[PATCH] drm/amdgpu/gfx11: need acquire mutex before access CP_VMID_RESET

[Zhang, Hawking]

[AMD Official Use Only - General]

+       /* release the gfx mutex */
+       tmp = REG_SET_FIELD(tmp, CP_GFX_INDEX_MUTEX, REQUEST, 0);
+       WREG32_SOC15(GC, 0, regCP_GFX_INDEX_MUTEX, tmp);

Shall we add a check by reading back CP_GFX_INDEX_MUTEX to ensure the release is done correctly?

Regards,
Hawking

-----Original Message-----
From: Xiao, Jack <Jack.Xiao at amd.com>
Sent: Tuesday, December 19, 2023 17:24
To: amd-gfx at lists.freedesktop.org; Deucher, Alexander <Alexander.Deucher at amd.com>; Zhang, Hawking <Hawking.Zhang at amd.com>
Cc: Xiao, Jack <Jack.Xiao at amd.com>
Subject: [PATCH] drm/amdgpu/gfx11: need acquire mutex before access CP_VMID_RESET

It's required to take the gfx mutex before access to CP_VMID_RESET, for there is a race condition with CP firmware to write the register.

Signed-off-by: Jack Xiao <Jack.Xiao at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
index bdcf96df69e6..ae3370d34d11 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
@@ -4518,6 +4518,22 @@ static int gfx_v11_0_soft_reset(void *handle)
                }
        }

+       /* Try to require the gfx mutex before access to CP_VMID_RESET */
+       for (i = 0; i < adev->usec_timeout; i++) {
+               /* Request with MeId=2, PipeId=0 */
+               tmp = REG_SET_FIELD(0, CP_GFX_INDEX_MUTEX, REQUEST, 1);
+               tmp = REG_SET_FIELD(tmp, CP_GFX_INDEX_MUTEX, CLIENTID, 4);
+               WREG32_SOC15(GC, 0, regCP_GFX_INDEX_MUTEX, tmp);
+               if (RREG32_SOC15(GC, 0, regCP_GFX_INDEX_MUTEX) == tmp)
+                       break;
+               udelay(1);
+       }
+
+       if (i >= adev->usec_timeout) {
+               printk("Failed to require the gfx mutex during soft reset\n");
+               return -EINVAL;
+       }
+
        WREG32_SOC15(GC, 0, regCP_VMID_RESET, 0xfffffffe);

        // Read CP_VMID_RESET register three times.
@@ -4526,6 +4542,10 @@ static int gfx_v11_0_soft_reset(void *handle)
        RREG32_SOC15(GC, 0, regCP_VMID_RESET);
        RREG32_SOC15(GC, 0, regCP_VMID_RESET);

+       /* release the gfx mutex */
+       tmp = REG_SET_FIELD(tmp, CP_GFX_INDEX_MUTEX, REQUEST, 0);
+       WREG32_SOC15(GC, 0, regCP_GFX_INDEX_MUTEX, tmp);
+
        for (i = 0; i < adev->usec_timeout; i++) {
                if (!RREG32_SOC15(GC, 0, regCP_HQD_ACTIVE) &&
                    !RREG32_SOC15(GC, 0, regCP_GFX_HQD_ACTIVE))
--
2.41.0