drm/panthor: Replace sleep locks with spinlocks in fdinfo path
authorAdrián Larumbe <adrian.larumbe@collabora.com>
Mon, 3 Mar 2025 19:08:45 +0000 (19:08 +0000)
committerSteven Price <steven.price@arm.com>
Wed, 5 Mar 2025 14:03:56 +0000 (14:03 +0000)
Commit 0590c94c3596 ("drm/panthor: Fix race condition when gathering fdinfo
group samples") introduced an xarray lock to deal with potential
use-after-free errors when accessing groups fdinfo figures. However, this
toggles the kernel's atomic context status, so the next nested mutex lock
will raise a warning when the kernel is compiled with mutex debug options:

CONFIG_DEBUG_RT_MUTEXES=y
CONFIG_DEBUG_MUTEXES=y

Replace Panthor's group fdinfo data mutex with a guarded spinlock.

Signed-off-by: Adrián Larumbe <adrian.larumbe@collabora.com>
Fixes: 0590c94c3596 ("drm/panthor: Fix race condition when gathering fdinfo group samples")
Reviewed-by: Liviu Dudau <liviu.dudau@arm.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Steven Price <steven.price@arm.com>
Signed-off-by: Steven Price <steven.price@arm.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20250303190923.1639985-1-adrian.larumbe@collabora.com
drivers/gpu/drm/panthor/panthor_sched.c

index 1a276db095ff609fcbbaee770dc57386e35f8cf6..4d31d1967716a5d413f88b6f9a95d7be4b25180f 100644 (file)
@@ -9,6 +9,7 @@
 #include <drm/panthor_drm.h>
 
 #include <linux/build_bug.h>
+#include <linux/cleanup.h>
 #include <linux/clk.h>
 #include <linux/delay.h>
 #include <linux/dma-mapping.h>
@@ -631,10 +632,10 @@ struct panthor_group {
                struct panthor_gpu_usage data;
 
                /**
-                * @lock: Mutex to govern concurrent access from drm file's fdinfo callback
-                * and job post-completion processing function
+                * @fdinfo.lock: Spinlock to govern concurrent access from drm file's fdinfo
+                * callback and job post-completion processing function
                 */
-               struct mutex lock;
+               spinlock_t lock;
 
                /** @fdinfo.kbo_sizes: Aggregate size of private kernel BO's held by the group. */
                size_t kbo_sizes;
@@ -910,8 +911,6 @@ static void group_release_work(struct work_struct *work)
                                                   release_work);
        u32 i;
 
-       mutex_destroy(&group->fdinfo.lock);
-
        for (i = 0; i < group->queue_count; i++)
                group_free_queue(group, group->queues[i]);
 
@@ -2861,12 +2860,12 @@ static void update_fdinfo_stats(struct panthor_job *job)
        struct panthor_job_profiling_data *slots = queue->profiling.slots->kmap;
        struct panthor_job_profiling_data *data = &slots[job->profiling.slot];
 
-       mutex_lock(&group->fdinfo.lock);
-       if (job->profiling.mask & PANTHOR_DEVICE_PROFILING_CYCLES)
-               fdinfo->cycles += data->cycles.after - data->cycles.before;
-       if (job->profiling.mask & PANTHOR_DEVICE_PROFILING_TIMESTAMP)
-               fdinfo->time += data->time.after - data->time.before;
-       mutex_unlock(&group->fdinfo.lock);
+       scoped_guard(spinlock, &group->fdinfo.lock) {
+               if (job->profiling.mask & PANTHOR_DEVICE_PROFILING_CYCLES)
+                       fdinfo->cycles += data->cycles.after - data->cycles.before;
+               if (job->profiling.mask & PANTHOR_DEVICE_PROFILING_TIMESTAMP)
+                       fdinfo->time += data->time.after - data->time.before;
+       }
 }
 
 void panthor_fdinfo_gather_group_samples(struct panthor_file *pfile)
@@ -2880,12 +2879,11 @@ void panthor_fdinfo_gather_group_samples(struct panthor_file *pfile)
 
        xa_lock(&gpool->xa);
        xa_for_each(&gpool->xa, i, group) {
-               mutex_lock(&group->fdinfo.lock);
+               guard(spinlock)(&group->fdinfo.lock);
                pfile->stats.cycles += group->fdinfo.data.cycles;
                pfile->stats.time += group->fdinfo.data.time;
                group->fdinfo.data.cycles = 0;
                group->fdinfo.data.time = 0;
-               mutex_unlock(&group->fdinfo.lock);
        }
        xa_unlock(&gpool->xa);
 }
@@ -3537,7 +3535,7 @@ int panthor_group_create(struct panthor_file *pfile,
        mutex_unlock(&sched->reset.lock);
 
        add_group_kbo_sizes(group->ptdev, group);
-       mutex_init(&group->fdinfo.lock);
+       spin_lock_init(&group->fdinfo.lock);
 
        return gid;