Skip to contentSkip to conversation

drm/i915: Initialize pm_intr_keep during intel_irq_init for GuC

PATCH In review Sagar Arun Kamble 3 participants 4 messages Jan 20, 2017

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

PATCH 1/2 drm/i915: Initialize pm_intr_keep during intel_irq_init for GuC

Driver needs to ensure that it doesn't mask the PM interrupts, which are unmasked/needed by GuC firmware. For that, Driver maintains a bitmask of interrupts to be kept unmasked, pm_intr_keep.

pm_intr_keep was determined across GuC load. GuC gets loaded in different scenarios and it is not going to change the pm_intr_keep so this patch moves its setup to intel_irq_init.

This patch fixes incorrect RPS masking and solves bunch of SKL GT3 performance regressions as Host RPS was taking frequency down when GuC is enabled. This was happening due to signed last_adj that becomes negative after 32 UP interrupts and makes cur_freq=min starting the ramp again. Next patch in the series avoids such frequency transitions.

Cc: Szwichtenberg, Radoslaw <[email protected]>
Cc: Tomi Sarvela <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97017

---
 drivers/gpu/drm/i915/i915_irq.c         | 24 ++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_reg.h         |  3 ++-
 drivers/gpu/drm/i915/intel_guc_loader.c | 26 --------------------------
 3 files changed, 26 insertions(+), 27 deletions(-)
drivers/gpu/drm/i915/i915_irq.c +24 -0
index ce5663d..95b4047 100644
@@ -4193,6 +4193,30 @@ void intel_irq_init(struct drm_i915_private *dev_priv)
 	if (INTEL_INFO(dev_priv)->gen >= 8)
 		dev_priv->rps.pm_intr_keep |= GEN8_PMINTR_REDIRECT_TO_GUC;
 
+	/*
+	 * The REDIRECT_TO_GUC bit of the PMINTRMSK register directs all
+	 * (unmasked) PM interrupts to the GuC. All other bits of this
+	 * register *disable* generation of a specific interrupt.
+	 *
+	 * 'pm_intr_keep' indicates bits that are NOT to be set when
+	 * writing to the PM interrupt mask register, i.e. interrupts
+	 * that must not be disabled.
+	 *
+	 * If the GuC is handling these interrupts, then we must not let
+	 * the PM code disable ANY interrupt that the GuC is expecting.
+	 * So for each ENABLED (0) bit in this register, we must SET the
+	 * bit in pm_intr_keep so that it's left enabled for the GuC.
+	 * GuC needs ARAT expired interrupt unmasked hence it is set in
+	 * pm_intr_keep.
+	 *
+	 * Here we CLEAR REDIRECT_TO_GUC bit in pm_intr_keep, which will
+	 * result in the register bit being left SET!
+	 */
+	if (HAS_GUC_SCHED(dev_priv)) {
+		dev_priv->rps.pm_intr_keep |= ARAT_EXPIRED_INTRMSK;
+		dev_priv->rps.pm_intr_keep &= ~GEN8_PMINTR_REDIRECT_TO_GUC;
+	}
+
 	if (IS_GEN2(dev_priv)) {
 		/* Gen2 doesn't have a hardware frame counter */
 		dev->max_vblank_count = 0;
drivers/gpu/drm/i915/i915_reg.h +2 -1
index 72f9f36..822c147 100644
@@ -7395,7 +7395,8 @@ enum {
 #define VLV_RCEDATA				_MMIO(0xA0BC)
 #define GEN6_RC6pp_THRESHOLD			_MMIO(0xA0C0)
 #define GEN6_PMINTRMSK				_MMIO(0xA168)
-#define   GEN8_PMINTR_REDIRECT_TO_GUC		  (1<<31)
+#define   GEN8_PMINTR_REDIRECT_TO_GUC		(1<<31)
+#define   ARAT_EXPIRED_INTRMSK			(1<<9)
 #define GEN8_MISC_CTRL0				_MMIO(0xA180)
 #define VLV_PWRDWNUPCTL				_MMIO(0xA294)
 #define GEN9_MEDIA_PG_IDLE_HYSTERESIS		_MMIO(0xA0C4)
drivers/gpu/drm/i915/intel_guc_loader.c +0 -26
index 2f1cf9a..0eaa8e4 100644
@@ -114,7 +114,6 @@ static void guc_interrupts_capture(struct drm_i915_private *dev_priv)
 	struct intel_engine_cs *engine;
 	enum intel_engine_id id;
 	int irqs;
-	u32 tmp;
 
 	/* tell all command streamers to forward interrupts (but not vblank) to GuC */
 	irqs = _MASKED_BIT_ENABLE(GFX_INTERRUPT_STEERING);
@@ -128,31 +127,6 @@ static void guc_interrupts_capture(struct drm_i915_private *dev_priv)
 	I915_WRITE(GUC_BCS_RCS_IER, ~irqs);
 	I915_WRITE(GUC_VCS2_VCS1_IER, ~irqs);
 	I915_WRITE(GUC_WD_VECS_IER, ~irqs);
-
-	/*
-	 * The REDIRECT_TO_GUC bit of the PMINTRMSK register directs all
-	 * (unmasked) PM interrupts to the GuC. All other bits of this
-	 * register *disable* generation of a specific interrupt.
-	 *
-	 * 'pm_intr_keep' indicates bits that are NOT to be set when
-	 * writing to the PM interrupt mask register, i.e. interrupts
-	 * that must not be disabled.
-	 *
-	 * If the GuC is handling these interrupts, then we must not let
-	 * the PM code disable ANY interrupt that the GuC is expecting.
-	 * So for each ENABLED (0) bit in this register, we must SET the
-	 * bit in pm_intr_keep so that it's left enabled for the GuC.
-	 *
-	 * OTOH the REDIRECT_TO_GUC bit is initially SET in pm_intr_keep
-	 * (so interrupts go to the DISPLAY unit at first); but here we
-	 * need to CLEAR that bit, which will result in the register bit
-	 * being left SET!
-	 */
-	tmp = I915_READ(GEN6_PMINTRMSK);
-	if (tmp & GEN8_PMINTR_REDIRECT_TO_GUC) {
-		dev_priv->rps.pm_intr_keep |= ~tmp;
-		dev_priv->rps.pm_intr_keep &= ~GEN8_PMINTR_REDIRECT_TO_GUC;
-	}
 }
 
 static u32 get_gttype(struct drm_i915_private *dev_priv)
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw

PATCH 2/2 drm/i915: Set adjustment to zero on Up/Down interrupts if freq is already max/min

When we reach the user's RPS limits, stop requesting an adjustment. Even though we will clamp the requested frequency later, we rely on interrupt masking to disable further adjustments in the same direction. Even though it is unlikely (one scenario is a bug in the driver, another is careful manipulation through the uAPI) if we keep exponentially increasing the adjustment value, it will wrap and cause a negative adjustment.

---
 drivers/gpu/drm/i915/i915_irq.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/gpu/drm/i915/i915_irq.c +6 -0
index 95b4047..eb972fc 100644
@@ -1170,6 +1170,9 @@ static void gen6_pm_rps_work(struct work_struct *work)
 			adj *= 2;
 		else /* CHV needs even encode values */
 			adj = IS_CHERRYVIEW(dev_priv) ? 2 : 1;
+
+		if (new_delay >= dev_priv->rps.max_freq_softlimit)
+			adj = 0;
 		/*
 		 * For better performance, jump directly
 		 * to RPe if we're below it.
@@ -1191,6 +1194,9 @@ static void gen6_pm_rps_work(struct work_struct *work)
 			adj *= 2;
 		else /* CHV needs even encode values */
 			adj = IS_CHERRYVIEW(dev_priv) ? -2 : -1;
+
+		if (new_delay <= dev_priv->rps.min_freq_softlimit)
+			adj = 0;
 	} else { /* unknown event */
 		adj = 0;
 	}
Signature · 6 lines
1.9.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#2 lore raw

Re: [PATCH 2/2] drm/i915: Set adjustment to zero on Up/Down interrupts if freq is already max/min

Quoted 10 lines · Sagar Arun Kamble

When we reach the user's RPS limits, stop requesting an adjustment. Even though we will clamp the requested frequency later, we rely on interrupt masking to disable further adjustments in the same direction. Even though it is unlikely (one scenario is a bug in the driver, another is careful manipulation through the uAPI) if we keep exponentially increasing the adjustment value, it will wrap and cause a negative adjustment.

Grabbed this one because I know it is safe. There's some underlying trouble fouling guc so judging results for patch 1 is tricky. A single run suggests that all guc machines are affected by an unrecoverable hang following that patch - but the same issue seems present without the patch just on isolated machines.
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#3 lore raw

✗ Fi.CI.BAT: failure for series starting with [1/2] drm/i915: Initialize pm_intr_keep during intel_irq_init for GuC

In reply to Sagar Arun Kamble
== Series Details ==

== Series Details ==

Series: series starting with [1/2] drm/i915: Initialize pm_intr_keep during intel_irq_init for GuC
URL : https://patchwork.freedesktop.org/series/18252/
State : failure

== Summary ==

Series 18252v1 Series without cover letter
https://patchwork.freedesktop.org/api/1.0/series/18252/revisions/1/mbox/

Test gem_busy:

        Subgroup basic-busy-default:
                pass       -> FAIL       (fi-ivb-3520m)
        Subgroup basic-hang-default:
                fail       -> PASS       (fi-ivb-3520m)
Test gem_ringfill:
        Subgroup basic-default-hang:
                pass       -> INCOMPLETE (fi-hsw-4770)
fi-bdw-5557u     total:246  pass:232  dwarn:0   dfail:0   fail:0   skip:14 
fi-bsw-n3050     total:246  pass:207  dwarn:0   dfail:0   fail:0   skip:39 
fi-bxt-j4205     total:246  pass:224  dwarn:0   dfail:0   fail:0   skip:22 
fi-bxt-t5700     total:79   pass:66   dwarn:0   dfail:0   fail:0   skip:12 
fi-byt-j1900     total:246  pass:219  dwarn:0   dfail:0   fail:0   skip:27 
fi-byt-n2820     total:246  pass:215  dwarn:0   dfail:0   fail:0   skip:31 
fi-hsw-4770      total:115  pass:106  dwarn:0   dfail:0   fail:0   skip:8  
fi-hsw-4770r     total:246  pass:227  dwarn:0   dfail:0   fail:0   skip:19 
fi-ivb-3520m     total:246  pass:224  dwarn:0   dfail:0   fail:1   skip:21 
fi-ivb-3770      total:246  pass:225  dwarn:0   dfail:0   fail:0   skip:21 
fi-kbl-7500u     total:246  pass:225  dwarn:0   dfail:0   fail:0   skip:21 
fi-skl-6260u     total:246  pass:233  dwarn:0   dfail:0   fail:0   skip:13 
fi-skl-6700hq    total:246  pass:226  dwarn:0   dfail:0   fail:0   skip:20 
fi-skl-6700k     total:246  pass:222  dwarn:3   dfail:0   fail:0   skip:21 
fi-skl-6770hq    total:246  pass:233  dwarn:0   dfail:0   fail:0   skip:13 
fi-snb-2520m     total:246  pass:215  dwarn:0   dfail:0   fail:0   skip:31 
fi-snb-2600      total:246  pass:214  dwarn:0   dfail:0   fail:0   skip:32 

6155f94d32e438a9656de680897060bd2245629a drm-tip: 2017y-01m-19d-18h-24m-49s UTC integration manifest e9264fd drm/i915: Set adjustment to zero on Up/Down interrupts if freq is already max/min bc26a26 drm/i915: Initialize pm_intr_keep during intel_irq_init for GuC

== Logs ==

For more details see: https://intel-gfx-ci.01.org/CI/Patchwork_3552/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx