tegra-profiler: convert spinlock to raw spinlock

Profiler kernel thread tries to acquire a spinlock for ring buffer
that was already locked before, this leads to BUG_ON in rt_mutex.
Convert this lock to raw lock, this prevents preemption during
the critical sections.
This commit fixes the bug on -rt kernels.

Bug 1843939

Change-Id: I3f4c0d28d13cb1c117dae4699fc79f466e72825f
Signed-off-by: Igor Nabirushkin <inabirushkin@nvidia.com>
Reviewed-on: http://git-master/r/1257247
(cherry picked from commit 7fb4665bde29ee5b4ab96618cc6b19c1ce9a59a6)
Reviewed-on: http://git-master/r/1259387
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Sumeet Gupta <sumeetg@nvidia.com>

diff --git a/drivers/misc/tegra-profiler/comm.c b/drivers/misc/tegra-profiler/comm.c
index 82f8d7e3a811e8dac6eeb0a9b62a59791711b872..7aa0f9d07da97be4dd2573673d2fb60fce7926e5 100644 (file)
--- a/drivers/misc/tegra-profiler/comm.c
+++ b/drivers/misc/tegra-profiler/comm.c
@@ -41,7 +41,7 @@ struct quadd_ring_buffer {
 
        struct quadd_mmap_area *mmap;
 
-       spinlock_t lock;
+       raw_spinlock_t lock;
 };
 
 struct quadd_comm_ctx {
@@ -222,7 +222,7 @@ put_sample(struct quadd_record_data *data,
 
        rb = &cc->rb;
 
-       spin_lock_irqsave(&rb->lock, flags);
+       raw_spin_lock_irqsave(&rb->lock, flags);
 
        err = write_sample(rb, data, vec, vec_count);
        if (err < 0) {
@@ -234,7 +234,7 @@ put_sample(struct quadd_record_data *data,
                        rb_hdr->skipped_samples++;
        }
 
-       spin_unlock_irqrestore(&rb->lock, flags);
+       raw_spin_unlock_irqrestore(&rb->lock, flags);
 
        return err;
 }
@@ -319,7 +319,7 @@ init_mmap_hdr(struct quadd_mmap_rb_info *mmap_rb,
 
        rb = &cc->rb;
 
-       spin_lock_irqsave(&rb->lock, flags);
+       raw_spin_lock_irqsave(&rb->lock, flags);
 
        mmap->rb = rb;
 
@@ -354,7 +354,7 @@ init_mmap_hdr(struct quadd_mmap_rb_info *mmap_rb,
 
        rb_hdr->state = QUADD_RB_STATE_ACTIVE;
 
-       spin_unlock_irqrestore(&rb->lock, flags);
+       raw_spin_unlock_irqrestore(&rb->lock, flags);
 
        pr_info("[cpu: %d] init_mmap_hdr: vma: %#lx - %#lx, data: %p - %p\n",
                cpu_id,
@@ -394,13 +394,13 @@ static void rb_reset(struct quadd_ring_buffer *rb)
        if (!rb)
                return;
 
-       spin_lock_irqsave(&rb->lock, flags);
+       raw_spin_lock_irqsave(&rb->lock, flags);
 
        rb->mmap = NULL;
        rb->buf = NULL;
        rb->rb_hdr = NULL;
 
-       spin_unlock_irqrestore(&rb->lock, flags);
+       raw_spin_unlock_irqrestore(&rb->lock, flags);
 }
 
 static int
@@ -906,7 +906,7 @@ static int comm_init(void)
                rb->max_fill_count = 0;
                rb->nr_skipped_samples = 0;
 
-               spin_lock_init(&rb->lock);
+               raw_spin_lock_init(&rb->lock);
        }
 
        reset_params_ok_flag();