drm/nouveau/device: fix changing endianess code to work on older GPUs

With this we try to detect if the endianess switch works and assume LE if
not. Suggested by Ben.

Fixes: 51c05340e407 ("drm/nouveau/device: detect if changing endianness failed")
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Cc: <stable@vger.kernel.org> # v5.8+
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>

diff --git a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
index dcb7067..7851bec 100644 (file)
--- a/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
+++ b/drivers/gpu/drm/nouveau/nvkm/engine/device/base.c
@@ -2924,17 +2924,34 @@ nvkm_device_del(struct nvkm_device **pdevice)
        }
 }
 
+/* returns true if the GPU is in the CPU native byte order */
 static inline bool
 nvkm_device_endianness(struct nvkm_device *device)
 {
-       u32 boot1 = nvkm_rd32(device, 0x000004) & 0x01000001;
 #ifdef __BIG_ENDIAN
-       if (!boot1)
-               return false;
+       const bool big_endian = true;
 #else
-       if (boot1)
-               return false;
+       const bool big_endian = false;
 #endif
+
+       /* Read NV_PMC_BOOT_1, and assume non-functional endian switch if it
+        * doesn't contain the expected values.
+        */
+       u32 pmc_boot_1 = nvkm_rd32(device, 0x000004);
+       if (pmc_boot_1 && pmc_boot_1 != 0x01000001)
+               return !big_endian; /* Assume GPU is LE in this case. */
+
+       /* 0 means LE and 0x01000001 means BE GPU. Condition is true when
+        * GPU/CPU endianness don't match.
+        */
+       if (big_endian == !pmc_boot_1) {
+               nvkm_wr32(device, 0x000004, 0x01000001);
+               nvkm_rd32(device, 0x000000);
+               if (nvkm_rd32(device, 0x000004) != (big_endian ? 0x01000001 : 0x00000000))
+                       return !big_endian; /* Assume GPU is LE on any unexpected read-back. */
+       }
+
+       /* CPU/GPU endianness should (hopefully) match. */
        return true;
 }
 
@@ -2987,14 +3004,10 @@ nvkm_device_ctor(const struct nvkm_device_func *func,
        if (detect) {
                /* switch mmio to cpu's native endianness */
                if (!nvkm_device_endianness(device)) {
-                       nvkm_wr32(device, 0x000004, 0x01000001);
-                       nvkm_rd32(device, 0x000000);
-                       if (!nvkm_device_endianness(device)) {
-                               nvdev_error(device,
-                                           "GPU not supported on big-endian\n");
-                               ret = -ENOSYS;
-                               goto done;
-                       }
+                       nvdev_error(device,
+                                   "Couldn't switch GPU to CPUs endianess\n");
+                       ret = -ENOSYS;
+                       goto done;
                }
 
                boot0 = nvkm_rd32(device, 0x000000);