drm/i915: Ensure a context is loaded before RC6
Subsystems (2)
PATCH 1/1 drm/i915: Ensure a context is loaded before RC6
RC6 works a lot like HW contexts in that when the GPU enters RC6 it saves away the state to a context, and loads it upon wake.
It's to be somewhat expected that BIOS will not set up valid GPU state. As a result, if loading bad state can cause the GPU to get angry, it would make sense then that we need to load state first. There are two ways in which we can do this:
1. Create 3d state in the driver, load it up, then enable RC6. 1b. Reuse a known good state, and just bind objects where needed. Then enable RC6
2. Hold off enabling RC6 until userspace has had a chance to complete batches.
This patch is a bad hack. It suffers two flaws. The first is, if the driver is loaded, but a batch is not submitted/completed, we'll never enter rc6. The other is, it expects userspace to submit a batch with 3d state first. Both of these things are not actual flaws for most users.
Technically, this tactic is required for all platforms, though I am not certain we've seen real failures.
--- drivers/gpu/drm/i915/i915_gem.c | 6 ++++++ drivers/gpu/drm/i915/intel_display.c | 4 ---- 2 files changed, 6 insertions(+), 4 deletions(-)
drivers/gpu/drm/i915/i915_gem.c +6 -0
@@ -2482,6 +2482,7 @@ void i915_gem_reset(struct drm_device *dev)void i915_gem_retire_requests_ring (struct intel_ring_buffer *ring) { +static bool rc6_enabled = false;uint32_t seqno;if (list_empty (&ring->request_list))
@@ -2505,6 +2506,11 @@ i915_gem_retire_requests_ring(struct intel_ring_buffer *ring)if (!i915_seqno_passed (seqno, obj->last_read_seqno))break ; +if (unlikely (!rc6_enabled) && ring->id ==RCS ) { +intel_enable_gt_powersave (ring->dev); + rc6_enabled = true; + } +i915_gem_object_move_to_inactive (obj); }
drivers/gpu/drm/i915/intel_display.c +0 -4
@@ -10982,10 +10982,6 @@ void intel_modeset_init_hw(struct drm_device *dev)intel_init_clock_gating (dev);intel_reset_dpio (dev); - -mutex_lock (&dev->struct_mutex); -intel_enable_gt_powersave (dev); -mutex_unlock (&dev->struct_mutex); }void intel_modeset_suspend_hw (struct drm_device *dev)
Signature · 1 line
1.8.5.3
Quoted 14 lines · Ben Widawsky
RC6 works a lot like HW contexts in that when the GPU enters RC6 it saves away the state to a context, and loads it upon wake.
It's to be somewhat expected that BIOS will not set up valid GPU state. As a result, if loading bad state can cause the GPU to get angry, it would make sense then that we need to load state first. There are two ways in which we can do this:
1. Create 3d state in the driver, load it up, then enable RC6. 1b. Reuse a known good state, and just bind objects where needed. Then enable RC6
2. Hold off enabling RC6 until userspace has had a chance to complete batches.
Quoted 4 lines
This patch is a bad hack. It suffers two flaws. The first is, if the driver is loaded, but a batch is not submitted/completed, we'll never enter rc6. The other is, it expects userspace to submit a batch with 3d state first. Both of these things are not actual flaws for most users.
There is another flaw actually. If the objects for the indirect state gets unbound before we enable RC6, or before we return from RC6.
I think for the case I want to test however, both of these are unlikely
- but it's another reason to not use a patch like this upstream.
Quoted 54 lines
Technically, this tactic is required for all platforms, though I am not certain we've seen real failures.
Cc David E. Box <[email protected]>Cc Kristen Carlson Accardi <[email protected]>Signed-off-by Ben Widawsky <[email protected]>--- drivers/gpu/drm/i915/i915_gem.c | 6 ++++++ drivers/gpu/drm/i915/intel_display.c | 4 ---- 2 files changed, 6 insertions(+), 4 deletions(-)drivers/gpu/drm/i915/i915_gem.c +6
-0@@ -2482,6 +2482,7 @@ void i915_gem_reset(struct drm_device *dev)void i915_gem_retire_requests_ring (struct intel_ring_buffer *ring) { +static bool rc6_enabled = false;uint32_t seqno;if (list_empty (&ring->request_list))@@ -2505,6 +2506,11 @@ i915_gem_retire_requests_ring(struct intel_ring_buffer *ring)if (!i915_seqno_passed (seqno, obj->last_read_seqno))break ; +if (unlikely (!rc6_enabled) && ring->id ==RCS ) { +intel_enable_gt_powersave (ring->dev); + rc6_enabled = true; + } +i915_gem_object_move_to_inactive (obj); }drivers/gpu/drm/i915/intel_display.c +0
-4@@ -10982,10 +10982,6 @@ void intel_modeset_init_hw(struct drm_device *dev)intel_init_clock_gating (dev);intel_reset_dpio (dev); - -mutex_lock (&dev->struct_mutex); -intel_enable_gt_powersave (dev); -mutex_unlock (&dev->struct_mutex); }void intel_modeset_suspend_hw (struct drm_device *dev)Signature · 1 line
1.8.5.3
Signature · 1 line
Ben Widawsky, Intel Open Source Technology Center