|
[AMD Official Use Only - AMD Internal Distribution Only]
Acked-by: Vitaly Prosyak <vitaly.prosyak@xxxxxxx>
The entire series of patches for gfx 7, 8, 9, 11 and compute gfx 10, 12
for queue_reset was tested using the new IGT test, amd_queue_reset, which has been merged upstream.
From: amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> on behalf of amd-gfx-request@xxxxxxxxxxxxxxxxxxxxx <amd-gfx-request@xxxxxxxxxxxxxxxxxxxxx>
Sent: Wednesday, July 17, 2024 4:40 PM To: amd-gfx@xxxxxxxxxxxxxxxxxxxxx <amd-gfx@xxxxxxxxxxxxxxxxxxxxx> Subject: amd-gfx Digest, Vol 98, Issue 218 Send amd-gfx mailing list submissions to
amd-gfx@xxxxxxxxxxxxxxxxxxxxx To subscribe or unsubscribe via the World Wide Web, visit https://lists.freedesktop.org/mailman/listinfo/amd-gfx or, via email, send a message with subject or body 'help' to amd-gfx-request@xxxxxxxxxxxxxxxxxxxxx You can reach the person managing the list at amd-gfx-owner@xxxxxxxxxxxxxxxxxxxxx When replying, please edit your Subject line so it is more specific than "Re: Contents of amd-gfx digest..." Today's Topics: 1. [PATCH 6/6] drm/amdgpu/gfx9.4.3: Enable bad opcode interrupt (Alex Deucher) 2. [PATCH 2/6] drm/amdgpu/gfx11: Enable bad opcode interrupt (Alex Deucher) 3. [PATCH 4/6] drm/amdgpu/gfx12: Enable bad opcode interrupt (Alex Deucher) ---------------------------------------------------------------------- Message: 1 Date: Wed, 17 Jul 2024 16:40:11 -0400 From: Alex Deucher <alexander.deucher@xxxxxxx> To: <amd-gfx@xxxxxxxxxxxxxxxxxxxxx> Cc: Alex Deucher <alexander.deucher@xxxxxxx> Subject: [PATCH 6/6] drm/amdgpu/gfx9.4.3: Enable bad opcode interrupt Message-ID: <20240717204011.15342-6-alexander.deucher@xxxxxxx> Content-Type: text/plain For the bad opcode case, it will cause CP/ME hang. The firmware will prevent the ME side from hanging by raising a bad opcode interrupt. And the driver needs to perform a vmid reset when receiving the interrupt. Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx> --- drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c | 69 +++++++++++++++++++++++++ 1 file changed, 69 insertions(+) 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 43a3ef276b5f..98fe6c40da64 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c @@ -901,6 +901,13 @@ static int gfx_v9_4_3_sw_init(void *handle) if (r) return r; + /* Bad opcode Event */ + r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, + GFX_9_0__SRCID__CP_BAD_OPCODE_ERROR, + &adev->gfx.bad_op_irq); + if (r) + return r; + /* Privileged reg */ r = amdgpu_irq_add_id(adev, SOC15_IH_CLIENTID_GRBM_CP, GFX_9_0__SRCID__CP_PRIV_REG_FAULT, &adev->gfx.priv_reg_irq); @@ -2162,6 +2169,7 @@ static int gfx_v9_4_3_hw_fini(void *handle) amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); + amdgpu_irq_put(adev, &adev->gfx.bad_op_irq, 0); num_xcc = NUM_XCC(adev->gfx.xcc_mask); for (i = 0; i < num_xcc; i++) { @@ -2327,6 +2335,10 @@ static int gfx_v9_4_3_late_init(void *handle) if (r) return r; + r = amdgpu_irq_get(adev, &adev->gfx.bad_op_irq, 0); + if (r) + return r; + if (adev->gfx.ras && adev->gfx.ras->enable_watchdog_timer) adev->gfx.ras->enable_watchdog_timer(adev); @@ -2964,6 +2976,46 @@ static int gfx_v9_4_3_set_priv_reg_fault_state(struct amdgpu_device *adev, return 0; } +static int gfx_v9_4_3_set_bad_op_fault_state(struct amdgpu_device *adev, + struct amdgpu_irq_src *source, + unsigned type, + enum amdgpu_interrupt_state state) +{ + 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++) { + WREG32_FIELD15_PREREG(GC, GET_INST(GC, i), CP_INT_CNTL_RING0, + OPCODE_ERROR_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, + OPCODE_ERROR_INT_ENABLE, + state == AMDGPU_IRQ_STATE_ENABLE ? + 1 : 0); + WREG32_XCC(mec_int_cntl_reg, mec_int_cntl, i); + } + } + } + } + break; + default: + break; + } + + return 0; +} + static int gfx_v9_4_3_set_priv_inst_fault_state(struct amdgpu_device *adev, struct amdgpu_irq_src *source, unsigned type, @@ -3116,6 +3168,15 @@ static int gfx_v9_4_3_priv_reg_irq(struct amdgpu_device *adev, return 0; } +static int gfx_v9_4_3_bad_op_irq(struct amdgpu_device *adev, + struct amdgpu_irq_src *source, + struct amdgpu_iv_entry *entry) +{ + DRM_ERROR("Illegal opcode in command stream\n"); + gfx_v9_4_3_fault(adev, entry); + return 0; +} + static int gfx_v9_4_3_priv_inst_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) @@ -4228,6 +4289,11 @@ static const struct amdgpu_irq_src_funcs gfx_v9_4_3_priv_reg_irq_funcs = { .process = gfx_v9_4_3_priv_reg_irq, }; +static const struct amdgpu_irq_src_funcs gfx_v9_4_3_bad_op_irq_funcs = { + .set = gfx_v9_4_3_set_bad_op_fault_state, + .process = gfx_v9_4_3_bad_op_irq, +}; + static const struct amdgpu_irq_src_funcs gfx_v9_4_3_priv_inst_irq_funcs = { .set = gfx_v9_4_3_set_priv_inst_fault_state, .process = gfx_v9_4_3_priv_inst_irq, @@ -4241,6 +4307,9 @@ static void gfx_v9_4_3_set_irq_funcs(struct amdgpu_device *adev) adev->gfx.priv_reg_irq.num_types = 1; adev->gfx.priv_reg_irq.funcs = &gfx_v9_4_3_priv_reg_irq_funcs; + adev->gfx.bad_op_irq.num_types = 1; + adev->gfx.bad_op_irq.funcs = &gfx_v9_4_3_bad_op_irq_funcs; + adev->gfx.priv_inst_irq.num_types = 1; adev->gfx.priv_inst_irq.funcs = &gfx_v9_4_3_priv_inst_irq_funcs; } -- 2.45.2 ------------------------------ Message: 2 Date: Wed, 17 Jul 2024 16:40:07 -0400 From: Alex Deucher <alexander.deucher@xxxxxxx> To: <amd-gfx@xxxxxxxxxxxxxxxxxxxxx> Cc: Jesse Zhang <jesse.zhang@xxxxxxx>, Jesse Zhang <Jesse.Zhang@xxxxxxx>, Prike Liang <Prike.Liang@xxxxxxx>, Alex Deucher <alexander.deucher@xxxxxxx> Subject: [PATCH 2/6] drm/amdgpu/gfx11: Enable bad opcode interrupt Message-ID: <20240717204011.15342-2-alexander.deucher@xxxxxxx> Content-Type: text/plain From: Jesse Zhang <jesse.zhang@xxxxxxx> For the bad opcode case, it will cause CP/ME hang. The firmware will prevent the ME side from hanging by raising a bad opcode interrupt. And the driver needs to perform a vmid reset when receiving the interrupt. v2: update irq naming (drop priv) (Alex) Signed-off-by: Jesse Zhang <Jesse.Zhang@xxxxxxx> Reviewed-by: Prike Liang <Prike.Liang@xxxxxxx> Reviewed-by: Alex Deucher <alexander.deucher@xxxxxxx> Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx> --- drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 73 ++++++++++++++++++++++++++ 1 file changed, 73 insertions(+) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c index 02efa475eb7e..ce5cb60b8628 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c @@ -1569,6 +1569,13 @@ static int gfx_v11_0_sw_init(void *handle) if (r) return r; + /* Bad opcode Event */ + r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP, + GFX_11_0_0__SRCID__CP_BAD_OPCODE_ERROR, + &adev->gfx.bad_op_irq); + if (r) + return r; + /* Privileged reg */ r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP, GFX_11_0_0__SRCID__CP_PRIV_REG_FAULT, @@ -4646,6 +4653,7 @@ static int gfx_v11_0_hw_fini(void *handle) amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); + amdgpu_irq_put(adev, &adev->gfx.bad_op_irq, 0); if (!adev->no_hw_access) { if (amdgpu_async_gfx_ring) { @@ -5002,6 +5010,9 @@ static int gfx_v11_0_late_init(void *handle) if (r) return r; + r = amdgpu_irq_get(adev, &adev->gfx.bad_op_irq, 0); + if (r) + return r; return 0; } @@ -6293,6 +6304,51 @@ static int gfx_v11_0_set_priv_reg_fault_state(struct amdgpu_device *adev, return 0; } +static int gfx_v11_0_set_bad_op_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: + for (i = 0; i < adev->gfx.me.num_me; i++) { + for (j = 0; j < adev->gfx.me.num_pipe_per_me; j++) { + cp_int_cntl_reg = gfx_v11_0_get_cpg_int_cntl(adev, i, 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_INT_CNTL_RING0, + OPCODE_ERROR_INT_ENABLE, + state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); + WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl); + } + } + } + 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_v11_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, + OPCODE_ERROR_INT_ENABLE, + state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); + WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl); + } + } + } + break; + default: + break; + } + return 0; +} + static int gfx_v11_0_set_priv_inst_fault_state(struct amdgpu_device *adev, struct amdgpu_irq_src *source, unsigned int type, @@ -6369,6 +6425,15 @@ static int gfx_v11_0_priv_reg_irq(struct amdgpu_device *adev, return 0; } +static int gfx_v11_0_bad_op_irq(struct amdgpu_device *adev, + struct amdgpu_irq_src *source, + struct amdgpu_iv_entry *entry) +{ + DRM_ERROR("Illegal opcode in command stream \n"); + gfx_v11_0_handle_priv_fault(adev, entry); + return 0; +} + static int gfx_v11_0_priv_inst_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) @@ -6747,6 +6812,11 @@ static const struct amdgpu_irq_src_funcs gfx_v11_0_priv_reg_irq_funcs = { .process = gfx_v11_0_priv_reg_irq, }; +static const struct amdgpu_irq_src_funcs gfx_v11_0_bad_op_irq_funcs = { + .set = gfx_v11_0_set_bad_op_fault_state, + .process = gfx_v11_0_bad_op_irq, +}; + static const struct amdgpu_irq_src_funcs gfx_v11_0_priv_inst_irq_funcs = { .set = gfx_v11_0_set_priv_inst_fault_state, .process = gfx_v11_0_priv_inst_irq, @@ -6764,6 +6834,9 @@ static void gfx_v11_0_set_irq_funcs(struct amdgpu_device *adev) adev->gfx.priv_reg_irq.num_types = 1; adev->gfx.priv_reg_irq.funcs = &gfx_v11_0_priv_reg_irq_funcs; + adev->gfx.bad_op_irq.num_types = 1; + adev->gfx.bad_op_irq.funcs = &gfx_v11_0_bad_op_irq_funcs; + adev->gfx.priv_inst_irq.num_types = 1; adev->gfx.priv_inst_irq.funcs = &gfx_v11_0_priv_inst_irq_funcs; -- 2.45.2 ------------------------------ Message: 3 Date: Wed, 17 Jul 2024 16:40:09 -0400 From: Alex Deucher <alexander.deucher@xxxxxxx> To: <amd-gfx@xxxxxxxxxxxxxxxxxxxxx> Cc: Jesse Zhang <jesse.zhang@xxxxxxx>, Alex Deucher <alexander.deucher@xxxxxxx> Subject: [PATCH 4/6] drm/amdgpu/gfx12: Enable bad opcode interrupt Message-ID: <20240717204011.15342-4-alexander.deucher@xxxxxxx> Content-Type: text/plain From: Jesse Zhang <jesse.zhang@xxxxxxx> For the bad opcode case, it will cause CP/ME hang. The firmware will prevent the ME side from hanging by raising a bad opcode interrupt. And the driver needs to perform a vmid reset when receiving the interrupt. v2: update irq naming (drop priv) (Alex) Signed-off-by: Jesse Zhang <jesse.zhang@xxxxxxx> Signed-off-by: Alex Deucher <alexander.deucher@xxxxxxx> --- drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 74 ++++++++++++++++++++++++++ 1 file changed, 74 insertions(+) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c index c74c8a60a23a..63b073fd4dc7 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c @@ -1349,6 +1349,13 @@ static int gfx_v12_0_sw_init(void *handle) if (r) return r; + /* Bad opcode Event */ + r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP, + GFX_11_0_0__SRCID__CP_BAD_OPCODE_ERROR, + &adev->gfx.bad_op_irq); + if (r) + return r; + /* Privileged reg */ r = amdgpu_irq_add_id(adev, SOC21_IH_CLIENTID_GRBM_CP, GFX_11_0_0__SRCID__CP_PRIV_REG_FAULT, @@ -3592,6 +3599,7 @@ static int gfx_v12_0_hw_fini(void *handle) amdgpu_irq_put(adev, &adev->gfx.priv_reg_irq, 0); amdgpu_irq_put(adev, &adev->gfx.priv_inst_irq, 0); + amdgpu_irq_put(adev, &adev->gfx.bad_op_irq, 0); if (!adev->no_hw_access) { if (amdgpu_async_gfx_ring) { @@ -3712,6 +3720,10 @@ static int gfx_v12_0_late_init(void *handle) if (r) return r; + r = amdgpu_irq_get(adev, &adev->gfx.bad_op_irq, 0); + if (r) + return r; + return 0; } @@ -4831,6 +4843,51 @@ static int gfx_v12_0_set_priv_reg_fault_state(struct amdgpu_device *adev, return 0; } +static int gfx_v12_0_set_bad_op_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: + for (i = 0; i < adev->gfx.me.num_me; i++) { + for (j = 0; j < adev->gfx.me.num_pipe_per_me; j++) { + cp_int_cntl_reg = gfx_v12_0_get_cpg_int_cntl(adev, i, 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_INT_CNTL_RING0, + OPCODE_ERROR_INT_ENABLE, + state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); + WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl); + } + } + } + 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_v12_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, + OPCODE_ERROR_INT_ENABLE, + state == AMDGPU_IRQ_STATE_ENABLE ? 1 : 0); + WREG32_SOC15_IP(GC, cp_int_cntl_reg, cp_int_cntl); + } + } + } + break; + default: + break; + } + return 0; +} + static int gfx_v12_0_set_priv_inst_fault_state(struct amdgpu_device *adev, struct amdgpu_irq_src *source, unsigned int type, @@ -4907,6 +4964,15 @@ static int gfx_v12_0_priv_reg_irq(struct amdgpu_device *adev, return 0; } +static int gfx_v12_0_bad_op_irq(struct amdgpu_device *adev, + struct amdgpu_irq_src *source, + struct amdgpu_iv_entry *entry) +{ + DRM_ERROR("Illegal opcode in command stream \n"); + gfx_v12_0_handle_priv_fault(adev, entry); + return 0; +} + static int gfx_v12_0_priv_inst_irq(struct amdgpu_device *adev, struct amdgpu_irq_src *source, struct amdgpu_iv_entry *entry) @@ -5219,6 +5285,11 @@ static const struct amdgpu_irq_src_funcs gfx_v12_0_priv_reg_irq_funcs = { .process = gfx_v12_0_priv_reg_irq, }; +static const struct amdgpu_irq_src_funcs gfx_v12_0_bad_op_irq_funcs = { + .set = gfx_v12_0_set_bad_op_fault_state, + .process = gfx_v12_0_bad_op_irq, +}; + static const struct amdgpu_irq_src_funcs gfx_v12_0_priv_inst_irq_funcs = { .set = gfx_v12_0_set_priv_inst_fault_state, .process = gfx_v12_0_priv_inst_irq, @@ -5232,6 +5303,9 @@ static void gfx_v12_0_set_irq_funcs(struct amdgpu_device *adev) adev->gfx.priv_reg_irq.num_types = 1; adev->gfx.priv_reg_irq.funcs = &gfx_v12_0_priv_reg_irq_funcs; + adev->gfx.bad_op_irq.num_types = 1; + adev->gfx.bad_op_irq.funcs = &gfx_v12_0_bad_op_irq_funcs; + adev->gfx.priv_inst_irq.num_types = 1; adev->gfx.priv_inst_irq.funcs = &gfx_v12_0_priv_inst_irq_funcs; } -- 2.45.2 ------------------------------ Subject: Digest Footer _______________________________________________ amd-gfx mailing list amd-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/amd-gfx ------------------------------ End of amd-gfx Digest, Vol 98, Issue 218 **************************************** |
