coresight: perf: Avoid unncessary CPU hotplug read lock
authorSuzuki K Poulose <suzuki.poulose@arm.com>
Thu, 20 Sep 2018 19:17:48 +0000 (13:17 -0600)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 25 Sep 2018 18:09:17 +0000 (20:09 +0200)
commitc48fb3bbe912a295e5b75eaabaf39874d5b9b773
tree36454dbec0a471b7fc8f5db5fa45baa148222d30
parent5ecabe4a76e8cdb61fa3e24862d9ca240a1c4ddf
coresight: perf: Avoid unncessary CPU hotplug read lock

We hold the read lock on CPU hotplug to simply copy the
online mask, which is not really needed. And this can
cause a lockdep warning, like :

[   54.632093] ======================================================
[   54.638207] WARNING: possible circular locking dependency detected
[   54.644322] 4.18.0-rc3-00042-g2d39e6356bb7-dirty #309 Not tainted
[   54.650350] ------------------------------------------------------
[   54.656464] perf/2862 is trying to acquire lock:
[   54.661031] 000000007e21d170 (&event->mmap_mutex){+.+.}, at: perf_event_set_output+0x98/0x138
[   54.669486]
[   54.669486] but task is already holding lock:
[   54.675256] 000000001080eb1b (&cpuctx_mutex){+.+.}, at: perf_event_ctx_lock_nested+0xf8/0x1f0
[   54.683704]
[   54.683704] which lock already depends on the new lock.
[   54.683704]
[   54.691797]
[   54.691797] the existing dependency chain (in reverse order) is:
[   54.699201]
[   54.699201] -> #3 (&cpuctx_mutex){+.+.}:
[   54.704556]        __mutex_lock+0x70/0x808
[   54.708608]        mutex_lock_nested+0x1c/0x28
[   54.713005]        perf_event_init_cpu+0x8c/0xd8
[   54.717574]        perf_event_init+0x194/0x1d4
[   54.721971]        start_kernel+0x2b8/0x42c
[   54.726107]
[   54.726107] -> #2 (pmus_lock){+.+.}:
[   54.731114]        __mutex_lock+0x70/0x808
[   54.735165]        mutex_lock_nested+0x1c/0x28
[   54.739560]        perf_event_init_cpu+0x30/0xd8
[   54.744129]        cpuhp_invoke_callback+0x84/0x248
[   54.748954]        _cpu_up+0xe8/0x1c8
[   54.752576]        do_cpu_up+0xa8/0xc8
[   54.756283]        cpu_up+0x10/0x18
[   54.759731]        smp_init+0xa0/0x114
[   54.763438]        kernel_init_freeable+0x120/0x288
[   54.768264]        kernel_init+0x10/0x108
[   54.772230]        ret_from_fork+0x10/0x18
[   54.776279]
[   54.776279] -> #1 (cpu_hotplug_lock.rw_sem){++++}:
[   54.782492]        cpus_read_lock+0x34/0xb0
[   54.786631]        etm_setup_aux+0x5c/0x308
[   54.790769]        rb_alloc_aux+0x1ec/0x300
[   54.794906]        perf_mmap+0x284/0x610
[   54.798787]        mmap_region+0x388/0x570
[   54.802838]        do_mmap+0x344/0x4f8
[   54.806544]        vm_mmap_pgoff+0xe4/0x110
[   54.810682]        ksys_mmap_pgoff+0xa8/0x240
[   54.814992]        sys_mmap+0x18/0x28
[   54.818613]        el0_svc_naked+0x30/0x34
[   54.822661]
[   54.822661] -> #0 (&event->mmap_mutex){+.+.}:
[   54.828445]        lock_acquire+0x48/0x68
[   54.832409]        __mutex_lock+0x70/0x808
[   54.836459]        mutex_lock_nested+0x1c/0x28
[   54.840855]        perf_event_set_output+0x98/0x138
[   54.845680]        _perf_ioctl+0x2a0/0x6a0
[   54.849731]        perf_ioctl+0x3c/0x68
[   54.853526]        do_vfs_ioctl+0xb8/0xa20
[   54.857577]        ksys_ioctl+0x80/0xb8
[   54.861370]        sys_ioctl+0xc/0x18
[   54.864990]        el0_svc_naked+0x30/0x34
[   54.869039]
[   54.869039] other info that might help us debug this:
[   54.869039]
[   54.876960] Chain exists of:
[   54.876960]   &event->mmap_mutex --> pmus_lock --> &cpuctx_mutex
[   54.876960]
[   54.887217]  Possible unsafe locking scenario:
[   54.887217]
[   54.893073]        CPU0                    CPU1
[   54.897552]        ----                    ----
[   54.902030]   lock(&cpuctx_mutex);
[   54.905396]                                lock(pmus_lock);
[   54.910911]                                lock(&cpuctx_mutex);
[   54.916770]   lock(&event->mmap_mutex);
[   54.920566]
[   54.920566]  *** DEADLOCK ***
[   54.920566]
[   54.926424] 1 lock held by perf/2862:
[   54.930042]  #0: 000000001080eb1b (&cpuctx_mutex){+.+.}, at: perf_event_ctx_lock_nested+0xf8/0x1f0

Since we have per-cpu array for the paths, we simply don't care about
the number of online CPUs. This patch gets rid of the
{get/put}_online_cpus().

Reported-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Suzuki K Poulose <suzuki.poulose@arm.com>
Signed-off-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/hwtracing/coresight/coresight-etm-perf.c