This is a note to let you know that I've just added the patch titled
drm/amdgpu/gfx9: properly handle error ints on all pipes
to the 6.10-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
drm-amdgpu-gfx9-properly-handle-error-ints-on-all-pi.patch
and it can be found in the queue-6.10 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@xxxxxxxxxxxxxxx> know about it.
commit 1b9adcb5f75046d24e96ecb4815fcead7e371fd6
Author: Alex Deucher <alexander.deucher@xxxxxxx>
Date: Tue Jul 2 10:24:59 2024 -0400
drm/amdgpu/gfx9: properly handle error ints on all pipes
[ Upstream commit 48695573d2feaf42812c1ad54e01caff0d1c2d71 ]
Need to handle the interrupt enables for all pipes.
Acked-by: Christian König <christian.koenig@xxxxxxx>
Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
index 0594eab666a9b..b278453cad6d4 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
@@ -2477,7 +2477,7 @@ static void gfx_v9_0_enable_gui_idle_interrupt(struct amdgpu_device *adev,
tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_BUSY_INT_ENABLE, enable ? 1 : 0);
tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CNTX_EMPTY_INT_ENABLE, enable ? 1 : 0);
tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, CMP_BUSY_INT_ENABLE, enable ? 1 : 0);
- if(adev->gfx.num_gfx_rings)
+ if (adev->gfx.num_gfx_rings)
tmp = REG_SET_FIELD(tmp, CP_INT_CNTL_RING0, GFX_IDLE_INT_ENABLE, enable ? 1 : 0);
WREG32_SOC15(GC, 0, mmCP_INT_CNTL_RING0, tmp);
@@ -5772,17 +5772,59 @@ static void gfx_v9_0_set_compute_eop_interrupt_state(struct amdgpu_device *adev,
}
}
+static u32 gfx_v9_0_get_cpc_int_cntl(struct amdgpu_device *adev,
+ int me, int pipe)
+{
+ /*
+ * amdgpu controls only the first MEC. That's why this function only
+ * handles the setting of interrupts for this specific MEC. All other
+ * pipes' interrupts are set by amdkfd.
+ */
+ if (me != 1)
+ return 0;
+
+ switch (pipe) {
+ case 0:
+ return SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE0_INT_CNTL);
+ case 1:
+ return SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE1_INT_CNTL);
+ case 2:
+ return SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE2_INT_CNTL);
+ case 3:
+ return SOC15_REG_OFFSET(GC, 0, mmCP_ME1_PIPE3_INT_CNTL);
+ default:
+ return 0;
+ }
+}
+
static int gfx_v9_0_set_priv_reg_fault_state(struct amdgpu_device *adev,
struct amdgpu_irq_src *source,
unsigned type,
enum amdgpu_interrupt_state state)
{
+ u32 cp_int_cntl_reg, cp_int_cntl;
+ int i, j;
+
switch (state) {
case AMDGPU_IRQ_STATE_DISABLE:
case AMDGPU_IRQ_STATE_ENABLE:
WREG32_FIELD15(GC, 0, CP_INT_CNTL_RING0,
PRIV_REG_INT_ENABLE,
state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
+ for (i = 0; i < adev->gfx.mec.num_mec; i++) {
+ for (j = 0; j < adev->gfx.mec.num_pipe_per_mec; j++) {
+ /* MECs start at 1 */
+ cp_int_cntl_reg = gfx_v9_0_get_cpc_int_cntl(adev, i + 1, j);
+
+ if (cp_int_cntl_reg) {
+ cp_int_cntl = RREG32_SOC15_IP(GC, cp_int_cntl_reg);
+ cp_int_cntl = REG_SET_FIELD(cp_int_cntl, CP_ME1_PIPE0_INT_CNTL,
+ PRIV_REG_INT_ENABLE,
+ state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
+ WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl);
+ }
+ }
+ }
break;
default:
break;
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
index f5b9f443cfdd7..2564a003526ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
@@ -2824,21 +2824,63 @@ static void gfx_v9_4_3_xcc_set_compute_eop_interrupt_state(
}
}
+static u32 gfx_v9_4_3_get_cpc_int_cntl(struct amdgpu_device *adev,
+ int xcc_id, int me, int pipe)
+{
+ /*
+ * amdgpu controls only the first MEC. That's why this function only
+ * handles the setting of interrupts for this specific MEC. All other
+ * pipes' interrupts are set by amdkfd.
+ */
+ if (me != 1)
+ return 0;
+
+ switch (pipe) {
+ case 0:
+ return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE0_INT_CNTL);
+ case 1:
+ return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE1_INT_CNTL);
+ case 2:
+ return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE2_INT_CNTL);
+ case 3:
+ return SOC15_REG_OFFSET(GC, GET_INST(GC, xcc_id), regCP_ME1_PIPE3_INT_CNTL);
+ default:
+ return 0;
+ }
+}
+
static int gfx_v9_4_3_set_priv_reg_fault_state(struct amdgpu_device *adev,
struct amdgpu_irq_src *source,
unsigned type,
enum amdgpu_interrupt_state state)
{
- int i, num_xcc;
+ u32 mec_int_cntl_reg, mec_int_cntl;
+ int i, j, k, num_xcc;
num_xcc = NUM_XCC(adev->gfx.xcc_mask);
switch (state) {
case AMDGPU_IRQ_STATE_DISABLE:
case AMDGPU_IRQ_STATE_ENABLE:
- for (i = 0; i < num_xcc; i++)
+ for (i = 0; i < num_xcc; i++) {
WREG32_FIELD15_PREREG(GC, GET_INST(GC, i), CP_INT_CNTL_RING0,
- PRIV_REG_INT_ENABLE,
- state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
+ PRIV_REG_INT_ENABLE,
+ state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0);
+ for (j = 0; j < adev->gfx.mec.num_mec; j++) {
+ for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
+ /* MECs start at 1 */
+ mec_int_cntl_reg = gfx_v9_4_3_get_cpc_int_cntl(adev, i, j + 1, k);
+
+ if (mec_int_cntl_reg) {
+ mec_int_cntl = RREG32_XCC(mec_int_cntl_reg, i);
+ mec_int_cntl = REG_SET_FIELD(mec_int_cntl, CP_ME1_PIPE0_INT_CNTL,
+ PRIV_REG_INT_ENABLE,
+ state == AMDGPU_IRQ_STATE_ENABLE ?
+ 1 : 0);
+ WREG32_XCC(mec_int_cntl_reg, mec_int_cntl, i);
+ }
+ }
+ }
+ }
break;
default:
break;
