[Intel-gfx] [PATCH] drm/i915: get a runtime PM ref in i915_wedged_set

[Michel Thierry]

On 3/14/2017 2:09 AM, Chris Wilson wrote:
> On Mon, Mar 13, 2017 at 05:26:06PM -0700, Michel Thierry wrote:
>> At least in bxt (with decoupled mmio), it prevents a warning while
>> capturing the reg state:
>>
>>  [  ] WARNING: assert_rpm_wakelock_held.part.4+0x1e/0x20 [i915]
>>  [  ] RPM wakelock ref not held during HW access
>>  [  ] Call Trace:
>>  [  ]  dump_stack+0x63/0x87
>>  [  ]  __warn+0xd1/0xf0
>>  [  ]  ? fwtable_write32+0x1a0/0x1a0 [i915]
>>  [  ]  warn_slowpath_fmt+0x4f/0x60
>>  [  ]  ? vprintk_default+0x29/0x50
>>  [  ]  assert_rpm_wakelock_held.part.4+0x1e/0x20 [i915]
>>  [  ]  gen9_decoupled_read32+0x18a/0x1a0 [i915]
>>  [  ]  ? fwtable_write32+0x1a0/0x1a0 [i915]
>>  [  ]  i915_handle_error+0x1d0/0x260 [i915]
>>  [  ]  ? __check_object_size+0x170/0x1aa
>>  [  ]  i915_wedged_set+0x25/0x30 [i915]
>>
>> These get/put were removed by commit 9c870d03674f ("drm/i915: Use RPM as the
>> barrier for controlling user mmap access"), before decoupled mmio was
>> added in commit 85ee17ebeedd ("drm/i915/bxt: Broxton decoupled MMIO).
>>
>> Reported-by: Antonio Argenziano <antonio.argenziano at intel.com>
>> Signed-off-by: Michel Thierry <michel.thierry at intel.com>
>> ---
>>  drivers/gpu/drm/i915/i915_debugfs.c | 4 ++++
>>  1 file changed, 4 insertions(+)
>>
>> diff --git a/drivers/gpu/drm/i915/i915_debugfs.c b/drivers/gpu/drm/i915/i915_debugfs.c
>> index 9fcc4f1a86fc..89f211391ef9 100644
>> --- a/drivers/gpu/drm/i915/i915_debugfs.c
>> +++ b/drivers/gpu/drm/i915/i915_debugfs.c
>> @@ -4155,9 +4155,13 @@ i915_wedged_set(void *data, u64 val)
>>  	if (i915_reset_in_progress(&dev_priv->gpu_error))
>>  		return -EAGAIN;
>>
>> +	intel_runtime_pm_get(dev_priv);
>
> They don't belong here. i915_handle_error is already taking rpm, it just
> forgot a path.

Oh, I should have looked at your tree first ;)
(https://cgit.freedesktop.org/~ickle/linux-2.6/commit/drivers/gpu/drm/i915?h=prescheduler&id=e190db9f6898b60dc8905d6c5bf2dcd7ec759d7f)

But the rpm still needs to cover capture_error_state.