Skip to contentSkip to conversation

drm/i915: Fix some NUM_RING iterators

PATCH In review Ben Widawsky 6 participants 9 messages Jun 27, 2014 – Jul 7, 2014

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/i915: Fix some NUM_RING iterators

There are some cases in the code where we need to know how many rings to iterate over, but cannot use for_each_ring(). These are always error cases which happen either before ring setup, or after ring teardown (or reset).

Note, a NUM_RINGS issue exists in semaphores, but this is fixed by the remaining semaphore patches which Rodrigo will resubmit shortly. I'd rather see those patches for fixing the problem than fix it here.

I found this initially for the BSD2 case where on the same platform we can have differing rings. AFAICT however this effects many platforms.

I'd CC stable on this, except I think all the issues have been around for multiple releases without bug reports.

Compile tested only for now.

---
 drivers/gpu/drm/i915/i915_gem_context.c | 6 +++---
 drivers/gpu/drm/i915/i915_gpu_error.c   | 2 +-
 drivers/gpu/drm/i915/intel_ringbuffer.h | 2 ++
 3 files changed, 6 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_gem_context.c +3 -3
index b9bac25..0c044a9 100644
@@ -403,7 +403,7 @@ void i915_gem_context_reset(struct drm_device *dev)
 
 	/* Prevent the hardware from restoring the last context (which hung) on
 	 * the next switch */
-	for (i = 0; i < I915_NUM_RINGS; i++) {
+	for (i = 0; i < I915_ACTIVE_RINGS(dev); i++) {
 		struct intel_engine_cs *ring = &dev_priv->ring[i];
 		struct intel_context *dctx = ring->default_context;
 
@@ -456,7 +456,7 @@ int i915_gem_context_init(struct drm_device *dev)
 	}
 
 	/* NB: RCS will hold a ref for all rings */
-	for (i = 0; i < I915_NUM_RINGS; i++)
+	for (i = 0; i < I915_ACTIVE_RINGS(dev); i++)
 		dev_priv->ring[i].default_context = ctx;
 
 	DRM_DEBUG_DRIVER("%s context support initialized\n", dev_priv->hw_context_size ? "HW" : "fake");
@@ -493,7 +493,7 @@ void i915_gem_context_fini(struct drm_device *dev)
 		i915_gem_object_ggtt_unpin(dctx->obj);
 	}
 
-	for (i = 0; i < I915_NUM_RINGS; i++) {
+	for (i = 0; i < I915_ACTIVE_RINGS(dev); i++) {
 		struct intel_engine_cs *ring = &dev_priv->ring[i];
 
 		if (ring->last_context)
drivers/gpu/drm/i915/i915_gpu_error.c +1 -1
index 86362de..6e5250d 100644
@@ -848,7 +848,7 @@ static uint32_t i915_error_generate_code(struct drm_i915_private *dev_priv,
 	 * synchronization commands which almost always appear in the case
 	 * strictly a client bug. Use instdone to differentiate those some.
 	 */
-	for (i = 0; i < I915_NUM_RINGS; i++) {
+	for (i = 0; i < I915_ACTIVE_RINGS(dev_priv->dev); i++) {
 		if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
 			if (ring_id)
 				*ring_id = i;
drivers/gpu/drm/i915/intel_ringbuffer.h +2 -0
index e72017b..67e2919 100644
@@ -90,6 +90,8 @@ struct  intel_engine_cs {
 	} id;
 #define I915_NUM_RINGS 5
 #define LAST_USER_RING (VECS + 1)
+#define I915_ACTIVE_RINGS(dev) hweight8(INTEL_INFO(dev)->ring_mask)
+
 	u32		mmio_base;
 	struct		drm_device *dev;
 	struct intel_ringbuffer *buffer;
Signature · 1 line
2.0.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 95 lines · Ben Widawsky

There are some cases in the code where we need to know how many rings to iterate over, but cannot use for_each_ring(). These are always error cases which happen either before ring setup, or after ring teardown (or reset).

Note, a NUM_RINGS issue exists in semaphores, but this is fixed by the remaining semaphore patches which Rodrigo will resubmit shortly. I'd rather see those patches for fixing the problem than fix it here.

I found this initially for the BSD2 case where on the same platform we can have differing rings. AFAICT however this effects many platforms.

I'd CC stable on this, except I think all the issues have been around for multiple releases without bug reports.

Compile tested only for now.

---
 drivers/gpu/drm/i915/i915_gem_context.c | 6 +++---
 drivers/gpu/drm/i915/i915_gpu_error.c   | 2 +-
 drivers/gpu/drm/i915/intel_ringbuffer.h | 2 ++
 3 files changed, 6 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_gem_context.c +0 -0
index b9bac25..0c044a9 100644
@@ -403,7 +403,7 @@ void i915_gem_context_reset(struct drm_device *dev)
 
        /* Prevent the hardware from restoring the last context (which

hung) on
* the next switch */

drivers/gpu/drm/i915/i915_gem_context.c +2 -2
-       for (i = 0; i < I915_NUM_RINGS; i++) {
+       for (i = 0; i < I915_ACTIVE_RINGS(dev); i++) {
                struct intel_engine_cs *ring = &dev_priv->ring[i];
                struct intel_context *dctx = ring->default_context;
@@ -456,7 +456,7 @@ int i915_gem_context_init(struct drm_device *dev)
        }
 
        /* NB: RCS will hold a ref for all rings */
-       for (i = 0; i < I915_NUM_RINGS; i++)
+       for (i = 0; i < I915_ACTIVE_RINGS(dev); i++)
                dev_priv->ring[i].default_context = ctx;
 
        DRM_DEBUG_DRIVER("%s context support initialized\n",

dev_priv->hw_context_size ? "HW" : "fake");

drivers/gpu/drm/i915/i915_gem_context.c +1 -1
@@ -493,7 +493,7 @@ void i915_gem_context_fini(struct drm_device *dev)
                i915_gem_object_ggtt_unpin(dctx->obj);
        }
 
-       for (i = 0; i < I915_NUM_RINGS; i++) {
+       for (i = 0; i < I915_ACTIVE_RINGS(dev); i++) {
                struct intel_engine_cs *ring = &dev_priv->ring[i];
 
                if (ring->last_context)
drivers/gpu/drm/i915/i915_gpu_error.c +1 -1
index 86362de..6e5250d 100644
@@ -848,7 +848,7 @@ static uint32_t i915_error_generate_code(struct
 drm_i915_private *dev_priv,
         * synchronization commands which almost always appear in the case
         * strictly a client bug. Use instdone to differentiate those some.
         */
-       for (i = 0; i < I915_NUM_RINGS; i++) {
+       for (i = 0; i < I915_ACTIVE_RINGS(dev_priv->dev); i++) {
                if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
                        if (ring_id)
                                *ring_id = i;
drivers/gpu/drm/i915/intel_ringbuffer.h +2 -1
index e72017b..67e2919 100644
@@ -90,6 +90,8 @@ struct  intel_engine_cs {
        } id;
 #define I915_NUM_RINGS 5
 #define LAST_USER_RING (VECS + 1)
+#define I915_ACTIVE_RINGS(dev) hweight8(INTEL_INFO(dev)->ring_mask)
+
        u32             mmio_base;
        struct          drm_device *dev;
        struct intel_ringbuffer *buffer;
--

2.0.0

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

Signature · 2 lines
Rodrigo Vivi
Blog: http://blog.vivi.eng.br

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#2 lore raw
Quoted 2 lines · Rodrigo Vivi

I have a couple of spots that I think are important to add, all in error state. I'll repost v2 after I can actually test it.

[snip]

Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center
#3 lore raw
In reply to Rodrigo Vivi
Quoted 83 lines · Rodrigo Vivi
   Reviewed-by: Rodrigo Vivi <[1][email protected]>

   On Fri, Jun 27, 2014 at 3:09 PM, Ben Widawsky
   <[2][email protected]> wrote:

     There are some cases in the code where we need to know how many rings to
     iterate over, but cannot use for_each_ring(). These are always error
     cases
     which happen either before ring setup, or after ring teardown (or
     reset).

     Note, a NUM_RINGS issue exists in semaphores, but this is fixed by the
     remaining semaphore patches which Rodrigo will resubmit shortly. I'd
     rather see those patches for fixing the problem than fix it here.

     I found this initially for the BSD2 case where on the same platform we
     can have differing rings. AFAICT however this effects many platforms.

     I'd CC stable on this, except I think all the issues have been around
     for multiple releases without bug reports.

     Compile tested only for now.

     Signed-off-by: Ben Widawsky <[3][email protected]>
     ---
      drivers/gpu/drm/i915/i915_gem_context.c | 6 +++---
      drivers/gpu/drm/i915/i915_gpu_error.c   | 2 +-
      drivers/gpu/drm/i915/intel_ringbuffer.h | 2 ++
      3 files changed, 6 insertions(+), 4 deletions(-)

     diff --git a/drivers/gpu/drm/i915/i915_gem_context.c
     b/drivers/gpu/drm/i915/i915_gem_context.c
     index b9bac25..0c044a9 100644
     --- a/drivers/gpu/drm/i915/i915_gem_context.c
     +++ b/drivers/gpu/drm/i915/i915_gem_context.c
     @@ -403,7 +403,7 @@ void i915_gem_context_reset(struct drm_device *dev)

             /* Prevent the hardware from restoring the last context (which
     hung) on
              * the next switch */
     -       for (i = 0; i < I915_NUM_RINGS; i++) {
     +       for (i = 0; i < I915_ACTIVE_RINGS(dev); i++) {
                     struct intel_engine_cs *ring = &dev_priv->ring[i];
                     struct intel_context *dctx = ring->default_context;

     @@ -456,7 +456,7 @@ int i915_gem_context_init(struct drm_device *dev)
             }

             /* NB: RCS will hold a ref for all rings */
     -       for (i = 0; i < I915_NUM_RINGS; i++)
     +       for (i = 0; i < I915_ACTIVE_RINGS(dev); i++)
                     dev_priv->ring[i].default_context = ctx;

             DRM_DEBUG_DRIVER("%s context support initialized\n",
     dev_priv->hw_context_size ? "HW" : "fake");
     @@ -493,7 +493,7 @@ void i915_gem_context_fini(struct drm_device *dev)
                     i915_gem_object_ggtt_unpin(dctx->obj);
             }

     -       for (i = 0; i < I915_NUM_RINGS; i++) {
     +       for (i = 0; i < I915_ACTIVE_RINGS(dev); i++) {
                     struct intel_engine_cs *ring = &dev_priv->ring[i];

                     if (ring->last_context)
     diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c
     b/drivers/gpu/drm/i915/i915_gpu_error.c
     index 86362de..6e5250d 100644
     --- a/drivers/gpu/drm/i915/i915_gpu_error.c
     +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
     @@ -848,7 +848,7 @@ static uint32_t i915_error_generate_code(struct
     drm_i915_private *dev_priv,
              * synchronization commands which almost always appear in the
     case
              * strictly a client bug. Use instdone to differentiate those
     some.
              */
     -       for (i = 0; i < I915_NUM_RINGS; i++) {
     +       for (i = 0; i < I915_ACTIVE_RINGS(dev_priv->dev); i++) {
                     if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
                             if (ring_id)
                                     *ring_id = i;
     diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h
     b/drivers/gpu/drm/i915/intel_ringbuffer.h
Quoted 8 lines
     index e72017b..67e2919 100644
     --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
     +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
     @@ -90,6 +90,8 @@ struct  intel_engine_cs {
             } id;
      #define I915_NUM_RINGS 5
      #define LAST_USER_RING (VECS + 1)
     +#define I915_ACTIVE_RINGS(dev) hweight8(INTEL_INFO(dev)->ring_mask)

What does the popcount of the mask have to do with the validity of the arrays being iterated over in this patch?
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#4 lore raw
Quoted 93 lines · Chris Wilson
Quoted 83 lines · Rodrigo Vivi
   Reviewed-by: Rodrigo Vivi <[1][email protected]>

   On Fri, Jun 27, 2014 at 3:09 PM, Ben Widawsky
   <[2][email protected]> wrote:

     There are some cases in the code where we need to know how many rings to
     iterate over, but cannot use for_each_ring(). These are always error
     cases
     which happen either before ring setup, or after ring teardown (or
     reset).

     Note, a NUM_RINGS issue exists in semaphores, but this is fixed by the
     remaining semaphore patches which Rodrigo will resubmit shortly. I'd
     rather see those patches for fixing the problem than fix it here.

     I found this initially for the BSD2 case where on the same platform we
     can have differing rings. AFAICT however this effects many platforms.

     I'd CC stable on this, except I think all the issues have been around
     for multiple releases without bug reports.

     Compile tested only for now.

     Signed-off-by: Ben Widawsky <[3][email protected]>
     ---
      drivers/gpu/drm/i915/i915_gem_context.c | 6 +++---
      drivers/gpu/drm/i915/i915_gpu_error.c   | 2 +-
      drivers/gpu/drm/i915/intel_ringbuffer.h | 2 ++
      3 files changed, 6 insertions(+), 4 deletions(-)

     diff --git a/drivers/gpu/drm/i915/i915_gem_context.c
     b/drivers/gpu/drm/i915/i915_gem_context.c
     index b9bac25..0c044a9 100644
     --- a/drivers/gpu/drm/i915/i915_gem_context.c
     +++ b/drivers/gpu/drm/i915/i915_gem_context.c
     @@ -403,7 +403,7 @@ void i915_gem_context_reset(struct drm_device *dev)

             /* Prevent the hardware from restoring the last context (which
     hung) on
              * the next switch */
     -       for (i = 0; i < I915_NUM_RINGS; i++) {
     +       for (i = 0; i < I915_ACTIVE_RINGS(dev); i++) {
                     struct intel_engine_cs *ring = &dev_priv->ring[i];
                     struct intel_context *dctx = ring->default_context;

     @@ -456,7 +456,7 @@ int i915_gem_context_init(struct drm_device *dev)
             }

             /* NB: RCS will hold a ref for all rings */
     -       for (i = 0; i < I915_NUM_RINGS; i++)
     +       for (i = 0; i < I915_ACTIVE_RINGS(dev); i++)
                     dev_priv->ring[i].default_context = ctx;

             DRM_DEBUG_DRIVER("%s context support initialized\n",
     dev_priv->hw_context_size ? "HW" : "fake");
     @@ -493,7 +493,7 @@ void i915_gem_context_fini(struct drm_device *dev)
                     i915_gem_object_ggtt_unpin(dctx->obj);
             }

     -       for (i = 0; i < I915_NUM_RINGS; i++) {
     +       for (i = 0; i < I915_ACTIVE_RINGS(dev); i++) {
                     struct intel_engine_cs *ring = &dev_priv->ring[i];

                     if (ring->last_context)
     diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c
     b/drivers/gpu/drm/i915/i915_gpu_error.c
     index 86362de..6e5250d 100644
     --- a/drivers/gpu/drm/i915/i915_gpu_error.c
     +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
     @@ -848,7 +848,7 @@ static uint32_t i915_error_generate_code(struct
     drm_i915_private *dev_priv,
              * synchronization commands which almost always appear in the
     case
              * strictly a client bug. Use instdone to differentiate those
     some.
              */
     -       for (i = 0; i < I915_NUM_RINGS; i++) {
     +       for (i = 0; i < I915_ACTIVE_RINGS(dev_priv->dev); i++) {
                     if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
                             if (ring_id)
                                     *ring_id = i;
     diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h
     b/drivers/gpu/drm/i915/intel_ringbuffer.h
Quoted 8 lines
     index e72017b..67e2919 100644
     --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
     +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
     @@ -90,6 +90,8 @@ struct  intel_engine_cs {
             } id;
      #define I915_NUM_RINGS 5
      #define LAST_USER_RING (VECS + 1)
     +#define I915_ACTIVE_RINGS(dev) hweight8(INTEL_INFO(dev)->ring_mask)
Quoted 3 lines

What does the popcount of the mask have to do with the validity of the arrays being iterated over in this patch?
-Chris

The popcount of the mask represents the number of rings available on the specific SKU, as opposed to the total number of rings on any SKU ever. It is not always correct to iterate on all rings in the system. Please note, the patch is incomplete. I have a couple of other, perhaps more interesting, cases which I've missed.

Signature · 5 lines
Ben Widawsky, Intel Open Source Technology Center
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#5 lore raw
Quoted 33 lines · Ben Widawsky
Quoted 28 lines · Chris Wilson
Quoted 19 lines
     diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c
     b/drivers/gpu/drm/i915/i915_gpu_error.c
     index 86362de..6e5250d 100644
     --- a/drivers/gpu/drm/i915/i915_gpu_error.c
     +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
     @@ -848,7 +848,7 @@ static uint32_t i915_error_generate_code(struct
     drm_i915_private *dev_priv,
              * synchronization commands which almost always appear in the
     case
              * strictly a client bug. Use instdone to differentiate those
     some.
              */
     -       for (i = 0; i < I915_NUM_RINGS; i++) {
     +       for (i = 0; i < I915_ACTIVE_RINGS(dev_priv->dev); i++) {
                     if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
                             if (ring_id)
                                     *ring_id = i;
     diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h
     b/drivers/gpu/drm/i915/intel_ringbuffer.h
Quoted 8 lines
     index e72017b..67e2919 100644
     --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
     +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
     @@ -90,6 +90,8 @@ struct  intel_engine_cs {
             } id;
      #define I915_NUM_RINGS 5
      #define LAST_USER_RING (VECS + 1)
     +#define I915_ACTIVE_RINGS(dev) hweight8(INTEL_INFO(dev)->ring_mask)
Quoted 3 lines

What does the popcount of the mask have to do with the validity of the arrays being iterated over in this patch?
-Chris

Quoted 5 lines

The popcount of the mask represents the number of rings available on the specific SKU, as opposed to the total number of rings on any SKU ever. It is not always correct to iterate on all rings in the system. Please note, the patch is incomplete. I have a couple of other, perhaps more interesting, cases which I've missed.

You still iterate over holes in the ring mask, and the iteration here is over a completely different array, not rings.
-Chris

Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
#6 lore raw
Quoted 40 lines · Chris Wilson
Quoted 33 lines · Ben Widawsky
Quoted 28 lines · Chris Wilson
Quoted 19 lines
     diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c
     b/drivers/gpu/drm/i915/i915_gpu_error.c
     index 86362de..6e5250d 100644
     --- a/drivers/gpu/drm/i915/i915_gpu_error.c
     +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
     @@ -848,7 +848,7 @@ static uint32_t i915_error_generate_code(struct
     drm_i915_private *dev_priv,
              * synchronization commands which almost always appear in the
     case
              * strictly a client bug. Use instdone to differentiate those
     some.
              */
     -       for (i = 0; i < I915_NUM_RINGS; i++) {
     +       for (i = 0; i < I915_ACTIVE_RINGS(dev_priv->dev); i++) {
                     if (error->ring[i].hangcheck_action == HANGCHECK_HUNG) {
                             if (ring_id)
                                     *ring_id = i;
     diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h
     b/drivers/gpu/drm/i915/intel_ringbuffer.h
Quoted 8 lines
     index e72017b..67e2919 100644
     --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
     +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
     @@ -90,6 +90,8 @@ struct  intel_engine_cs {
             } id;
      #define I915_NUM_RINGS 5
      #define LAST_USER_RING (VECS + 1)
     +#define I915_ACTIVE_RINGS(dev) hweight8(INTEL_INFO(dev)->ring_mask)
Quoted 3 lines

What does the popcount of the mask have to do with the validity of the arrays being iterated over in this patch?
-Chris

Quoted 5 lines

The popcount of the mask represents the number of rings available on the specific SKU, as opposed to the total number of rings on any SKU ever. It is not always correct to iterate on all rings in the system. Please note, the patch is incomplete. I have a couple of other, perhaps more interesting, cases which I've missed.

Quoted 3 lines

You still iterate over holes in the ring mask, and the iteration here is over a completely different array, not rings.
-Chris

For the holes, I mentioned that in the commit message of the yet to be submitted v2; it's not really an issue in the way things are today. When/if we add a new ring, it will be.

What you're asking for has already been submitted multiple times with seemingly no traction. I do realize the fixes (with my v2) are due to bugs introduced in patches I've not yet submitted, so I think for that reason, it's fair to drop this patch.

I'd rather the other patch get in (for_each_active_ring), but it's tied up with execlists atm, and I continue to think this is a useful way to iterate over the rings in error conditions and during reset.

As for your second point, assuming it's the code above, I don't quite follow what you mean. Error code generation shouldn't be based upon inactive rings. As for whether it changes any of the functionality, it does not - but that wasn't the point of that hunk.

Quoted 2 lines
Signature · 1 line
Chris Wilson, Intel Open Source Technology Centre
Signature · 5 lines
Ben Widawsky, Intel Open Source Technology Center
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#7 lore raw

---------------------------------------------------------------------
Intel Corporation (UK) Limited
Registered No. 1134945 (England)
Registered Office: Pipers Way, Swindon SN3 1RJ
VAT No: 860 2173 47

Quoted 61 lines

-----Original Message-----
From: Intel-gfx [mailto:[email protected]] On Behalf Of Ben Widawsky
Sent: Saturday, June 28, 2014 10:56 PM
To: Chris Wilson; Rodrigo Vivi; Widawsky, Benjamin; Intel GFX
Subject: Re: [Intel-gfx] [PATCH] drm/i915: Fix some NUM_RING iterators

Quoted 8 lines · Chris Wilson
Quoted 7 lines · Ben Widawsky
Quoted 6 lines · Chris Wilson
Quoted 6 lines
     diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c
     b/drivers/gpu/drm/i915/i915_gpu_error.c
     index 86362de..6e5250d 100644
     --- a/drivers/gpu/drm/i915/i915_gpu_error.c
     +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
     @@ -848,7 +848,7 @@ static uint32_t

i915_error_generate_code(struct

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

drm_i915_private *dev_priv,
         * synchronization commands which almost always appear in

the

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

case
         * strictly a client bug. Use instdone to differentiate those some.
         */
-       for (i = 0; i < I915_NUM_RINGS; i++) {
+       for (i = 0; i < I915_ACTIVE_RINGS(dev_priv->dev); i++) { if (error->ring[i].hangcheck_action ==

HANGCHECK_HUNG) {

Quoted 14 lines
Quoted 14 lines
Quoted 14 lines
Quoted 13 lines
                             if (ring_id)
                                     *ring_id = i;
     diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h
     b/drivers/gpu/drm/i915/intel_ringbuffer.h
     index e72017b..67e2919 100644
     --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
     +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
     @@ -90,6 +90,8 @@ struct  intel_engine_cs {
             } id;
      #define I915_NUM_RINGS 5
      #define LAST_USER_RING (VECS + 1)
     +#define I915_ACTIVE_RINGS(dev)

hweight8(INTEL_INFO(dev)->ring_mask)

Quoted 6 lines
Quoted 6 lines
Quoted 3 lines

What does the popcount of the mask have to do with the validity of the arrays being iterated over in this patch?
-Chris

The popcount of the mask represents the number of rings available on the specific SKU, as opposed to the total number of rings on any SKU

ever.

Quoted 7 lines
Quoted 3 lines

It is not always correct to iterate on all rings in the system. Please note, the patch is incomplete. I have a couple of other, perhaps more interesting, cases which I've missed.

You still iterate over holes in the ring mask, and the iteration here is over a completely different array, not rings.
-Chris

For the holes, I mentioned that in the commit message of the yet to be submitted v2; it's not really an issue in the way things are today. When/if we add a new ring, it will be.

Quoted 8 lines

What you're asking for has already been submitted multiple times with seemingly no traction. I do realize the fixes (with my v2) are due to bugs introduced in patches I've not yet submitted, so I think for that reason, it's fair to drop this patch.

I'd rather the other patch get in (for_each_active_ring), but it's tied up with execlists atm, and I continue to think this is a useful way to iterate over the rings in error conditions and during reset.

I dropped that patch, since it received some resistance and I couldn´t really justify it on the Execlists series anymore (on the latest versions we don´t introduce new for i < I915_NUM_RINGS). I imagine the patch could be sent again as a standalone?

-- Oscar
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

#8 lore raw
Quoted 74 lines · Oscar

---------------------------------------------------------------------
Intel Corporation (UK) Limited
Registered No. 1134945 (England)
Registered Office: Pipers Way, Swindon SN3 1RJ
VAT No: 860 2173 47

Quoted 66 lines

-----Original Message-----
From: Intel-gfx [mailto:[email protected]] On Behalf Of Ben Widawsky
Sent: Saturday, June 28, 2014 10:56 PM
To: Chris Wilson; Rodrigo Vivi; Widawsky, Benjamin; Intel GFX
Subject: Re: [Intel-gfx] [PATCH] drm/i915: Fix some NUM_RING iterators

Quoted 8 lines · Chris Wilson
Quoted 7 lines · Ben Widawsky
Quoted 6 lines · Chris Wilson
Quoted 6 lines
     diff --git a/drivers/gpu/drm/i915/i915_gpu_error.c
     b/drivers/gpu/drm/i915/i915_gpu_error.c
     index 86362de..6e5250d 100644
     --- a/drivers/gpu/drm/i915/i915_gpu_error.c
     +++ b/drivers/gpu/drm/i915/i915_gpu_error.c
     @@ -848,7 +848,7 @@ static uint32_t

i915_error_generate_code(struct

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

drm_i915_private *dev_priv,
         * synchronization commands which almost always appear in

the

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

case
         * strictly a client bug. Use instdone to differentiate those some.
         */
-       for (i = 0; i < I915_NUM_RINGS; i++) {
+       for (i = 0; i < I915_ACTIVE_RINGS(dev_priv->dev); i++) { if (error->ring[i].hangcheck_action ==

HANGCHECK_HUNG) {

Quoted 14 lines
Quoted 14 lines
Quoted 14 lines
Quoted 13 lines
                             if (ring_id)
                                     *ring_id = i;
     diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h
     b/drivers/gpu/drm/i915/intel_ringbuffer.h
     index e72017b..67e2919 100644
     --- a/drivers/gpu/drm/i915/intel_ringbuffer.h
     +++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
     @@ -90,6 +90,8 @@ struct  intel_engine_cs {
             } id;
      #define I915_NUM_RINGS 5
      #define LAST_USER_RING (VECS + 1)
     +#define I915_ACTIVE_RINGS(dev)

hweight8(INTEL_INFO(dev)->ring_mask)

Quoted 6 lines
Quoted 6 lines
Quoted 3 lines

What does the popcount of the mask have to do with the validity of the arrays being iterated over in this patch?
-Chris

The popcount of the mask represents the number of rings available on the specific SKU, as opposed to the total number of rings on any SKU

ever.

Quoted 7 lines
Quoted 3 lines

It is not always correct to iterate on all rings in the system. Please note, the patch is incomplete. I have a couple of other, perhaps more interesting, cases which I've missed.

You still iterate over holes in the ring mask, and the iteration here is over a completely different array, not rings.
-Chris

For the holes, I mentioned that in the commit message of the yet to be submitted v2; it's not really an issue in the way things are today. When/if we add a new ring, it will be.

What you're asking for has already been submitted multiple times with seemingly no traction. I do realize the fixes (with my v2) are due to bugs introduced in patches I've not yet submitted, so I think for that reason, it's fair to drop this patch.

Quoted 8 lines
Quoted 3 lines

I'd rather the other patch get in (for_each_active_ring), but it's tied up with execlists atm, and I continue to think this is a useful way to iterate over the rings in error conditions and during reset.

I dropped that patch, since it received some resistance and I couldn´t really justify it on the Execlists series anymore (on the latest versions we don´t introduce new for i < I915_NUM_RINGS). I imagine the patch could be sent again as a standalone?

With Chris' patch to no longer tear down ring structures over reset/system suspend we should be able to always use ring_for_each. If not that means we still have some fun to look at.

In any case I'm always happy to merge such drive-by cleanup patches, no need to have a big patch series to justify it. Well as long as it's indeed a step forward, which occasionally is a contentions topic ... -Daniel

Signature · 3 lines
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch