On 23/03/2023 10:12, Badal Nilawar wrote:
From: Don Hiatt <don.hiatt@xxxxxxxxx>
Add the ability to enable/disable rc6 on a per tile basis.
Some words on justification?
To enable rc6 on a tile:
echo 1 > /sys/class/drm/card0/gt/gtX/rc6_enable
To disable rc6 on a tile:
echo 0 > /sys/class/drm/card0/gt/gtX/rc6_enable
To display the rc6 state of a tile:
cat /sys/class/drm/card0/gt/gtX/rc6_enable
Cc: Andi Shyti <andi.shyti@xxxxxxxxx>
Co-developed-by: Badal Nilawar <badal.nilawar@xxxxxxxxx>
Signed-off-by: Badal Nilawar <badal.nilawar@xxxxxxxxx>
Signed-off-by: Don Hiatt <don.hiatt@xxxxxxxxx>
---
drivers/gpu/drm/i915/gt/intel_gt_sysfs_pm.c | 84 ++++++++++++++++++---
drivers/gpu/drm/i915/gt/intel_rc6.c | 14 ++--
drivers/gpu/drm/i915/gt/intel_rc6.h | 3 +
3 files changed, 82 insertions(+), 19 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_sysfs_pm.c b/drivers/gpu/drm/i915/gt/intel_gt_sysfs_pm.c
index 28f27091cd3b..e8f77dab8f5f 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt_sysfs_pm.c
+++ b/drivers/gpu/drm/i915/gt/intel_gt_sysfs_pm.c
@@ -176,18 +176,78 @@ static u32 get_residency(struct intel_gt *gt, enum intel_rc6_res_type id)
return DIV_ROUND_CLOSEST_ULL(res, 1000);
}
-static u8 get_rc6_mask(struct intel_gt *gt)
+static ssize_t rc6_enable_store(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ const char *buff, size_t count)
{
- u8 mask = 0;
+ struct intel_gt *gt = intel_gt_sysfs_get_drvdata(kobj, attr->attr.name);
+ intel_wakeref_t wakeref;
+ ssize_t ret;
+ u32 val;
+
+ ret = kstrtou32(buff, 0, &val);
+ if (ret)
+ return ret;
+
+ wakeref = intel_runtime_pm_get(gt->uncore->rpm);
+
+ if (val) {
+ if (gt->rc6.enabled)
+ goto unlock;
Access to rc6.enabled is not serialized should multiple threads write to
this file.
+
+ if (!gt->rc6.wakeref)
+ intel_rc6_rpm_get(>->rc6);
Similar with rc6.wakeref.
I think some locking of the internal state will be needed.
+
+ intel_rc6_enable(>->rc6);
+ intel_rc6_unpark(>->rc6);
+ } else {
+ intel_rc6_disable(>->rc6);
I am noticing the lack of symmetry with the above enable, unpark - it is
okay to not do the reverse park, disable here?
+
+ if (gt->rc6.wakeref)
+ intel_rc6_rpm_put(>->rc6);
+ }
+
+unlock:
+ intel_runtime_pm_put(gt->uncore->rpm, wakeref);
+
+ return count;
+}
+
+static ssize_t rc6_enable_dev_store(struct device *dev,
+ struct device_attribute *attr,
+ const char *buff, size_t count)
+{
+ struct intel_gt *gt = intel_gt_sysfs_get_drvdata(&dev->kobj, attr->attr.name);
+ intel_wakeref_t wakeref;
+ ssize_t ret;
+ u32 val;
+
+ ret = kstrtou32(buff, 0, &val);
+ if (ret)
+ return ret;
+
+ wakeref = intel_runtime_pm_get(gt->uncore->rpm);
+
+ if (val) {
+ if (gt->rc6.enabled)
+ goto unlock;
+
+ if (!gt->rc6.wakeref)
+ intel_rc6_rpm_get(>->rc6);
+
+ intel_rc6_enable(>->rc6);
+ intel_rc6_unpark(>->rc6);
+ } else {
+ intel_rc6_disable(>->rc6);
+
+ if (gt->rc6.wakeref)
+ intel_rc6_rpm_put(>->rc6);
+ }
- if (HAS_RC6(gt->i915))
- mask |= BIT(0);
- if (HAS_RC6p(gt->i915))
- mask |= BIT(1);
- if (HAS_RC6pp(gt->i915))
- mask |= BIT(2);
+unlock:
+ intel_runtime_pm_put(gt->uncore->rpm, wakeref);
- return mask;
+ return count;
rc6_enable_store and rc6_enable_dev_store end up pretty much almost
identical? Common helper that takes the gt?
Which gt will the device one work on on say MTL?
}
static ssize_t rc6_enable_show(struct kobject *kobj,
@@ -196,7 +256,7 @@ static ssize_t rc6_enable_show(struct kobject *kobj,
{
struct intel_gt *gt = intel_gt_sysfs_get_drvdata(kobj, attr->attr.name);
- return sysfs_emit(buff, "%x\n", get_rc6_mask(gt));
+ return sysfs_emit(buff, "%u\n", gt->rc6.enabled);
Change from bitmask to boolean needs to be mentioned and explained in
the commit message how it doesn't break any existing userspace.
Regards,
Tvrtko
}
static ssize_t rc6_enable_dev_show(struct device *dev,
@@ -205,7 +265,7 @@ static ssize_t rc6_enable_dev_show(struct device *dev,
{
struct intel_gt *gt = intel_gt_sysfs_get_drvdata(&dev->kobj, attr->attr.name);
- return sysfs_emit(buff, "%x\n", get_rc6_mask(gt));
+ return sysfs_emit(buff, "%u\n", gt->rc6.enabled);
}
static u32 __rc6_residency_ms_show(struct intel_gt *gt)
@@ -233,7 +293,7 @@ INTEL_GT_SYSFS_SHOW_MIN(rc6p_residency_ms);
INTEL_GT_SYSFS_SHOW_MIN(rc6pp_residency_ms);
INTEL_GT_SYSFS_SHOW_MIN(media_rc6_residency_ms);
-INTEL_GT_DUAL_ATTR_RO(rc6_enable);
+INTEL_GT_DUAL_ATTR_RW(rc6_enable);
INTEL_GT_DUAL_ATTR_RO(rc6_residency_ms);
INTEL_GT_DUAL_ATTR_RO(rc6p_residency_ms);
INTEL_GT_DUAL_ATTR_RO(rc6pp_residency_ms);
diff --git a/drivers/gpu/drm/i915/gt/intel_rc6.c b/drivers/gpu/drm/i915/gt/intel_rc6.c
index f4150f61f39c..ea7420271849 100644
--- a/drivers/gpu/drm/i915/gt/intel_rc6.c
+++ b/drivers/gpu/drm/i915/gt/intel_rc6.c
@@ -513,14 +513,14 @@ static bool rc6_supported(struct intel_rc6 *rc6)
return true;
}
-static void rpm_get(struct intel_rc6 *rc6)
+void intel_rc6_rpm_get(struct intel_rc6 *rc6)
{
GEM_BUG_ON(rc6->wakeref);
pm_runtime_get_sync(rc6_to_i915(rc6)->drm.dev);
rc6->wakeref = true;
}
-static void rpm_put(struct intel_rc6 *rc6)
+void intel_rc6_rpm_put(struct intel_rc6 *rc6)
{
GEM_BUG_ON(!rc6->wakeref);
pm_runtime_put(rc6_to_i915(rc6)->drm.dev);
@@ -582,7 +582,7 @@ void intel_rc6_init(struct intel_rc6 *rc6)
int err;
/* Disable runtime-pm until we can save the GPU state with rc6 pctx */
- rpm_get(rc6);
+ intel_rc6_rpm_get(rc6);
if (!rc6_supported(rc6))
return;
@@ -607,7 +607,7 @@ void intel_rc6_sanitize(struct intel_rc6 *rc6)
memset(rc6->prev_hw_residency, 0, sizeof(rc6->prev_hw_residency));
if (rc6->enabled) { /* unbalanced suspend/resume */
- rpm_get(rc6);
+ intel_rc6_rpm_get(rc6);
rc6->enabled = false;
}
@@ -650,7 +650,7 @@ void intel_rc6_enable(struct intel_rc6 *rc6)
return;
/* rc6 is ready, runtime-pm is go! */
- rpm_put(rc6);
+ intel_rc6_rpm_put(rc6);
rc6->enabled = true;
}
@@ -698,7 +698,7 @@ void intel_rc6_disable(struct intel_rc6 *rc6)
if (!rc6->enabled)
return;
- rpm_get(rc6);
+ intel_rc6_rpm_get(rc6);
rc6->enabled = false;
__intel_rc6_disable(rc6);
@@ -715,7 +715,7 @@ void intel_rc6_fini(struct intel_rc6 *rc6)
i915_gem_object_put(pctx);
if (rc6->wakeref)
- rpm_put(rc6);
+ intel_rc6_rpm_put(rc6);
}
static u64 vlv_residency_raw(struct intel_uncore *uncore, const i915_reg_t reg)
diff --git a/drivers/gpu/drm/i915/gt/intel_rc6.h b/drivers/gpu/drm/i915/gt/intel_rc6.h
index 456fa668a276..8771fbd9a6cf 100644
--- a/drivers/gpu/drm/i915/gt/intel_rc6.h
+++ b/drivers/gpu/drm/i915/gt/intel_rc6.h
@@ -12,6 +12,9 @@ enum intel_rc6_res_type;
struct intel_rc6;
struct seq_file;
+void intel_rc6_rpm_get(struct intel_rc6 *rc6);
+void intel_rc6_rpm_put(struct intel_rc6 *rc6);
+
void intel_rc6_init(struct intel_rc6 *rc6);
void intel_rc6_fini(struct intel_rc6 *rc6);
