drm/nouveau/gr/gp107,gp108: implement workaround for HW hanging during init

[ Upstream commit 028a12f5aa829b4ba6ac011530b815eda4960e89 ]

Certain boards with GP107/GP108 chipsets hang (often, but randomly) for
unknown reasons during GR initialisation.

The first tell-tale symptom of this issue is:

nouveau 0000:01:00.0: bus: MMIO read of 00000000 FAULT at 409800 [ TIMEOUT ]

appearing in dmesg, likely followed by many other failures being logged.

Karol found this WAR for the issue a while back, but efforts to isolate
the root cause and proper fix have not yielded success so far.  I've
modified the original patch to include a few more details, limit it to
GP107/GP108 by default, and added a config option to override this choice.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

1 files changed, 26 insertions, 0 deletions

diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
index dd8f85b8b3a7..f2f5636efac4 100644
--- a/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/gr/gf100.c
@@ -1981,8 +1981,34 @@ gf100_gr_init_(struct nvkm_gr *base)
 {
 	struct gf100_gr *gr = gf100_gr(base);
 	struct nvkm_subdev *subdev = &base->engine.subdev;
+	struct nvkm_device *device = subdev->device;
+	bool reset = device->chipset == 0x137 || device->chipset == 0x138;
 	u32 ret;
 
+	/* On certain GP107/GP108 boards, we trigger a weird issue where
+	 * GR will stop responding to PRI accesses after we've asked the
+	 * SEC2 RTOS to boot the GR falcons.  This happens with far more
+	 * frequency when cold-booting a board (ie. returning from D3).
+	 *
+	 * The root cause for this is not known and has proven difficult
+	 * to isolate, with many avenues being dead-ends.
+	 *
+	 * A workaround was discovered by Karol, whereby putting GR into
+	 * reset for an extended period right before initialisation
+	 * prevents the problem from occuring.
+	 *
+	 * XXX: As RM does not require any such workaround, this is more
+	 *      of a hack than a true fix.
+	 */
+	reset = nvkm_boolopt(device->cfgopt, "NvGrResetWar", reset);
+	if (reset) {
+		nvkm_mask(device, 0x000200, 0x00001000, 0x00000000);
+		nvkm_rd32(device, 0x000200);
+		msleep(50);
+		nvkm_mask(device, 0x000200, 0x00001000, 0x00001000);
+		nvkm_rd32(device, 0x000200);
+	}
+
 	nvkm_pmu_pgob(gr->base.engine.subdev.device->pmu, false);
 
 	ret = nvkm_falcon_get(&gr->fecs.falcon, subdev);