Gently ping for network developers, could you look at ref_tracker patches,
as the ref_tracker library was developed for network.
This is revived patchset improving ref_tracker library and converting
i915 internal tracker to ref_tracker.
The old thread ended without consensus about small kernel allocations,
which are performed under spinlock.
I have tried to solve the problem by splitting the calls, but it results
in complicated API, so I went back to original solution.
If there are better solutions I am glad to discuss them.
Meanwhile I send original patchset with addressed remaining comments.
To: Jani Nikula <[email protected]>
To: Joonas Lahtinen <[email protected]>
To: Rodrigo Vivi <[email protected]>
To: Tvrtko Ursulin <[email protected]>
To: David Airlie <[email protected]>
To: Daniel Vetter <[email protected]>
Cc: [email protected]
Cc: [email protected]
Cc: [email protected]
Cc: Chris Wilson <[email protected]>
Cc: [email protected]
Cc: Eric Dumazet <[email protected]>
Cc: Jakub Kicinski <[email protected]>
Cc: Dmitry Vyukov <[email protected]>
Cc: "David S. Miller" <[email protected]>
Cc: Andi Shyti <[email protected]>
Cc: Das, Nirmoy <[email protected]>
Signed-off-by: Andrzej Hajda <[email protected]>
---
Changes in v7:
- removed 8th patch (hold wakeref), as it was already merged
- added tags (thx Andi)
- Link to v6: https://lore.kernel.org/r/[email protected]
Changes in v6:
- rebased to solve minor conflict and allow CI testing
- Link to v5: https://lore.kernel.org/r/[email protected]
Changes in v5 (thx Andi for review):
- use *_locked convention instead of __*,
- improved commit messages,
- re-worked i915 patches, squashed separation and conversion patches,
- added tags,
- Link to v4: https://lore.kernel.org/r/[email protected]
Changes in v4:
- split "Separate wakeref tracking" to smaller parts
- fixed typos,
- Link to v1-v3: https://patchwork.freedesktop.org/series/100327/
---
Andrzej Hajda (7):
lib/ref_tracker: add unlocked leak print helper
lib/ref_tracker: improve printing stats
lib/ref_tracker: add printing to memory buffer
lib/ref_tracker: remove warnings in case of allocation failure
drm/i915: Correct type of wakeref variable
drm/i915: Replace custom intel runtime_pm tracker with ref_tracker library
drm/i915: track gt pm wakerefs
drivers/gpu/drm/i915/Kconfig.debug | 19 ++
drivers/gpu/drm/i915/display/intel_display_power.c | 2 +-
drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c | 7 +-
.../drm/i915/gem/selftests/i915_gem_coherency.c | 10 +-
drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c | 14 +-
drivers/gpu/drm/i915/gt/intel_breadcrumbs.c | 13 +-
drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h | 3 +-
drivers/gpu/drm/i915/gt/intel_context.h | 4 +-
drivers/gpu/drm/i915/gt/intel_context_types.h | 2 +
drivers/gpu/drm/i915/gt/intel_engine_pm.c | 7 +-
drivers/gpu/drm/i915/gt/intel_engine_types.h | 2 +
.../gpu/drm/i915/gt/intel_execlists_submission.c | 2 +-
drivers/gpu/drm/i915/gt/intel_gt_pm.c | 12 +-
drivers/gpu/drm/i915/gt/intel_gt_pm.h | 38 +++-
drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.c | 4 +-
drivers/gpu/drm/i915/gt/selftest_engine_cs.c | 20 +-
drivers/gpu/drm/i915/gt/selftest_gt_pm.c | 5 +-
drivers/gpu/drm/i915/gt/selftest_reset.c | 10 +-
drivers/gpu/drm/i915/gt/selftest_rps.c | 17 +-
drivers/gpu/drm/i915/gt/selftest_slpc.c | 5 +-
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 11 +-
drivers/gpu/drm/i915/i915_driver.c | 2 +-
drivers/gpu/drm/i915/i915_pmu.c | 16 +-
drivers/gpu/drm/i915/intel_runtime_pm.c | 221 ++-------------------
drivers/gpu/drm/i915/intel_runtime_pm.h | 11 +-
drivers/gpu/drm/i915/intel_wakeref.c | 7 +-
drivers/gpu/drm/i915/intel_wakeref.h | 99 ++++++++-
include/linux/ref_tracker.h | 31 ++-
lib/ref_tracker.c | 179 ++++++++++++++---
29 files changed, 444 insertions(+), 329 deletions(-)
---
base-commit: d224485cc77189981322356d4ca6b8c42f7bf9fa
change-id: 20230224-track_gt-1b3da8bdacd7
Best regards,
--
Andrzej Hajda <[email protected]>
To have reliable detection of leaks, caller must be able to check under the same
lock both: tracked counter and the leaks. dir.lock is natural candidate for such
lock and unlocked print helper can be called with this lock taken.
As a bonus we can reuse this helper in ref_tracker_dir_exit.
Signed-off-by: Andrzej Hajda <[email protected]>
Reviewed-by: Andi Shyti <[email protected]>
---
include/linux/ref_tracker.h | 8 ++++++
lib/ref_tracker.c | 66 ++++++++++++++++++++++++++-------------------
2 files changed, 46 insertions(+), 28 deletions(-)
diff --git a/include/linux/ref_tracker.h b/include/linux/ref_tracker.h
index 9ca353ab712b5e..87a92f2bec1b88 100644
--- a/include/linux/ref_tracker.h
+++ b/include/linux/ref_tracker.h
@@ -36,6 +36,9 @@ static inline void ref_tracker_dir_init(struct ref_tracker_dir *dir,
void ref_tracker_dir_exit(struct ref_tracker_dir *dir);
+void ref_tracker_dir_print_locked(struct ref_tracker_dir *dir,
+ unsigned int display_limit);
+
void ref_tracker_dir_print(struct ref_tracker_dir *dir,
unsigned int display_limit);
@@ -56,6 +59,11 @@ static inline void ref_tracker_dir_exit(struct ref_tracker_dir *dir)
{
}
+static inline void ref_tracker_dir_print_locked(struct ref_tracker_dir *dir,
+ unsigned int display_limit)
+{
+}
+
static inline void ref_tracker_dir_print(struct ref_tracker_dir *dir,
unsigned int display_limit)
{
diff --git a/lib/ref_tracker.c b/lib/ref_tracker.c
index dc7b14aa3431e2..d4eb0929af8f96 100644
--- a/lib/ref_tracker.c
+++ b/lib/ref_tracker.c
@@ -14,6 +14,38 @@ struct ref_tracker {
depot_stack_handle_t free_stack_handle;
};
+void ref_tracker_dir_print_locked(struct ref_tracker_dir *dir,
+ unsigned int display_limit)
+{
+ struct ref_tracker *tracker;
+ unsigned int i = 0;
+
+ lockdep_assert_held(&dir->lock);
+
+ list_for_each_entry(tracker, &dir->list, head) {
+ if (i < display_limit) {
+ pr_err("leaked reference.\n");
+ if (tracker->alloc_stack_handle)
+ stack_depot_print(tracker->alloc_stack_handle);
+ i++;
+ } else {
+ break;
+ }
+ }
+}
+EXPORT_SYMBOL(ref_tracker_dir_print_locked);
+
+void ref_tracker_dir_print(struct ref_tracker_dir *dir,
+ unsigned int display_limit)
+{
+ unsigned long flags;
+
+ spin_lock_irqsave(&dir->lock, flags);
+ ref_tracker_dir_print_locked(dir, display_limit);
+ spin_unlock_irqrestore(&dir->lock, flags);
+}
+EXPORT_SYMBOL(ref_tracker_dir_print);
+
void ref_tracker_dir_exit(struct ref_tracker_dir *dir)
{
struct ref_tracker *tracker, *n;
@@ -27,13 +59,13 @@ void ref_tracker_dir_exit(struct ref_tracker_dir *dir)
kfree(tracker);
dir->quarantine_avail++;
}
- list_for_each_entry_safe(tracker, n, &dir->list, head) {
- pr_err("leaked reference.\n");
- if (tracker->alloc_stack_handle)
- stack_depot_print(tracker->alloc_stack_handle);
+ if (!list_empty(&dir->list)) {
+ ref_tracker_dir_print_locked(dir, 16);
leak = true;
- list_del(&tracker->head);
- kfree(tracker);
+ list_for_each_entry_safe(tracker, n, &dir->list, head) {
+ list_del(&tracker->head);
+ kfree(tracker);
+ }
}
spin_unlock_irqrestore(&dir->lock, flags);
WARN_ON_ONCE(leak);
@@ -42,28 +74,6 @@ void ref_tracker_dir_exit(struct ref_tracker_dir *dir)
}
EXPORT_SYMBOL(ref_tracker_dir_exit);
-void ref_tracker_dir_print(struct ref_tracker_dir *dir,
- unsigned int display_limit)
-{
- struct ref_tracker *tracker;
- unsigned long flags;
- unsigned int i = 0;
-
- spin_lock_irqsave(&dir->lock, flags);
- list_for_each_entry(tracker, &dir->list, head) {
- if (i < display_limit) {
- pr_err("leaked reference.\n");
- if (tracker->alloc_stack_handle)
- stack_depot_print(tracker->alloc_stack_handle);
- i++;
- } else {
- break;
- }
- }
- spin_unlock_irqrestore(&dir->lock, flags);
-}
-EXPORT_SYMBOL(ref_tracker_dir_print);
-
int ref_tracker_alloc(struct ref_tracker_dir *dir,
struct ref_tracker **trackerp,
gfp_t gfp)
--
2.34.1
In case the library is tracking busy subsystem, simply
printing stack for every active reference will spam log
with long, hard to read, redundant stack traces. To improve
readabilty following changes have been made:
- reports are printed per stack_handle - log is more compact,
- added display name for ref_tracker_dir - it will differentiate
multiple subsystems,
- stack trace is printed indented, in the same printk call,
- info about dropped references is printed as well.
Signed-off-by: Andrzej Hajda <[email protected]>
Reviewed-by: Andi Shyti <[email protected]>
---
include/linux/ref_tracker.h | 15 ++++++--
lib/ref_tracker.c | 90 +++++++++++++++++++++++++++++++++++++++------
2 files changed, 91 insertions(+), 14 deletions(-)
diff --git a/include/linux/ref_tracker.h b/include/linux/ref_tracker.h
index 87a92f2bec1b88..fc9ef9952f01fd 100644
--- a/include/linux/ref_tracker.h
+++ b/include/linux/ref_tracker.h
@@ -17,12 +17,19 @@ struct ref_tracker_dir {
bool dead;
struct list_head list; /* List of active trackers */
struct list_head quarantine; /* List of dead trackers */
+ char name[32];
#endif
};
#ifdef CONFIG_REF_TRACKER
-static inline void ref_tracker_dir_init(struct ref_tracker_dir *dir,
- unsigned int quarantine_count)
+
+/* Temporary allow two and three arguments, until consumers are converted */
+#define ref_tracker_dir_init(_d, _q, args...) _ref_tracker_dir_init(_d, _q, ##args, #_d)
+#define _ref_tracker_dir_init(_d, _q, _n, ...) __ref_tracker_dir_init(_d, _q, _n)
+
+static inline void __ref_tracker_dir_init(struct ref_tracker_dir *dir,
+ unsigned int quarantine_count,
+ const char *name)
{
INIT_LIST_HEAD(&dir->list);
INIT_LIST_HEAD(&dir->quarantine);
@@ -31,6 +38,7 @@ static inline void ref_tracker_dir_init(struct ref_tracker_dir *dir,
dir->dead = false;
refcount_set(&dir->untracked, 1);
refcount_set(&dir->no_tracker, 1);
+ strlcpy(dir->name, name, sizeof(dir->name));
stack_depot_init();
}
@@ -51,7 +59,8 @@ int ref_tracker_free(struct ref_tracker_dir *dir,
#else /* CONFIG_REF_TRACKER */
static inline void ref_tracker_dir_init(struct ref_tracker_dir *dir,
- unsigned int quarantine_count)
+ unsigned int quarantine_count,
+ ...)
{
}
diff --git a/lib/ref_tracker.c b/lib/ref_tracker.c
index d4eb0929af8f96..2ffe79c90c1771 100644
--- a/lib/ref_tracker.c
+++ b/lib/ref_tracker.c
@@ -1,11 +1,16 @@
// SPDX-License-Identifier: GPL-2.0-or-later
+
+#define pr_fmt(fmt) "ref_tracker: " fmt
+
#include <linux/export.h>
+#include <linux/list_sort.h>
#include <linux/ref_tracker.h>
#include <linux/slab.h>
#include <linux/stacktrace.h>
#include <linux/stackdepot.h>
#define REF_TRACKER_STACK_ENTRIES 16
+#define STACK_BUF_SIZE 1024
struct ref_tracker {
struct list_head head; /* anchor into dir->list or dir->quarantine */
@@ -14,24 +19,87 @@ struct ref_tracker {
depot_stack_handle_t free_stack_handle;
};
-void ref_tracker_dir_print_locked(struct ref_tracker_dir *dir,
- unsigned int display_limit)
+struct ref_tracker_dir_stats {
+ int total;
+ int count;
+ struct {
+ depot_stack_handle_t stack_handle;
+ unsigned int count;
+ } stacks[];
+};
+
+static struct ref_tracker_dir_stats *
+ref_tracker_get_stats(struct ref_tracker_dir *dir, unsigned int limit)
{
+ struct ref_tracker_dir_stats *stats;
struct ref_tracker *tracker;
- unsigned int i = 0;
- lockdep_assert_held(&dir->lock);
+ stats = kmalloc(struct_size(stats, stacks, limit),
+ GFP_NOWAIT | __GFP_NOWARN);
+ if (!stats)
+ return ERR_PTR(-ENOMEM);
+ stats->total = 0;
+ stats->count = 0;
list_for_each_entry(tracker, &dir->list, head) {
- if (i < display_limit) {
- pr_err("leaked reference.\n");
- if (tracker->alloc_stack_handle)
- stack_depot_print(tracker->alloc_stack_handle);
- i++;
- } else {
- break;
+ depot_stack_handle_t stack = tracker->alloc_stack_handle;
+ int i;
+
+ ++stats->total;
+ for (i = 0; i < stats->count; ++i)
+ if (stats->stacks[i].stack_handle == stack)
+ break;
+ if (i >= limit)
+ continue;
+ if (i >= stats->count) {
+ stats->stacks[i].stack_handle = stack;
+ stats->stacks[i].count = 0;
+ ++stats->count;
}
+ ++stats->stacks[i].count;
+ }
+
+ return stats;
+}
+
+void ref_tracker_dir_print_locked(struct ref_tracker_dir *dir,
+ unsigned int display_limit)
+{
+ struct ref_tracker_dir_stats *stats;
+ unsigned int i = 0, skipped;
+ depot_stack_handle_t stack;
+ char *sbuf;
+
+ lockdep_assert_held(&dir->lock);
+
+ if (list_empty(&dir->list))
+ return;
+
+ stats = ref_tracker_get_stats(dir, display_limit);
+ if (IS_ERR(stats)) {
+ pr_err("%s@%pK: couldn't get stats, error %pe\n",
+ dir->name, dir, stats);
+ return;
}
+
+ sbuf = kmalloc(STACK_BUF_SIZE, GFP_NOWAIT | __GFP_NOWARN);
+
+ for (i = 0, skipped = stats->total; i < stats->count; ++i) {
+ stack = stats->stacks[i].stack_handle;
+ if (sbuf && !stack_depot_snprint(stack, sbuf, STACK_BUF_SIZE, 4))
+ sbuf[0] = 0;
+ pr_err("%s@%pK has %d/%d users at\n%s\n", dir->name, dir,
+ stats->stacks[i].count, stats->total, sbuf);
+ skipped -= stats->stacks[i].count;
+ }
+
+ if (skipped)
+ pr_err("%s@%pK skipped reports about %d/%d users.\n",
+ dir->name, dir, skipped, stats->total);
+
+ kfree(sbuf);
+
+ kfree(stats);
}
EXPORT_SYMBOL(ref_tracker_dir_print_locked);
--
2.34.1
Similar to stack_(depot|trace)_snprint the patch
adds helper to printing stats to memory buffer.
It will be helpful in case of debugfs.
Signed-off-by: Andrzej Hajda <[email protected]>
Reviewed-by: Andi Shyti <[email protected]>
---
include/linux/ref_tracker.h | 8 +++++++
lib/ref_tracker.c | 56 ++++++++++++++++++++++++++++++++++++++-------
2 files changed, 56 insertions(+), 8 deletions(-)
diff --git a/include/linux/ref_tracker.h b/include/linux/ref_tracker.h
index fc9ef9952f01fd..4cd260efc4023f 100644
--- a/include/linux/ref_tracker.h
+++ b/include/linux/ref_tracker.h
@@ -50,6 +50,8 @@ void ref_tracker_dir_print_locked(struct ref_tracker_dir *dir,
void ref_tracker_dir_print(struct ref_tracker_dir *dir,
unsigned int display_limit);
+int ref_tracker_dir_snprint(struct ref_tracker_dir *dir, char *buf, size_t size);
+
int ref_tracker_alloc(struct ref_tracker_dir *dir,
struct ref_tracker **trackerp, gfp_t gfp);
@@ -78,6 +80,12 @@ static inline void ref_tracker_dir_print(struct ref_tracker_dir *dir,
{
}
+static inline int ref_tracker_dir_snprint(struct ref_tracker_dir *dir,
+ char *buf, size_t size)
+{
+ return 0;
+}
+
static inline int ref_tracker_alloc(struct ref_tracker_dir *dir,
struct ref_tracker **trackerp,
gfp_t gfp)
diff --git a/lib/ref_tracker.c b/lib/ref_tracker.c
index 2ffe79c90c1771..cce4614b07940f 100644
--- a/lib/ref_tracker.c
+++ b/lib/ref_tracker.c
@@ -62,8 +62,27 @@ ref_tracker_get_stats(struct ref_tracker_dir *dir, unsigned int limit)
return stats;
}
-void ref_tracker_dir_print_locked(struct ref_tracker_dir *dir,
- unsigned int display_limit)
+struct ostream {
+ char *buf;
+ int size, used;
+};
+
+#define pr_ostream(stream, fmt, args...) \
+({ \
+ struct ostream *_s = (stream); \
+\
+ if (!_s->buf) { \
+ pr_err(fmt, ##args); \
+ } else { \
+ int ret, len = _s->size - _s->used; \
+ ret = snprintf(_s->buf + _s->used, len, pr_fmt(fmt), ##args); \
+ _s->used += min(ret, len); \
+ } \
+})
+
+static void
+__ref_tracker_dir_pr_ostream(struct ref_tracker_dir *dir,
+ unsigned int display_limit, struct ostream *s)
{
struct ref_tracker_dir_stats *stats;
unsigned int i = 0, skipped;
@@ -77,8 +96,8 @@ void ref_tracker_dir_print_locked(struct ref_tracker_dir *dir,
stats = ref_tracker_get_stats(dir, display_limit);
if (IS_ERR(stats)) {
- pr_err("%s@%pK: couldn't get stats, error %pe\n",
- dir->name, dir, stats);
+ pr_ostream(s, "%s@%pK: couldn't get stats, error %pe\n",
+ dir->name, dir, stats);
return;
}
@@ -88,19 +107,27 @@ void ref_tracker_dir_print_locked(struct ref_tracker_dir *dir,
stack = stats->stacks[i].stack_handle;
if (sbuf && !stack_depot_snprint(stack, sbuf, STACK_BUF_SIZE, 4))
sbuf[0] = 0;
- pr_err("%s@%pK has %d/%d users at\n%s\n", dir->name, dir,
- stats->stacks[i].count, stats->total, sbuf);
+ pr_ostream(s, "%s@%pK has %d/%d users at\n%s\n", dir->name, dir,
+ stats->stacks[i].count, stats->total, sbuf);
skipped -= stats->stacks[i].count;
}
if (skipped)
- pr_err("%s@%pK skipped reports about %d/%d users.\n",
- dir->name, dir, skipped, stats->total);
+ pr_ostream(s, "%s@%pK skipped reports about %d/%d users.\n",
+ dir->name, dir, skipped, stats->total);
kfree(sbuf);
kfree(stats);
}
+
+void ref_tracker_dir_print_locked(struct ref_tracker_dir *dir,
+ unsigned int display_limit)
+{
+ struct ostream os = {};
+
+ __ref_tracker_dir_pr_ostream(dir, display_limit, &os);
+}
EXPORT_SYMBOL(ref_tracker_dir_print_locked);
void ref_tracker_dir_print(struct ref_tracker_dir *dir,
@@ -114,6 +141,19 @@ void ref_tracker_dir_print(struct ref_tracker_dir *dir,
}
EXPORT_SYMBOL(ref_tracker_dir_print);
+int ref_tracker_dir_snprint(struct ref_tracker_dir *dir, char *buf, size_t size)
+{
+ struct ostream os = { .buf = buf, .size = size };
+ unsigned long flags;
+
+ spin_lock_irqsave(&dir->lock, flags);
+ __ref_tracker_dir_pr_ostream(dir, 16, &os);
+ spin_unlock_irqrestore(&dir->lock, flags);
+
+ return os.used;
+}
+EXPORT_SYMBOL(ref_tracker_dir_snprint);
+
void ref_tracker_dir_exit(struct ref_tracker_dir *dir)
{
struct ref_tracker *tracker, *n;
--
2.34.1
Library can handle allocation failures. To avoid allocation warnings
__GFP_NOWARN has been added everywhere. Moreover GFP_ATOMIC has been
replaced with GFP_NOWAIT in case of stack allocation on tracker free
call.
Signed-off-by: Andrzej Hajda <[email protected]>
Reviewed-by: Andi Shyti <[email protected]>
---
lib/ref_tracker.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/lib/ref_tracker.c b/lib/ref_tracker.c
index cce4614b07940f..cf5609b1ca7936 100644
--- a/lib/ref_tracker.c
+++ b/lib/ref_tracker.c
@@ -189,7 +189,7 @@ int ref_tracker_alloc(struct ref_tracker_dir *dir,
unsigned long entries[REF_TRACKER_STACK_ENTRIES];
struct ref_tracker *tracker;
unsigned int nr_entries;
- gfp_t gfp_mask = gfp;
+ gfp_t gfp_mask = gfp | __GFP_NOWARN;
unsigned long flags;
WARN_ON_ONCE(dir->dead);
@@ -237,7 +237,8 @@ int ref_tracker_free(struct ref_tracker_dir *dir,
return -EEXIST;
}
nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 1);
- stack_handle = stack_depot_save(entries, nr_entries, GFP_ATOMIC);
+ stack_handle = stack_depot_save(entries, nr_entries,
+ GFP_NOWAIT | __GFP_NOWARN);
spin_lock_irqsave(&dir->lock, flags);
if (tracker->dead) {
--
2.34.1
Wakeref has dedicated type. Assumption it will be int
compatible forever is incorrect.
Signed-off-by: Andrzej Hajda <[email protected]>
Reviewed-by: Andi Shyti <[email protected]>
---
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
index 88e881b100cf0a..74d28a3af2d57b 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
@@ -3248,7 +3248,7 @@ static void destroyed_worker_func(struct work_struct *w)
struct intel_guc *guc = container_of(w, struct intel_guc,
submission_state.destroyed_worker);
struct intel_gt *gt = guc_to_gt(guc);
- int tmp;
+ intel_wakeref_t tmp;
with_intel_gt_pm(gt, tmp)
deregister_destroyed_contexts(guc);
--
2.34.1
Beside reusing existing code, the main advantage of ref_tracker is
tracking per instance of wakeref. It allows also to catch double
put.
On the other side we lose information about the first acquire and
the last release, but the advantages outweigh it.
Signed-off-by: Andrzej Hajda <[email protected]>
---
drivers/gpu/drm/i915/Kconfig.debug | 4 +
drivers/gpu/drm/i915/display/intel_display_power.c | 2 +-
drivers/gpu/drm/i915/i915_driver.c | 2 +-
drivers/gpu/drm/i915/intel_runtime_pm.c | 221 ++-------------------
drivers/gpu/drm/i915/intel_runtime_pm.h | 11 +-
drivers/gpu/drm/i915/intel_wakeref.h | 61 +++++-
6 files changed, 84 insertions(+), 217 deletions(-)
diff --git a/drivers/gpu/drm/i915/Kconfig.debug b/drivers/gpu/drm/i915/Kconfig.debug
index 47e845353ffad8..76454fcbf65228 100644
--- a/drivers/gpu/drm/i915/Kconfig.debug
+++ b/drivers/gpu/drm/i915/Kconfig.debug
@@ -24,7 +24,9 @@ config DRM_I915_DEBUG
select DEBUG_FS
select PREEMPT_COUNT
select I2C_CHARDEV
+ select REF_TRACKER
select STACKDEPOT
+ select STACKTRACE
select DRM_DP_AUX_CHARDEV
select X86_MSR # used by igt/pm_rpm
select DRM_VGEM # used by igt/prime_vgem (dmabuf interop checks)
@@ -230,7 +232,9 @@ config DRM_I915_DEBUG_RUNTIME_PM
bool "Enable extra state checking for runtime PM"
depends on DRM_I915
default n
+ select REF_TRACKER
select STACKDEPOT
+ select STACKTRACE
help
Choose this option to turn on extra state checking for the
runtime PM functionality. This may introduce overhead during
diff --git a/drivers/gpu/drm/i915/display/intel_display_power.c b/drivers/gpu/drm/i915/display/intel_display_power.c
index 7c9f4288329ede..6ebdcfaa009906 100644
--- a/drivers/gpu/drm/i915/display/intel_display_power.c
+++ b/drivers/gpu/drm/i915/display/intel_display_power.c
@@ -406,7 +406,7 @@ print_async_put_domains_state(struct i915_power_domains *power_domains)
struct drm_i915_private,
display.power.domains);
- drm_dbg(&i915->drm, "async_put_wakeref %u\n",
+ drm_dbg(&i915->drm, "async_put_wakeref %lu\n",
power_domains->async_put_wakeref);
print_power_domains(power_domains, "async_put_domains[0]",
diff --git a/drivers/gpu/drm/i915/i915_driver.c b/drivers/gpu/drm/i915/i915_driver.c
index a52db8a8090065..89d5e4cdfb4386 100644
--- a/drivers/gpu/drm/i915/i915_driver.c
+++ b/drivers/gpu/drm/i915/i915_driver.c
@@ -1019,7 +1019,7 @@ void i915_driver_shutdown(struct drm_i915_private *i915)
intel_power_domains_driver_remove(i915);
enable_rpm_wakeref_asserts(&i915->runtime_pm);
- intel_runtime_pm_driver_release(&i915->runtime_pm);
+ intel_runtime_pm_driver_last_release(&i915->runtime_pm);
}
static bool suspend_to_idle(struct drm_i915_private *dev_priv)
diff --git a/drivers/gpu/drm/i915/intel_runtime_pm.c b/drivers/gpu/drm/i915/intel_runtime_pm.c
index cf5122299b6b8c..c5785ddaa717e4 100644
--- a/drivers/gpu/drm/i915/intel_runtime_pm.c
+++ b/drivers/gpu/drm/i915/intel_runtime_pm.c
@@ -52,182 +52,37 @@
#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_RUNTIME_PM)
-#include <linux/sort.h>
-
-#define STACKDEPTH 8
-
-static noinline depot_stack_handle_t __save_depot_stack(void)
-{
- unsigned long entries[STACKDEPTH];
- unsigned int n;
-
- n = stack_trace_save(entries, ARRAY_SIZE(entries), 1);
- return stack_depot_save(entries, n, GFP_NOWAIT | __GFP_NOWARN);
-}
-
static void init_intel_runtime_pm_wakeref(struct intel_runtime_pm *rpm)
{
- spin_lock_init(&rpm->debug.lock);
- stack_depot_init();
+ ref_tracker_dir_init(&rpm->debug, INTEL_REFTRACK_DEAD_COUNT, dev_name(rpm->kdev));
}
-static noinline depot_stack_handle_t
+static intel_wakeref_t
track_intel_runtime_pm_wakeref(struct intel_runtime_pm *rpm)
{
- depot_stack_handle_t stack, *stacks;
- unsigned long flags;
-
- if (rpm->no_wakeref_tracking)
- return -1;
-
- stack = __save_depot_stack();
- if (!stack)
+ if (!rpm->available || rpm->no_wakeref_tracking)
return -1;
- spin_lock_irqsave(&rpm->debug.lock, flags);
-
- if (!rpm->debug.count)
- rpm->debug.last_acquire = stack;
-
- stacks = krealloc(rpm->debug.owners,
- (rpm->debug.count + 1) * sizeof(*stacks),
- GFP_NOWAIT | __GFP_NOWARN);
- if (stacks) {
- stacks[rpm->debug.count++] = stack;
- rpm->debug.owners = stacks;
- } else {
- stack = -1;
- }
-
- spin_unlock_irqrestore(&rpm->debug.lock, flags);
-
- return stack;
+ return intel_ref_tracker_alloc(&rpm->debug);
}
static void untrack_intel_runtime_pm_wakeref(struct intel_runtime_pm *rpm,
- depot_stack_handle_t stack)
+ intel_wakeref_t wakeref)
{
- struct drm_i915_private *i915 = container_of(rpm,
- struct drm_i915_private,
- runtime_pm);
- unsigned long flags, n;
- bool found = false;
-
- if (unlikely(stack == -1))
+ if (!rpm->available || rpm->no_wakeref_tracking)
return;
- spin_lock_irqsave(&rpm->debug.lock, flags);
- for (n = rpm->debug.count; n--; ) {
- if (rpm->debug.owners[n] == stack) {
- memmove(rpm->debug.owners + n,
- rpm->debug.owners + n + 1,
- (--rpm->debug.count - n) * sizeof(stack));
- found = true;
- break;
- }
- }
- spin_unlock_irqrestore(&rpm->debug.lock, flags);
-
- if (drm_WARN(&i915->drm, !found,
- "Unmatched wakeref (tracking %lu), count %u\n",
- rpm->debug.count, atomic_read(&rpm->wakeref_count))) {
- char *buf;
-
- buf = kmalloc(PAGE_SIZE, GFP_NOWAIT | __GFP_NOWARN);
- if (!buf)
- return;
-
- stack_depot_snprint(stack, buf, PAGE_SIZE, 2);
- DRM_DEBUG_DRIVER("wakeref %x from\n%s", stack, buf);
-
- stack = READ_ONCE(rpm->debug.last_release);
- if (stack) {
- stack_depot_snprint(stack, buf, PAGE_SIZE, 2);
- DRM_DEBUG_DRIVER("wakeref last released at\n%s", buf);
- }
-
- kfree(buf);
- }
+ intel_ref_tracker_free(&rpm->debug, wakeref);
}
-static int cmphandle(const void *_a, const void *_b)
+static void untrack_all_intel_runtime_pm_wakerefs(struct intel_runtime_pm *rpm)
{
- const depot_stack_handle_t * const a = _a, * const b = _b;
-
- if (*a < *b)
- return -1;
- else if (*a > *b)
- return 1;
- else
- return 0;
-}
-
-static void
-__print_intel_runtime_pm_wakeref(struct drm_printer *p,
- const struct intel_runtime_pm_debug *dbg)
-{
- unsigned long i;
- char *buf;
-
- buf = kmalloc(PAGE_SIZE, GFP_NOWAIT | __GFP_NOWARN);
- if (!buf)
- return;
-
- if (dbg->last_acquire) {
- stack_depot_snprint(dbg->last_acquire, buf, PAGE_SIZE, 2);
- drm_printf(p, "Wakeref last acquired:\n%s", buf);
- }
-
- if (dbg->last_release) {
- stack_depot_snprint(dbg->last_release, buf, PAGE_SIZE, 2);
- drm_printf(p, "Wakeref last released:\n%s", buf);
- }
-
- drm_printf(p, "Wakeref count: %lu\n", dbg->count);
-
- sort(dbg->owners, dbg->count, sizeof(*dbg->owners), cmphandle, NULL);
-
- for (i = 0; i < dbg->count; i++) {
- depot_stack_handle_t stack = dbg->owners[i];
- unsigned long rep;
-
- rep = 1;
- while (i + 1 < dbg->count && dbg->owners[i + 1] == stack)
- rep++, i++;
- stack_depot_snprint(stack, buf, PAGE_SIZE, 2);
- drm_printf(p, "Wakeref x%lu taken at:\n%s", rep, buf);
- }
-
- kfree(buf);
-}
-
-static noinline void
-__untrack_all_wakerefs(struct intel_runtime_pm_debug *debug,
- struct intel_runtime_pm_debug *saved)
-{
- *saved = *debug;
-
- debug->owners = NULL;
- debug->count = 0;
- debug->last_release = __save_depot_stack();
-}
-
-static void
-dump_and_free_wakeref_tracking(struct intel_runtime_pm_debug *debug)
-{
- if (debug->count) {
- struct drm_printer p = drm_debug_printer("i915");
-
- __print_intel_runtime_pm_wakeref(&p, debug);
- }
-
- kfree(debug->owners);
+ ref_tracker_dir_exit(&rpm->debug);
}
static noinline void
__intel_wakeref_dec_and_check_tracking(struct intel_runtime_pm *rpm)
{
- struct intel_runtime_pm_debug dbg = {};
unsigned long flags;
if (!atomic_dec_and_lock_irqsave(&rpm->wakeref_count,
@@ -235,60 +90,14 @@ __intel_wakeref_dec_and_check_tracking(struct intel_runtime_pm *rpm)
flags))
return;
- __untrack_all_wakerefs(&rpm->debug, &dbg);
+ ref_tracker_dir_print_locked(&rpm->debug, INTEL_REFTRACK_PRINT_LIMIT);
spin_unlock_irqrestore(&rpm->debug.lock, flags);
-
- dump_and_free_wakeref_tracking(&dbg);
-}
-
-static noinline void
-untrack_all_intel_runtime_pm_wakerefs(struct intel_runtime_pm *rpm)
-{
- struct intel_runtime_pm_debug dbg = {};
- unsigned long flags;
-
- spin_lock_irqsave(&rpm->debug.lock, flags);
- __untrack_all_wakerefs(&rpm->debug, &dbg);
- spin_unlock_irqrestore(&rpm->debug.lock, flags);
-
- dump_and_free_wakeref_tracking(&dbg);
}
void print_intel_runtime_pm_wakeref(struct intel_runtime_pm *rpm,
struct drm_printer *p)
{
- struct intel_runtime_pm_debug dbg = {};
-
- do {
- unsigned long alloc = dbg.count;
- depot_stack_handle_t *s;
-
- spin_lock_irq(&rpm->debug.lock);
- dbg.count = rpm->debug.count;
- if (dbg.count <= alloc) {
- memcpy(dbg.owners,
- rpm->debug.owners,
- dbg.count * sizeof(*s));
- }
- dbg.last_acquire = rpm->debug.last_acquire;
- dbg.last_release = rpm->debug.last_release;
- spin_unlock_irq(&rpm->debug.lock);
- if (dbg.count <= alloc)
- break;
-
- s = krealloc(dbg.owners,
- dbg.count * sizeof(*s),
- GFP_NOWAIT | __GFP_NOWARN);
- if (!s)
- goto out;
-
- dbg.owners = s;
- } while (1);
-
- __print_intel_runtime_pm_wakeref(p, &dbg);
-
-out:
- kfree(dbg.owners);
+ intel_wakeref_tracker_show(&rpm->debug, p);
}
#else
@@ -297,14 +106,14 @@ static void init_intel_runtime_pm_wakeref(struct intel_runtime_pm *rpm)
{
}
-static depot_stack_handle_t
+static intel_wakeref_t
track_intel_runtime_pm_wakeref(struct intel_runtime_pm *rpm)
{
return -1;
}
static void untrack_intel_runtime_pm_wakeref(struct intel_runtime_pm *rpm,
- intel_wakeref_t wref)
+ intel_wakeref_t wakeref)
{
}
@@ -639,7 +448,11 @@ void intel_runtime_pm_driver_release(struct intel_runtime_pm *rpm)
"i915 raw-wakerefs=%d wakelocks=%d on cleanup\n",
intel_rpm_raw_wakeref_count(count),
intel_rpm_wakelock_count(count));
+}
+void intel_runtime_pm_driver_last_release(struct intel_runtime_pm *rpm)
+{
+ intel_runtime_pm_driver_release(rpm);
untrack_all_intel_runtime_pm_wakerefs(rpm);
}
diff --git a/drivers/gpu/drm/i915/intel_runtime_pm.h b/drivers/gpu/drm/i915/intel_runtime_pm.h
index e592e8d6499a1f..2f81d685bdb4d1 100644
--- a/drivers/gpu/drm/i915/intel_runtime_pm.h
+++ b/drivers/gpu/drm/i915/intel_runtime_pm.h
@@ -83,15 +83,7 @@ struct intel_runtime_pm {
* paired rpm_put) we can remove corresponding pairs of and keep
* the array trimmed to active wakerefs.
*/
- struct intel_runtime_pm_debug {
- spinlock_t lock;
-
- depot_stack_handle_t last_acquire;
- depot_stack_handle_t last_release;
-
- depot_stack_handle_t *owners;
- unsigned long count;
- } debug;
+ struct ref_tracker_dir debug;
#endif
};
@@ -195,6 +187,7 @@ void intel_runtime_pm_init_early(struct intel_runtime_pm *rpm);
void intel_runtime_pm_enable(struct intel_runtime_pm *rpm);
void intel_runtime_pm_disable(struct intel_runtime_pm *rpm);
void intel_runtime_pm_driver_release(struct intel_runtime_pm *rpm);
+void intel_runtime_pm_driver_last_release(struct intel_runtime_pm *rpm);
intel_wakeref_t intel_runtime_pm_get(struct intel_runtime_pm *rpm);
intel_wakeref_t intel_runtime_pm_get_if_in_use(struct intel_runtime_pm *rpm);
diff --git a/drivers/gpu/drm/i915/intel_wakeref.h b/drivers/gpu/drm/i915/intel_wakeref.h
index 0b6b4852ab236a..99ee6254976ebf 100644
--- a/drivers/gpu/drm/i915/intel_wakeref.h
+++ b/drivers/gpu/drm/i915/intel_wakeref.h
@@ -7,16 +7,25 @@
#ifndef INTEL_WAKEREF_H
#define INTEL_WAKEREF_H
+#include <drm/drm_print.h>
+
#include <linux/atomic.h>
#include <linux/bitfield.h>
#include <linux/bits.h>
#include <linux/lockdep.h>
#include <linux/mutex.h>
#include <linux/refcount.h>
+#include <linux/ref_tracker.h>
+#include <linux/slab.h>
#include <linux/stackdepot.h>
#include <linux/timer.h>
#include <linux/workqueue.h>
+typedef unsigned long intel_wakeref_t;
+
+#define INTEL_REFTRACK_DEAD_COUNT 16
+#define INTEL_REFTRACK_PRINT_LIMIT 16
+
#if IS_ENABLED(CONFIG_DRM_I915_DEBUG)
#define INTEL_WAKEREF_BUG_ON(expr) BUG_ON(expr)
#else
@@ -26,8 +35,6 @@
struct intel_runtime_pm;
struct intel_wakeref;
-typedef depot_stack_handle_t intel_wakeref_t;
-
struct intel_wakeref_ops {
int (*get)(struct intel_wakeref *wf);
int (*put)(struct intel_wakeref *wf);
@@ -261,6 +268,56 @@ __intel_wakeref_defer_park(struct intel_wakeref *wf)
*/
int intel_wakeref_wait_for_idle(struct intel_wakeref *wf);
+#define INTEL_WAKEREF_DEF ((intel_wakeref_t)(-1))
+
+static inline intel_wakeref_t intel_ref_tracker_alloc(struct ref_tracker_dir *dir)
+{
+ struct ref_tracker *user = NULL;
+
+ ref_tracker_alloc(dir, &user, GFP_NOWAIT);
+
+ return (intel_wakeref_t)user ?: INTEL_WAKEREF_DEF;
+}
+
+static inline void intel_ref_tracker_free(struct ref_tracker_dir *dir,
+ intel_wakeref_t handle)
+{
+ struct ref_tracker *user;
+
+ user = (handle == INTEL_WAKEREF_DEF) ? NULL : (void *)handle;
+
+ ref_tracker_free(dir, &user);
+}
+
+static inline void
+intel_wakeref_tracker_show(struct ref_tracker_dir *dir,
+ struct drm_printer *p)
+{
+ const size_t buf_size = PAGE_SIZE;
+ char *buf, *sb, *se;
+ size_t count;
+
+ buf = kmalloc(buf_size, GFP_NOWAIT);
+ if (!buf)
+ return;
+
+ count = ref_tracker_dir_snprint(dir, buf, buf_size);
+ if (!count)
+ goto free;
+ /* printk does not like big buffers, so we split it */
+ for (sb = buf; *sb; sb = se + 1) {
+ se = strchrnul(sb, '\n');
+ drm_printf(p, "%.*s", (int)(se - sb + 1), sb);
+ if (!*se)
+ break;
+ }
+ if (count >= buf_size)
+ drm_printf(p, "\n...dropped %zd extra bytes of leak report.\n",
+ count + 1 - buf_size);
+free:
+ kfree(buf);
+}
+
struct intel_wakeref_auto {
struct intel_runtime_pm *rpm;
struct timer_list timer;
--
2.34.1
Track every intel_gt_pm_get() until its corresponding release in
intel_gt_pm_put() by returning a cookie to the caller for acquire that
must be passed by on released. When there is an imbalance, we can see who
either tried to free a stale wakeref, or who forgot to free theirs.
Signed-off-by: Andrzej Hajda <[email protected]>
---
drivers/gpu/drm/i915/Kconfig.debug | 15 +++++++++
drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c | 7 ++--
.../drm/i915/gem/selftests/i915_gem_coherency.c | 10 +++---
drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c | 14 ++++----
drivers/gpu/drm/i915/gt/intel_breadcrumbs.c | 13 +++++---
drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h | 3 +-
drivers/gpu/drm/i915/gt/intel_context.h | 4 +--
drivers/gpu/drm/i915/gt/intel_context_types.h | 2 ++
drivers/gpu/drm/i915/gt/intel_engine_pm.c | 7 ++--
drivers/gpu/drm/i915/gt/intel_engine_types.h | 2 ++
.../gpu/drm/i915/gt/intel_execlists_submission.c | 2 +-
drivers/gpu/drm/i915/gt/intel_gt_pm.c | 12 ++++---
drivers/gpu/drm/i915/gt/intel_gt_pm.h | 38 +++++++++++++++++-----
drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.c | 4 +--
drivers/gpu/drm/i915/gt/selftest_engine_cs.c | 20 +++++++-----
drivers/gpu/drm/i915/gt/selftest_gt_pm.c | 5 +--
drivers/gpu/drm/i915/gt/selftest_reset.c | 10 +++---
drivers/gpu/drm/i915/gt/selftest_rps.c | 17 ++++++----
drivers/gpu/drm/i915/gt/selftest_slpc.c | 5 +--
drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c | 9 ++---
drivers/gpu/drm/i915/i915_pmu.c | 16 +++++----
drivers/gpu/drm/i915/intel_wakeref.c | 7 +++-
drivers/gpu/drm/i915/intel_wakeref.h | 38 ++++++++++++++++++++--
23 files changed, 182 insertions(+), 78 deletions(-)
diff --git a/drivers/gpu/drm/i915/Kconfig.debug b/drivers/gpu/drm/i915/Kconfig.debug
index 76454fcbf65228..52e9083a50282e 100644
--- a/drivers/gpu/drm/i915/Kconfig.debug
+++ b/drivers/gpu/drm/i915/Kconfig.debug
@@ -40,6 +40,7 @@ config DRM_I915_DEBUG
select DRM_I915_DEBUG_GEM_ONCE
select DRM_I915_DEBUG_MMIO
select DRM_I915_DEBUG_RUNTIME_PM
+ select DRM_I915_DEBUG_WAKEREF
select DRM_I915_SW_FENCE_DEBUG_OBJECTS
select DRM_I915_SELFTEST
default n
@@ -243,3 +244,17 @@ config DRM_I915_DEBUG_RUNTIME_PM
Recommended for driver developers only.
If in doubt, say "N"
+
+config DRM_I915_DEBUG_WAKEREF
+ bool "Enable extra tracking for wakerefs"
+ depends on DRM_I915
+ default n
+ select REF_TRACKER
+ select STACKDEPOT
+ select STACKTRACE
+ help
+ Choose this option to turn on extra state checking and usage
+ tracking for the wakerefPM functionality. This may introduce
+ overhead during driver runtime.
+
+ If in doubt, say "N"
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index 3aeede6aee4dcc..33a034a9c42f11 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -253,6 +253,7 @@ struct i915_execbuffer {
struct intel_gt *gt; /* gt for the execbuf */
struct intel_context *context; /* logical state for the request */
struct i915_gem_context *gem_context; /** caller's context */
+ intel_wakeref_t wakeref;
/** our requests to build */
struct i915_request *requests[MAX_ENGINE_INSTANCE + 1];
@@ -2709,7 +2710,7 @@ eb_select_engine(struct i915_execbuffer *eb)
for_each_child(ce, child)
intel_context_get(child);
- intel_gt_pm_get(ce->engine->gt);
+ eb->wakeref = intel_gt_pm_get(ce->engine->gt);
if (!test_bit(CONTEXT_ALLOC_BIT, &ce->flags)) {
err = intel_context_alloc_state(ce);
@@ -2748,7 +2749,7 @@ eb_select_engine(struct i915_execbuffer *eb)
return err;
err:
- intel_gt_pm_put(ce->engine->gt);
+ intel_gt_pm_put(ce->engine->gt, eb->wakeref);
for_each_child(ce, child)
intel_context_put(child);
intel_context_put(ce);
@@ -2761,7 +2762,7 @@ eb_put_engine(struct i915_execbuffer *eb)
struct intel_context *child;
i915_vm_put(eb->context->vm);
- intel_gt_pm_put(eb->gt);
+ intel_gt_pm_put(eb->context->engine->gt, eb->wakeref);
for_each_child(eb->context, child)
intel_context_put(child);
intel_context_put(eb->context);
diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_coherency.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_coherency.c
index 3bef1beec7cbb5..3fd68a099a85ef 100644
--- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_coherency.c
+++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_coherency.c
@@ -85,6 +85,7 @@ static int cpu_get(struct context *ctx, unsigned long offset, u32 *v)
static int gtt_set(struct context *ctx, unsigned long offset, u32 v)
{
+ intel_wakeref_t wakeref;
struct i915_vma *vma;
u32 __iomem *map;
int err = 0;
@@ -99,7 +100,7 @@ static int gtt_set(struct context *ctx, unsigned long offset, u32 v)
if (IS_ERR(vma))
return PTR_ERR(vma);
- intel_gt_pm_get(vma->vm->gt);
+ wakeref = intel_gt_pm_get(vma->vm->gt);
map = i915_vma_pin_iomap(vma);
i915_vma_unpin(vma);
@@ -112,12 +113,13 @@ static int gtt_set(struct context *ctx, unsigned long offset, u32 v)
i915_vma_unpin_iomap(vma);
out_rpm:
- intel_gt_pm_put(vma->vm->gt);
+ intel_gt_pm_put(vma->vm->gt, wakeref);
return err;
}
static int gtt_get(struct context *ctx, unsigned long offset, u32 *v)
{
+ intel_wakeref_t wakeref;
struct i915_vma *vma;
u32 __iomem *map;
int err = 0;
@@ -132,7 +134,7 @@ static int gtt_get(struct context *ctx, unsigned long offset, u32 *v)
if (IS_ERR(vma))
return PTR_ERR(vma);
- intel_gt_pm_get(vma->vm->gt);
+ wakeref = intel_gt_pm_get(vma->vm->gt);
map = i915_vma_pin_iomap(vma);
i915_vma_unpin(vma);
@@ -145,7 +147,7 @@ static int gtt_get(struct context *ctx, unsigned long offset, u32 *v)
i915_vma_unpin_iomap(vma);
out_rpm:
- intel_gt_pm_put(vma->vm->gt);
+ intel_gt_pm_put(vma->vm->gt, wakeref);
return err;
}
diff --git a/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c b/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c
index 56279908ed305b..f6f36a85688814 100644
--- a/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c
+++ b/drivers/gpu/drm/i915/gem/selftests/i915_gem_mman.c
@@ -630,14 +630,14 @@ static bool assert_mmap_offset(struct drm_i915_private *i915,
static void disable_retire_worker(struct drm_i915_private *i915)
{
i915_gem_driver_unregister__shrinker(i915);
- intel_gt_pm_get(to_gt(i915));
+ intel_gt_pm_get_untracked(to_gt(i915));
cancel_delayed_work_sync(&to_gt(i915)->requests.retire_work);
}
static void restore_retire_worker(struct drm_i915_private *i915)
{
igt_flush_test(i915);
- intel_gt_pm_put(to_gt(i915));
+ intel_gt_pm_put_untracked(to_gt(i915));
i915_gem_driver_register__shrinker(i915);
}
@@ -778,6 +778,7 @@ static int igt_mmap_offset_exhaustion(void *arg)
static int gtt_set(struct drm_i915_gem_object *obj)
{
+ intel_wakeref_t wakeref;
struct i915_vma *vma;
void __iomem *map;
int err = 0;
@@ -786,7 +787,7 @@ static int gtt_set(struct drm_i915_gem_object *obj)
if (IS_ERR(vma))
return PTR_ERR(vma);
- intel_gt_pm_get(vma->vm->gt);
+ wakeref = intel_gt_pm_get(vma->vm->gt);
map = i915_vma_pin_iomap(vma);
i915_vma_unpin(vma);
if (IS_ERR(map)) {
@@ -798,12 +799,13 @@ static int gtt_set(struct drm_i915_gem_object *obj)
i915_vma_unpin_iomap(vma);
out:
- intel_gt_pm_put(vma->vm->gt);
+ intel_gt_pm_put(vma->vm->gt, wakeref);
return err;
}
static int gtt_check(struct drm_i915_gem_object *obj)
{
+ intel_wakeref_t wakeref;
struct i915_vma *vma;
void __iomem *map;
int err = 0;
@@ -812,7 +814,7 @@ static int gtt_check(struct drm_i915_gem_object *obj)
if (IS_ERR(vma))
return PTR_ERR(vma);
- intel_gt_pm_get(vma->vm->gt);
+ wakeref = intel_gt_pm_get(vma->vm->gt);
map = i915_vma_pin_iomap(vma);
i915_vma_unpin(vma);
if (IS_ERR(map)) {
@@ -828,7 +830,7 @@ static int gtt_check(struct drm_i915_gem_object *obj)
i915_vma_unpin_iomap(vma);
out:
- intel_gt_pm_put(vma->vm->gt);
+ intel_gt_pm_put(vma->vm->gt, wakeref);
return err;
}
diff --git a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
index ecc990ec1b9526..d650beb8ed22f6 100644
--- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs.c
@@ -28,11 +28,14 @@ static void irq_disable(struct intel_breadcrumbs *b)
static void __intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
{
+ intel_wakeref_t wakeref;
+
/*
* Since we are waiting on a request, the GPU should be busy
* and should have its own rpm reference.
*/
- if (GEM_WARN_ON(!intel_gt_pm_get_if_awake(b->irq_engine->gt)))
+ wakeref = intel_gt_pm_get_if_awake(b->irq_engine->gt);
+ if (GEM_WARN_ON(!wakeref))
return;
/*
@@ -41,7 +44,7 @@ static void __intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
* which we can add a new waiter and avoid the cost of re-enabling
* the irq.
*/
- WRITE_ONCE(b->irq_armed, true);
+ WRITE_ONCE(b->irq_armed, wakeref);
/* Requests may have completed before we could enable the interrupt. */
if (!b->irq_enabled++ && b->irq_enable(b))
@@ -61,12 +64,14 @@ static void intel_breadcrumbs_arm_irq(struct intel_breadcrumbs *b)
static void __intel_breadcrumbs_disarm_irq(struct intel_breadcrumbs *b)
{
+ intel_wakeref_t wakeref = b->irq_armed;
+
GEM_BUG_ON(!b->irq_enabled);
if (!--b->irq_enabled)
b->irq_disable(b);
- WRITE_ONCE(b->irq_armed, false);
- intel_gt_pm_put_async(b->irq_engine->gt);
+ WRITE_ONCE(b->irq_armed, 0);
+ intel_gt_pm_put_async(b->irq_engine->gt, wakeref);
}
static void intel_breadcrumbs_disarm_irq(struct intel_breadcrumbs *b)
diff --git a/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h b/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h
index 72dfd3748c4c33..bdf09fd67b6e70 100644
--- a/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_breadcrumbs_types.h
@@ -13,6 +13,7 @@
#include <linux/types.h>
#include "intel_engine_types.h"
+#include "intel_wakeref.h"
/*
* Rather than have every client wait upon all user interrupts,
@@ -43,7 +44,7 @@ struct intel_breadcrumbs {
spinlock_t irq_lock; /* protects the interrupt from hardirq context */
struct irq_work irq_work; /* for use from inside irq_lock */
unsigned int irq_enabled;
- bool irq_armed;
+ intel_wakeref_t irq_armed;
/* Not all breadcrumbs are attached to physical HW */
intel_engine_mask_t engine_mask;
diff --git a/drivers/gpu/drm/i915/gt/intel_context.h b/drivers/gpu/drm/i915/gt/intel_context.h
index 48f888c3da083b..582faa21181e58 100644
--- a/drivers/gpu/drm/i915/gt/intel_context.h
+++ b/drivers/gpu/drm/i915/gt/intel_context.h
@@ -212,7 +212,7 @@ static inline void intel_context_enter(struct intel_context *ce)
return;
ce->ops->enter(ce);
- intel_gt_pm_get(ce->vm->gt);
+ ce->wakeref = intel_gt_pm_get(ce->vm->gt);
}
static inline void intel_context_mark_active(struct intel_context *ce)
@@ -229,7 +229,7 @@ static inline void intel_context_exit(struct intel_context *ce)
if (--ce->active_count)
return;
- intel_gt_pm_put_async(ce->vm->gt);
+ intel_gt_pm_put_async(ce->vm->gt, ce->wakeref);
ce->ops->exit(ce);
}
diff --git a/drivers/gpu/drm/i915/gt/intel_context_types.h b/drivers/gpu/drm/i915/gt/intel_context_types.h
index e36670f2e6260b..5dc39a9d7a501c 100644
--- a/drivers/gpu/drm/i915/gt/intel_context_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_context_types.h
@@ -17,6 +17,7 @@
#include "i915_utils.h"
#include "intel_engine_types.h"
#include "intel_sseu.h"
+#include "intel_wakeref.h"
#include "uc/intel_guc_fwif.h"
@@ -110,6 +111,7 @@ struct intel_context {
u32 ring_size;
struct intel_ring *ring;
struct intel_timeline *timeline;
+ intel_wakeref_t wakeref;
unsigned long flags;
#define CONTEXT_BARRIER_BIT 0
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_pm.c b/drivers/gpu/drm/i915/gt/intel_engine_pm.c
index ee531a5c142c77..bb021d83d3f882 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_pm.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_pm.c
@@ -63,7 +63,7 @@ static int __engine_unpark(struct intel_wakeref *wf)
ENGINE_TRACE(engine, "\n");
- intel_gt_pm_get(engine->gt);
+ engine->wakeref_track = intel_gt_pm_get(engine->gt);
/* Discard stale context state from across idling */
ce = engine->kernel_context;
@@ -122,6 +122,7 @@ __queue_and_release_pm(struct i915_request *rq,
*/
GEM_BUG_ON(rq->context->active_count != 1);
__intel_gt_pm_get(engine->gt);
+ rq->context->wakeref = intel_wakeref_track(&engine->gt->wakeref);
/*
* We have to serialise all potential retirement paths with our
@@ -285,7 +286,7 @@ static int __engine_park(struct intel_wakeref *wf)
engine->park(engine);
/* While gt calls i915_vma_parked(), we have to break the lock cycle */
- intel_gt_pm_put_async(engine->gt);
+ intel_gt_pm_put_async(engine->gt, engine->wakeref_track);
return 0;
}
@@ -298,7 +299,7 @@ void intel_engine_init__pm(struct intel_engine_cs *engine)
{
struct intel_runtime_pm *rpm = engine->uncore->rpm;
- intel_wakeref_init(&engine->wakeref, rpm, &wf_ops);
+ intel_wakeref_init(&engine->wakeref, rpm, &wf_ops, engine->name);
intel_engine_init_heartbeat(engine);
intel_gsc_idle_msg_enable(engine);
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_types.h b/drivers/gpu/drm/i915/gt/intel_engine_types.h
index 960291f88fd6cf..f55b524e518035 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_types.h
+++ b/drivers/gpu/drm/i915/gt/intel_engine_types.h
@@ -433,7 +433,9 @@ struct intel_engine_cs {
unsigned long serial;
unsigned long wakeref_serial;
+ intel_wakeref_t wakeref_track;
struct intel_wakeref wakeref;
+
struct file *default_state;
struct {
diff --git a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
index 750326434677f1..f7d674c3c6ee65 100644
--- a/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
+++ b/drivers/gpu/drm/i915/gt/intel_execlists_submission.c
@@ -630,7 +630,7 @@ static void __execlists_schedule_out(struct i915_request * const rq,
execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_OUT);
if (engine->fw_domain && !--engine->fw_active)
intel_uncore_forcewake_put(engine->uncore, engine->fw_domain);
- intel_gt_pm_put_async(engine->gt);
+ intel_gt_pm_put_async_untracked(engine->gt);
/*
* If this is part of a virtual engine, its next request may
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.c b/drivers/gpu/drm/i915/gt/intel_gt_pm.c
index e02cb90723ae03..76378dd2873dfc 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt_pm.c
+++ b/drivers/gpu/drm/i915/gt/intel_gt_pm.c
@@ -27,19 +27,20 @@
static void user_forcewake(struct intel_gt *gt, bool suspend)
{
int count = atomic_read(>->user_wakeref);
+ intel_wakeref_t wakeref;
/* Inside suspend/resume so single threaded, no races to worry about. */
if (likely(!count))
return;
- intel_gt_pm_get(gt);
+ wakeref = intel_gt_pm_get(gt);
if (suspend) {
GEM_BUG_ON(count > atomic_read(>->wakeref.count));
atomic_sub(count, >->wakeref.count);
} else {
atomic_add(count, >->wakeref.count);
}
- intel_gt_pm_put(gt);
+ intel_gt_pm_put(gt, wakeref);
}
static void runtime_begin(struct intel_gt *gt)
@@ -137,7 +138,7 @@ void intel_gt_pm_init_early(struct intel_gt *gt)
* runtime_pm is per-device rather than per-tile, so this is still the
* correct structure.
*/
- intel_wakeref_init(>->wakeref, >->i915->runtime_pm, &wf_ops);
+ intel_wakeref_init(>->wakeref, >->i915->runtime_pm, &wf_ops, "GT");
seqcount_mutex_init(>->stats.lock, >->wakeref.mutex);
}
@@ -220,6 +221,7 @@ int intel_gt_resume(struct intel_gt *gt)
{
struct intel_engine_cs *engine;
enum intel_engine_id id;
+ intel_wakeref_t wakeref;
int err;
err = intel_gt_has_unrecoverable_error(gt);
@@ -236,7 +238,7 @@ int intel_gt_resume(struct intel_gt *gt)
*/
gt_sanitize(gt, true);
- intel_gt_pm_get(gt);
+ wakeref = intel_gt_pm_get(gt);
intel_uncore_forcewake_get(gt->uncore, FORCEWAKE_ALL);
intel_rc6_sanitize(>->rc6);
@@ -279,7 +281,7 @@ int intel_gt_resume(struct intel_gt *gt)
out_fw:
intel_uncore_forcewake_put(gt->uncore, FORCEWAKE_ALL);
- intel_gt_pm_put(gt);
+ intel_gt_pm_put(gt, wakeref);
return err;
err_wedged:
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm.h b/drivers/gpu/drm/i915/gt/intel_gt_pm.h
index 6c9a4645236467..2ae5fbbede6ca8 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt_pm.h
+++ b/drivers/gpu/drm/i915/gt/intel_gt_pm.h
@@ -16,19 +16,28 @@ static inline bool intel_gt_pm_is_awake(const struct intel_gt *gt)
return intel_wakeref_is_active(>->wakeref);
}
-static inline void intel_gt_pm_get(struct intel_gt *gt)
+static inline void intel_gt_pm_get_untracked(struct intel_gt *gt)
{
intel_wakeref_get(>->wakeref);
}
+static inline intel_wakeref_t intel_gt_pm_get(struct intel_gt *gt)
+{
+ intel_gt_pm_get_untracked(gt);
+ return intel_wakeref_track(>->wakeref);
+}
+
static inline void __intel_gt_pm_get(struct intel_gt *gt)
{
__intel_wakeref_get(>->wakeref);
}
-static inline bool intel_gt_pm_get_if_awake(struct intel_gt *gt)
+static inline intel_wakeref_t intel_gt_pm_get_if_awake(struct intel_gt *gt)
{
- return intel_wakeref_get_if_active(>->wakeref);
+ if (!intel_wakeref_get_if_active(>->wakeref))
+ return 0;
+
+ return intel_wakeref_track(>->wakeref);
}
static inline void intel_gt_pm_might_get(struct intel_gt *gt)
@@ -36,12 +45,18 @@ static inline void intel_gt_pm_might_get(struct intel_gt *gt)
intel_wakeref_might_get(>->wakeref);
}
-static inline void intel_gt_pm_put(struct intel_gt *gt)
+static inline void intel_gt_pm_put_untracked(struct intel_gt *gt)
{
intel_wakeref_put(>->wakeref);
}
-static inline void intel_gt_pm_put_async(struct intel_gt *gt)
+static inline void intel_gt_pm_put(struct intel_gt *gt, intel_wakeref_t handle)
+{
+ intel_wakeref_untrack(>->wakeref, handle);
+ intel_gt_pm_put_untracked(gt);
+}
+
+static inline void intel_gt_pm_put_async_untracked(struct intel_gt *gt)
{
intel_wakeref_put_async(>->wakeref);
}
@@ -51,9 +66,14 @@ static inline void intel_gt_pm_might_put(struct intel_gt *gt)
intel_wakeref_might_put(>->wakeref);
}
-#define with_intel_gt_pm(gt, tmp) \
- for (tmp = 1, intel_gt_pm_get(gt); tmp; \
- intel_gt_pm_put(gt), tmp = 0)
+static inline void intel_gt_pm_put_async(struct intel_gt *gt, intel_wakeref_t handle)
+{
+ intel_wakeref_untrack(>->wakeref, handle);
+ intel_gt_pm_put_async_untracked(gt);
+}
+
+#define with_intel_gt_pm(gt, wf) \
+ for (wf = intel_gt_pm_get(gt); wf; intel_gt_pm_put(gt, wf), wf = 0)
/**
* with_intel_gt_pm_if_awake - if GT is PM awake, get a reference to prevent
@@ -64,7 +84,7 @@ static inline void intel_gt_pm_might_put(struct intel_gt *gt)
* @wf: pointer to a temporary wakeref.
*/
#define with_intel_gt_pm_if_awake(gt, wf) \
- for (wf = intel_gt_pm_get_if_awake(gt); wf; intel_gt_pm_put_async(gt), wf = 0)
+ for (wf = intel_gt_pm_get_if_awake(gt); wf; intel_gt_pm_put_async(gt, wf), wf = 0)
static inline int intel_gt_pm_wait_for_idle(struct intel_gt *gt)
{
diff --git a/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.c b/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.c
index 80dbbef86b1dbf..0960fae5a0b205 100644
--- a/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.c
+++ b/drivers/gpu/drm/i915/gt/intel_gt_pm_debugfs.c
@@ -27,7 +27,7 @@
void intel_gt_pm_debugfs_forcewake_user_open(struct intel_gt *gt)
{
atomic_inc(>->user_wakeref);
- intel_gt_pm_get(gt);
+ intel_gt_pm_get_untracked(gt);
if (GRAPHICS_VER(gt->i915) >= 6)
intel_uncore_forcewake_user_get(gt->uncore);
}
@@ -36,7 +36,7 @@ void intel_gt_pm_debugfs_forcewake_user_release(struct intel_gt *gt)
{
if (GRAPHICS_VER(gt->i915) >= 6)
intel_uncore_forcewake_user_put(gt->uncore);
- intel_gt_pm_put(gt);
+ intel_gt_pm_put_untracked(gt);
atomic_dec(>->user_wakeref);
}
diff --git a/drivers/gpu/drm/i915/gt/selftest_engine_cs.c b/drivers/gpu/drm/i915/gt/selftest_engine_cs.c
index 542ce6d2de1922..f3d21bdfd7f639 100644
--- a/drivers/gpu/drm/i915/gt/selftest_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/selftest_engine_cs.c
@@ -21,20 +21,22 @@ static int cmp_u32(const void *A, const void *B)
return *a - *b;
}
-static void perf_begin(struct intel_gt *gt)
+static intel_wakeref_t perf_begin(struct intel_gt *gt)
{
- intel_gt_pm_get(gt);
+ intel_wakeref_t wakeref = intel_gt_pm_get(gt);
/* Boost gpufreq to max [waitboost] and keep it fixed */
atomic_inc(>->rps.num_waiters);
schedule_work(>->rps.work);
flush_work(>->rps.work);
+
+ return wakeref;
}
-static int perf_end(struct intel_gt *gt)
+static int perf_end(struct intel_gt *gt, intel_wakeref_t wakeref)
{
atomic_dec(>->rps.num_waiters);
- intel_gt_pm_put(gt);
+ intel_gt_pm_put(gt, wakeref);
return igt_flush_test(gt->i915);
}
@@ -133,12 +135,13 @@ static int perf_mi_bb_start(void *arg)
struct intel_gt *gt = arg;
struct intel_engine_cs *engine;
enum intel_engine_id id;
+ intel_wakeref_t wakeref;
int err = 0;
if (GRAPHICS_VER(gt->i915) < 4) /* Any CS_TIMESTAMP? */
return 0;
- perf_begin(gt);
+ wakeref = perf_begin(gt);
for_each_engine(engine, gt, id) {
struct intel_context *ce = engine->kernel_context;
struct i915_vma *batch;
@@ -207,7 +210,7 @@ static int perf_mi_bb_start(void *arg)
pr_info("%s: MI_BB_START cycles: %u\n",
engine->name, trifilter(cycles));
}
- if (perf_end(gt))
+ if (perf_end(gt, wakeref))
err = -EIO;
return err;
@@ -260,12 +263,13 @@ static int perf_mi_noop(void *arg)
struct intel_gt *gt = arg;
struct intel_engine_cs *engine;
enum intel_engine_id id;
+ intel_wakeref_t wakeref;
int err = 0;
if (GRAPHICS_VER(gt->i915) < 4) /* Any CS_TIMESTAMP? */
return 0;
- perf_begin(gt);
+ wakeref = perf_begin(gt);
for_each_engine(engine, gt, id) {
struct intel_context *ce = engine->kernel_context;
struct i915_vma *base, *nop;
@@ -364,7 +368,7 @@ static int perf_mi_noop(void *arg)
pr_info("%s: 16K MI_NOOP cycles: %u\n",
engine->name, trifilter(cycles));
}
- if (perf_end(gt))
+ if (perf_end(gt, wakeref))
err = -EIO;
return err;
diff --git a/drivers/gpu/drm/i915/gt/selftest_gt_pm.c b/drivers/gpu/drm/i915/gt/selftest_gt_pm.c
index 0971241707ce83..33351deeea4f0b 100644
--- a/drivers/gpu/drm/i915/gt/selftest_gt_pm.c
+++ b/drivers/gpu/drm/i915/gt/selftest_gt_pm.c
@@ -81,6 +81,7 @@ static int live_gt_clocks(void *arg)
struct intel_gt *gt = arg;
struct intel_engine_cs *engine;
enum intel_engine_id id;
+ intel_wakeref_t wakeref;
int err = 0;
if (!gt->clock_frequency) { /* unknown */
@@ -91,7 +92,7 @@ static int live_gt_clocks(void *arg)
if (GRAPHICS_VER(gt->i915) < 4) /* Any CS_TIMESTAMP? */
return 0;
- intel_gt_pm_get(gt);
+ wakeref = intel_gt_pm_get(gt);
intel_uncore_forcewake_get(gt->uncore, FORCEWAKE_ALL);
for_each_engine(engine, gt, id) {
@@ -128,7 +129,7 @@ static int live_gt_clocks(void *arg)
}
intel_uncore_forcewake_put(gt->uncore, FORCEWAKE_ALL);
- intel_gt_pm_put(gt);
+ intel_gt_pm_put(gt, wakeref);
return err;
}
diff --git a/drivers/gpu/drm/i915/gt/selftest_reset.c b/drivers/gpu/drm/i915/gt/selftest_reset.c
index a9e0a91bc0e026..ab464cd72b8c39 100644
--- a/drivers/gpu/drm/i915/gt/selftest_reset.c
+++ b/drivers/gpu/drm/i915/gt/selftest_reset.c
@@ -257,11 +257,12 @@ static int igt_atomic_reset(void *arg)
{
struct intel_gt *gt = arg;
const typeof(*igt_atomic_phases) *p;
+ intel_wakeref_t wakeref;
int err = 0;
/* Check that the resets are usable from atomic context */
- intel_gt_pm_get(gt);
+ wakeref = intel_gt_pm_get(gt);
igt_global_reset_lock(gt);
/* Flush any requests before we get started and check basics */
@@ -292,7 +293,7 @@ static int igt_atomic_reset(void *arg)
unlock:
igt_global_reset_unlock(gt);
- intel_gt_pm_put(gt);
+ intel_gt_pm_put(gt, wakeref);
return err;
}
@@ -303,6 +304,7 @@ static int igt_atomic_engine_reset(void *arg)
const typeof(*igt_atomic_phases) *p;
struct intel_engine_cs *engine;
enum intel_engine_id id;
+ intel_wakeref_t wakeref;
int err = 0;
/* Check that the resets are usable from atomic context */
@@ -313,7 +315,7 @@ static int igt_atomic_engine_reset(void *arg)
if (intel_uc_uses_guc_submission(>->uc))
return 0;
- intel_gt_pm_get(gt);
+ wakeref = intel_gt_pm_get(gt);
igt_global_reset_lock(gt);
/* Flush any requests before we get started and check basics */
@@ -361,7 +363,7 @@ static int igt_atomic_engine_reset(void *arg)
out_unlock:
igt_global_reset_unlock(gt);
- intel_gt_pm_put(gt);
+ intel_gt_pm_put(gt, wakeref);
return err;
}
diff --git a/drivers/gpu/drm/i915/gt/selftest_rps.c b/drivers/gpu/drm/i915/gt/selftest_rps.c
index 84e77e8dbba19c..514ad27c3654ba 100644
--- a/drivers/gpu/drm/i915/gt/selftest_rps.c
+++ b/drivers/gpu/drm/i915/gt/selftest_rps.c
@@ -223,6 +223,7 @@ int live_rps_clock_interval(void *arg)
struct intel_engine_cs *engine;
enum intel_engine_id id;
struct igt_spinner spin;
+ intel_wakeref_t wakeref;
int err = 0;
if (!intel_rps_is_enabled(rps) || GRAPHICS_VER(gt->i915) < 6)
@@ -235,7 +236,7 @@ int live_rps_clock_interval(void *arg)
saved_work = rps->work.func;
rps->work.func = dummy_rps_work;
- intel_gt_pm_get(gt);
+ wakeref = intel_gt_pm_get(gt);
intel_rps_disable(>->rps);
intel_gt_check_clock_frequency(gt);
@@ -354,7 +355,7 @@ int live_rps_clock_interval(void *arg)
}
intel_rps_enable(>->rps);
- intel_gt_pm_put(gt);
+ intel_gt_pm_put(gt, wakeref);
igt_spinner_fini(&spin);
@@ -375,6 +376,7 @@ int live_rps_control(void *arg)
struct intel_engine_cs *engine;
enum intel_engine_id id;
struct igt_spinner spin;
+ intel_wakeref_t wakeref;
int err = 0;
/*
@@ -397,7 +399,7 @@ int live_rps_control(void *arg)
saved_work = rps->work.func;
rps->work.func = dummy_rps_work;
- intel_gt_pm_get(gt);
+ wakeref = intel_gt_pm_get(gt);
for_each_engine(engine, gt, id) {
struct i915_request *rq;
ktime_t min_dt, max_dt;
@@ -487,7 +489,7 @@ int live_rps_control(void *arg)
break;
}
}
- intel_gt_pm_put(gt);
+ intel_gt_pm_put(gt, wakeref);
igt_spinner_fini(&spin);
@@ -1022,6 +1024,7 @@ int live_rps_interrupt(void *arg)
struct intel_engine_cs *engine;
enum intel_engine_id id;
struct igt_spinner spin;
+ intel_wakeref_t wakeref;
u32 pm_events;
int err = 0;
@@ -1032,9 +1035,9 @@ int live_rps_interrupt(void *arg)
if (!intel_rps_has_interrupts(rps) || GRAPHICS_VER(gt->i915) < 6)
return 0;
- intel_gt_pm_get(gt);
- pm_events = rps->pm_events;
- intel_gt_pm_put(gt);
+ pm_events = 0;
+ with_intel_gt_pm(gt, wakeref)
+ pm_events = rps->pm_events;
if (!pm_events) {
pr_err("No RPS PM events registered, but RPS is enabled?\n");
return -ENODEV;
diff --git a/drivers/gpu/drm/i915/gt/selftest_slpc.c b/drivers/gpu/drm/i915/gt/selftest_slpc.c
index bd44ce73a5044e..81fa891eac19fc 100644
--- a/drivers/gpu/drm/i915/gt/selftest_slpc.c
+++ b/drivers/gpu/drm/i915/gt/selftest_slpc.c
@@ -241,6 +241,7 @@ static int run_test(struct intel_gt *gt, int test_type)
struct intel_rps *rps = >->rps;
struct intel_engine_cs *engine;
enum intel_engine_id id;
+ intel_wakeref_t wakeref;
struct igt_spinner spin;
u32 slpc_min_freq, slpc_max_freq;
int err = 0;
@@ -278,7 +279,7 @@ static int run_test(struct intel_gt *gt, int test_type)
}
intel_gt_pm_wait_for_idle(gt);
- intel_gt_pm_get(gt);
+ wakeref = intel_gt_pm_get(gt);
for_each_engine(engine, gt, id) {
struct i915_request *rq;
u32 max_act_freq;
@@ -365,7 +366,7 @@ static int run_test(struct intel_gt *gt, int test_type)
if (igt_flush_test(gt->i915))
err = -EIO;
- intel_gt_pm_put(gt);
+ intel_gt_pm_put(gt, wakeref);
igt_spinner_fini(&spin);
intel_gt_pm_wait_for_idle(gt);
diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
index 74d28a3af2d57b..a1960d46a50902 100644
--- a/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_submission.c
@@ -1106,7 +1106,7 @@ static void scrub_guc_desc_for_outstanding_g2h(struct intel_guc *guc)
if (deregister)
guc_signal_context_fence(ce);
if (destroyed) {
- intel_gt_pm_put_async(guc_to_gt(guc));
+ intel_gt_pm_put_async_untracked(guc_to_gt(guc));
release_guc_id(guc, ce);
__guc_context_destroy(ce);
}
@@ -1302,6 +1302,7 @@ static ktime_t guc_engine_busyness(struct intel_engine_cs *engine, ktime_t *now)
unsigned long flags;
u32 reset_count;
bool in_reset;
+ intel_wakeref_t wakeref;
spin_lock_irqsave(&guc->timestamp.lock, flags);
@@ -1324,7 +1325,7 @@ static ktime_t guc_engine_busyness(struct intel_engine_cs *engine, ktime_t *now)
* start_gt_clk is derived from GuC state. To get a consistent
* view of activity, we query the GuC state only if gt is awake.
*/
- if (!in_reset && intel_gt_pm_get_if_awake(gt)) {
+ if (!in_reset && (wakeref = intel_gt_pm_get_if_awake(gt))) {
stats_saved = *stats;
gt_stamp_saved = guc->timestamp.gt_stamp;
/*
@@ -1333,7 +1334,7 @@ static ktime_t guc_engine_busyness(struct intel_engine_cs *engine, ktime_t *now)
*/
guc_update_engine_gt_clks(engine);
guc_update_pm_timestamp(guc, now);
- intel_gt_pm_put_async(gt);
+ intel_gt_pm_put_async(gt, wakeref);
if (i915_reset_count(gpu_error) != reset_count) {
*stats = stats_saved;
guc->timestamp.gt_stamp = gt_stamp_saved;
@@ -4604,7 +4605,7 @@ int intel_guc_deregister_done_process_msg(struct intel_guc *guc,
intel_context_put(ce);
} else if (context_destroyed(ce)) {
/* Context has been destroyed */
- intel_gt_pm_put_async(guc_to_gt(guc));
+ intel_gt_pm_put_async_untracked(guc_to_gt(guc));
release_guc_id(guc, ce);
__guc_context_destroy(ce);
}
diff --git a/drivers/gpu/drm/i915/i915_pmu.c b/drivers/gpu/drm/i915/i915_pmu.c
index 7ece883a7d9567..12d858e17902ec 100644
--- a/drivers/gpu/drm/i915/i915_pmu.c
+++ b/drivers/gpu/drm/i915/i915_pmu.c
@@ -167,19 +167,19 @@ static u64 get_rc6(struct intel_gt *gt)
{
struct drm_i915_private *i915 = gt->i915;
struct i915_pmu *pmu = &i915->pmu;
+ intel_wakeref_t wakeref;
unsigned long flags;
- bool awake = false;
u64 val;
- if (intel_gt_pm_get_if_awake(gt)) {
+ wakeref = intel_gt_pm_get_if_awake(gt);
+ if (wakeref) {
val = __get_rc6(gt);
- intel_gt_pm_put_async(gt);
- awake = true;
+ intel_gt_pm_put_async(gt, wakeref);
}
spin_lock_irqsave(&pmu->lock, flags);
- if (awake) {
+ if (wakeref) {
pmu->sample[__I915_SAMPLE_RC6].cur = val;
} else {
/*
@@ -372,12 +372,14 @@ frequency_sample(struct intel_gt *gt, unsigned int period_ns)
struct drm_i915_private *i915 = gt->i915;
struct i915_pmu *pmu = &i915->pmu;
struct intel_rps *rps = >->rps;
+ intel_wakeref_t wakeref;
if (!frequency_sampling_enabled(pmu))
return;
/* Report 0/0 (actual/requested) frequency while parked. */
- if (!intel_gt_pm_get_if_awake(gt))
+ wakeref = intel_gt_pm_get_if_awake(gt);
+ if (!wakeref)
return;
if (pmu->enable & config_mask(I915_PMU_ACTUAL_FREQUENCY)) {
@@ -406,7 +408,7 @@ frequency_sample(struct intel_gt *gt, unsigned int period_ns)
period_ns / 1000);
}
- intel_gt_pm_put_async(gt);
+ intel_gt_pm_put_async(gt, wakeref);
}
static enum hrtimer_restart i915_sample(struct hrtimer *hrtimer)
diff --git a/drivers/gpu/drm/i915/intel_wakeref.c b/drivers/gpu/drm/i915/intel_wakeref.c
index dfd87d08221807..61f974d97a3757 100644
--- a/drivers/gpu/drm/i915/intel_wakeref.c
+++ b/drivers/gpu/drm/i915/intel_wakeref.c
@@ -96,7 +96,8 @@ static void __intel_wakeref_put_work(struct work_struct *wrk)
void __intel_wakeref_init(struct intel_wakeref *wf,
struct intel_runtime_pm *rpm,
const struct intel_wakeref_ops *ops,
- struct intel_wakeref_lockclass *key)
+ struct intel_wakeref_lockclass *key,
+ const char *name)
{
wf->rpm = rpm;
wf->ops = ops;
@@ -108,6 +109,10 @@ void __intel_wakeref_init(struct intel_wakeref *wf,
INIT_DELAYED_WORK(&wf->work, __intel_wakeref_put_work);
lockdep_init_map(&wf->work.work.lockdep_map,
"wakeref.work", &key->work, 0);
+
+#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_WAKEREF)
+ ref_tracker_dir_init(&wf->debug, INTEL_REFTRACK_DEAD_COUNT, name);
+#endif
}
int intel_wakeref_wait_for_idle(struct intel_wakeref *wf)
diff --git a/drivers/gpu/drm/i915/intel_wakeref.h b/drivers/gpu/drm/i915/intel_wakeref.h
index 99ee6254976ebf..19a71241473dfe 100644
--- a/drivers/gpu/drm/i915/intel_wakeref.h
+++ b/drivers/gpu/drm/i915/intel_wakeref.h
@@ -50,6 +50,10 @@ struct intel_wakeref {
const struct intel_wakeref_ops *ops;
struct delayed_work work;
+
+#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_WAKEREF)
+ struct ref_tracker_dir debug;
+#endif
};
struct intel_wakeref_lockclass {
@@ -60,11 +64,12 @@ struct intel_wakeref_lockclass {
void __intel_wakeref_init(struct intel_wakeref *wf,
struct intel_runtime_pm *rpm,
const struct intel_wakeref_ops *ops,
- struct intel_wakeref_lockclass *key);
-#define intel_wakeref_init(wf, rpm, ops) do { \
+ struct intel_wakeref_lockclass *key,
+ const char *name);
+#define intel_wakeref_init(wf, rpm, ops, name) do { \
static struct intel_wakeref_lockclass __key; \
\
- __intel_wakeref_init((wf), (rpm), (ops), &__key); \
+ __intel_wakeref_init((wf), (rpm), (ops), &__key, name); \
} while (0)
int __intel_wakeref_get_first(struct intel_wakeref *wf);
@@ -318,6 +323,33 @@ intel_wakeref_tracker_show(struct ref_tracker_dir *dir,
kfree(buf);
}
+#if IS_ENABLED(CONFIG_DRM_I915_DEBUG_WAKEREF)
+
+static inline intel_wakeref_t intel_wakeref_track(struct intel_wakeref *wf)
+{
+ return intel_ref_tracker_alloc(&wf->debug);
+}
+
+static inline void intel_wakeref_untrack(struct intel_wakeref *wf,
+ intel_wakeref_t handle)
+{
+ intel_ref_tracker_free(&wf->debug, handle);
+}
+
+#else
+
+static inline intel_wakeref_t intel_wakeref_track(struct intel_wakeref *wf)
+{
+ return -1;
+}
+
+static inline void intel_wakeref_untrack(struct intel_wakeref *wf,
+ intel_wakeref_t handle)
+{
+}
+
+#endif
+
struct intel_wakeref_auto {
struct intel_runtime_pm *rpm;
struct timer_list timer;
--
2.34.1
On Fri, 21 Apr 2023 13:35:04 +0200 Andrzej Hajda wrote:
> Gently ping for network developers, could you look at ref_tracker patches,
> as the ref_tracker library was developed for network.
Putting Eric in the To: field, I know email so hard and confusing...
On Fri, Apr 21, 2023 at 1:35 PM Andrzej Hajda <[email protected]> wrote:
>
> In case the library is tracking busy subsystem, simply
> printing stack for every active reference will spam log
> with long, hard to read, redundant stack traces. To improve
> readabilty following changes have been made:
> - reports are printed per stack_handle - log is more compact,
> - added display name for ref_tracker_dir - it will differentiate
> multiple subsystems,
> - stack trace is printed indented, in the same printk call,
> - info about dropped references is printed as well.
>
> Signed-off-by: Andrzej Hajda <[email protected]>
> Reviewed-by: Andi Shyti <[email protected]>
> ---
> include/linux/ref_tracker.h | 15 ++++++--
> lib/ref_tracker.c | 90 +++++++++++++++++++++++++++++++++++++++------
> 2 files changed, 91 insertions(+), 14 deletions(-)
>
> diff --git a/include/linux/ref_tracker.h b/include/linux/ref_tracker.h
> index 87a92f2bec1b88..fc9ef9952f01fd 100644
> --- a/include/linux/ref_tracker.h
> +++ b/include/linux/ref_tracker.h
> @@ -17,12 +17,19 @@ struct ref_tracker_dir {
> bool dead;
> struct list_head list; /* List of active trackers */
> struct list_head quarantine; /* List of dead trackers */
> + char name[32];
> #endif
> };
>
> #ifdef CONFIG_REF_TRACKER
> -static inline void ref_tracker_dir_init(struct ref_tracker_dir *dir,
> - unsigned int quarantine_count)
> +
> +/* Temporary allow two and three arguments, until consumers are converted */
> +#define ref_tracker_dir_init(_d, _q, args...) _ref_tracker_dir_init(_d, _q, ##args, #_d)
> +#define _ref_tracker_dir_init(_d, _q, _n, ...) __ref_tracker_dir_init(_d, _q, _n)
> +
We only have four callers of ref_tracker_dir_init() .
Why not simply add a name on them, and avoid this magic ?
On Fri, Apr 21, 2023 at 1:35 PM Andrzej Hajda <[email protected]> wrote:
>
> Library can handle allocation failures. To avoid allocation warnings
> __GFP_NOWARN has been added everywhere. Moreover GFP_ATOMIC has been
> replaced with GFP_NOWAIT in case of stack allocation on tracker free
> call.
>
> Signed-off-by: Andrzej Hajda <[email protected]>
> Reviewed-by: Andi Shyti <[email protected]>
> ---
Reviewed-by: Eric Dumazet <[email protected]>
Thanks.
On Fri, Apr 21, 2023 at 1:35 PM Andrzej Hajda <[email protected]> wrote:
>
> To have reliable detection of leaks, caller must be able to check under the same
> lock both: tracked counter and the leaks. dir.lock is natural candidate for such
> lock and unlocked print helper can be called with this lock taken.
> As a bonus we can reuse this helper in ref_tracker_dir_exit.
>
> Signed-off-by: Andrzej Hajda <[email protected]>
> Reviewed-by: Andi Shyti <[email protected]>
> ---
SGTM, thanks.
Reviewed-by: Eric Dumazet <[email protected]>
On 21.04.2023 16:21, Eric Dumazet wrote:
> On Fri, Apr 21, 2023 at 1:35 PM Andrzej Hajda <[email protected]> wrote:
>> In case the library is tracking busy subsystem, simply
>> printing stack for every active reference will spam log
>> with long, hard to read, redundant stack traces. To improve
>> readabilty following changes have been made:
>> - reports are printed per stack_handle - log is more compact,
>> - added display name for ref_tracker_dir - it will differentiate
>> multiple subsystems,
>> - stack trace is printed indented, in the same printk call,
>> - info about dropped references is printed as well.
>>
>> Signed-off-by: Andrzej Hajda <[email protected]>
>> Reviewed-by: Andi Shyti <[email protected]>
>> ---
>> include/linux/ref_tracker.h | 15 ++++++--
>> lib/ref_tracker.c | 90 +++++++++++++++++++++++++++++++++++++++------
>> 2 files changed, 91 insertions(+), 14 deletions(-)
>>
>> diff --git a/include/linux/ref_tracker.h b/include/linux/ref_tracker.h
>> index 87a92f2bec1b88..fc9ef9952f01fd 100644
>> --- a/include/linux/ref_tracker.h
>> +++ b/include/linux/ref_tracker.h
>> @@ -17,12 +17,19 @@ struct ref_tracker_dir {
>> bool dead;
>> struct list_head list; /* List of active trackers */
>> struct list_head quarantine; /* List of dead trackers */
>> + char name[32];
>> #endif
>> };
>>
>> #ifdef CONFIG_REF_TRACKER
>> -static inline void ref_tracker_dir_init(struct ref_tracker_dir *dir,
>> - unsigned int quarantine_count)
>> +
>> +/* Temporary allow two and three arguments, until consumers are converted */
>> +#define ref_tracker_dir_init(_d, _q, args...) _ref_tracker_dir_init(_d, _q, ##args, #_d)
>> +#define _ref_tracker_dir_init(_d, _q, _n, ...) __ref_tracker_dir_init(_d, _q, _n)
>> +
> We only have four callers of ref_tracker_dir_init() .
>
> Why not simply add a name on them, and avoid this magic ?
If this can be done in one patch, that's great.
Regards
Andrzej
If in doubt, say "N"
> +
> +config DRM_I915_DEBUG_WAKEREF
> + bool "Enable extra tracking for wakerefs"
> + depends on DRM_I915
> + default n
'default n' is not need
> +
> +static inline void
> +intel_wakeref_tracker_show(struct ref_tracker_dir *dir,
> + struct drm_printer *p)
> +{
> + const size_t buf_size = PAGE_SIZE;
> + char *buf, *sb, *se;
> + size_t count;
> +
> + buf = kmalloc(buf_size, GFP_NOWAIT);
> + if (!buf)
> + return;
> +
> + count = ref_tracker_dir_snprint(dir, buf, buf_size);
> + if (!count)
> + goto free;
> + /* printk does not like big buffers, so we split it */
> + for (sb = buf; *sb; sb = se + 1) {
> + se = strchrnul(sb, '\n');
> + drm_printf(p, "%.*s", (int)(se - sb + 1), sb);
> + if (!*se)
> + break;
> + }
> + if (count >= buf_size)
> + drm_printf(p, "\n...dropped %zd extra bytes of leak report.\n",
> + count + 1 - buf_size);
> +free:
> + kfree(buf);
> +}
> +
move to c source?