drm/i915: Hold a wakeref for probing the ring registers
authorChris Wilson <chris@chris-wilson.co.uk>
Tue, 30 May 2017 12:13:33 +0000 (13:13 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Tue, 30 May 2017 16:18:32 +0000 (17:18 +0100)
Allow intel_engine_is_idle() to be called outside of the GT wakeref by
acquiring the device runtime pm for ourselves. This allows the function
to act as check after we assume the engine is idle and we release the GT
wakeref held whilst we have requests. At the moment, we do not call it
outside of an awake context but taking the wakeref as required makes it
more convenient to use for quick debugging in future.

[ 2613.401647] RPM wakelock ref not held during HW access
[ 2613.401684] ------------[ cut here ]------------
[ 2613.401720] WARNING: CPU: 5 PID: 7739 at drivers/gpu/drm/i915/intel_drv.h:1787 gen6_read32+0x21f/0x2b0 [i915]
[ 2613.401731] Modules linked in: snd_hda_intel i915 vgem snd_hda_codec_hdmi x86_pkg_temp_thermal intel_powerclamp snd_hda_codec_realtek coretemp snd_hda_codec_generic crct10dif_pclmul crc32_pclmul ghash_clmulni_intel snd_hda_codec snd_hwdep snd_hda_core snd_pcm r8169 mii mei_me lpc_ich mei prime_numbers [last unloaded: i915]
[ 2613.401823] CPU: 5 PID: 7739 Comm: drv_missed_irq Tainted: G     U          4.12.0-rc2-CI-CI_DRM_421+ #1
[ 2613.401825] Hardware name: MSI MS-7924/Z97M-G43(MS-7924), BIOS V1.12 02/15/2016
[ 2613.401840] task: ffff880409e3a740 task.stack: ffffc900084dc000
[ 2613.401861] RIP: 0010:gen6_read32+0x21f/0x2b0 [i915]
[ 2613.401863] RSP: 0018:ffffc900084dfce8 EFLAGS: 00010292
[ 2613.401869] RAX: 000000000000002a RBX: ffff8804016a8000 RCX: 0000000000000006
[ 2613.401871] RDX: 0000000000000006 RSI: ffffffff81cbf2d9 RDI: ffffffff81c9e3a7
[ 2613.401874] RBP: ffffc900084dfd18 R08: ffff880409e3afc8 R09: 0000000000000000
[ 2613.401877] R10: 000000008a1c483f R11: 0000000000000000 R12: 000000000000209c
[ 2613.401879] R13: 0000000000000001 R14: ffff8804016a8000 R15: ffff8804016ac150
[ 2613.401882] FS:  00007f39ef3dd8c0(0000) GS:ffff88041fb40000(0000) knlGS:0000000000000000
[ 2613.401885] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 2613.401887] CR2: 00000000023717c8 CR3: 00000002e7b34000 CR4: 00000000001406e0
[ 2613.401889] Call Trace:
[ 2613.401912]  intel_engine_is_idle+0x76/0x90 [i915]
[ 2613.401931]  i915_gem_wait_for_idle+0xe6/0x1e0 [i915]
[ 2613.401951]  fault_irq_set+0x40/0x90 [i915]
[ 2613.401970]  i915_ring_test_irq_set+0x42/0x50 [i915]
[ 2613.401976]  simple_attr_write+0xc7/0xe0
[ 2613.401981]  full_proxy_write+0x4f/0x70
[ 2613.401987]  __vfs_write+0x23/0x120
[ 2613.401992]  ? rcu_read_lock_sched_held+0x75/0x80
[ 2613.401996]  ? rcu_sync_lockdep_assert+0x2a/0x50
[ 2613.401999]  ? __sb_start_write+0xfa/0x1f0
[ 2613.402004]  vfs_write+0xc5/0x1d0
[ 2613.402008]  ? trace_hardirqs_on_caller+0xe7/0x1c0
[ 2613.402013]  SyS_write+0x44/0xb0
[ 2613.402020]  entry_SYSCALL_64_fastpath+0x1c/0xb1
[ 2613.402022] RIP: 0033:0x7f39eded6670
[ 2613.402025] RSP: 002b:00007fffdcdcb1a8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
[ 2613.402030] RAX: ffffffffffffffda RBX: ffffffff81470203 RCX: 00007f39eded6670
[ 2613.402033] RDX: 0000000000000001 RSI: 000000000041bc33 RDI: 0000000000000006
[ 2613.402036] RBP: ffffc900084dff88 R08: 00007f39ef3dd8c0 R09: 0000000000000001
[ 2613.402038] R10: 0000000000000000 R11: 0000000000000246 R12: 000000000041bc33
[ 2613.402041] R13: 0000000000000006 R14: 0000000000000000 R15: 0000000000000000
[ 2613.402046]  ? __this_cpu_preempt_check+0x13/0x20
[ 2613.402052] Code: 01 9b fa e0 0f ff e9 28 fe ff ff 80 3d 6a dd 0e 00 00 0f 85 29 fe ff ff 48 c7 c7 48 19 29 a0 c6 05 56 dd 0e 00 01 e8 da 9a fa e0 <0f> ff e9 0f fe ff ff b9 01 00 00 00 ba 01 00 00 00 44 89 e6 48
[ 2613.402199] ---[ end trace 31f0cfa93ab632bf ]---

Fixes: 5400367a864d ("drm/i915: Ensure the engine is idle before manually changing HWS")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170530121334.17364-2-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
drivers/gpu/drm/i915/intel_engine_cs.c

index 413bfd8d4bf489ec73d4ce7677a223bf4465dca3..699f2d3861c71f4fd3e09c6ed3627d97c49157e9 100644 (file)
@@ -1205,6 +1205,22 @@ int intel_ring_workarounds_emit(struct drm_i915_gem_request *req)
        return 0;
 }
 
+static bool ring_is_idle(struct intel_engine_cs *engine)
+{
+       struct drm_i915_private *dev_priv = engine->i915;
+       bool idle = true;
+
+       intel_runtime_pm_get(dev_priv);
+
+       /* No bit for gen2, so assume the CS parser is idle */
+       if (INTEL_GEN(dev_priv) > 2 && !(I915_READ_MODE(engine) & MODE_IDLE))
+               idle = false;
+
+       intel_runtime_pm_put(dev_priv);
+
+       return idle;
+}
+
 /**
  * intel_engine_is_idle() - Report if the engine has finished process all work
  * @engine: the intel_engine_cs
@@ -1237,7 +1253,7 @@ bool intel_engine_is_idle(struct intel_engine_cs *engine)
                return false;
 
        /* Ring stopped? */
-       if (INTEL_GEN(dev_priv) > 2 && !(I915_READ_MODE(engine) & MODE_IDLE))
+       if (!ring_is_idle(engine))
                return false;
 
        return true;