drm/i915: protect RPS/RC6 related accesses (including PCU) with a new mutex
This allows the power related code to run independently of the rest of
the pipeline, extending the resume and init time improvements into
userspace, which would otherwise have been blocked on the struct mutex
if we were doing PCU communication.
v2: Also convert the locking for the rps sysfs interface.
Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org> (v1)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
index 598987f..09cd7d7 100644
--- a/drivers/gpu/drm/i915/i915_debugfs.c
+++ b/drivers/gpu/drm/i915/i915_debugfs.c
@@ -1280,7 +1280,7 @@
return 0;
}
- ret = mutex_lock_interruptible(&dev->struct_mutex);
+ ret = mutex_lock_interruptible(&dev_priv->rps.hw_lock);
if (ret)
return ret;
@@ -1296,7 +1296,7 @@
seq_printf(m, "%d\t\t%d\n", gpu_freq * GT_FREQUENCY_MULTIPLIER, ia_freq * 100);
}
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
return 0;
}
@@ -1713,13 +1713,13 @@
if (!(IS_GEN6(dev) || IS_GEN7(dev)))
return -ENODEV;
- ret = mutex_lock_interruptible(&dev->struct_mutex);
+ ret = mutex_lock_interruptible(&dev_priv->rps.hw_lock);
if (ret)
return ret;
len = snprintf(buf, sizeof(buf),
"max freq: %d\n", dev_priv->rps.max_delay * GT_FREQUENCY_MULTIPLIER);
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
if (len > sizeof(buf))
len = sizeof(buf);
@@ -1754,7 +1754,7 @@
DRM_DEBUG_DRIVER("Manually setting max freq to %d\n", val);
- ret = mutex_lock_interruptible(&dev->struct_mutex);
+ ret = mutex_lock_interruptible(&dev_priv->rps.hw_lock);
if (ret)
return ret;
@@ -1764,7 +1764,7 @@
dev_priv->rps.max_delay = val / GT_FREQUENCY_MULTIPLIER;
gen6_set_rps(dev, val / GT_FREQUENCY_MULTIPLIER);
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
return cnt;
}
@@ -1789,13 +1789,13 @@
if (!(IS_GEN6(dev) || IS_GEN7(dev)))
return -ENODEV;
- ret = mutex_lock_interruptible(&dev->struct_mutex);
+ ret = mutex_lock_interruptible(&dev_priv->rps.hw_lock);
if (ret)
return ret;
len = snprintf(buf, sizeof(buf),
"min freq: %d\n", dev_priv->rps.min_delay * GT_FREQUENCY_MULTIPLIER);
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
if (len > sizeof(buf))
len = sizeof(buf);
@@ -1828,7 +1828,7 @@
DRM_DEBUG_DRIVER("Manually setting min freq to %d\n", val);
- ret = mutex_lock_interruptible(&dev->struct_mutex);
+ ret = mutex_lock_interruptible(&dev_priv->rps.hw_lock);
if (ret)
return ret;
@@ -1838,7 +1838,7 @@
dev_priv->rps.min_delay = val / GT_FREQUENCY_MULTIPLIER;
gen6_set_rps(dev, val / GT_FREQUENCY_MULTIPLIER);
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
return cnt;
}
diff --git a/drivers/gpu/drm/i915/i915_dma.c b/drivers/gpu/drm/i915/i915_dma.c
index 530cf90..4f3c933 100644
--- a/drivers/gpu/drm/i915/i915_dma.c
+++ b/drivers/gpu/drm/i915/i915_dma.c
@@ -1625,6 +1625,8 @@
spin_lock_init(&dev_priv->rps.lock);
spin_lock_init(&dev_priv->dpio_lock);
+ mutex_init(&dev_priv->rps.hw_lock);
+
if (IS_IVYBRIDGE(dev) || IS_HASWELL(dev))
dev_priv->num_pipe = 3;
else if (IS_MOBILE(dev) || !IS_GEN2(dev))
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index 8db7bb9..c4339c2 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -564,6 +564,12 @@
u8 max_delay;
struct delayed_work delayed_resume_work;
+
+ /*
+ * Protects RPS/RC6 register access and PCU communication.
+ * Must be taken after struct_mutex if nested.
+ */
+ struct mutex hw_lock;
};
struct intel_ilk_power_mgmt {
diff --git a/drivers/gpu/drm/i915/i915_irq.c b/drivers/gpu/drm/i915/i915_irq.c
index 2c401275..2604867 100644
--- a/drivers/gpu/drm/i915/i915_irq.c
+++ b/drivers/gpu/drm/i915/i915_irq.c
@@ -378,7 +378,7 @@
if ((pm_iir & GEN6_PM_DEFERRED_EVENTS) == 0)
return;
- mutex_lock(&dev_priv->dev->struct_mutex);
+ mutex_lock(&dev_priv->rps.hw_lock);
if (pm_iir & GEN6_PM_RP_UP_THRESHOLD)
new_delay = dev_priv->rps.cur_delay + 1;
@@ -393,7 +393,7 @@
gen6_set_rps(dev_priv->dev, new_delay);
}
- mutex_unlock(&dev_priv->dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
}
diff --git a/drivers/gpu/drm/i915/i915_sysfs.c b/drivers/gpu/drm/i915/i915_sysfs.c
index 9700ae9..3bf51d5 100644
--- a/drivers/gpu/drm/i915/i915_sysfs.c
+++ b/drivers/gpu/drm/i915/i915_sysfs.c
@@ -211,12 +211,9 @@
struct drm_i915_private *dev_priv = dev->dev_private;
int ret;
- ret = i915_mutex_lock_interruptible(dev);
- if (ret)
- return ret;
-
+ mutex_lock(&dev_priv->rps.hw_lock);
ret = dev_priv->rps.cur_delay * GT_FREQUENCY_MULTIPLIER;
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
return snprintf(buf, PAGE_SIZE, "%d", ret);
}
@@ -228,12 +225,9 @@
struct drm_i915_private *dev_priv = dev->dev_private;
int ret;
- ret = i915_mutex_lock_interruptible(dev);
- if (ret)
- return ret;
-
+ mutex_lock(&dev_priv->rps.hw_lock);
ret = dev_priv->rps.max_delay * GT_FREQUENCY_MULTIPLIER;
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
return snprintf(buf, PAGE_SIZE, "%d", ret);
}
@@ -254,16 +248,14 @@
val /= GT_FREQUENCY_MULTIPLIER;
- ret = mutex_lock_interruptible(&dev->struct_mutex);
- if (ret)
- return ret;
+ mutex_lock(&dev_priv->rps.hw_lock);
rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
hw_max = (rp_state_cap & 0xff);
hw_min = ((rp_state_cap & 0xff0000) >> 16);
if (val < hw_min || val > hw_max || val < dev_priv->rps.min_delay) {
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
return -EINVAL;
}
@@ -272,7 +264,7 @@
dev_priv->rps.max_delay = val;
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
return count;
}
@@ -284,12 +276,9 @@
struct drm_i915_private *dev_priv = dev->dev_private;
int ret;
- ret = i915_mutex_lock_interruptible(dev);
- if (ret)
- return ret;
-
+ mutex_lock(&dev_priv->rps.hw_lock);
ret = dev_priv->rps.min_delay * GT_FREQUENCY_MULTIPLIER;
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
return snprintf(buf, PAGE_SIZE, "%d", ret);
}
@@ -310,16 +299,14 @@
val /= GT_FREQUENCY_MULTIPLIER;
- ret = mutex_lock_interruptible(&dev->struct_mutex);
- if (ret)
- return ret;
+ mutex_lock(&dev_priv->rps.hw_lock);
rp_state_cap = I915_READ(GEN6_RP_STATE_CAP);
hw_max = (rp_state_cap & 0xff);
hw_min = ((rp_state_cap & 0xff0000) >> 16);
if (val < hw_min || val > hw_max || val > dev_priv->rps.max_delay) {
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
return -EINVAL;
}
@@ -328,7 +315,7 @@
dev_priv->rps.min_delay = val;
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
return count;
diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
index 169b416..8f15616 100644
--- a/drivers/gpu/drm/i915/intel_pm.c
+++ b/drivers/gpu/drm/i915/intel_pm.c
@@ -2323,7 +2323,7 @@
struct drm_i915_private *dev_priv = dev->dev_private;
u32 limits = gen6_rps_limits(dev_priv, &val);
- WARN_ON(!mutex_is_locked(&dev->struct_mutex));
+ WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
WARN_ON(val > dev_priv->rps.max_delay);
WARN_ON(val < dev_priv->rps.min_delay);
@@ -2409,7 +2409,7 @@
int rc6_mode;
int i, ret;
- WARN_ON(!mutex_is_locked(&dev->struct_mutex));
+ WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
/* Here begins a magic sequence of register writes to enable
* auto-downclocking.
@@ -2550,7 +2550,7 @@
int gpu_freq, ia_freq, max_ia_freq;
int scaling_factor = 180;
- WARN_ON(!mutex_is_locked(&dev->struct_mutex));
+ WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
max_ia_freq = cpufreq_quick_get_max(0);
/*
@@ -3311,7 +3311,9 @@
ironlake_disable_rc6(dev);
} else if (INTEL_INFO(dev)->gen >= 6 && !IS_VALLEYVIEW(dev)) {
cancel_delayed_work_sync(&dev_priv->rps.delayed_resume_work);
+ mutex_lock(&dev_priv->rps.hw_lock);
gen6_disable_rps(dev);
+ mutex_unlock(&dev_priv->rps.hw_lock);
}
}
@@ -3322,10 +3324,10 @@
rps.delayed_resume_work.work);
struct drm_device *dev = dev_priv->dev;
- mutex_lock(&dev->struct_mutex);
+ mutex_lock(&dev_priv->rps.hw_lock);
gen6_enable_rps(dev);
gen6_update_ring_freq(dev);
- mutex_unlock(&dev->struct_mutex);
+ mutex_unlock(&dev_priv->rps.hw_lock);
}
void intel_enable_gt_powersave(struct drm_device *dev)
@@ -4245,7 +4247,7 @@
int sandybridge_pcode_read(struct drm_i915_private *dev_priv, u8 mbox, u32 *val)
{
- WARN_ON(!mutex_is_locked(&dev_priv->dev->struct_mutex));
+ WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
DRM_DEBUG_DRIVER("warning: pcode (read) mailbox access failed\n");
@@ -4269,7 +4271,7 @@
int sandybridge_pcode_write(struct drm_i915_private *dev_priv, u8 mbox, u32 val)
{
- WARN_ON(!mutex_is_locked(&dev_priv->dev->struct_mutex));
+ WARN_ON(!mutex_is_locked(&dev_priv->rps.hw_lock));
if (I915_READ(GEN6_PCODE_MAILBOX) & GEN6_PCODE_READY) {
DRM_DEBUG_DRIVER("warning: pcode (write) mailbox access failed\n");