drm/i915/pmu: avoid an maybe-uninitialized warning
authorArnd Bergmann <arnd@arndb.de>
Wed, 27 May 2020 14:05:08 +0000 (16:05 +0200)
committerJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
Mon, 15 Jun 2020 09:36:02 +0000 (12:36 +0300)
Conditional spinlocks make it hard for gcc and for lockdep to
follow the code flow. This one causes a warning with at least
gcc-9 and higher:

In file included from include/linux/irq.h:14,
                 from drivers/gpu/drm/i915/i915_pmu.c:7:
drivers/gpu/drm/i915/i915_pmu.c: In function 'i915_sample':
include/linux/spinlock.h:289:3: error: 'flags' may be used uninitialized in this function [-Werror=maybe-uninitialized]
  289 |   _raw_spin_unlock_irqrestore(lock, flags); \
      |   ^~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/gpu/drm/i915/i915_pmu.c:288:17: note: 'flags' was declared here
  288 |   unsigned long flags;
      |                 ^~~~~

Split out the part between the locks into a separate function
for readability and to let the compiler figure out what the
logic actually is.

Fixes: d79e1bd676f0 ("drm/i915/pmu: Only use exclusive mmio access for gen7")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20200527140526.1458215-1-arnd@arndb.de
(cherry picked from commit 6ec81b82732e2b4a5ac0853fd33919ff1ca94238)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
drivers/gpu/drm/i915/i915_pmu.c

index e991a70..962ded9 100644 (file)
@@ -269,12 +269,48 @@ static bool exclusive_mmio_access(const struct drm_i915_private *i915)
        return IS_GEN(i915, 7);
 }
 
+static void engine_sample(struct intel_engine_cs *engine, unsigned int period_ns)
+{
+       struct intel_engine_pmu *pmu = &engine->pmu;
+       bool busy;
+       u32 val;
+
+       val = ENGINE_READ_FW(engine, RING_CTL);
+       if (val == 0) /* powerwell off => engine idle */
+               return;
+
+       if (val & RING_WAIT)
+               add_sample(&pmu->sample[I915_SAMPLE_WAIT], period_ns);
+       if (val & RING_WAIT_SEMAPHORE)
+               add_sample(&pmu->sample[I915_SAMPLE_SEMA], period_ns);
+
+       /* No need to sample when busy stats are supported. */
+       if (intel_engine_supports_stats(engine))
+               return;
+
+       /*
+        * While waiting on a semaphore or event, MI_MODE reports the
+        * ring as idle. However, previously using the seqno, and with
+        * execlists sampling, we account for the ring waiting as the
+        * engine being busy. Therefore, we record the sample as being
+        * busy if either waiting or !idle.
+        */
+       busy = val & (RING_WAIT_SEMAPHORE | RING_WAIT);
+       if (!busy) {
+               val = ENGINE_READ_FW(engine, RING_MI_MODE);
+               busy = !(val & MODE_IDLE);
+       }
+       if (busy)
+               add_sample(&pmu->sample[I915_SAMPLE_BUSY], period_ns);
+}
+
 static void
 engines_sample(struct intel_gt *gt, unsigned int period_ns)
 {
        struct drm_i915_private *i915 = gt->i915;
        struct intel_engine_cs *engine;
        enum intel_engine_id id;
+       unsigned long flags;
 
        if ((i915->pmu.enable & ENGINE_SAMPLE_MASK) == 0)
                return;
@@ -283,53 +319,17 @@ engines_sample(struct intel_gt *gt, unsigned int period_ns)
                return;
 
        for_each_engine(engine, gt, id) {
-               struct intel_engine_pmu *pmu = &engine->pmu;
-               spinlock_t *mmio_lock;
-               unsigned long flags;
-               bool busy;
-               u32 val;
-
                if (!intel_engine_pm_get_if_awake(engine))
                        continue;
 
-               mmio_lock = NULL;
-               if (exclusive_mmio_access(i915))
-                       mmio_lock = &engine->uncore->lock;
-
-               if (unlikely(mmio_lock))
-                       spin_lock_irqsave(mmio_lock, flags);
-
-               val = ENGINE_READ_FW(engine, RING_CTL);
-               if (val == 0) /* powerwell off => engine idle */
-                       goto skip;
-
-               if (val & RING_WAIT)
-                       add_sample(&pmu->sample[I915_SAMPLE_WAIT], period_ns);
-               if (val & RING_WAIT_SEMAPHORE)
-                       add_sample(&pmu->sample[I915_SAMPLE_SEMA], period_ns);
-
-               /* No need to sample when busy stats are supported. */
-               if (intel_engine_supports_stats(engine))
-                       goto skip;
-
-               /*
-                * While waiting on a semaphore or event, MI_MODE reports the
-                * ring as idle. However, previously using the seqno, and with
-                * execlists sampling, we account for the ring waiting as the
-                * engine being busy. Therefore, we record the sample as being
-                * busy if either waiting or !idle.
-                */
-               busy = val & (RING_WAIT_SEMAPHORE | RING_WAIT);
-               if (!busy) {
-                       val = ENGINE_READ_FW(engine, RING_MI_MODE);
-                       busy = !(val & MODE_IDLE);
+               if (exclusive_mmio_access(i915)) {
+                       spin_lock_irqsave(&engine->uncore->lock, flags);
+                       engine_sample(engine, period_ns);
+                       spin_unlock_irqrestore(&engine->uncore->lock, flags);
+               } else {
+                       engine_sample(engine, period_ns);
                }
-               if (busy)
-                       add_sample(&pmu->sample[I915_SAMPLE_BUSY], period_ns);
 
-skip:
-               if (unlikely(mmio_lock))
-                       spin_unlock_irqrestore(mmio_lock, flags);
                intel_engine_pm_put_async(engine);
        }
 }