By disabling a power domain asynchronously we can restrict holding a
reference on that power domain to the actual code sequence that
requires the power to be on for the HW access it's doing, by also
avoiding unneeded on-off-on togglings of the power domain (since the
disabling happens with a delay).
One benefit is potential power saving due to the following two reasons:
1. The fact that we will now be holding the reference only for the
necessary duration by the end of the patchset. While simply not
delaying the disabling has the same benefit, it has the problem that
frequent on-off-on power switching has its own power cost (see the 2.
point below) and the debug trace for power well on/off events will
cause a lot of dmesg spam (see details about this further below).
2. Avoiding the power cost of freuqent on-off-on power switching. This
requires us to find the optimal disabling delay based on the measured
power cost of on->off and off->on switching of each power well vs.
the power of keeping the given power well on.
In this patchset I'm not providing this optimal delay for two
reasons:
a) I don't have the means yet to perform the measurement (with high
enough signal-to-noise ratio, or with the help of an energy
counter that takes switching into account). I'm currently looking
for a way to measure this.
b) Before reducing the disabling delay we need an alternative way for
debug tracing powerwell on/off events. Simply avoiding/throttling
the debug messages is not a solution, see further below.
Note that even in the case where we can't measure any considerable
power cost of frequent on-off switching of powerwells, it still would
make sense to do the disabling asynchronously (with 0 delay) to avoid
blocking on the disabling. On VLV I measured this disabling time
overhead to be 1ms on average with a worst case of 4ms.
In the case of the AUX power domains on ICL we would also need to keep
the sequence where we hold the power reference short, the way it would
be by the end of this patchset where we hold it only for the actual AUX
transfer. Anything else would make the locking we need for ICL TypeC
ports (whenever we hold a reference on any AUX power domain) rather
problematic, adding for instance unnecessary lockdep dependencies to
the required TypeC port lock.
I chose the disabling delay to be 100msec for now to avoid the unneeded
toggling (and so not to introduce dmesg spamming) in the DP MST sideband
signaling code. We could optimize this delay later, once we have the
means to measure the switching power cost (see above).
Note that simply removing/throttling the debug tracing for power well
on/off events is not a solution. We need to know the exact spots of
these events and cannot rely only on incorrect register accesses caught
(due to not holding a wakeref at the time of access). Incorrect
powerwell enabling/disabling could lead to other problems, for instance
we need to keep certain powerwells enabled for the duration of modesets
and AUX transfers.
v2:
- Clarify the commit log parts about power cost measurement and the
problem of simply removing/throttling debug tracing. (Chris)
Cc: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Cc: Ville Syrjala <ville.syrjala@xxxxxxxxxxxxxxx>
Signed-off-by: Imre Deak <imre.deak@xxxxxxxxx>
---
drivers/gpu/drm/i915/i915_drv.h | 5 +
drivers/gpu/drm/i915/intel_runtime_pm.c | 316 +++++++++++++++++++++++-
drivers/gpu/drm/i915/intel_runtime_pm.h | 4 +
3 files changed, 315 insertions(+), 10 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index a162d57f0b58..8d18d124f496 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -835,6 +835,11 @@ struct i915_power_domains {
struct mutex lock;
int domain_use_count[POWER_DOMAIN_NUM];
+
+ struct delayed_work async_put_work;
+ intel_wakeref_t async_put_wakeref;
+ u64 async_put_domains[2];
+
struct i915_power_well *power_wells;
};
diff --git a/drivers/gpu/drm/i915/intel_runtime_pm.c b/drivers/gpu/drm/i915/intel_runtime_pm.c
index b4b17cdc6f26..4079b1378f2d 100644
--- a/drivers/gpu/drm/i915/intel_runtime_pm.c
+++ b/drivers/gpu/drm/i915/intel_runtime_pm.c
@@ -1884,6 +1884,130 @@ static void chv_pipe_power_well_disable(struct drm_i915_private *dev_priv,
chv_set_pipe_power_well(dev_priv, power_well, false);
}
+static intel_wakeref_t
+intel_runtime_pm_get_raw(struct drm_i915_private *i915);
+static void
+intel_runtime_pm_put_raw(struct drm_i915_private *i915, intel_wakeref_t wref);
+
+static u64 __async_put_domains_mask(struct i915_power_domains *power_domains)
+{
+ return power_domains->async_put_domains[0] |
+ power_domains->async_put_domains[1];
+}
+
+#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_RUNTIME_PM)
+
+static bool
+assert_async_put_domain_masks_disjoint(struct i915_power_domains *power_domains)
+{
+ return !WARN_ON(power_domains->async_put_domains[0] &
+ power_domains->async_put_domains[1]);
+}
+
+static bool
+__async_put_domains_state_ok(struct i915_power_domains *power_domains)
+{
+ enum intel_display_power_domain domain;
+ bool err = false;
+
+ err |= !assert_async_put_domain_masks_disjoint(power_domains);
+ err |= WARN_ON(!!power_domains->async_put_wakeref !=
+ !!__async_put_domains_mask(power_domains));
+
+ for_each_power_domain(domain, __async_put_domains_mask(power_domains))
+ err |= WARN_ON(power_domains->domain_use_count[domain] != 1);
+
+ return !err;
+}
+
+static void print_power_domains(struct i915_power_domains *power_domains,
+ const char *prefix, u64 mask)
+{
+ enum intel_display_power_domain domain;
+
+ DRM_DEBUG_DRIVER("%s (%lu):\n", prefix, hweight64(mask));
+ for_each_power_domain(domain, mask)
+ DRM_DEBUG_DRIVER("%s use_count %d\n",
+ intel_display_power_domain_str(domain),
+ power_domains->domain_use_count[domain]);
+}
+
+static void
+print_async_put_domains_state(struct i915_power_domains *power_domains)
+{
+ DRM_DEBUG_DRIVER("async_put_wakeref %u\n",
+ power_domains->async_put_wakeref);
+
+ print_power_domains(power_domains, "async_put_domains[0]",
+ power_domains->async_put_domains[0]);
+ print_power_domains(power_domains, "async_put_domains[1]",
+ power_domains->async_put_domains[1]);
+}
+
+static void
+verify_async_put_domains_state(struct i915_power_domains *power_domains)
+{
+ if (!__async_put_domains_state_ok(power_domains))
+ print_async_put_domains_state(power_domains);
+}
+
+#else
+
+static void
+assert_async_put_domain_masks_disjoint(struct i915_power_domains *power_domains)
+{
+}
+
+static void
+verify_async_put_domains_state(struct i915_power_domains *power_domains)
+{
+}
+
+#endif /* CONFIG_DRM_I915_DEBUG_RUNTIME_PM */
+
+static u64 async_put_domains_mask(struct i915_power_domains *power_domains)
+{
+ assert_async_put_domain_masks_disjoint(power_domains);
+
+ return __async_put_domains_mask(power_domains);
+}
+
+static void
+async_put_domains_clear_domain(struct i915_power_domains *power_domains,
+ enum intel_display_power_domain domain)
+{
+ assert_async_put_domain_masks_disjoint(power_domains);
+
+ power_domains->async_put_domains[0] &= ~BIT_ULL(domain);
+ power_domains->async_put_domains[1] &= ~BIT_ULL(domain);
+}
+
+static bool
+intel_display_power_grab_async_put_ref(struct drm_i915_private *dev_priv,
+ enum intel_display_power_domain domain)
+{
+ struct i915_power_domains *power_domains = &dev_priv->power_domains;
+ bool ret = false;
+
+ if (!(async_put_domains_mask(power_domains) & BIT_ULL(domain)))
+ goto out_verify;
+
+ async_put_domains_clear_domain(power_domains, domain);
+
+ ret = true;
+
+ if (async_put_domains_mask(power_domains))
+ goto out_verify;
+
+ cancel_delayed_work(&power_domains->async_put_work);
+ intel_runtime_pm_put_raw(dev_priv,
+ fetch_and_zero(&power_domains->async_put_wakeref));
+out_verify:
+ verify_async_put_domains_state(power_domains);
+
+ return ret;
+}
+
static void
__intel_display_power_get_domain(struct drm_i915_private *dev_priv,
enum intel_display_power_domain domain)
@@ -1891,6 +2015,9 @@ __intel_display_power_get_domain(struct drm_i915_private *dev_priv,
struct i915_power_domains *power_domains = &dev_priv->power_domains;
struct i915_power_well *power_well;
+ if (intel_display_power_grab_async_put_ref(dev_priv, domain))
+ return;
+
for_each_power_domain_well(dev_priv, power_well, BIT_ULL(domain))
intel_power_well_get(dev_priv, power_well);
@@ -1916,9 +2043,7 @@ intel_wakeref_t intel_display_power_get(struct drm_i915_private *dev_priv,
intel_wakeref_t wakeref = intel_runtime_pm_get(dev_priv);
mutex_lock(&power_domains->lock);
-
__intel_display_power_get_domain(dev_priv, domain);
-
mutex_unlock(&power_domains->lock);
return wakeref;
@@ -1967,24 +2092,36 @@ intel_display_power_get_if_enabled(struct drm_i915_private *dev_priv,
return wakeref;
}
-static void __intel_display_power_put(struct drm_i915_private *dev_priv,
- enum intel_display_power_domain domain)
+static void
+__intel_display_power_put_domain(struct drm_i915_private *dev_priv,
+ enum intel_display_power_domain domain)
{
struct i915_power_domains *power_domains;
struct i915_power_well *power_well;
+ const char *name = intel_display_power_domain_str(domain);
power_domains = &dev_priv->power_domains;
- mutex_lock(&power_domains->lock);
-
WARN(!power_domains->domain_use_count[domain],
"Use count on domain %s is already zero\n",
- intel_display_power_domain_str(domain));
+ name);
+ WARN(async_put_domains_mask(power_domains) & BIT_ULL(domain),
+ "Async disabling of domain %s is pending\n",
+ name);
+
power_domains->domain_use_count[domain]--;
for_each_power_domain_well_reverse(dev_priv, power_well, BIT_ULL(domain))
intel_power_well_put(dev_priv, power_well);
+}
+static void __intel_display_power_put(struct drm_i915_private *dev_priv,
+ enum intel_display_power_domain domain)
+{
+ struct i915_power_domains *power_domains = &dev_priv->power_domains;
+
+ mutex_lock(&power_domains->lock);
+ __intel_display_power_put_domain(dev_priv, domain);
mutex_unlock(&power_domains->lock);
}
@@ -2004,6 +2141,159 @@ void intel_display_power_put_unchecked(struct drm_i915_private *dev_priv,
intel_runtime_pm_put_unchecked(dev_priv);
}
+static void
+queue_async_put_domains_work(struct i915_power_domains *power_domains,
+ intel_wakeref_t wakeref)
+{
+ WARN_ON(power_domains->async_put_wakeref);
+ power_domains->async_put_wakeref = wakeref;
+ WARN_ON(!queue_delayed_work(system_unbound_wq,
+ &power_domains->async_put_work,
+ msecs_to_jiffies(100)));
+}
+
+static void
+release_async_put_domains(struct i915_power_domains *power_domains, u64 mask)
+{
+ struct drm_i915_private *dev_priv =
+ container_of(power_domains, struct drm_i915_private,
+ power_domains);
+ enum intel_display_power_domain domain;
+ intel_wakeref_t wakeref;
+
+ /*
+ * The caller must hold already raw wakeref, upgrade that to a proper
+ * wakeref to make the state checker happy about the HW access during
+ * power well disabling.
+ */
+ assert_raw_rpm_wakelock_held(dev_priv);
+ wakeref = intel_runtime_pm_get(dev_priv);
+
+ for_each_power_domain(domain, mask) {
+ /* Clear before put, so put's sanity check is happy. */
+ async_put_domains_clear_domain(power_domains, domain);
+ __intel_display_power_put_domain(dev_priv, domain);
+ }
+
+ intel_runtime_pm_put(dev_priv, wakeref);
+}
+
+static void
+intel_display_power_put_async_work(struct work_struct *work)
+{
+ struct drm_i915_private *dev_priv =
+ container_of(work, struct drm_i915_private,
+ power_domains.async_put_work.work);
+ struct i915_power_domains *power_domains = &dev_priv->power_domains;
+ intel_wakeref_t new_work_wakeref = intel_runtime_pm_get_raw(dev_priv);
+ intel_wakeref_t old_work_wakeref = 0;
+
+ mutex_lock(&power_domains->lock);
+
+ /*
+ * Bail out if all the domain refs pending to be released were grabbed
+ * by subsequent gets or a flush_work.
+ */
+ old_work_wakeref = fetch_and_zero(&power_domains->async_put_wakeref);
+ if (!old_work_wakeref)
+ goto out_verify;
+
+ release_async_put_domains(power_domains,
+ power_domains->async_put_domains[0]);
+
+ /* Requeue the work if more domains were async put meanwhile. */
+ if (power_domains->async_put_domains[1]) {
+ power_domains->async_put_domains[0] =
+ fetch_and_zero(&power_domains->async_put_domains[1]);
+ queue_async_put_domains_work(power_domains,
+ fetch_and_zero(&new_work_wakeref));
+ }
+
+out_verify:
+ verify_async_put_domains_state(power_domains);
+
+ mutex_unlock(&power_domains->lock);
+
+ if (old_work_wakeref)
+ intel_runtime_pm_put_raw(dev_priv, old_work_wakeref);
+ if (new_work_wakeref)
+ intel_runtime_pm_put_raw(dev_priv, new_work_wakeref);
+}
+
+void intel_display_power_put_async(struct drm_i915_private *dev_priv,
+ enum intel_display_power_domain domain,
+ intel_wakeref_t wakeref)
+{
+ struct i915_power_domains *power_domains = &dev_priv->power_domains;
+ intel_wakeref_t work_wakeref = intel_runtime_pm_get_raw(dev_priv);
+
+ mutex_lock(&power_domains->lock);
+
+ if (power_domains->domain_use_count[domain] > 1) {
+ __intel_display_power_put_domain(dev_priv, domain);
+
+ goto out_verify;
+ }
+
+ WARN_ON(power_domains->domain_use_count[domain] != 1);
+
+ /* Let a pending work requeue itself or queue a new one. */
+ if (power_domains->async_put_wakeref) {
+ power_domains->async_put_domains[1] |= BIT_ULL(domain);
+ } else {
+ power_domains->async_put_domains[0] |= BIT_ULL(domain);
+ queue_async_put_domains_work(power_domains,
+ fetch_and_zero(&work_wakeref));
+ }
+
+out_verify:
+ verify_async_put_domains_state(power_domains);
+
+ mutex_unlock(&power_domains->lock);
+
+ if (work_wakeref)
+ intel_runtime_pm_put_raw(dev_priv, work_wakeref);
+
+ intel_runtime_pm_put(dev_priv, wakeref);
+}
+
+void intel_display_power_flush_work(struct drm_i915_private *dev_priv)
+{
+ struct i915_power_domains *power_domains = &dev_priv->power_domains;
+ intel_wakeref_t work_wakeref;
+
+ mutex_lock(&power_domains->lock);
+
+ work_wakeref = fetch_and_zero(&power_domains->async_put_wakeref);
+ if (!work_wakeref)
+ goto out_verify;
+
+ release_async_put_domains(power_domains,
+ async_put_domains_mask(power_domains));
+ cancel_delayed_work(&power_domains->async_put_work);
+
+out_verify:
+ verify_async_put_domains_state(power_domains);
+
+ mutex_unlock(&power_domains->lock);
+
+ if (work_wakeref)
+ intel_runtime_pm_put_raw(dev_priv, work_wakeref);
+}
+
+static void
+intel_display_power_flush_work_sync(struct drm_i915_private *dev_priv)
+{
+ struct i915_power_domains *power_domains = &dev_priv->power_domains;
+
+ intel_display_power_flush_work(dev_priv);
+ cancel_delayed_work_sync(&power_domains->async_put_work);
+
+ verify_async_put_domains_state(power_domains);
+
+ WARN_ON(power_domains->async_put_wakeref);
+}
+
#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_RUNTIME_PM)
void intel_display_power_put(struct drm_i915_private *dev_priv,
enum intel_display_power_domain domain,
@@ -3492,6 +3782,9 @@ int intel_power_domains_init(struct drm_i915_private *dev_priv)
mutex_init(&power_domains->lock);
+ INIT_DELAYED_WORK(&power_domains->async_put_work,
+ intel_display_power_put_async_work);
+
/*
* The enabling order will be from lower to higher indexed wells,
* the disabling order is reversed.
@@ -4169,6 +4462,8 @@ void intel_power_domains_fini_hw(struct drm_i915_private *i915)
if (!i915_modparams.disable_power_well)
intel_display_power_put_unchecked(i915, POWER_DOMAIN_INIT);
+ intel_display_power_flush_work_sync(i915);
+
intel_power_domains_verify_state(i915);
/* Keep the power well enabled, but cancel its rpm wakeref. */
@@ -4244,6 +4539,7 @@ void intel_power_domains_suspend(struct drm_i915_private *i915,
if (!(i915->csr.allowed_dc_mask & DC_STATE_EN_DC9) &&
suspend_mode == I915_DRM_SUSPEND_IDLE &&
i915->csr.dmc_payload) {
+ intel_display_power_flush_work(i915);
intel_power_domains_verify_state(i915);
return;
}
@@ -4255,6 +4551,7 @@ void intel_power_domains_suspend(struct drm_i915_private *i915,
if (!i915_modparams.disable_power_well)
intel_display_power_put_unchecked(i915, POWER_DOMAIN_INIT);
+ intel_display_power_flush_work(i915);
intel_power_domains_verify_state(i915);
if (INTEL_GEN(i915) >= 11)
@@ -4333,6 +4630,8 @@ static void intel_power_domains_verify_state(struct drm_i915_private *i915)
mutex_lock(&power_domains->lock);
+ verify_async_put_domains_state(power_domains);
+
dump_domain_info = false;
for_each_power_well(i915, power_well) {
enum intel_display_power_domain domain;
@@ -4401,7 +4700,6 @@ static void __intel_runtime_pm_get(struct drm_i915_private *i915)
WARN_ONCE(ret < 0, "pm_runtime_get_sync() failed: %d\n", ret);
}
-__attribute__((__used__))
static intel_wakeref_t intel_runtime_pm_get_raw(struct drm_i915_private *i915)
{
__intel_runtime_pm_get(i915);
@@ -4509,7 +4807,6 @@ void intel_runtime_pm_put_unchecked(struct drm_i915_private *i915)
}
#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_RUNTIME_PM)
-__attribute__((__used__))
static void intel_runtime_pm_put_raw(struct drm_i915_private *i915,
intel_wakeref_t wref)
{
@@ -4523,7 +4820,6 @@ void intel_runtime_pm_put(struct drm_i915_private *i915, intel_wakeref_t wref)
intel_runtime_pm_put_unchecked(i915);
}
#else
-__attribute__((__used__))
static void intel_runtime_pm_put_raw(struct drm_i915_private *i915,
intel_wakeref_t wref)
{
diff --git a/drivers/gpu/drm/i915/intel_runtime_pm.h b/drivers/gpu/drm/i915/intel_runtime_pm.h
index 69227756de3e..cabf6f900273 100644
--- a/drivers/gpu/drm/i915/intel_runtime_pm.h
+++ b/drivers/gpu/drm/i915/intel_runtime_pm.h
@@ -57,6 +57,10 @@ intel_display_power_get_if_enabled(struct drm_i915_private *dev_priv,
enum intel_display_power_domain domain);
void intel_display_power_put_unchecked(struct drm_i915_private *dev_priv,
enum intel_display_power_domain domain);
+void intel_display_power_put_async(struct drm_i915_private *dev_priv,
+ enum intel_display_power_domain domain,
+ intel_wakeref_t wakeref);
+void intel_display_power_flush_work(struct drm_i915_private *dev_priv);
#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_RUNTIME_PM)
void intel_display_power_put(struct drm_i915_private *dev_priv,
enum intel_display_power_domain domain,
--
2.17.1
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
