drm/i915: IVB missed irqs w/a
Subsystems (2)
PATCH 1/1 drm/i915: IVB missed irqs w/a
Request for testing.
This is the proposed workaround from the HW team for the missed IRQ issue. I do not have a machine to test on, so it's just compile tested. This is very similar to another workaround I implemented, but I never tried it on IVB, and don't recall if I ever mailed those out.
Anyone who is capable of testing this, please undo the previous workaround, and try this patch.
Meanwhile, I'll try to find a similar workaround for SNB.
Thanks.
--- drivers/gpu/drm/i915/i915_irq.c | 15 +++++++++++++++ drivers/gpu/drm/i915/i915_reg.h | 4 ++++ 2 files changed, 19 insertions(+)
drivers/gpu/drm/i915/i915_irq.c +15 -0
@@ -479,11 +479,26 @@ static irqreturn_t ivybridge_irq_handler(DRM_IRQ_ARGS)atomic_inc (&dev_priv->irq_received); +again: +/* Clear out message status bits */ +I915_WRITE (GEN7_CPU_MSG_SENT_STATUS ,0xFFFFFFFF ); +I915_WRITE (GEN7_CPU_MSG_RCVD_STATUS1 ,0xFFFFFFFF ); +I915_WRITE (GEN7_CPU_MSG_RCVD_STATUS2 ,0xFFFFFFFF ); +POSTING_READ (GEN7_CPU_MSG_SENT_STATUS ); +/* disable master interrupt before clearing iir */ de_ier =I915_READ (DEIER );I915_WRITE (DEIER , de_ier & ~DE_MASTER_IRQ_CONTROL );POSTING_READ (DEIER ); +/* Check if any render response activity occurred */ +if ((I915_READ (GEN7_CPU_MSG_RCVD_STATUS1 ) &0xfff1c ) || + (I915_READ (GEN7_CPU_MSG_RCVD_STATUS2 ) &0x7e00 )) { +I915_WRITE (DEIER , de_ier |DE_MASTER_IRQ_CONTROL ); +POSTING_READ (DEIER ); + goto again; + } + de_iir =I915_READ (DEIIR ); gt_iir =I915_READ (GTIIR ); pch_iir =I915_READ (SDEIIR );
drivers/gpu/drm/i915/i915_reg.h +4 -0
@@ -3776,6 +3776,10 @@#define GEN6_RC6pp_THRESHOLD 0xA0C0#define GEN6_PMINTRMSK 0xA168 +#define GEN7_CPU_MSG_SENT_STATUS 0x42030 +#define GEN7_CPU_MSG_RCVD_STATUS1 0x42034 +#define GEN7_CPU_MSG_RCVD_STATUS2 0x42038 +#define GEN6_PMISR 0x44020#define GEN6_PMIMR 0x44024/* rps_lock */ #define GEN6_PMIIR 0x44028
Signature · 1 line
1.7.9.4
Quoted 7 lines · On Mon, 19 Mar 2012 08:48:44 -0700 Ben Widawsky <[email protected]> wrote:
Request for testing.
This is the proposed workaround from the HW team for the missed IRQ issue. I do not have a machine to test on, so it's just compile tested. This is very similar to another workaround I implemented, but I never tried it on IVB, and don't recall if I ever mailed those out.
Quoted 8 lines
Anyone who is capable of testing this, please undo the previous workaround, and try this patch.
Meanwhile, I'll try to find a similar workaround for SNB.
Thanks.
Signed-off-by Ben Widawsky <[email protected]>
So it's no longer clear whether or not we actually want this patch. Furthermore I've found out there is no equivalent SNB patch.