]> git.karo-electronics.de Git - linux-beck.git/commitdiff
drm/nouveau/pmu: be more strict about locking
authorKarol Herbst <nouveau@karolherbst.de>
Mon, 11 Jan 2016 01:58:03 +0000 (02:58 +0100)
committerBen Skeggs <bskeggs@redhat.com>
Fri, 20 May 2016 04:43:04 +0000 (14:43 +1000)
When we start communicating with the pmu a bit more, the current code is
a real issue. I encountered a dead lock here, while testing my dynamic
reclocking code

Signed-off-by: Karol Herbst <nouveau@karolherbst.de>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
drivers/gpu/drm/nouveau/nvkm/subdev/pmu/base.c

index d95eb8659d1bdb6279d256b6e90f49d04cff2215..6e6d2ef598c8c9f8af43234704accfaedd6a7ebb 100644 (file)
@@ -40,21 +40,23 @@ nvkm_pmu_send(struct nvkm_pmu *pmu, u32 reply[2],
        struct nvkm_device *device = subdev->device;
        u32 addr;
 
+       mutex_lock(&subdev->mutex);
        /* wait for a free slot in the fifo */
        addr  = nvkm_rd32(device, 0x10a4a0);
        if (nvkm_msec(device, 2000,
                u32 tmp = nvkm_rd32(device, 0x10a4b0);
                if (tmp != (addr ^ 8))
                        break;
-       ) < 0)
+       ) < 0) {
+               mutex_unlock(&subdev->mutex);
                return -EBUSY;
+       }
 
        /* we currently only support a single process at a time waiting
         * on a synchronous reply, take the PMU mutex and tell the
         * receive handler what we're waiting for
         */
        if (reply) {
-               mutex_lock(&subdev->mutex);
                pmu->recv.message = message;
                pmu->recv.process = process;
        }
@@ -81,9 +83,9 @@ nvkm_pmu_send(struct nvkm_pmu *pmu, u32 reply[2],
                wait_event(pmu->recv.wait, (pmu->recv.process == 0));
                reply[0] = pmu->recv.data[0];
                reply[1] = pmu->recv.data[1];
-               mutex_unlock(&subdev->mutex);
        }
 
+       mutex_unlock(&subdev->mutex);
        return 0;
 }