[PATCH 1/2] drm/amdgpu: Enable IH retry CAM on GFX9
Mukul Joshi
mukul.joshi at amd.com
Mon Dec 12 17:10:54 UTC 2022
This patch enables the IH retry CAM on GFX9 series cards. This
retry filter is used to prevent sending lots of retry interrupts
in a short span of time and overflowing the IH ring buffer. This
will also help reduce CPU interrupt workload.
Signed-off-by: Mukul Joshi <mukul.joshi at amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h | 2 +
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 51 ++++++++++++-------
drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c | 2 +-
drivers/gpu/drm/amd/amdgpu/vega20_ih.c | 46 ++++++-----------
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 4 +-
.../asic_reg/oss/osssys_4_2_0_offset.h | 6 +++
.../asic_reg/oss/osssys_4_2_0_sh_mask.h | 13 +++++
7 files changed, 75 insertions(+), 49 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h
index e9f2c11ea416..be243adf3e65 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h
@@ -98,6 +98,8 @@ struct amdgpu_irq {
struct irq_domain *domain; /* GPU irq controller domain */
unsigned virq[AMDGPU_MAX_IRQ_SRC_ID];
uint32_t srbm_soft_reset;
+ u32 retry_cam_doorbell_index;
+ bool retry_cam_enabled;
};
void amdgpu_irq_disable_all(struct amdgpu_device *adev);
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index 08d6cf79fb15..e5ab84f3e072 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -537,32 +537,49 @@ static int gmc_v9_0_process_interrupt(struct amdgpu_device *adev,
const char *mmhub_cid;
const char *hub_name;
u64 addr;
+ uint32_t cam_index = 0;
+ int ret;
addr = (u64)entry->src_data[0] << 12;
addr |= ((u64)entry->src_data[1] & 0xf) << 44;
if (retry_fault) {
- /* Returning 1 here also prevents sending the IV to the KFD */
+ if (adev->irq.retry_cam_enabled) {
+ /* Delegate it to a different ring if the hardware hasn't
+ * already done it.
+ */
+ if (entry->ih == &adev->irq.ih) {
+ amdgpu_irq_delegate(adev, entry, 8);
+ return 1;
+ }
+
+ cam_index = entry->src_data[2] & 0x3ff;
- /* Process it onyl if it's the first fault for this address */
- if (entry->ih != &adev->irq.ih_soft &&
- amdgpu_gmc_filter_faults(adev, entry->ih, addr, entry->pasid,
+ ret = amdgpu_vm_handle_fault(adev, entry->pasid, addr, write_fault);
+ WDOORBELL32(adev->irq.retry_cam_doorbell_index, cam_index);
+ if (ret)
+ return 1;
+ } else {
+ /* Process it onyl if it's the first fault for this address */
+ if (entry->ih != &adev->irq.ih_soft &&
+ amdgpu_gmc_filter_faults(adev, entry->ih, addr, entry->pasid,
entry->timestamp))
- return 1;
+ return 1;
- /* Delegate it to a different ring if the hardware hasn't
- * already done it.
- */
- if (entry->ih == &adev->irq.ih) {
- amdgpu_irq_delegate(adev, entry, 8);
- return 1;
- }
+ /* Delegate it to a different ring if the hardware hasn't
+ * already done it.
+ */
+ if (entry->ih == &adev->irq.ih) {
+ amdgpu_irq_delegate(adev, entry, 8);
+ return 1;
+ }
- /* Try to handle the recoverable page faults by filling page
- * tables
- */
- if (amdgpu_vm_handle_fault(adev, entry->pasid, addr, write_fault))
- return 1;
+ /* Try to handle the recoverable page faults by filling page
+ * tables
+ */
+ if (amdgpu_vm_handle_fault(adev, entry->pasid, addr, write_fault))
+ return 1;
+ }
}
if (!printk_ratelimit())
diff --git a/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c b/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c
index 19455a725939..685abf57ffdd 100644
--- a/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c
+++ b/drivers/gpu/drm/amd/amdgpu/nbio_v7_4.c
@@ -238,7 +238,7 @@ static void nbio_v7_4_ih_doorbell_range(struct amdgpu_device *adev,
if (use_doorbell) {
ih_doorbell_range = REG_SET_FIELD(ih_doorbell_range, BIF_IH_DOORBELL_RANGE, OFFSET, doorbell_index);
- ih_doorbell_range = REG_SET_FIELD(ih_doorbell_range, BIF_IH_DOORBELL_RANGE, SIZE, 4);
+ ih_doorbell_range = REG_SET_FIELD(ih_doorbell_range, BIF_IH_DOORBELL_RANGE, SIZE, 8);
} else
ih_doorbell_range = REG_SET_FIELD(ih_doorbell_range, BIF_IH_DOORBELL_RANGE, SIZE, 0);
diff --git a/drivers/gpu/drm/amd/amdgpu/vega20_ih.c b/drivers/gpu/drm/amd/amdgpu/vega20_ih.c
index 1706081d054d..ed41f96b251f 100644
--- a/drivers/gpu/drm/amd/amdgpu/vega20_ih.c
+++ b/drivers/gpu/drm/amd/amdgpu/vega20_ih.c
@@ -251,36 +251,14 @@ static int vega20_ih_enable_ring(struct amdgpu_device *adev,
return 0;
}
-/**
- * vega20_ih_reroute_ih - reroute VMC/UTCL2 ih to an ih ring
- *
- * @adev: amdgpu_device pointer
- *
- * Reroute VMC and UMC interrupts on primary ih ring to
- * ih ring 1 so they won't lose when bunches of page faults
- * interrupts overwhelms the interrupt handler(VEGA20)
- */
-static void vega20_ih_reroute_ih(struct amdgpu_device *adev)
+static uint32_t vega20_setup_retry_doorbell(u32 doorbell_index)
{
- uint32_t tmp;
+ u32 val = 0;
- /* vega20 ih reroute will go through psp this
- * function is used for newer asics starting arcturus
- */
- if (adev->ip_versions[OSSSYS_HWIP][0] >= IP_VERSION(4, 2, 1)) {
- /* Reroute to IH ring 1 for VMC */
- WREG32_SOC15(OSSSYS, 0, mmIH_CLIENT_CFG_INDEX, 0x12);
- tmp = RREG32_SOC15(OSSSYS, 0, mmIH_CLIENT_CFG_DATA);
- tmp = REG_SET_FIELD(tmp, IH_CLIENT_CFG_DATA, CLIENT_TYPE, 1);
- tmp = REG_SET_FIELD(tmp, IH_CLIENT_CFG_DATA, RING_ID, 1);
- WREG32_SOC15(OSSSYS, 0, mmIH_CLIENT_CFG_DATA, tmp);
-
- /* Reroute IH ring 1 for UTCL2 */
- WREG32_SOC15(OSSSYS, 0, mmIH_CLIENT_CFG_INDEX, 0x1B);
- tmp = RREG32_SOC15(OSSSYS, 0, mmIH_CLIENT_CFG_DATA);
- tmp = REG_SET_FIELD(tmp, IH_CLIENT_CFG_DATA, RING_ID, 1);
- WREG32_SOC15(OSSSYS, 0, mmIH_CLIENT_CFG_DATA, tmp);
- }
+ val = REG_SET_FIELD(val, IH_DOORBELL_RPTR, OFFSET, doorbell_index);
+ val = REG_SET_FIELD(val, IH_DOORBELL_RPTR, ENABLE, 1);
+
+ return val;
}
/**
@@ -332,8 +310,6 @@ static int vega20_ih_irq_init(struct amdgpu_device *adev)
for (i = 0; i < ARRAY_SIZE(ih); i++) {
if (ih[i]->ring_size) {
- if (i == 1)
- vega20_ih_reroute_ih(adev);
ret = vega20_ih_enable_ring(adev, ih[i]);
if (ret)
return ret;
@@ -346,6 +322,16 @@ static int vega20_ih_irq_init(struct amdgpu_device *adev)
pci_set_master(adev->pdev);
+ /* Enable IH Retry CAM */
+ WREG32_FIELD15(OSSSYS, 0, IH_RETRY_INT_CAM_CNTL, ENABLE, 1);
+
+ /* Allocate the doorbell for IH Retry CAM */
+ adev->irq.retry_cam_doorbell_index = (adev->doorbell_index.ih + 3) << 1;
+ WREG32_SOC15(OSSSYS, 0, mmIH_DOORBELL_RETRY_CAM,
+ vega20_setup_retry_doorbell(adev->irq.retry_cam_doorbell_index));
+
+ adev->irq.retry_cam_enabled = true;
+
/* enable interrupts */
ret = vega20_ih_toggle_interrupts(adev, true);
if (ret)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 2dc3b04064bd..747adad7e0d8 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -2162,7 +2162,9 @@ static void svm_range_drain_retry_fault(struct svm_range_list *svms)
pr_debug("drain retry fault gpu %d svms %p\n", i, svms);
amdgpu_ih_wait_on_checkpoint_process_ts(pdd->dev->adev,
- &pdd->dev->adev->irq.ih1);
+ pdd->dev->adev->irq.retry_cam_enabled ?
+ &pdd->dev->adev->irq.ih_soft :
+ &pdd->dev->adev->irq.ih1);
pr_debug("drain retry fault gpu %d svms 0x%p done\n", i, svms);
}
if (atomic_cmpxchg(&svms->drain_pagefaults, drain, 0) != drain)
diff --git a/drivers/gpu/drm/amd/include/asic_reg/oss/osssys_4_2_0_offset.h b/drivers/gpu/drm/amd/include/asic_reg/oss/osssys_4_2_0_offset.h
index bd129266ebfd..a84a7cfaf71e 100644
--- a/drivers/gpu/drm/amd/include/asic_reg/oss/osssys_4_2_0_offset.h
+++ b/drivers/gpu/drm/amd/include/asic_reg/oss/osssys_4_2_0_offset.h
@@ -135,6 +135,8 @@
#define mmIH_RB_WPTR_ADDR_LO_BASE_IDX 0
#define mmIH_DOORBELL_RPTR 0x0087
#define mmIH_DOORBELL_RPTR_BASE_IDX 0
+#define mmIH_DOORBELL_RETRY_CAM 0x0088
+#define mmIH_DOORBELL_RETRY_CAM_BASE_IDX 0
#define mmIH_RB_CNTL_RING1 0x008c
#define mmIH_RB_CNTL_RING1_BASE_IDX 0
#define mmIH_RB_BASE_RING1 0x008d
@@ -159,6 +161,8 @@
#define mmIH_RB_WPTR_RING2_BASE_IDX 0
#define mmIH_DOORBELL_RPTR_RING2 0x009f
#define mmIH_DOORBELL_RPTR_RING2_BASE_IDX 0
+#define mmIH_RETRY_CAM_ACK 0x00a4
+#define mmIH_RETRY_CAM_ACK_BASE_IDX 0
#define mmIH_VERSION 0x00a5
#define mmIH_VERSION_BASE_IDX 0
#define mmIH_CNTL 0x00c0
@@ -235,6 +239,8 @@
#define mmIH_MMHUB_ERROR_BASE_IDX 0
#define mmIH_MEM_POWER_CTRL 0x00e8
#define mmIH_MEM_POWER_CTRL_BASE_IDX 0
+#define mmIH_RETRY_INT_CAM_CNTL 0x00e9
+#define mmIH_RETRY_INT_CAM_CNTL_BASE_IDX 0
#define mmIH_REGISTER_LAST_PART2 0x00ff
#define mmIH_REGISTER_LAST_PART2_BASE_IDX 0
#define mmSEM_CLK_CTRL 0x0100
diff --git a/drivers/gpu/drm/amd/include/asic_reg/oss/osssys_4_2_0_sh_mask.h b/drivers/gpu/drm/amd/include/asic_reg/oss/osssys_4_2_0_sh_mask.h
index 3ea83ea9ce3a..a0cc5ecfa161 100644
--- a/drivers/gpu/drm/amd/include/asic_reg/oss/osssys_4_2_0_sh_mask.h
+++ b/drivers/gpu/drm/amd/include/asic_reg/oss/osssys_4_2_0_sh_mask.h
@@ -349,6 +349,19 @@
#define IH_DOORBELL_RPTR_RING2__ENABLE__SHIFT 0x1c
#define IH_DOORBELL_RPTR_RING2__OFFSET_MASK 0x03FFFFFFL
#define IH_DOORBELL_RPTR_RING2__ENABLE_MASK 0x10000000L
+//IH_RETRY_INT_CAM_CNTL
+#define IH_RETRY_INT_CAM_CNTL__CAM_SIZE__SHIFT 0x0
+#define IH_RETRY_INT_CAM_CNTL__BACK_PRESSURE_SKID_VALUE__SHIFT 0x8
+#define IH_RETRY_INT_CAM_CNTL__ENABLE__SHIFT 0x10
+#define IH_RETRY_INT_CAM_CNTL__MM_BACK_PRESSURE_ENABLE__SHIFT 0x11
+#define IH_RETRY_INT_CAM_CNTL__GC_BACK_PRESSURE_ENABLE__SHIFT 0x12
+#define IH_RETRY_INT_CAM_CNTL__PER_VF_ENTRY_SIZE__SHIFT 0x14
+#define IH_RETRY_INT_CAM_CNTL__CAM_SIZE_MASK 0x0000001FL
+#define IH_RETRY_INT_CAM_CNTL__BACK_PRESSURE_SKID_VALUE_MASK 0x00003F00L
+#define IH_RETRY_INT_CAM_CNTL__ENABLE_MASK 0x00010000L
+#define IH_RETRY_INT_CAM_CNTL__MM_BACK_PRESSURE_ENABLE_MASK 0x00020000L
+#define IH_RETRY_INT_CAM_CNTL__GC_BACK_PRESSURE_ENABLE_MASK 0x00040000L
+#define IH_RETRY_INT_CAM_CNTL__PER_VF_ENTRY_SIZE_MASK 0x00300000L
//IH_VERSION
#define IH_VERSION__MINVER__SHIFT 0x0
#define IH_VERSION__MAJVER__SHIFT 0x8
--
2.35.1
More information about the amd-gfx
mailing list