke (dev_priv);
@@ -288,7 +304,7 @@ static void gen7_fbc_enable(struct intel_crtc *crtc)I915_WRITE (SNB_DPFC_CTL_SA ,SNB_CPU_FENCE_ENABLE | obj->fence_reg); -I915_WRITE (DPFC_CPU_FENCE_OFFSET , crtc->base.y); +I915_WRITE (DPFC_CPU_FENCE_OFFSET ,get_crtc_fence_y_offset (crtc));intel_fbc_nuke (dev_priv);
Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 4/6 drm/i915: set ILK_DPFC_FENCE_YOFF to 0 on SNB
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
The doc is pretty clear that this register should be set to 0 on SNB. We already write y_offset to DPFC_CPU_FENCE_OFFSET a few lines below.
--- drivers/gpu/drm/i915/intel_fbc.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)
drivers/gpu/drm/i915/intel_fbc.c +6 -1
@@ -216,7 +216,12 @@ static void ilk_fbc_enable(struct intel_crtc *crtc) dpfc_ctl |= obj->fence_reg; y_offset =get_crtc_fence_y_offset (crtc); -I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); + +if (IS_GEN5 (dev_priv)) +I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); +else +I915_WRITE (ILK_DPFC_FENCE_YOFF ,0 ); +I915_WRITE (ILK_FBC_RT_BASE ,i915_gem_obj_ggtt_offset (obj) |ILK_FBC_RT_VALID );/* enable it... */ I915_WRITE (ILK_DPFC_CONTROL , dpfc_ctl |DPFC_CTL_EN );
Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 5/6 drm/i915: check for the supported strides on HSW+ FBC
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
I could only find the restrictions for HSW+, but I think it's safe to assume that the older platforms also can't support the configurations HSW can't support. The older platforms probably have additional restrictions, so we need to figure out those and implement them later. Let's not block HSW+ FBC based on missing information for the older platforms.
This may fix kms_frontbuffer_tracking failures depending on your monitor configuration.
--- drivers/gpu/drm/i915/i915_drv.h | 1 + drivers/gpu/drm/i915/intel_fbc.c | 10 ++++++++++ 2 files changed, 11 insertions(+)
drivers/gpu/drm/i915/i915_drv.h +1 -0
@@ -936,6 +936,7 @@ struct i915_fbc {FBC_CHIP_DEFAULT ,/* disabled by default on this chip */ FBC_ROTATION ,/* rotation is not supported */ FBC_IN_DBG_MASTER ,/* kernel debugger is active */ +FBC_BAD_STRIDE ,/* stride is not supported */ } no_fbc_reason;bool (*fbc_enabled)(struct drm_i915_private *dev_priv);
drivers/gpu/drm/i915/intel_fbc.c +10 -0
@@ -501,6 +501,8 @@ const char *intel_no_fbc_reason_str(enum no_fbc_reason reason)return "rotation unsupported" ;case FBC_IN_DBG_MASTER :return "Kernel debugger is active" ; +case FBC_BAD_STRIDE : +return "framebuffer stride not supported" ;default :MISSING_CASE (reason);return "unknown reason" ;
@@ -804,6 +806,14 @@ static void __intel_fbc_update(struct drm_i915_private *dev_priv) goto out_disable; } +/* TODO: we need to figure out what are the stride restrictions for the +* older platforms. */ +if (fb->pitches[0 ] <512 || fb->pitches[0 ] >16384 || + (fb->pitches[0 ] & (64 -1 )) !=0 ) { +set_no_fbc_reason (dev_priv,FBC_BAD_STRIDE ); + goto out_disable; + } +/* If the kernel debugger is active, always disable compression */ if (in_dbg_master ()) {set_no_fbc_reason (dev_priv,FBC_IN_DBG_MASTER );
Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 6/6 drm/i915: try a little harder to find an FBC CRTC
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
Keep searching in case the candidate has a NULL primary fb. This is only relevant for the platforms that don't have the pipe_a_only restriction.
--- drivers/gpu/drm/i915/intel_fbc.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/intel_fbc.c +2 -4
@@ -532,16 +532,14 @@ static struct drm_crtc *intel_fbc_find_crtc(struct drm_i915_private *dev_priv) tmp_crtc = dev_priv->pipe_to_crtc_mapping[pipe];if (intel_crtc_active (tmp_crtc) && -to_intel_plane_state (tmp_crtc->primary->state)->visible) +to_intel_plane_state (tmp_crtc->primary->state)->visible && + tmp_crtc->primary->fb !=NULL ) crtc = tmp_crtc;if (pipe_a_only)break ; } -if (!crtc || crtc->primary->fb ==NULL ) -return NULL ; -return crtc; }
Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 2/6] drm/i915: fix the FBC work allocation failure path
In reply to Paulo ZanoniQuoted 4 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
We were forgetting to set fbc.crtc, fbc.fb_id and fbc.y when the kzalloc call failed. Luckily, this is not a very common case, I hope.
That did not prepare me for the patch that lies ahead.
"Always update the currrent crtc, fb and vertical offset after calling enable_fbc. We were forgetting to do so along the failure paths when enabling fbc synchronously. Fix this with a new helper to enable_fbc() and update the state simultaneously."
-Chris
Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 4/6] drm/i915: set ILK_DPFC_FENCE_YOFF to 0 on SNB
In reply to Paulo ZanoniQuoted 6 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
The doc is pretty clear that this register should be set to 0 on SNB. We already write
y_offsettoDPFC_CPU_FENCE_OFFSETa few lines below.Signed-off-by Paulo Zanoni <[email protected]>
Hm, do we have testcases where we have a sufficiently big y offset? We can just allocate 128 lines more and use that as the offset, that should be big enough everywhere. Actually make that 129 lines to check the tile-size rounding ;-)
Ofc this means we need to have two sets of testcases for all the affected tests (i.e. everything that tries to test the gtt hw tracking).
Another funny corner case (which we're getting wrong on skl even without fbc) is x offsets > 2048 pixels (since x/y offset registers don't hold bigger values and then it wraps).
I.e. I'd like this patch (and the others) to be augmented with a Testcase:
tag.
Cheers, Daniel
Quoted 29 lines
--- drivers/gpu/drm/i915/intel_fbc.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/intel_fbc.c +6
-1@@ -216,7 +216,12 @@ static void ilk_fbc_enable(struct intel_crtc *crtc) dpfc_ctl |= obj->fence_reg; y_offset =get_crtc_fence_y_offset (crtc); -I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); + +if (IS_GEN5 (dev_priv)) +I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); +else +I915_WRITE (ILK_DPFC_FENCE_YOFF ,0 ); +I915_WRITE (ILK_FBC_RT_BASE ,i915_gem_obj_ggtt_offset (obj) |ILK_FBC_RT_VALID );/* enable it... */ I915_WRITE (ILK_DPFC_CONTROL , dpfc_ctl |DPFC_CTL_EN );Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 5/6] drm/i915: check for the supported strides on HSW+ FBC
In reply to Paulo ZanoniQuoted 44 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
I could only find the restrictions for HSW+, but I think it's safe to assume that the older platforms also can't support the configurations HSW can't support. The older platforms probably have additional restrictions, so we need to figure out those and implement them later. Let's not block HSW+ FBC based on missing information for the older platforms.
This may fix
kms_frontbuffer_trackingfailures depending on your monitor configuration.Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_drv.h | 1 + drivers/gpu/drm/i915/intel_fbc.c | 10 ++++++++++ 2 files changed, 11 insertions(+)drivers/gpu/drm/i915/i915_drv.h +1
-0@@ -936,6 +936,7 @@ struct i915_fbc {FBC_CHIP_DEFAULT ,/* disabled by default on this chip */ FBC_ROTATION ,/* rotation is not supported */ FBC_IN_DBG_MASTER ,/* kernel debugger is active */ +FBC_BAD_STRIDE ,/* stride is not supported */ } no_fbc_reason;bool (*fbc_enabled)(struct drm_i915_private *dev_priv);drivers/gpu/drm/i915/intel_fbc.c +2
-0@@ -501,6 +501,8 @@ const char *intel_no_fbc_reason_str(enum no_fbc_reason reason)return "rotation unsupported" ;case FBC_IN_DBG_MASTER :return "Kernel debugger is active" ; +case FBC_BAD_STRIDE : +return "framebuffer stride not supported" ;default :MISSING_CASE (reason);return "unknown reason" ;
drivers/gpu/drm/i915/intel_fbc.c static void __intel_fbc_update(struct drm_i915_private *dev_priv) · line 811
Quoted 6 lines
drivers/gpu/drm/i915/intel_fbc.c +3
-0@@ -804,6 +806,14 @@ static void __intel_fbc_update(struct drm_i915_private *dev_priv) goto out_disable; } +/* TODO: we need to figure out what are the stride restrictions for the +* older platforms. */ + if (fb->pitches[0] < 512 || fb->pitches[0] > 16384 ||
< 512 is already take care of by the X-tiling restriction (requires a multiple of 512 bytes stride). > 16kb might be the general stride limit, we should probably reject that in intel_framebuffer_init. 64b alignment is strange since X-tiled is stricter, and we that already in intel_fb_stride_alignment.
So summary: Sounds like we need to add per-platform stride limit checks to fb create code. Plus igt testcases to make sure we check for this (since right now it seems like we don't). Additional checks here in the fbc code don't seem required. But if you want to I guess you could convert them to WARN_ON (without bailing out).
-Daniel
Quoted 15 lines
drivers/gpu/drm/i915/intel_fbc.c +5
-0+ (fb->pitches[0] & (64 - 1)) != 0) { +set_no_fbc_reason (dev_priv,FBC_BAD_STRIDE ); + goto out_disable; + } +/* If the kernel debugger is active, always disable compression */ if (in_dbg_master ()) {set_no_fbc_reason (dev_priv,FBC_IN_DBG_MASTER );Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 4/6] drm/i915: set ILK_DPFC_FENCE_YOFF to 0 on SNB
In reply to Daniel Vetter2015-07-09 14:10 GMT-03:00 Daniel Vetter <[email protected]>:
Quoted 6 lines · Paulo Zanoni
Quoted 5 lines
From: Paulo Zanoni <[email protected]>
The doc is pretty clear that this register should be set to 0 on SNB. We already write
y_offsettoDPFC_CPU_FENCE_OFFSETa few lines below.
Quoted 6 lines
Quoted 1 line
Signed-off-by Paulo Zanoni <[email protected]>Hm, do we have testcases where we have a sufficiently big y offset? We can just allocate 128 lines more and use that as the offset, that should be big enough everywhere. Actually make that 129 lines to check the tile-size rounding ;-)
Yes, it's 500x500. See FBS_MULTI on kms_frontbuffer_tracking. Subtests with "sfb" on their names create a Single frontbuffer for each possible primary plane (primary, secondary, offscreen), while subtests with "mfb" have Multiple pipes pointing to the same frontbuffer. See the small drawing inside kms_frontbuffer_tracking, at the top of create_big_fb(). By the way, it's on my TODO list to change that arrangement a little bit in order to avoid super-huge strides.
Quoted 3 lines
Ofc this means we need to have two sets of testcases for all the affected tests (i.e. everything that tries to test the gtt hw tracking).
We do.
Quoted 4 lines
Another funny corner case (which we're getting wrong on skl even without fbc) is x offsets > 2048 pixels (since x/y offset registers don't hold bigger values and then it wraps).
Can you please clarify the sentence above? For the dual-screen subtests, if we use 2 screens of 1024 + the 500 pixel offset, we'll already be bigger than 2048 pixels.
Quoted 2 lines
I.e. I'd like this patch (and the others) to be augmented with a Testcase:
This one doesn't have a Testcase tag because I'm not testing SNB right now, so I can't confirm anything here. Patch 3 has the useful Testcases tags.
Quoted 38 lines
tag.
Cheers, Daniel
Quoted 25 lines
--- drivers/gpu/drm/i915/intel_fbc.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/intel_fbc.c +6
-2@@ -216,7 +216,12 @@ static void ilk_fbc_enable(struct intel_crtc *crtc) dpfc_ctl |= obj->fence_reg; y_offset =get_crtc_fence_y_offset (crtc); -I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); + +if (IS_GEN5 (dev_priv)) +I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); +else +I915_WRITE (ILK_DPFC_FENCE_YOFF ,0 ); +I915_WRITE (ILK_FBC_RT_BASE ,i915_gem_obj_ggtt_offset (obj) |ILK_FBC_RT_VALID );/* enable it... */ I915_WRITE (ILK_DPFC_CONTROL , dpfc_ctl |DPFC_CTL_EN ); --2.1.4
Quoted 4 lines
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
Signature · 5 lines
Paulo Zanoni _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 4/6] drm/i915: set ILK_DPFC_FENCE_YOFF to 0 on SNB
2015-07-09 14:15 GMT-03:00 Paulo Zanoni <[email protected]>:
Quoted 14 lines
2015-07-09 14:10 GMT-03:00 Daniel Vetter <[email protected]>:
Quoted 6 lines · Paulo Zanoni
Quoted 5 lines
From: Paulo Zanoni <[email protected]>
The doc is pretty clear that this register should be set to 0 on SNB. We already write
y_offsettoDPFC_CPU_FENCE_OFFSETa few lines below.Quoted 6 lines
Quoted 1 line
Signed-off-by Paulo Zanoni <[email protected]>Hm, do we have testcases where we have a sufficiently big y offset? We can just allocate 128 lines more and use that as the offset, that should be big enough everywhere. Actually make that 129 lines to check the tile-size rounding ;-)
Quoted 7 lines
Yes, it's 500x500. See
FBS_MULTIonkms_frontbuffer_tracking. Subtests with "sfb" on their names create a Single frontbuffer for each possible primary plane (primary, secondary, offscreen), while subtests with "mfb" have Multiple pipes pointing to the same frontbuffer. See the small drawing insidekms_frontbuffer_tracking, at the top ofcreate_big_fb(). By the way, it's on my TODO list to change that arrangement a little bit in order to avoid super-huge strides.
Also notice that I tested values other than 500x500 while developing patch 3. 500x500 seems like a nice value since it's not aligned with any tiles and never part of the first tile.
Quoted 66 lines
Quoted 4 lines
Ofc this means we need to have two sets of testcases for all the affected tests (i.e. everything that tries to test the gtt hw tracking).
We do.
Quoted 4 lines
Another funny corner case (which we're getting wrong on skl even without fbc) is x offsets > 2048 pixels (since x/y offset registers don't hold bigger values and then it wraps).
Can you please clarify the sentence above? For the dual-screen subtests, if we use 2 screens of 1024 + the 500 pixel offset, we'll already be bigger than 2048 pixels.
Quoted 2 lines
I.e. I'd like this patch (and the others) to be augmented with a Testcase:
This one doesn't have a Testcase tag because I'm not testing SNB right now, so I can't confirm anything here. Patch 3 has the useful Testcases tags.
Quoted 29 lines
tag.
Cheers, Daniel
Quoted 25 lines
--- drivers/gpu/drm/i915/intel_fbc.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/intel_fbc.c +6
-2@@ -216,7 +216,12 @@ static void ilk_fbc_enable(struct intel_crtc *crtc) dpfc_ctl |= obj->fence_reg; y_offset =get_crtc_fence_y_offset (crtc); -I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); + +if (IS_GEN5 (dev_priv)) +I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); +else +I915_WRITE (ILK_DPFC_FENCE_YOFF ,0 ); +I915_WRITE (ILK_FBC_RT_BASE ,i915_gem_obj_ggtt_offset (obj) |ILK_FBC_RT_VALID );/* enable it... */ I915_WRITE (ILK_DPFC_CONTROL , dpfc_ctl |DPFC_CTL_EN ); --2.1.4
Quoted 9 lines
Quoted 4 lines
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch--
Paulo Zanoni
Signature · 5 lines
Paulo Zanoni _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 4/6] drm/i915: set ILK_DPFC_FENCE_YOFF to 0 on SNB
In reply to Daniel VetterQuoted 16 lines · Daniel Vetter
Quoted 6 lines · Paulo Zanoni
From: Paulo Zanoni <[email protected]>
The doc is pretty clear that this register should be set to 0 on SNB. We already write
y_offsettoDPFC_CPU_FENCE_OFFSETa few lines below.Signed-off-by Paulo Zanoni <[email protected]>Hm, do we have testcases where we have a sufficiently big y offset? We can just allocate 128 lines more and use that as the offset, that should be big enough everywhere. Actually make that 129 lines to check the tile-size rounding ;-)
Ofc this means we need to have two sets of testcases for all the affected tests (i.e. everything that tries to test the gtt hw tracking).
Quoted 6 lines
Another funny corner case (which we're getting wrong on skl even without fbc) is x offsets > 2048 pixels (since x/y offset registers don't hold bigger values and then it wraps).
I.e. I'd like this patch (and the others) to be augmented with a Testcase:
tag.
I think the entire Y offset thing is currently being misprogrammed. IIRC the offset is from the display base address but we program in the offset from the start of the FB.
Quoted 41 lines
Cheers, Daniel
Quoted 29 lines
--- drivers/gpu/drm/i915/intel_fbc.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/intel_fbc.c +6
-1@@ -216,7 +216,12 @@ static void ilk_fbc_enable(struct intel_crtc *crtc) dpfc_ctl |= obj->fence_reg; y_offset =get_crtc_fence_y_offset (crtc); -I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); + +if (IS_GEN5 (dev_priv)) +I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); +else +I915_WRITE (ILK_DPFC_FENCE_YOFF ,0 ); +I915_WRITE (ILK_FBC_RT_BASE ,i915_gem_obj_ggtt_offset (obj) |ILK_FBC_RT_VALID );/* enable it... */ I915_WRITE (ILK_DPFC_CONTROL , dpfc_ctl |DPFC_CTL_EN );Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfxSignature · 7 lines
Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Signature · 6 lines
Ville Syrjälä Intel OTC _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 5/6] drm/i915: check for the supported strides on HSW+ FBC
In reply to Daniel Vetter2015-07-09 14:15 GMT-03:00 Daniel Vetter <[email protected]>:
Quoted 48 lines · Paulo Zanoni
Quoted 47 lines
From: Paulo Zanoni <[email protected]>
I could only find the restrictions for HSW+, but I think it's safe to assume that the older platforms also can't support the configurations HSW can't support. The older platforms probably have additional restrictions, so we need to figure out those and implement them later. Let's not block HSW+ FBC based on missing information for the older platforms.
This may fix
kms_frontbuffer_trackingfailures depending on your monitor configuration.Signed-off-by Paulo Zanoni <[email protected]>--- drivers/gpu/drm/i915/i915_drv.h | 1 + drivers/gpu/drm/i915/intel_fbc.c | 10 ++++++++++ 2 files changed, 11 insertions(+)drivers/gpu/drm/i915/i915_drv.h +1
-0@@ -936,6 +936,7 @@ struct i915_fbc {FBC_CHIP_DEFAULT ,/* disabled by default on this chip */ FBC_ROTATION ,/* rotation is not supported */ FBC_IN_DBG_MASTER ,/* kernel debugger is active */ +FBC_BAD_STRIDE ,/* stride is not supported */ } no_fbc_reason;bool (*fbc_enabled)(struct drm_i915_private *dev_priv);drivers/gpu/drm/i915/intel_fbc.c +2
-0@@ -501,6 +501,8 @@ const char *intel_no_fbc_reason_str(enum no_fbc_reason reason)return "rotation unsupported" ;case FBC_IN_DBG_MASTER :return "Kernel debugger is active" ; +case FBC_BAD_STRIDE : +return "framebuffer stride not supported" ;default :MISSING_CASE (reason);return "unknown reason" ;@@ -804,6 +806,14 @@ static void __intel_fbc_update(struct drm_i915_private *dev_priv) goto out_disable; }
Quoted 6 lines
drivers/gpu/drm/i915/intel_fbc.c static void __intel_fbc_update(struct drm_i915_private *dev_priv) · line 811
Quoted 3 lines
drivers/gpu/drm/i915/intel_fbc.c +3
-0+/* TODO: we need to figure out what are the stride restrictions for the +* older platforms. */ + if (fb->pitches[0] < 512 || fb->pitches[0] > 16384 ||< 512 is already take care of by the X-tiling restriction (requires a multiple of 512 bytes stride)
. > 16kb might be the general stride limit,
Quoted 3 lines
we should probably reject that in
intel_framebuffer_init. 64b alignment is strange since X-tiled is stricter, and we that already inintel_fb_stride_alignment.
16384 bytes is just 4096 32bit pixels wide. If you have 2 modern monitors configured in wide mode you already beat that (or maybe 3 not-so-fancy monitors). So if you want FBC you'll have to "xrandr --output DP1 --above eDP1 --auto", insead of "--left-of" (also, pray that the guys developing your desktop environment actually tested this case). So this is certainly a check that we want in the FBC code but not in other places. I can remove the other checks if you want.
I also did find that gen3 requires specifically 4k or 8k strides:
nothing except for that. But I'll only write this patch much later (gen3 is not a priority for now, as you know).
Quoted 3 lines
So summary: Sounds like we need to add per-platform stride limit checks to fb create code.
As mentioned above: no. FBC can be more strict.
Quoted 2 lines
Plus igt testcases to make sure we check for this (since right now it seems like we don't).
It's on the TODO list but it's not a priority since the Kernel checks are very straightforward. One of the problems is that different platforms have different requirements, so I'll have to hardcode those requirements on IGT too. And I'll have to stop using igt_create_fb for that case since it only use powers of 2 for stride.
So yeah, one day we may have the test, but not a priority right now since it's very easy to look at the Kernel code and make sure it's correct.
Quoted 25 lines
Additional checks here in the fbc code
don't seem required. But if you want to I guess you could convert them toWARN_ON(without bailing out).
-DanielQuoted 11 lines
drivers/gpu/drm/i915/intel_fbc.c +5
-1+ (fb->pitches[0] & (64 - 1)) != 0) { +set_no_fbc_reason (dev_priv,FBC_BAD_STRIDE ); + goto out_disable; + } +/* If the kernel debugger is active, always disable compression */ if (in_dbg_master ()) {set_no_fbc_reason (dev_priv,FBC_IN_DBG_MASTER ); --2.1.4
Quoted 4 lines
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
Signature · 5 lines
Paulo Zanoni _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 4/6] drm/i915: set ILK_DPFC_FENCE_YOFF to 0 on SNB
In reply to Ville Syrjälä2015-07-09 14:22 GMT-03:00 Ville Syrjälä <[email protected]>:
Quoted 21 lines · Daniel Vetter
Quoted 20 lines · Paulo Zanoni
Quoted 6 lines
From: Paulo Zanoni <[email protected]>
The doc is pretty clear that this register should be set to 0 on SNB. We already write
y_offsettoDPFC_CPU_FENCE_OFFSETa few lines below.Signed-off-by Paulo Zanoni <[email protected]>Hm, do we have testcases where we have a sufficiently big y offset? We can just allocate 128 lines more and use that as the offset, that should be big enough everywhere. Actually make that 129 lines to check the tile-size rounding ;-)
Ofc this means we need to have two sets of testcases for all the affected tests (i.e. everything that tries to test the gtt hw tracking).
Another funny corner case (which we're getting wrong on skl even without fbc) is x offsets > 2048 pixels (since x/y offset registers don't hold bigger values and then it wraps).
Quoted 6 lines
Quoted 2 lines
I.e. I'd like this patch (and the others) to be augmented with a Testcase:
tag.I think the entire Y offset thing is currently being misprogrammed. IIRC the offset is from the display base address but we program in the offset from the start of the FB.
After patch 3, all the current tests pass on BDW. Can you suggest a different test that won't pass?
Quoted 46 lines
Quoted 29 lines
Cheers, Daniel
Quoted 25 lines
--- drivers/gpu/drm/i915/intel_fbc.c | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-)drivers/gpu/drm/i915/intel_fbc.c +6
-2@@ -216,7 +216,12 @@ static void ilk_fbc_enable(struct intel_crtc *crtc) dpfc_ctl |= obj->fence_reg; y_offset =get_crtc_fence_y_offset (crtc); -I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); + +if (IS_GEN5 (dev_priv)) +I915_WRITE (ILK_DPFC_FENCE_YOFF , y_offset); +else +I915_WRITE (ILK_DPFC_FENCE_YOFF ,0 ); +I915_WRITE (ILK_FBC_RT_BASE ,i915_gem_obj_ggtt_offset (obj) |ILK_FBC_RT_VALID );/* enable it... */ I915_WRITE (ILK_DPFC_CONTROL , dpfc_ctl |DPFC_CTL_EN ); --2.1.4
Quoted 13 lines
Quoted 4 lines
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx--
Ville Syrjälä
Intel OTC
Signature · 5 lines
Paulo Zanoni _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 4/6] drm/i915: set ILK_DPFC_FENCE_YOFF to 0 on SNB
Quoted 25 lines · Paulo Zanoni
2015-07-09 14:22 GMT-03:00 Ville Syrjälä <[email protected]>:
Quoted 24 lines · Daniel Vetter
Quoted 22 lines · Paulo Zanoni
Quoted 6 lines
From: Paulo Zanoni <[email protected]>
The doc is pretty clear that this register should be set to 0 on SNB. We already write
y_offsettoDPFC_CPU_FENCE_OFFSETa few lines below.Signed-off-by Paulo Zanoni <[email protected]>Hm, do we have testcases where we have a sufficiently big y offset? We can just allocate 128 lines more and use that as the offset, that should be big enough everywhere. Actually make that 129 lines to check the tile-size rounding ;-)
Ofc this means we need to have two sets of testcases for all the affected tests (i.e. everything that tries to test the gtt hw tracking).
Another funny corner case (which we're getting wrong on skl even without fbc) is x offsets > 2048 pixels (since x/y offset registers don't hold bigger values and then it wraps).
I.e. I'd like this patch (and the others) to be augmented with a Testcase:
tag.
Quoted 6 lines
Quoted 3 lines
I think the entire Y offset thing is currently being misprogrammed. IIRC the offset is from the display base address but we program in the offset from the start of the FB.
After patch 3, all the current tests pass on BDW. Can you suggest a different test that won't pass?
Ah patch 3 tries to fix it. It's not entirely accurate though since it simply relies on an implementation detail of intel_gen4_compute_page_offset(). Well, assuming my recollection of the hardware details is correct.
Also IIRC intel_gen4_compute_page_offset() isn't even used on SKL/BXT currently, so it should fail on those platforms.
Signature · 6 lines
Ville Syrjälä Intel OTC _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 2/6 drm/i915: fix the FBC work allocation failure path
In reply to Chris WilsonFrom: Paulo Zanoni <[email protected]>
Always update the currrent crtc, fb and vertical offset after calling enable_fbc. We were forgetting to do so along the failure paths when enabling fbc synchronously. Fix this with a new helper to enable_fbc() and update the state simultaneously.
v2: Improve commit message (Chris).
--- drivers/gpu/drm/i915/intel_fbc.c | 27 +++++++++++++++++---------- 1 file changed, 17 insertions(+), 10 deletions(-)
drivers/gpu/drm/i915/intel_fbc.c +17 -10
@@ -308,6 +308,18 @@ bool intel_fbc_enabled(struct drm_i915_private *dev_priv)return dev_priv->fbc.enabled; } +static void intel_fbc_enable (struct intel_crtc *crtc, +struct drm_framebuffer *fb) +{ +struct drm_i915_private *dev_priv = crtc->base.dev->dev_private; + + dev_priv->fbc.enable_fbc (crtc); + + dev_priv->fbc.crtc = crtc; + dev_priv->fbc.fb_id = fb->base.id; + dev_priv->fbc.y = crtc->base.y; +} +static void intel_fbc_work_fn (struct work_struct *__work) { struct intel_fbc_work *work =
@@ -321,13 +333,8 @@ static void intel_fbc_work_fn(struct work_struct *__work)/* Double check that we haven't switched fb without cancelling * the prior work. */ -if (crtc_fb == work->fb) { - dev_priv->fbc.enable_fbc (work->crtc); - - dev_priv->fbc.crtc = work->crtc; - dev_priv->fbc.fb_id = crtc_fb->base.id; - dev_priv->fbc.y = work->crtc->base.y; - } +if (crtc_fb == work->fb) +intel_fbc_enable (work->crtc, work->fb); dev_priv->fbc.fbc_work =NULL ; }
@@ -361,7 +368,7 @@ static void intel_fbc_cancel_work(struct drm_i915_private *dev_priv) dev_priv->fbc.fbc_work =NULL ; } -static void intel_fbc_enable (struct intel_crtc *crtc) +static void intel_fbc_schedule_enable (struct intel_crtc *crtc) {struct intel_fbc_work *work;struct drm_i915_private *dev_priv = crtc->base.dev->dev_private;
@@ -373,7 +380,7 @@ static void intel_fbc_enable(struct intel_crtc *crtc) work =kzalloc (sizeof (*work),GFP_KERNEL );if (work ==NULL ) {DRM_ERROR ("Failed to allocate FBC work structure\n" ); - dev_priv->fbc.enable_fbc (crtc); +intel_fbc_enable (crtc, crtc->base.primary->fb);return ; }
@@ -826,7 +833,7 @@ static void __intel_fbc_update(struct drm_i915_private *dev_priv)__intel_fbc_disable (dev_priv); } -intel_fbc_enable (intel_crtc); +intel_fbc_schedule_enable (intel_crtc); dev_priv->fbc.no_fbc_reason =FBC_OK ;return ;
Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
PATCH 5/6 drm/i915: check for the supported strides on HSW+ FBC
In reply to Paulo ZanoniFrom: Paulo Zanoni <[email protected]>
I could only find the restrictions for HSW+, but I think it's safe to assume that the older platforms also can't support the configurations HSW can't support. The older platforms probably have additional restrictions, so we need to figure out those and implement them later. Let's not block HSW+ FBC based on missing information for the older platforms.
v2
- Just WARN_ON() the strides that should have been caught earlier (Daniel)
- Make it a new function since I expect this to grow more.
--- drivers/gpu/drm/i915/i915_drv.h | 1 + drivers/gpu/drm/i915/intel_fbc.c | 23 +++++++++++++++++++++++ 2 files changed, 24 insertions(+)
drivers/gpu/drm/i915/i915_drv.h +1 -0
@@ -937,6 +937,7 @@ struct i915_fbc {FBC_CHIP_DEFAULT ,/* disabled by default on this chip */ FBC_ROTATION ,/* rotation is not supported */ FBC_IN_DBG_MASTER ,/* kernel debugger is active */ +FBC_BAD_STRIDE ,/* stride is not supported */ } no_fbc_reason;bool (*fbc_enabled)(struct drm_i915_private *dev_priv);
drivers/gpu/drm/i915/intel_fbc.c +23 -0
@@ -501,6 +501,8 @@ const char *intel_no_fbc_reason_str(enum no_fbc_reason reason)return "rotation unsupported" ;case FBC_IN_DBG_MASTER :return "Kernel debugger is active" ; +case FBC_BAD_STRIDE : +return "framebuffer stride not supported" ;default :MISSING_CASE (reason);return "unknown reason" ;
@@ -694,6 +696,22 @@ static int intel_fbc_setup_cfb(struct drm_i915_private *dev_priv, int size,return intel_fbc_alloc_cfb (dev_priv, size, fb_cpp); } +static bool stride_is_valid (unsigned int stride) +{ +/* TODO: we need to figure out what are the stride restrictions for the +* older platforms. */ + +/* These should have been caught earlier. */ +WARN_ON (stride <512 ); +WARN_ON ((stride & (64 -1 )) !=0 ); + +/* These are additional FBC restrictions. */ +if (stride >16385 ) +return false; + +return true; +} +/** * __intel_fbc_update - enable/disable FBC as needed, unlocked * @dev_priv: i915 device instance
@@ -804,6 +822,11 @@ static void __intel_fbc_update(struct drm_i915_private *dev_priv) goto out_disable; } +if (!stride_is_valid (fb->pitches[0 ])) { +set_no_fbc_reason (dev_priv,FBC_BAD_STRIDE ); + goto out_disable; + } +/* If the kernel debugger is active, always disable compression */ if (in_dbg_master ()) {set_no_fbc_reason (dev_priv,FBC_IN_DBG_MASTER );
Signature · 6 lines
2.1.4 _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 5/6] drm/i915: check for the supported strides on HSW+ FBC
In reply to Paulo Zanoni2015-07-09 14:28 GMT-03:00 Paulo Zanoni <[email protected]>:
Quoted 10 lines
2015-07-09 14:15 GMT-03:00 Daniel Vetter <[email protected]>:
Quoted 2 lines
Plus igt testcases to make sure we check for this (since right now it seems like we don't).
It's on the TODO list but it's not a priority since the Kernel checks are very straightforward. One of the problems is that different platforms have different requirements, so I'll have to hardcode those requirements on IGT too. And I'll have to stop using
igt_create_fbfor that case since it only use powers of 2 for stride.
Quoted 3 lines
So yeah, one day we may have the test, but not a priority right now since it's very easy to look at the Kernel code and make sure it's correct.
Ok, so I implemented fbc-badstride with the restrictions mentioned in this patch:
If we add more restrictions we'll have to update the test.
Quoted 31 lines
Quoted 22 lines
Additional checks here in the fbc code
don't seem required. But if you want to I guess you could convert them toWARN_ON(without bailing out).
-DanielQuoted 11 lines
drivers/gpu/drm/i915/intel_fbc.c +5
-1+ (fb->pitches[0] & (64 - 1)) != 0) { +set_no_fbc_reason (dev_priv,FBC_BAD_STRIDE ); + goto out_disable; + } +/* If the kernel debugger is active, always disable compression */ if (in_dbg_master ()) {set_no_fbc_reason (dev_priv,FBC_IN_DBG_MASTER ); --2.1.4
Quoted 4 lines
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfxQuoted 4 lines
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch--
Paulo Zanoni
Signature · 5 lines
Paulo Zanoni _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx
Re: [PATCH 4/6] drm/i915: set ILK_DPFC_FENCE_YOFF to 0 on SNB
In reply to Ville Syrjälä2015-07-09 14:39 GMT-03:00 Ville Syrjälä <[email protected]>:
Quoted 33 lines · Paulo Zanoni
Quoted 25 lines
2015-07-09 14:22 GMT-03:00 Ville Syrjälä <[email protected]>:
Quoted 24 lines · Daniel Vetter
Quoted 22 lines · Paulo Zanoni
Quoted 6 lines
From: Paulo Zanoni <[email protected]>
The doc is pretty clear that this register should be set to 0 on SNB. We already write
y_offsettoDPFC_CPU_FENCE_OFFSETa few lines below.Signed-off-by Paulo Zanoni <[email protected]>Hm, do we have testcases where we have a sufficiently big y offset? We can just allocate 128 lines more and use that as the offset, that should be big enough everywhere. Actually make that 129 lines to check the tile-size rounding ;-)
Ofc this means we need to have two sets of testcases for all the affected tests (i.e. everything that tries to test the gtt hw tracking).
Another funny corner case (which we're getting wrong on skl even without fbc) is x offsets > 2048 pixels (since x/y offset registers don't hold bigger values and then it wraps).
I.e. I'd like this patch (and the others) to be augmented with a Testcase:
tag.Quoted 6 lines
Quoted 3 lines
I think the entire Y offset thing is currently being misprogrammed. IIRC the offset is from the display base address but we program in the offset from the start of the FB.
After patch 3, all the current tests pass on BDW. Can you suggest a different test that won't pass?
Quoted 6 lines
Ah patch 3 tries to fix it. It's not entirely accurate though since it simply relies on an implementation detail of
intel_gen4_compute_page_offset(). Well, assuming my recollection of the hardware details is correct.Also IIRC
intel_gen4_compute_page_offset()isn't even used on SKL/BXT currently, so it should fail on those platforms.
Daniel clarified the problem to me, so I implemented the following test case:
http://cgit.freedesktop.org/xorg/app/intel-gpu-tools/commit/?id=04d1311fc3d2127d609b5c5e670bf9887652cb17
I hope this exercises the problem you're mentioning. So far I only tested BDW and it passes.
Quoted 4 lines
--
Ville Syrjälä
Intel OTC
Signature · 5 lines
Paulo Zanoni _______________________________________________ Intel-gfx mailing list [email protected] http://lists.freedesktop.org/mailman/listinfo/intel-gfx