From f1f6f48a338cdab96efef712dbef6b1e279583e2 Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Tue, 25 Apr 2023 14:13:36 -0400
Subject: drm/amdgpu: Set GTT size equal to TTM mem limit
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Use the helper function in TTM to get TTM mem limit and
set GTT size to be equal to TTL mem limit.
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 25 ++++++-------------------
1 file changed, 6 insertions(+), 19 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 2cd081cbf706..f61f07575f63 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1803,26 +1803,13 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
DRM_INFO("amdgpu: %uM of VRAM memory ready\n",
(unsigned) (adev->gmc.real_vram_size / (1024 * 1024)));
- /* Compute GTT size, either based on 1/2 the size of RAM size
- * or whatever the user passed on module init */
- if (amdgpu_gtt_size == -1) {
- struct sysinfo si;
-
- si_meminfo(&si);
- /* Certain GL unit tests for large textures can cause problems
- * with the OOM killer since there is no way to link this memory
- * to a process. This was originally mitigated (but not necessarily
- * eliminated) by limiting the GTT size. The problem is this limit
- * is often too low for many modern games so just make the limit 1/2
- * of system memory which aligns with TTM. The OOM accounting needs
- * to be addressed, but we shouldn't prevent common 3D applications
- * from being usable just to potentially mitigate that corner case.
- */
- gtt_size = max((AMDGPU_DEFAULT_GTT_SIZE_MB << 20),
- (u64)si.totalram * si.mem_unit / 2);
- } else {
+ /* Compute GTT size, either based on TTM limit
+ * or whatever the user passed on module init.
+ */
+ if (amdgpu_gtt_size == -1)
+ gtt_size = ttm_tt_pages_limit() << PAGE_SHIFT;
+ else
gtt_size = (uint64_t)amdgpu_gtt_size << 20;
- }
/* Initialize GTT memory pool */
r = amdgpu_gtt_mgr_init(adev, gtt_size);
--
cgit v1.2.3
From 2f77b9a242a2e01822efc80c8b63eaa31df0f8b4 Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Mon, 9 May 2022 21:45:50 -0400
Subject: drm/amdkfd: Update MQD management on multi XCC setup
Update MQD management for both HIQ and user-mode compute
queues on a multi XCC setup. MQDs needs to be allocated,
initialized, loaded and destroyed for each XCC in the KFD
node.
v2: squash in fix "drm/amdkfd: Fix SDMA+HIQ HQD allocation on GFX9.4.3"
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Tested-by: Amber Lin <Amber.Lin@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 51 ++--
.../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 3 +-
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c | 28 +-
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h | 8 +
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c | 3 +
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c | 3 +
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c | 292 ++++++++++++++++++---
drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c | 3 +
.../gpu/drm/amd/amdkfd/kfd_process_queue_manager.c | 16 +-
drivers/gpu/drm/amd/include/v9_structs.h | 30 ++-
10 files changed, 380 insertions(+), 57 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index f61f07575f63..6bbe3b89aef5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -800,6 +800,41 @@ static void amdgpu_ttm_tt_unpin_userptr(struct ttm_device *bdev,
sg_free_table(ttm->sg);
}
+/*
+ * total_pages is constructed as MQD0+CtrlStack0 + MQD1+CtrlStack1 + ...
+ * MQDn+CtrlStackn where n is the number of XCCs per partition.
+ * pages_per_xcc is the size of one MQD+CtrlStack. The first page is MQD
+ * and uses memory type default, UC. The rest of pages_per_xcc are
+ * Ctrl stack and modify their memory type to NC.
+ */
+static void amdgpu_ttm_gart_bind_gfx9_mqd(struct amdgpu_device *adev,
+ struct ttm_tt *ttm, uint64_t flags)
+{
+ struct amdgpu_ttm_tt *gtt = (void *)ttm;
+ uint64_t total_pages = ttm->num_pages;
+ int num_xcc = max(1U, adev->gfx.num_xcc_per_xcp);
+ uint64_t page_idx, pages_per_xcc = total_pages / num_xcc;
+ int i;
+ uint64_t ctrl_flags = (flags & ~AMDGPU_PTE_MTYPE_VG10_MASK) |
+ AMDGPU_PTE_MTYPE_VG10(AMDGPU_MTYPE_NC);
+
+ for (i = 0, page_idx = 0; i < num_xcc; i++, page_idx += pages_per_xcc) {
+ /* MQD page: use default flags */
+ amdgpu_gart_bind(adev,
+ gtt->offset + (page_idx << PAGE_SHIFT),
+ 1, >t->ttm.dma_address[page_idx], flags);
+ /*
+ * Ctrl pages - modify the memory type to NC (ctrl_flags) from
+ * the second page of the BO onward.
+ */
+ amdgpu_gart_bind(adev,
+ gtt->offset + ((page_idx + 1) << PAGE_SHIFT),
+ pages_per_xcc - 1,
+ >t->ttm.dma_address[page_idx + 1],
+ ctrl_flags);
+ }
+}
+
static void amdgpu_ttm_gart_bind(struct amdgpu_device *adev,
struct ttm_buffer_object *tbo,
uint64_t flags)
@@ -812,21 +847,7 @@ static void amdgpu_ttm_gart_bind(struct amdgpu_device *adev,
flags |= AMDGPU_PTE_TMZ;
if (abo->flags & AMDGPU_GEM_CREATE_CP_MQD_GFX9) {
- uint64_t page_idx = 1;
-
- amdgpu_gart_bind(adev, gtt->offset, page_idx,
- gtt->ttm.dma_address, flags);
-
- /* The memory type of the first page defaults to UC. Now
- * modify the memory type to NC from the second page of
- * the BO onward.
- */
- flags &= ~AMDGPU_PTE_MTYPE_VG10_MASK;
- flags |= AMDGPU_PTE_MTYPE_VG10(AMDGPU_MTYPE_NC);
-
- amdgpu_gart_bind(adev, gtt->offset + (page_idx << PAGE_SHIFT),
- ttm->num_pages - page_idx,
- &(gtt->ttm.dma_address[page_idx]), flags);
+ amdgpu_ttm_gart_bind_gfx9_mqd(adev, ttm, flags);
} else {
amdgpu_gart_bind(adev, gtt->offset, ttm->num_pages,
gtt->ttm.dma_address, flags);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index 6ee17100c333..9afd3295ca85 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -2247,7 +2247,8 @@ static int allocate_hiq_sdma_mqd(struct device_queue_manager *dqm)
uint32_t size = dqm->mqd_mgrs[KFD_MQD_TYPE_SDMA]->mqd_size *
get_num_all_sdma_engines(dqm) *
dev->kfd->device_info.num_sdma_queues_per_engine +
- dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size;
+ (dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size *
+ dqm->dev->num_xcc_per_node);
retval = amdgpu_amdkfd_alloc_gtt_mem(dev->adev, size,
&(mem_obj->gtt_mem), &(mem_obj->gpu_addr),
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
index 61f6dd68c84b..074f6075ccc7 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.c
@@ -76,7 +76,8 @@ struct kfd_mem_obj *allocate_sdma_mqd(struct kfd_node *dev,
q->sdma_queue_id) *
dev->dqm->mqd_mgrs[KFD_MQD_TYPE_SDMA]->mqd_size;
- offset += dev->dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size;
+ offset += dev->dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size *
+ dev->num_xcc_per_node;
mqd_mem_obj->gtt_mem = (void *)((uint64_t)dev->dqm->hiq_sdma_mqd.gtt_mem
+ offset);
@@ -246,3 +247,28 @@ bool kfd_is_occupied_sdma(struct mqd_manager *mm, void *mqd,
{
return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->adev, mqd);
}
+
+uint64_t kfd_hiq_mqd_stride(struct kfd_node *dev)
+{
+ return dev->dqm->mqd_mgrs[KFD_MQD_TYPE_HIQ]->mqd_size;
+}
+
+void kfd_get_hiq_xcc_mqd(struct kfd_node *dev, struct kfd_mem_obj *mqd_mem_obj,
+ uint32_t virtual_xcc_id)
+{
+ uint64_t offset;
+
+ offset = kfd_hiq_mqd_stride(dev) * virtual_xcc_id;
+
+ mqd_mem_obj->gtt_mem = (virtual_xcc_id == 0) ?
+ dev->dqm->hiq_sdma_mqd.gtt_mem : NULL;
+ mqd_mem_obj->gpu_addr = dev->dqm->hiq_sdma_mqd.gpu_addr + offset;
+ mqd_mem_obj->cpu_ptr = (uint32_t *)((uintptr_t)
+ dev->dqm->hiq_sdma_mqd.cpu_ptr + offset);
+}
+
+uint64_t kfd_mqd_stride(struct mqd_manager *mm,
+ struct queue_properties *q)
+{
+ return mm->mqd_size;
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h
index 46fc3f273d0d..eb18be74f559 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager.h
@@ -119,6 +119,8 @@ struct mqd_manager {
int (*debugfs_show_mqd)(struct seq_file *m, void *data);
#endif
uint32_t (*read_doorbell_id)(void *mqd);
+ uint64_t (*mqd_stride)(struct mqd_manager *mm,
+ struct queue_properties *p);
struct mutex mqd_mutex;
struct kfd_node *dev;
@@ -164,4 +166,10 @@ bool kfd_is_occupied_sdma(struct mqd_manager *mm, void *mqd,
uint64_t queue_address, uint32_t pipe_id,
uint32_t queue_id);
+void kfd_get_hiq_xcc_mqd(struct kfd_node *dev,
+ struct kfd_mem_obj *mqd_mem_obj, uint32_t virtual_xcc_id);
+
+uint64_t kfd_hiq_mqd_stride(struct kfd_node *dev);
+uint64_t kfd_mqd_stride(struct mqd_manager *mm,
+ struct queue_properties *q);
#endif /* KFD_MQD_MANAGER_H_ */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c
index 03e04d5e5a11..ca1966466759 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_cik.c
@@ -428,6 +428,7 @@ struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
mqd->destroy_mqd = kfd_destroy_mqd_cp;
mqd->is_occupied = kfd_is_occupied_cp;
mqd->mqd_size = sizeof(struct cik_mqd);
+ mqd->mqd_stride = kfd_mqd_stride;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd;
#endif
@@ -442,6 +443,7 @@ struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
mqd->destroy_mqd = kfd_destroy_mqd_cp;
mqd->is_occupied = kfd_is_occupied_cp;
mqd->mqd_size = sizeof(struct cik_mqd);
+ mqd->mqd_stride = kfd_mqd_stride;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd;
#endif
@@ -457,6 +459,7 @@ struct mqd_manager *mqd_manager_init_cik(enum KFD_MQD_TYPE type,
mqd->checkpoint_mqd = checkpoint_mqd_sdma;
mqd->restore_mqd = restore_mqd_sdma;
mqd->mqd_size = sizeof(struct cik_sdma_rlc_registers);
+ mqd->mqd_stride = kfd_mqd_stride;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
#endif
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c
index 7a93be0ebb19..c9565ea99df5 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c
@@ -432,6 +432,7 @@ struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type,
mqd->get_wave_state = get_wave_state;
mqd->checkpoint_mqd = checkpoint_mqd;
mqd->restore_mqd = restore_mqd;
+ mqd->mqd_stride = kfd_mqd_stride;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd;
#endif
@@ -447,6 +448,7 @@ struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type,
mqd->destroy_mqd = kfd_destroy_mqd_cp;
mqd->is_occupied = kfd_is_occupied_cp;
mqd->mqd_size = sizeof(struct v10_compute_mqd);
+ mqd->mqd_stride = kfd_mqd_stride;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd;
#endif
@@ -478,6 +480,7 @@ struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type,
mqd->checkpoint_mqd = checkpoint_mqd_sdma;
mqd->restore_mqd = restore_mqd_sdma;
mqd->mqd_size = sizeof(struct v10_sdma_mqd);
+ mqd->mqd_stride = kfd_mqd_stride;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
#endif
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
index 943a738e73f9..c677322057dd 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v9.c
@@ -33,6 +33,21 @@
#include "sdma0/sdma0_4_0_sh_mask.h"
#include "amdgpu_amdkfd.h"
+static void update_mqd(struct mqd_manager *mm, void *mqd,
+ struct queue_properties *q,
+ struct mqd_update_info *minfo);
+
+static uint64_t mqd_stride_v9(struct mqd_manager *mm,
+ struct queue_properties *q)
+{
+ if (mm->dev->kfd->cwsr_enabled &&
+ q->type == KFD_QUEUE_TYPE_COMPUTE)
+ return ALIGN(q->ctl_stack_size, PAGE_SIZE) +
+ ALIGN(sizeof(struct v9_mqd), PAGE_SIZE);
+
+ return mm->mqd_size;
+}
+
static inline struct v9_mqd *get_mqd(void *mqd)
{
return (struct v9_mqd *)mqd;
@@ -110,8 +125,9 @@ static struct kfd_mem_obj *allocate_mqd(struct kfd_node *node,
if (!mqd_mem_obj)
return NULL;
retval = amdgpu_amdkfd_alloc_gtt_mem(node->adev,
- ALIGN(q->ctl_stack_size, PAGE_SIZE) +
- ALIGN(sizeof(struct v9_mqd), PAGE_SIZE),
+ (ALIGN(q->ctl_stack_size, PAGE_SIZE) +
+ ALIGN(sizeof(struct v9_mqd), PAGE_SIZE)) *
+ node->num_xcc_per_node,
&(mqd_mem_obj->gtt_mem),
&(mqd_mem_obj->gpu_addr),
(void *)&(mqd_mem_obj->cpu_ptr), true);
@@ -165,24 +181,9 @@ static void init_mqd(struct mqd_manager *mm, void **mqd,
1 << CP_HQD_QUANTUM__QUANTUM_SCALE__SHIFT |
1 << CP_HQD_QUANTUM__QUANTUM_DURATION__SHIFT;
- if (q->format == KFD_QUEUE_FORMAT_AQL) {
+ if (q->format == KFD_QUEUE_FORMAT_AQL)
m->cp_hqd_aql_control =
1 << CP_HQD_AQL_CONTROL__CONTROL0__SHIFT;
- if (KFD_GC_VERSION(mm->dev) == IP_VERSION(9, 4, 3)) {
- /* On GC 9.4.3, DW 41 is re-purposed as
- * compute_tg_chunk_size.
- * TODO: review this setting when active CUs in the
- * partition play a role
- */
- m->compute_static_thread_mgmt_se6 = 1;
- }
- } else {
- /* PM4 queue */
- if (KFD_GC_VERSION(mm->dev) == IP_VERSION(9, 4, 3)) {
- m->compute_static_thread_mgmt_se6 = 0;
- /* TODO: program pm4_target_xcc */
- }
- }
if (q->tba_addr) {
m->compute_pgm_rsrc2 |=
@@ -205,7 +206,7 @@ static void init_mqd(struct mqd_manager *mm, void **mqd,
*mqd = m;
if (gart_addr)
*gart_addr = addr;
- mm->update_mqd(mm, m, q, NULL);
+ update_mqd(mm, m, q, NULL);
}
static int load_mqd(struct mqd_manager *mm, void *mqd,
@@ -269,13 +270,10 @@ static void update_mqd(struct mqd_manager *mm, void *mqd,
m->cp_hqd_vmid = q->vmid;
if (q->format == KFD_QUEUE_FORMAT_AQL) {
- m->cp_hqd_pq_control |=
+ m->cp_hqd_pq_control |= CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK |
2 << CP_HQD_PQ_CONTROL__SLOT_BASED_WPTR__SHIFT |
1 << CP_HQD_PQ_CONTROL__QUEUE_FULL_EN__SHIFT |
1 << CP_HQD_PQ_CONTROL__WPP_CLAMP_EN__SHIFT;
- if (KFD_GC_VERSION(mm->dev) != IP_VERSION(9, 4, 3))
- m->cp_hqd_pq_control |=
- CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK;
m->cp_hqd_pq_doorbell_control |= 1 <<
CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_BIF_DROP__SHIFT;
}
@@ -466,6 +464,224 @@ static void restore_mqd_sdma(struct mqd_manager *mm, void **mqd,
qp->is_active = 0;
}
+static void init_mqd_hiq_v9_4_3(struct mqd_manager *mm, void **mqd,
+ struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr,
+ struct queue_properties *q)
+{
+ struct v9_mqd *m;
+ int xcc = 0;
+ struct kfd_mem_obj xcc_mqd_mem_obj;
+ uint64_t xcc_gart_addr = 0;
+
+ memset(&xcc_mqd_mem_obj, 0x0, sizeof(struct kfd_mem_obj));
+
+ for (xcc = 0; xcc < mm->dev->num_xcc_per_node; xcc++) {
+ kfd_get_hiq_xcc_mqd(mm->dev, &xcc_mqd_mem_obj, xcc);
+
+ init_mqd(mm, (void **)&m, &xcc_mqd_mem_obj, &xcc_gart_addr, q);
+
+ m->cp_hqd_pq_control |= CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK |
+ 1 << CP_HQD_PQ_CONTROL__PRIV_STATE__SHIFT |
+ 1 << CP_HQD_PQ_CONTROL__KMD_QUEUE__SHIFT;
+ m->cp_mqd_stride_size = kfd_hiq_mqd_stride(mm->dev);
+ if (xcc == 0) {
+ /* Set no_update_rptr = 0 in Master XCC */
+ m->cp_hqd_pq_control &= ~CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK;
+
+ /* Set the MQD pointer and gart address to XCC0 MQD */
+ *mqd = m;
+ *gart_addr = xcc_gart_addr;
+ }
+ }
+}
+
+static int hiq_load_mqd_kiq_v9_4_3(struct mqd_manager *mm, void *mqd,
+ uint32_t pipe_id, uint32_t queue_id,
+ struct queue_properties *p, struct mm_struct *mms)
+{
+ int xcc, err;
+ void *xcc_mqd;
+ uint64_t hiq_mqd_size = kfd_hiq_mqd_stride(mm->dev);
+
+ for (xcc = 0; xcc < mm->dev->num_xcc_per_node; xcc++) {
+ xcc_mqd = mqd + hiq_mqd_size * xcc;
+ err = mm->dev->kfd2kgd->hiq_mqd_load(mm->dev->adev, xcc_mqd,
+ pipe_id, queue_id,
+ p->doorbell_off);
+ if (err) {
+ pr_debug("Failed to load HIQ MQD for XCC: %d\n", xcc);
+ break;
+ }
+ }
+
+ return err;
+}
+
+static int destroy_hiq_mqd_v9_4_3(struct mqd_manager *mm, void *mqd,
+ enum kfd_preempt_type type, unsigned int timeout,
+ uint32_t pipe_id, uint32_t queue_id)
+{
+ int xcc = 0, err;
+ void *xcc_mqd;
+ uint64_t hiq_mqd_size = kfd_hiq_mqd_stride(mm->dev);
+
+ for (xcc = 0; xcc < mm->dev->num_xcc_per_node; xcc++) {
+ xcc_mqd = mqd + hiq_mqd_size * xcc;
+ err = mm->dev->kfd2kgd->hqd_destroy(mm->dev->adev, xcc_mqd,
+ type, timeout, pipe_id,
+ queue_id);
+ if (err) {
+ pr_debug("Destroy MQD failed for xcc: %d\n", xcc);
+ break;
+ }
+ }
+
+ return err;
+}
+
+static void get_xcc_mqd(struct kfd_mem_obj *mqd_mem_obj,
+ struct kfd_mem_obj *xcc_mqd_mem_obj,
+ uint64_t offset)
+{
+ xcc_mqd_mem_obj->gtt_mem = (offset == 0) ?
+ mqd_mem_obj->gtt_mem : NULL;
+ xcc_mqd_mem_obj->gpu_addr = mqd_mem_obj->gpu_addr + offset;
+ xcc_mqd_mem_obj->cpu_ptr = (uint32_t *)((uintptr_t)mqd_mem_obj->cpu_ptr
+ + offset);
+}
+
+static void init_mqd_v9_4_3(struct mqd_manager *mm, void **mqd,
+ struct kfd_mem_obj *mqd_mem_obj, uint64_t *gart_addr,
+ struct queue_properties *q)
+{
+ struct v9_mqd *m;
+ int xcc = 0;
+ struct kfd_mem_obj xcc_mqd_mem_obj;
+ uint64_t xcc_gart_addr = 0;
+ uint64_t offset = mm->mqd_stride(mm, q);
+
+ memset(&xcc_mqd_mem_obj, 0x0, sizeof(struct kfd_mem_obj));
+ for (xcc = 0; xcc < mm->dev->num_xcc_per_node; xcc++) {
+ get_xcc_mqd(mqd_mem_obj, &xcc_mqd_mem_obj, offset*xcc);
+
+ init_mqd(mm, (void **)&m, &xcc_mqd_mem_obj, &xcc_gart_addr, q);
+
+ m->cp_mqd_stride_size = offset;
+ if (q->format == KFD_QUEUE_FORMAT_AQL) {
+ m->compute_tg_chunk_size = 1;
+
+ switch (xcc) {
+ case 0:
+ /* Master XCC */
+ m->cp_hqd_pq_control &=
+ ~CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK;
+ m->compute_current_logic_xcc_id =
+ mm->dev->num_xcc_per_node - 1;
+ break;
+ default:
+ m->compute_current_logic_xcc_id =
+ xcc - 1;
+ break;
+ }
+ } else {
+ /* PM4 Queue */
+ m->compute_current_logic_xcc_id = 0;
+ m->compute_tg_chunk_size = 0;
+ }
+
+ if (xcc == 0) {
+ /* Set the MQD pointer and gart address to XCC0 MQD */
+ *mqd = m;
+ *gart_addr = xcc_gart_addr;
+ }
+ }
+}
+
+static void update_mqd_v9_4_3(struct mqd_manager *mm, void *mqd,
+ struct queue_properties *q, struct mqd_update_info *minfo)
+{
+ struct v9_mqd *m;
+ int xcc = 0;
+ uint64_t size = mm->mqd_stride(mm, q);
+
+ for (xcc = 0; xcc < mm->dev->num_xcc_per_node; xcc++) {
+ m = get_mqd(mqd + size * xcc);
+ update_mqd(mm, m, q, minfo);
+
+ if (q->format == KFD_QUEUE_FORMAT_AQL) {
+ switch (xcc) {
+ case 0:
+ /* Master XCC */
+ m->cp_hqd_pq_control &=
+ ~CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK;
+ m->compute_current_logic_xcc_id =
+ mm->dev->num_xcc_per_node - 1;
+ break;
+ default:
+ m->compute_current_logic_xcc_id =
+ xcc - 1;
+ break;
+ }
+ m->compute_tg_chunk_size = 1;
+ } else {
+ /* PM4 Queue */
+ m->compute_current_logic_xcc_id = 0;
+ m->compute_tg_chunk_size = 0;
+ }
+ }
+}
+
+static int destroy_mqd_v9_4_3(struct mqd_manager *mm, void *mqd,
+ enum kfd_preempt_type type, unsigned int timeout,
+ uint32_t pipe_id, uint32_t queue_id)
+{
+ int xcc = 0, err;
+ void *xcc_mqd;
+ struct v9_mqd *m;
+ uint64_t mqd_offset;
+
+ m = get_mqd(mqd);
+ mqd_offset = m->cp_mqd_stride_size;
+
+ for (xcc = 0; xcc < mm->dev->num_xcc_per_node; xcc++) {
+ xcc_mqd = mqd + mqd_offset * xcc;
+ err = mm->dev->kfd2kgd->hqd_destroy(mm->dev->adev, xcc_mqd,
+ type, timeout, pipe_id,
+ queue_id);
+ if (err) {
+ pr_debug("Destroy MQD failed for xcc: %d\n", xcc);
+ break;
+ }
+ }
+
+ return err;
+}
+
+static int load_mqd_v9_4_3(struct mqd_manager *mm, void *mqd,
+ uint32_t pipe_id, uint32_t queue_id,
+ struct queue_properties *p, struct mm_struct *mms)
+{
+ /* AQL write pointer counts in 64B packets, PM4/CP counts in dwords. */
+ uint32_t wptr_shift = (p->format == KFD_QUEUE_FORMAT_AQL ? 4 : 0);
+ int xcc = 0, err;
+ void *xcc_mqd;
+ uint64_t mqd_stride_size = mm->mqd_stride(mm, p);
+
+ for (xcc = 0; xcc < mm->dev->num_xcc_per_node; xcc++) {
+ xcc_mqd = mqd + mqd_stride_size * xcc;
+ err = mm->dev->kfd2kgd->hqd_load(mm->dev->adev, xcc_mqd,
+ pipe_id, queue_id,
+ (uint32_t __user *)p->write_ptr,
+ wptr_shift, 0, mms);
+ if (err) {
+ pr_debug("Load MQD failed for xcc: %d\n", xcc);
+ break;
+ }
+ }
+
+ return err;
+}
+
#if defined(CONFIG_DEBUG_FS)
static int debugfs_show_mqd(struct seq_file *m, void *data)
@@ -501,34 +717,49 @@ struct mqd_manager *mqd_manager_init_v9(enum KFD_MQD_TYPE type,
switch (type) {
case KFD_MQD_TYPE_CP:
mqd->allocate_mqd = allocate_mqd;
- mqd->init_mqd = init_mqd;
mqd->free_mqd = kfd_free_mqd_cp;
- mqd->load_mqd = load_mqd;
- mqd->update_mqd = update_mqd;
- mqd->destroy_mqd = kfd_destroy_mqd_cp;
mqd->is_occupied = kfd_is_occupied_cp;
mqd->get_wave_state = get_wave_state;
mqd->get_checkpoint_info = get_checkpoint_info;
mqd->checkpoint_mqd = checkpoint_mqd;
mqd->restore_mqd = restore_mqd;
mqd->mqd_size = sizeof(struct v9_mqd);
+ mqd->mqd_stride = mqd_stride_v9;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd;
#endif
+ if (KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 3)) {
+ mqd->init_mqd = init_mqd_v9_4_3;
+ mqd->load_mqd = load_mqd_v9_4_3;
+ mqd->update_mqd = update_mqd_v9_4_3;
+ mqd->destroy_mqd = destroy_mqd_v9_4_3;
+ } else {
+ mqd->init_mqd = init_mqd;
+ mqd->load_mqd = load_mqd;
+ mqd->update_mqd = update_mqd;
+ mqd->destroy_mqd = kfd_destroy_mqd_cp;
+ }
break;
case KFD_MQD_TYPE_HIQ:
mqd->allocate_mqd = allocate_hiq_mqd;
- mqd->init_mqd = init_mqd_hiq;
mqd->free_mqd = free_mqd_hiq_sdma;
- mqd->load_mqd = kfd_hiq_load_mqd_kiq;
mqd->update_mqd = update_mqd;
- mqd->destroy_mqd = kfd_destroy_mqd_cp;
mqd->is_occupied = kfd_is_occupied_cp;
mqd->mqd_size = sizeof(struct v9_mqd);
+ mqd->mqd_stride = kfd_mqd_stride;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd;
#endif
mqd->read_doorbell_id = read_doorbell_id;
+ if (KFD_GC_VERSION(dev) == IP_VERSION(9, 4, 3)) {
+ mqd->init_mqd = init_mqd_hiq_v9_4_3;
+ mqd->load_mqd = hiq_load_mqd_kiq_v9_4_3;
+ mqd->destroy_mqd = destroy_hiq_mqd_v9_4_3;
+ } else {
+ mqd->init_mqd = init_mqd_hiq;
+ mqd->load_mqd = kfd_hiq_load_mqd_kiq;
+ mqd->destroy_mqd = kfd_destroy_mqd_cp;
+ }
break;
case KFD_MQD_TYPE_DIQ:
mqd->allocate_mqd = allocate_mqd;
@@ -554,6 +785,7 @@ struct mqd_manager *mqd_manager_init_v9(enum KFD_MQD_TYPE type,
mqd->checkpoint_mqd = checkpoint_mqd_sdma;
mqd->restore_mqd = restore_mqd_sdma;
mqd->mqd_size = sizeof(struct v9_sdma_mqd);
+ mqd->mqd_stride = kfd_mqd_stride;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
#endif
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c
index f6b4a5686dcb..8736a3cdbe1e 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_vi.c
@@ -486,6 +486,7 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
mqd->destroy_mqd = kfd_destroy_mqd_cp;
mqd->is_occupied = kfd_is_occupied_cp;
mqd->mqd_size = sizeof(struct vi_mqd);
+ mqd->mqd_stride = kfd_mqd_stride;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd;
#endif
@@ -500,6 +501,7 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
mqd->destroy_mqd = kfd_destroy_mqd_cp;
mqd->is_occupied = kfd_is_occupied_cp;
mqd->mqd_size = sizeof(struct vi_mqd);
+ mqd->mqd_stride = kfd_mqd_stride;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd;
#endif
@@ -515,6 +517,7 @@ struct mqd_manager *mqd_manager_init_vi(enum KFD_MQD_TYPE type,
mqd->checkpoint_mqd = checkpoint_mqd_sdma;
mqd->restore_mqd = restore_mqd_sdma;
mqd->mqd_size = sizeof(struct vi_sdma_mqd);
+ mqd->mqd_stride = kfd_mqd_stride;
#if defined(CONFIG_DEBUG_FS)
mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
#endif
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
index 5602498e713f..b1fb017b2ef8 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
@@ -927,7 +927,9 @@ int pqm_debugfs_mqds(struct seq_file *m, void *data)
struct queue *q;
enum KFD_MQD_TYPE mqd_type;
struct mqd_manager *mqd_mgr;
- int r = 0;
+ int r = 0, xcc, num_xccs = 1;
+ void *mqd;
+ uint64_t size = 0;
list_for_each_entry(pqn, &pqm->queues, process_queue_list) {
if (pqn->q) {
@@ -943,6 +945,7 @@ int pqm_debugfs_mqds(struct seq_file *m, void *data)
seq_printf(m, " Compute queue on device %x\n",
q->device->id);
mqd_type = KFD_MQD_TYPE_CP;
+ num_xccs = q->device->num_xcc_per_node;
break;
default:
seq_printf(m,
@@ -951,6 +954,8 @@ int pqm_debugfs_mqds(struct seq_file *m, void *data)
continue;
}
mqd_mgr = q->device->dqm->mqd_mgrs[mqd_type];
+ size = mqd_mgr->mqd_stride(mqd_mgr,
+ &q->properties);
} else if (pqn->kq) {
q = pqn->kq->queue;
mqd_mgr = pqn->kq->mqd_mgr;
@@ -972,9 +977,12 @@ int pqm_debugfs_mqds(struct seq_file *m, void *data)
continue;
}
- r = mqd_mgr->debugfs_show_mqd(m, q->mqd);
- if (r != 0)
- break;
+ for (xcc = 0; xcc < num_xccs; xcc++) {
+ mqd = q->mqd + size * xcc;
+ r = mqd_mgr->debugfs_show_mqd(m, mqd);
+ if (r != 0)
+ break;
+ }
}
return r;
diff --git a/drivers/gpu/drm/amd/include/v9_structs.h b/drivers/gpu/drm/amd/include/v9_structs.h
index a0c672889fe4..a2f81b9c38af 100644
--- a/drivers/gpu/drm/amd/include/v9_structs.h
+++ b/drivers/gpu/drm/amd/include/v9_structs.h
@@ -196,10 +196,20 @@ struct v9_mqd {
uint32_t compute_wave_restore_addr_lo;
uint32_t compute_wave_restore_addr_hi;
uint32_t compute_wave_restore_control;
- uint32_t compute_static_thread_mgmt_se4;
- uint32_t compute_static_thread_mgmt_se5;
- uint32_t compute_static_thread_mgmt_se6;
- uint32_t compute_static_thread_mgmt_se7;
+ union {
+ struct {
+ uint32_t compute_static_thread_mgmt_se4;
+ uint32_t compute_static_thread_mgmt_se5;
+ uint32_t compute_static_thread_mgmt_se6;
+ uint32_t compute_static_thread_mgmt_se7;
+ };
+ struct {
+ uint32_t compute_current_logic_xcc_id; // offset: 39 (0x27)
+ uint32_t compute_restart_cg_tg_id; // offset: 40 (0x28)
+ uint32_t compute_tg_chunk_size; // offset: 41 (0x29)
+ uint32_t compute_restore_tg_chunk_size; // offset: 42 (0x2A)
+ };
+ };
uint32_t reserved_43;
uint32_t reserved_44;
uint32_t reserved_45;
@@ -382,8 +392,16 @@ struct v9_mqd {
uint32_t iqtimer_pkt_dw29;
uint32_t iqtimer_pkt_dw30;
uint32_t iqtimer_pkt_dw31;
- uint32_t reserved_225;
- uint32_t reserved_226;
+ union {
+ struct {
+ uint32_t reserved_225;
+ uint32_t reserved_226;
+ };
+ struct {
+ uint32_t pm4_target_xcc_in_xcp; // offset: 225 (0xE1)
+ uint32_t cp_mqd_stride_size; // offset: 226 (0xE2)
+ };
+ };
uint32_t reserved_227;
uint32_t set_resources_header;
uint32_t set_resources_dw1;
--
cgit v1.2.3
From 228ce176434b0f61451019065393040d58e1668d Mon Sep 17 00:00:00 2001
From: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>
Date: Fri, 27 Jan 2023 21:57:00 -0500
Subject: drm/amdgpu: Handle VRAM dependencies on GFXIP9.4.3
[For 1P NPS1 mode driver bringup]
Changes required to initialize the amdgpu driver with frontdoor firmware
loading and discovery=2 with the native mode SBIOS that enables CPU GPU
unified interleaved memory.
sudo modprobe amdgpu discovery=2
Once PSP TMR region is reported via the ACPI interface, the dependency
on the ip_discovery.bin will be removed.
Choice of where to allocate driver table is given to each IP version. In
general, both GTT and VRAM domains will be considered. If one of the
tables has a strict restriction for VRAM domain, then only VRAM domain
is considered.
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
(lijo: Modified the handling for SMU Tables)
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 3 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 6 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 89 +++++++++++++---------
drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c | 7 +-
drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c | 3 +-
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 9 ++-
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 6 ++
drivers/gpu/drm/amd/amdkfd/kfd_crat.c | 5 ++
drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c | 10 ++-
.../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c | 6 +-
11 files changed, 99 insertions(+), 47 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index af37f2ef4438..4e179e50de25 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -2292,8 +2292,9 @@ int amdgpu_amdkfd_gpuvm_import_dmabuf(struct amdgpu_device *adev,
(*mem)->dmabuf = dma_buf;
(*mem)->bo = bo;
(*mem)->va = va;
- (*mem)->domain = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) ?
+ (*mem)->domain = (bo->preferred_domains & AMDGPU_GEM_DOMAIN_VRAM) && !adev->gmc.is_app_apu ?
AMDGPU_GEM_DOMAIN_VRAM : AMDGPU_GEM_DOMAIN_GTT;
+
(*mem)->mapped_to_gpu_memory = 0;
(*mem)->process_info = avm->process_info;
add_kgd_mem_to_kfd_bo_list(*mem, avm->process_info, false);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 9f0d5f02119e..f431205e1077 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -1044,7 +1044,7 @@ static const char * const amdgpu_vram_names[] = {
int amdgpu_bo_init(struct amdgpu_device *adev)
{
/* On A+A platform, VRAM can be mapped as WB */
- if (!adev->gmc.xgmi.connected_to_cpu) {
+ if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu) {
/* reserve PAT memory space to WC for VRAM */
int r = arch_io_reserve_memtype_wc(adev->gmc.aper_base,
adev->gmc.aper_size);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
index 863fa331e6ff..4395c53d09d8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c
@@ -476,7 +476,8 @@ static int psp_sw_init(void *handle)
return ret;
ret = amdgpu_bo_create_kernel(adev, PSP_FENCE_BUFFER_SIZE, PAGE_SIZE,
- AMDGPU_GEM_DOMAIN_VRAM,
+ AMDGPU_GEM_DOMAIN_VRAM |
+ AMDGPU_GEM_DOMAIN_GTT,
&psp->fence_buf_bo,
&psp->fence_buf_mc_addr,
&psp->fence_buf);
@@ -484,7 +485,8 @@ static int psp_sw_init(void *handle)
goto failed1;
ret = amdgpu_bo_create_kernel(adev, PSP_CMD_BUFFER_SIZE, PAGE_SIZE,
- AMDGPU_GEM_DOMAIN_VRAM,
+ AMDGPU_GEM_DOMAIN_VRAM |
+ AMDGPU_GEM_DOMAIN_GTT,
&psp->cmd_buf_bo, &psp->cmd_buf_mc_addr,
(void **)&psp->cmd_buf_mem);
if (ret)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 6bbe3b89aef5..bc11ae56bba5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1708,15 +1708,20 @@ static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev)
ctx->init = PSP_MEM_TRAIN_RESERVE_SUCCESS;
}
- ret = amdgpu_bo_create_kernel_at(adev,
- adev->gmc.real_vram_size - adev->mman.discovery_tmr_size,
- adev->mman.discovery_tmr_size,
- &adev->mman.discovery_memory,
- NULL);
- if (ret) {
- DRM_ERROR("alloc tmr failed(%d)!\n", ret);
- amdgpu_bo_free_kernel(&adev->mman.discovery_memory, NULL, NULL);
- return ret;
+ if (!adev->gmc.is_app_apu) {
+ ret = amdgpu_bo_create_kernel_at(adev,
+ adev->gmc.real_vram_size -
+ adev->mman.discovery_tmr_size,
+ adev->mman.discovery_tmr_size,
+ &adev->mman.discovery_memory,
+ NULL);
+ if (ret) {
+ DRM_ERROR("alloc tmr failed(%d)!\n", ret);
+ amdgpu_bo_free_kernel(&adev->mman.discovery_memory, NULL, NULL);
+ return ret;
+ }
+ } else {
+ DRM_DEBUG_DRIVER("backdoor fw loading path for PSP TMR, no reservation needed\n");
}
return 0;
@@ -1765,10 +1770,12 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
adev->mman.aper_base_kaddr = ioremap_cache(adev->gmc.aper_base,
adev->gmc.visible_vram_size);
- else
+ else if (!adev->gmc.is_app_apu)
#endif
adev->mman.aper_base_kaddr = ioremap_wc(adev->gmc.aper_base,
adev->gmc.visible_vram_size);
+ else
+ DRM_DEBUG_DRIVER("No need to ioremap when real vram size is 0\n");
#endif
/*
@@ -1803,23 +1810,32 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
* This is used for VGA emulation and pre-OS scanout buffers to
* avoid display artifacts while transitioning between pre-OS
* and driver. */
- r = amdgpu_bo_create_kernel_at(adev, 0, adev->mman.stolen_vga_size,
- &adev->mman.stolen_vga_memory,
- NULL);
- if (r)
- return r;
- r = amdgpu_bo_create_kernel_at(adev, adev->mman.stolen_vga_size,
- adev->mman.stolen_extended_size,
- &adev->mman.stolen_extended_memory,
- NULL);
- if (r)
- return r;
- r = amdgpu_bo_create_kernel_at(adev, adev->mman.stolen_reserved_offset,
- adev->mman.stolen_reserved_size,
- &adev->mman.stolen_reserved_memory,
- NULL);
- if (r)
- return r;
+ if (!adev->gmc.is_app_apu) {
+ r = amdgpu_bo_create_kernel_at(adev, 0,
+ adev->mman.stolen_vga_size,
+ &adev->mman.stolen_vga_memory,
+ NULL);
+ if (r)
+ return r;
+
+ r = amdgpu_bo_create_kernel_at(adev, adev->mman.stolen_vga_size,
+ adev->mman.stolen_extended_size,
+ &adev->mman.stolen_extended_memory,
+ NULL);
+
+ if (r)
+ return r;
+
+ r = amdgpu_bo_create_kernel_at(adev,
+ adev->mman.stolen_reserved_offset,
+ adev->mman.stolen_reserved_size,
+ &adev->mman.stolen_reserved_memory,
+ NULL);
+ if (r)
+ return r;
+ } else {
+ DRM_DEBUG_DRIVER("Skipped stolen memory reservation\n");
+ }
DRM_INFO("amdgpu: %uM of VRAM memory ready\n",
(unsigned) (adev->gmc.real_vram_size / (1024 * 1024)));
@@ -1866,7 +1882,6 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
DRM_ERROR("Failed initializing oa heap.\n");
return r;
}
-
if (amdgpu_bo_create_kernel(adev, PAGE_SIZE, PAGE_SIZE,
AMDGPU_GEM_DOMAIN_GTT,
&adev->mman.sdma_access_bo, NULL,
@@ -1887,13 +1902,15 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev)
amdgpu_ttm_training_reserve_vram_fini(adev);
/* return the stolen vga memory back to VRAM */
- amdgpu_bo_free_kernel(&adev->mman.stolen_vga_memory, NULL, NULL);
- amdgpu_bo_free_kernel(&adev->mman.stolen_extended_memory, NULL, NULL);
- /* return the IP Discovery TMR memory back to VRAM */
- amdgpu_bo_free_kernel(&adev->mman.discovery_memory, NULL, NULL);
- if (adev->mman.stolen_reserved_size)
- amdgpu_bo_free_kernel(&adev->mman.stolen_reserved_memory,
- NULL, NULL);
+ if (!adev->gmc.is_app_apu) {
+ amdgpu_bo_free_kernel(&adev->mman.stolen_vga_memory, NULL, NULL);
+ amdgpu_bo_free_kernel(&adev->mman.stolen_extended_memory, NULL, NULL);
+ /* return the IP Discovery TMR memory back to VRAM */
+ amdgpu_bo_free_kernel(&adev->mman.discovery_memory, NULL, NULL);
+ if (adev->mman.stolen_reserved_size)
+ amdgpu_bo_free_kernel(&adev->mman.stolen_reserved_memory,
+ NULL, NULL);
+ }
amdgpu_bo_free_kernel(&adev->mman.sdma_access_bo, NULL,
&adev->mman.sdma_access_ptr);
amdgpu_ttm_fw_reserve_vram_fini(adev);
@@ -1935,7 +1952,7 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
int r;
if (!adev->mman.initialized || amdgpu_in_reset(adev) ||
- adev->mman.buffer_funcs_enabled == enable)
+ adev->mman.buffer_funcs_enabled == enable || adev->gmc.is_app_apu)
return;
if (enable) {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
index df63dc3bca18..bc5d126b600b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
@@ -512,7 +512,12 @@ int amdgpu_vm_pt_create(struct amdgpu_device *adev, struct amdgpu_vm *vm,
bp.size = amdgpu_vm_pt_size(adev, level);
bp.byte_align = AMDGPU_GPU_PAGE_SIZE;
- bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+
+ if (!adev->gmc.is_app_apu)
+ bp.domain = AMDGPU_GEM_DOMAIN_VRAM;
+ else
+ bp.domain = AMDGPU_GEM_DOMAIN_GTT;
+
bp.domain = amdgpu_bo_get_preferred_domain(adev, bp.domain);
bp.flags = AMDGPU_GEM_CREATE_VRAM_CONTIGUOUS |
AMDGPU_GEM_CREATE_CPU_GTT_USWC;
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 1f1268cd5e09..42877c4505f1 100644
--- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
+++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
@@ -459,7 +459,8 @@ static int gfx_v9_4_3_mec_init(struct amdgpu_device *adev)
adev->gfx.num_compute_rings * num_xcc * GFX9_MEC_HPD_SIZE;
if (mec_hpd_size) {
r = amdgpu_bo_create_reserved(adev, mec_hpd_size, PAGE_SIZE,
- AMDGPU_GEM_DOMAIN_VRAM,
+ AMDGPU_GEM_DOMAIN_VRAM |
+ AMDGPU_GEM_DOMAIN_GTT,
&adev->gfx.mec.hpd_eop_obj,
&adev->gfx.mec.hpd_eop_gpu_addr,
(void **)&hpd);
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index 16634a791e10..245de27c7540 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -1593,8 +1593,13 @@ static int gmc_v9_0_mc_init(struct amdgpu_device *adev)
int r;
/* size in MB on si */
- adev->gmc.mc_vram_size =
- adev->nbio.funcs->get_memsize(adev) * 1024ULL * 1024ULL;
+ if (!adev->gmc.is_app_apu) {
+ adev->gmc.mc_vram_size =
+ adev->nbio.funcs->get_memsize(adev) * 1024ULL * 1024ULL;
+ } else {
+ DRM_DEBUG("Set mc_vram_size = 0 for APP APU\n");
+ adev->gmc.mc_vram_size = 0;
+ }
adev->gmc.real_vram_size = adev->gmc.mc_vram_size;
if (!(adev->flags & AMD_IS_APU) &&
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 8b9accecf49b..f85ac4dbc673 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -1026,6 +1026,12 @@ bool kfd_dev_is_large_bar(struct kfd_node *dev)
if (dev->kfd->local_mem_info.local_mem_size_private == 0 &&
dev->kfd->local_mem_info.local_mem_size_public > 0)
return true;
+
+ if (dev->kfd->local_mem_info.local_mem_size_public == 0 && dev->kfd->adev->gmc.is_app_apu) {
+ pr_debug("APP APU, Consider like a large bar system\n");
+ return true;
+ }
+
return false;
}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
index 16475921587b..1aaf933f9f48 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
@@ -30,6 +30,9 @@
#include "amdgpu.h"
#include "amdgpu_amdkfd.h"
+/* Fixme: Fake 32GB for 1PNPS1 mode bringup */
+#define DUMMY_VRAM_SIZE 31138512896
+
/* GPU Processor ID base for dGPUs for which VCRAT needs to be created.
* GPU processor ID are expressed with Bit[31]=1.
* The base is set to 0x8000_0000 + 0x1000 to avoid collision with GPU IDs
@@ -1053,6 +1056,8 @@ static int kfd_parse_subtype_mem(struct crat_subtype_memory *mem,
props->heap_type = heap_type;
props->flags = flags;
+ if (size_in_bytes == 0)
+ size_in_bytes = DUMMY_VRAM_SIZE; /* Fixme: TBD */
props->size_in_bytes = size_in_bytes;
props->width = width;
diff --git a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
index 2ddf5198e5c4..4dea79a0c5b5 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/amdgpu_smu.c
@@ -822,11 +822,20 @@ static int smu_init_fb_allocations(struct smu_context *smu)
}
}
+ driver_table->domain = AMDGPU_GEM_DOMAIN_VRAM | AMDGPU_GEM_DOMAIN_GTT;
/* VRAM allocation for driver table */
for (i = 0; i < SMU_TABLE_COUNT; i++) {
if (tables[i].size == 0)
continue;
+ /* If one of the tables has VRAM domain restriction, keep it in
+ * VRAM
+ */
+ if ((tables[i].domain &
+ (AMDGPU_GEM_DOMAIN_VRAM | AMDGPU_GEM_DOMAIN_GTT)) ==
+ AMDGPU_GEM_DOMAIN_VRAM)
+ driver_table->domain = AMDGPU_GEM_DOMAIN_VRAM;
+
if (i == SMU_TABLE_PMSTATUSLOG)
continue;
@@ -836,7 +845,6 @@ static int smu_init_fb_allocations(struct smu_context *smu)
driver_table->size = max_table_size;
driver_table->align = PAGE_SIZE;
- driver_table->domain = AMDGPU_GEM_DOMAIN_VRAM;
ret = amdgpu_bo_create_kernel(adev,
driver_table->size,
diff --git a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c
index ea8f3d6fb98b..8969b3ff5c8f 100644
--- a/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c
+++ b/drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0_6_ppt.c
@@ -220,10 +220,12 @@ static int smu_v13_0_6_tables_init(struct smu_context *smu)
PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM);
SMU_TABLE_INIT(tables, SMU_TABLE_SMU_METRICS, sizeof(MetricsTable_t),
- PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM);
+ PAGE_SIZE,
+ AMDGPU_GEM_DOMAIN_VRAM | AMDGPU_GEM_DOMAIN_GTT);
SMU_TABLE_INIT(tables, SMU_TABLE_I2C_COMMANDS, sizeof(SwI2cRequest_t),
- PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM);
+ PAGE_SIZE,
+ AMDGPU_GEM_DOMAIN_VRAM | AMDGPU_GEM_DOMAIN_GTT);
smu_table->metrics_table = kzalloc(sizeof(MetricsTable_t), GFP_KERNEL);
if (!smu_table->metrics_table)
--
cgit v1.2.3
From db3b5cb64a9ca301d14ed027e470834316720e42 Mon Sep 17 00:00:00 2001
From: Lijo Lazar <lijo.lazar@amd.com>
Date: Fri, 24 Feb 2023 18:01:38 +0530
Subject: drm/amdgpu: Use apt name for FW reserved region
Use the generic term fw_reserved_memory for FW reserve region. This
region may also hold discovery TMR in addition to other reserve
regions. This region size could be larger than discovery tmr size, hence
don't change the discovery tmr size based on this.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Le Ma <le.ma@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 36 +++++++++++++++++----------------
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 3 ++-
2 files changed, 21 insertions(+), 18 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index bc11ae56bba5..c8a2d030f226 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1644,14 +1644,15 @@ static int amdgpu_ttm_training_reserve_vram_fini(struct amdgpu_device *adev)
return 0;
}
-static void amdgpu_ttm_training_data_block_init(struct amdgpu_device *adev)
+static void amdgpu_ttm_training_data_block_init(struct amdgpu_device *adev,
+ uint32_t reserve_size)
{
struct psp_memory_training_context *ctx = &adev->psp.mem_train_ctx;
memset(ctx, 0, sizeof(*ctx));
ctx->c2p_train_data_offset =
- ALIGN((adev->gmc.mc_vram_size - adev->mman.discovery_tmr_size - SZ_1M), SZ_1M);
+ ALIGN((adev->gmc.mc_vram_size - reserve_size - SZ_1M), SZ_1M);
ctx->p2c_train_data_offset =
(adev->gmc.mc_vram_size - GDDR6_MEM_TRAINING_OFFSET);
ctx->train_data_size =
@@ -1669,9 +1670,10 @@ static void amdgpu_ttm_training_data_block_init(struct amdgpu_device *adev)
*/
static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev)
{
- int ret;
struct psp_memory_training_context *ctx = &adev->psp.mem_train_ctx;
bool mem_train_support = false;
+ uint32_t reserve_size = 0;
+ int ret;
if (!amdgpu_sriov_vf(adev)) {
if (amdgpu_atomfirmware_mem_training_supported(adev))
@@ -1687,14 +1689,15 @@ static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev)
* Otherwise, fallback to legacy approach to check and reserve tmr block for ip
* discovery data and G6 memory training data respectively
*/
- adev->mman.discovery_tmr_size =
- amdgpu_atomfirmware_get_fw_reserved_fb_size(adev);
- if (!adev->mman.discovery_tmr_size)
- adev->mman.discovery_tmr_size = DISCOVERY_TMR_OFFSET;
+ if (adev->bios)
+ reserve_size =
+ amdgpu_atomfirmware_get_fw_reserved_fb_size(adev);
+ if (!reserve_size)
+ reserve_size = DISCOVERY_TMR_OFFSET;
if (mem_train_support) {
/* reserve vram for mem train according to TMR location */
- amdgpu_ttm_training_data_block_init(adev);
+ amdgpu_ttm_training_data_block_init(adev, reserve_size);
ret = amdgpu_bo_create_kernel_at(adev,
ctx->c2p_train_data_offset,
ctx->train_data_size,
@@ -1709,15 +1712,13 @@ static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev)
}
if (!adev->gmc.is_app_apu) {
- ret = amdgpu_bo_create_kernel_at(adev,
- adev->gmc.real_vram_size -
- adev->mman.discovery_tmr_size,
- adev->mman.discovery_tmr_size,
- &adev->mman.discovery_memory,
- NULL);
+ ret = amdgpu_bo_create_kernel_at(
+ adev, adev->gmc.real_vram_size - reserve_size,
+ reserve_size, &adev->mman.fw_reserved_memory, NULL);
if (ret) {
DRM_ERROR("alloc tmr failed(%d)!\n", ret);
- amdgpu_bo_free_kernel(&adev->mman.discovery_memory, NULL, NULL);
+ amdgpu_bo_free_kernel(&adev->mman.fw_reserved_memory,
+ NULL, NULL);
return ret;
}
} else {
@@ -1905,8 +1906,9 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev)
if (!adev->gmc.is_app_apu) {
amdgpu_bo_free_kernel(&adev->mman.stolen_vga_memory, NULL, NULL);
amdgpu_bo_free_kernel(&adev->mman.stolen_extended_memory, NULL, NULL);
- /* return the IP Discovery TMR memory back to VRAM */
- amdgpu_bo_free_kernel(&adev->mman.discovery_memory, NULL, NULL);
+ /* return the FW reserved memory back to VRAM */
+ amdgpu_bo_free_kernel(&adev->mman.fw_reserved_memory, NULL,
+ NULL);
if (adev->mman.stolen_reserved_size)
amdgpu_bo_free_kernel(&adev->mman.stolen_reserved_memory,
NULL, NULL);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index e2cd5894afc9..da6544fdc8dd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -78,7 +78,8 @@ struct amdgpu_mman {
/* discovery */
uint8_t *discovery_bin;
uint32_t discovery_tmr_size;
- struct amdgpu_bo *discovery_memory;
+ /* fw reserved memory */
+ struct amdgpu_bo *fw_reserved_memory;
/* firmware VRAM reservation */
u64 fw_vram_usage_start_offset;
--
cgit v1.2.3
From 1e03322cfef9b83aa87ea0a508588f9f05a47dfc Mon Sep 17 00:00:00 2001
From: Philip Yang <Philip.Yang@amd.com>
Date: Mon, 27 Feb 2023 11:16:09 -0500
Subject: drm/amdgpu: Set TTM pools for memory partitions
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
For native mode only, create TTM pool for each memory partition to store
the NUMA node id, then the TTM pool will be selected using memory
partition id to allocate memory from the correct partition.
Acked-by: Christian König <christian.koenig@amd.com>
(rajneesh: changed need_swiotlb and need_dma32 to false for pool init)
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-and-tested-by: Mukul Joshi <mukul.joshi@amd.com>
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 61 +++++++++++++++++++++++++++++++--
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 +
2 files changed, 60 insertions(+), 2 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index c8a2d030f226..7674109810b0 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -631,6 +631,7 @@ struct amdgpu_ttm_tt {
struct task_struct *usertask;
uint32_t userflags;
bool bound;
+ int32_t pool_id;
};
#define ttm_to_amdgpu_ttm_tt(ptr) container_of(ptr, struct amdgpu_ttm_tt, ttm)
@@ -1059,6 +1060,7 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
return NULL;
}
gtt->gobj = &bo->base;
+ gtt->pool_id = NUMA_NO_NODE;
if (abo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
caching = ttm_write_combined;
@@ -1085,6 +1087,7 @@ static int amdgpu_ttm_tt_populate(struct ttm_device *bdev,
{
struct amdgpu_device *adev = amdgpu_ttm_adev(bdev);
struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
+ struct ttm_pool *pool;
pgoff_t i;
int ret;
@@ -1099,7 +1102,11 @@ static int amdgpu_ttm_tt_populate(struct ttm_device *bdev,
if (ttm->page_flags & TTM_TT_FLAG_EXTERNAL)
return 0;
- ret = ttm_pool_alloc(&adev->mman.bdev.pool, ttm, ctx);
+ if (adev->mman.ttm_pools && gtt->pool_id >= 0)
+ pool = &adev->mman.ttm_pools[gtt->pool_id];
+ else
+ pool = &adev->mman.bdev.pool;
+ ret = ttm_pool_alloc(pool, ttm, ctx);
if (ret)
return ret;
@@ -1120,6 +1127,7 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_device *bdev,
{
struct amdgpu_ttm_tt *gtt = ttm_to_amdgpu_ttm_tt(ttm);
struct amdgpu_device *adev;
+ struct ttm_pool *pool;
pgoff_t i;
amdgpu_ttm_backend_unbind(bdev, ttm);
@@ -1138,7 +1146,13 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_device *bdev,
ttm->pages[i]->mapping = NULL;
adev = amdgpu_ttm_adev(bdev);
- return ttm_pool_free(&adev->mman.bdev.pool, ttm);
+
+ if (adev->mman.ttm_pools && gtt->pool_id >= 0)
+ pool = &adev->mman.ttm_pools[gtt->pool_id];
+ else
+ pool = &adev->mman.bdev.pool;
+
+ return ttm_pool_free(pool, ttm);
}
/**
@@ -1728,6 +1742,41 @@ static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev)
return 0;
}
+static int amdgpu_ttm_pools_init(struct amdgpu_device *adev)
+{
+ int i;
+
+ if (!adev->gmc.is_app_apu || !adev->gmc.num_mem_partitions)
+ return 0;
+
+ adev->mman.ttm_pools = kcalloc(adev->gmc.num_mem_partitions,
+ sizeof(*adev->mman.ttm_pools),
+ GFP_KERNEL);
+ if (!adev->mman.ttm_pools)
+ return -ENOMEM;
+
+ for (i = 0; i < adev->gmc.num_mem_partitions; i++) {
+ ttm_pool_init(&adev->mman.ttm_pools[i], adev->dev,
+ adev->gmc.mem_partitions[i].numa.node,
+ false, false);
+ }
+ return 0;
+}
+
+static void amdgpu_ttm_pools_fini(struct amdgpu_device *adev)
+{
+ int i;
+
+ if (!adev->gmc.is_app_apu || !adev->mman.ttm_pools)
+ return;
+
+ for (i = 0; i < adev->gmc.num_mem_partitions; i++)
+ ttm_pool_fini(&adev->mman.ttm_pools[i]);
+
+ kfree(adev->mman.ttm_pools);
+ adev->mman.ttm_pools = NULL;
+}
+
/*
* amdgpu_ttm_init - Init the memory management (ttm) as well as various
* gtt/vram related fields.
@@ -1754,6 +1803,12 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
DRM_ERROR("failed initializing buffer object driver(%d).\n", r);
return r;
}
+
+ r = amdgpu_ttm_pools_init(adev);
+ if (r) {
+ DRM_ERROR("failed to init ttm pools(%d).\n", r);
+ return r;
+ }
adev->mman.initialized = true;
/* Initialize VRAM pool with all of VRAM divided into pages */
@@ -1901,6 +1956,8 @@ void amdgpu_ttm_fini(struct amdgpu_device *adev)
if (!adev->mman.initialized)
return;
+ amdgpu_ttm_pools_fini(adev);
+
amdgpu_ttm_training_reserve_vram_fini(adev);
/* return the stolen vga memory back to VRAM */
if (!adev->gmc.is_app_apu) {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index da6544fdc8dd..8ef048a0a33e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -49,6 +49,7 @@ struct amdgpu_gtt_mgr {
struct amdgpu_mman {
struct ttm_device bdev;
+ struct ttm_pool *ttm_pools;
bool initialized;
void __iomem *aper_base_kaddr;
--
cgit v1.2.3
From fcfefd85f18a0004c7c7b499f0701fd2c76d4c68 Mon Sep 17 00:00:00 2001
From: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>
Date: Mon, 27 Feb 2023 20:08:29 -0500
Subject: drm/amdkfd: Native mode memory partition support
For native mode, after amdgpu_bo is created on CPU domain, then call
amdgpu_ttm_tt_set_mem_pool to select the TTM pool using bo->mem_id.
ttm_bo_validate will allocate the memory to the correct memory partition
before mapping to GPUs.
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Acked-and-tested-by: Mukul Joshi <mukul.joshi@amd.com>
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 7 +++++++
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 18 ++++++++++++++++++
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 +
3 files changed, 26 insertions(+)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index bbdd5e3aa18e..59404b3e6b87 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -1643,6 +1643,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
uint64_t aligned_size;
u64 alloc_flags;
int ret;
+ int mem_id = 0; /* Fixme : to be changed when mem_id support patch lands, until then NPS1, SPX only */
/*
* Check on which domain to allocate BO
@@ -1750,6 +1751,11 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
((*mem)->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM)) {
bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
bo->preferred_domains = AMDGPU_GEM_DOMAIN_GTT;
+ ret = amdgpu_ttm_tt_set_mem_pool(&bo->tbo, mem_id);
+ if (ret) {
+ pr_debug("failed to set ttm mem pool %d\n", ret);
+ goto err_set_mem_partition;
+ }
}
add_kgd_mem_to_kfd_bo_list(*mem, avm->process_info, user_addr);
@@ -1778,6 +1784,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
allocate_init_user_pages_failed:
err_pin_bo:
remove_kgd_mem_from_kfd_bo_list(*mem, avm->process_info);
+err_set_mem_partition:
drm_vma_node_revoke(&gobj->vma_node, drm_priv);
err_node_allow:
/* Don't unreserve system mem limit twice */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 7674109810b0..3933432daaac 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1155,6 +1155,24 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_device *bdev,
return ttm_pool_free(pool, ttm);
}
+/**
+ * amdgpu_ttm_tt_set_mem_pool - Set the TTM memory pool for the TTM BO
+ * @tbo: The ttm_buffer_object that backs the VRAM bo
+ * @mem_id: to select the initialized ttm pool corresponding to the memory partition
+ */
+int amdgpu_ttm_tt_set_mem_pool(struct ttm_buffer_object *tbo, int mem_id)
+{
+ struct ttm_tt *ttm = tbo->ttm;
+ struct amdgpu_ttm_tt *gtt;
+
+ if (!ttm && !ttm_tt_is_populated(ttm))
+ return -EINVAL;
+
+ gtt = ttm_to_amdgpu_ttm_tt(ttm);
+ gtt->pool_id = mem_id;
+ return 0;
+}
+
/**
* amdgpu_ttm_tt_get_userptr - Return the userptr GTT ttm_tt for the current
* task
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index 8ef048a0a33e..fe32de1bf4d5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -192,6 +192,7 @@ bool amdgpu_ttm_tt_has_userptr(struct ttm_tt *ttm);
struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm);
bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
unsigned long end, unsigned long *userptr);
+int amdgpu_ttm_tt_set_mem_pool(struct ttm_buffer_object *tbo, int mem_id);
bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
int *last_invalidated);
bool amdgpu_ttm_tt_is_userptr(struct ttm_tt *ttm);
--
cgit v1.2.3
From a0ba127960982b8827ba8b410c272ec8f3ee7e6a Mon Sep 17 00:00:00 2001
From: Lijo Lazar <lijo.lazar@amd.com>
Date: Fri, 3 Mar 2023 18:03:00 +0530
Subject: drm/amdgpu: Fix unmapping of aperture
When aperture size is zero, there is no mapping done.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 3 +--
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 7 ++++---
3 files changed, 6 insertions(+), 6 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 7f62826fcaca..19a3f9ba545b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -4121,7 +4121,7 @@ static void amdgpu_device_unmap_mmio(struct amdgpu_device *adev)
adev->mman.aper_base_kaddr = NULL;
/* Memory manager related */
- if (!adev->gmc.xgmi.connected_to_cpu) {
+ if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu) {
arch_phys_wc_del(adev->gmc.vram_mtrr);
arch_io_free_memtype_wc(adev->gmc.aper_base, adev->gmc.aper_size);
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index f431205e1077..725530eb4e0c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -1080,8 +1080,7 @@ void amdgpu_bo_fini(struct amdgpu_device *adev)
amdgpu_ttm_fini(adev);
if (drm_dev_enter(adev_to_drm(adev), &idx)) {
-
- if (!adev->gmc.xgmi.connected_to_cpu) {
+ if (!adev->gmc.xgmi.connected_to_cpu && !adev->gmc.is_app_apu) {
arch_phys_wc_del(adev->gmc.vram_mtrr);
arch_io_free_memtype_wc(adev->gmc.aper_base, adev->gmc.aper_size);
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 3933432daaac..09d1a98bd11e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1844,12 +1844,13 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
adev->mman.aper_base_kaddr = ioremap_cache(adev->gmc.aper_base,
adev->gmc.visible_vram_size);
- else if (!adev->gmc.is_app_apu)
+ else if (adev->gmc.is_app_apu)
+ DRM_DEBUG_DRIVER(
+ "No need to ioremap when real vram size is 0\n");
+ else
#endif
adev->mman.aper_base_kaddr = ioremap_wc(adev->gmc.aper_base,
adev->gmc.visible_vram_size);
- else
- DRM_DEBUG_DRIVER("No need to ioremap when real vram size is 0\n");
#endif
/*
--
cgit v1.2.3
From 53c5692e7a3c8e8eed3ec6b876a3c982d217a5d7 Mon Sep 17 00:00:00 2001
From: Philip Yang <Philip.Yang@amd.com>
Date: Thu, 26 Jan 2023 18:50:09 -0500
Subject: drm/amdkfd: Alloc memory of GPU support memory partition
For dGPU mode VRAM allocation, create amdgpu_bo from amdgpu_vm->mem_id,
to alloc from the correct memory range.
For APU mode VRAM allocation, set alloc domain to GTT, and set
bp->mem_id_plus1 from amdgpu_vm->mem_id + 1 to create amdgpu_bo, to
allocate system memory from correct NUMA node.
For GTT allocation, use mem_id -1 to allocate system memory from any
NUMA nodes.
Remove amdgpu_ttm_tt_set_mem_pool, to avoid the confusion that memory
maybe allocated from different mem_id.
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 24 +++++++-----------------
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 20 +-------------------
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 1 -
3 files changed, 8 insertions(+), 37 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index 71f1e1990925..c234dc0db799 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -1641,9 +1641,9 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
struct drm_gem_object *gobj = NULL;
u32 domain, alloc_domain;
uint64_t aligned_size;
+ int8_t mem_id = -1;
u64 alloc_flags;
int ret;
- int mem_id = 0; /* Fixme : to be changed when mem_id support patch lands, until then NPS1, SPX only */
/*
* Check on which domain to allocate BO
@@ -1653,13 +1653,14 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
if (adev->gmc.is_app_apu) {
domain = AMDGPU_GEM_DOMAIN_GTT;
- alloc_domain = AMDGPU_GEM_DOMAIN_CPU;
+ alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
alloc_flags = 0;
} else {
alloc_flags = AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE;
alloc_flags |= (flags & KFD_IOC_ALLOC_MEM_FLAGS_PUBLIC) ?
AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED : 0;
}
+ mem_id = avm->mem_id;
} else if (flags & KFD_IOC_ALLOC_MEM_FLAGS_GTT) {
domain = alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
alloc_flags = 0;
@@ -1717,11 +1718,12 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
goto err_reserve_limit;
}
- pr_debug("\tcreate BO VA 0x%llx size 0x%llx domain %s\n",
- va, (*mem)->aql_queue ? size << 1 : size, domain_string(alloc_domain));
+ pr_debug("\tcreate BO VA 0x%llx size 0x%llx domain %s mem_id %d\n",
+ va, (*mem)->aql_queue ? size << 1 : size,
+ domain_string(alloc_domain), mem_id);
ret = amdgpu_gem_object_create(adev, aligned_size, 1, alloc_domain, alloc_flags,
- bo_type, NULL, &gobj, 0);
+ bo_type, NULL, &gobj, mem_id + 1);
if (ret) {
pr_debug("Failed to create BO on domain %s. ret %d\n",
domain_string(alloc_domain), ret);
@@ -1747,17 +1749,6 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
(*mem)->mapped_to_gpu_memory = 0;
(*mem)->process_info = avm->process_info;
- if (adev->gmc.is_app_apu &&
- ((*mem)->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_VRAM)) {
- bo->allowed_domains = AMDGPU_GEM_DOMAIN_GTT;
- bo->preferred_domains = AMDGPU_GEM_DOMAIN_GTT;
- ret = amdgpu_ttm_tt_set_mem_pool(&bo->tbo, mem_id);
- if (ret) {
- pr_debug("failed to set ttm mem pool %d\n", ret);
- goto err_set_mem_partition;
- }
- }
-
add_kgd_mem_to_kfd_bo_list(*mem, avm->process_info, user_addr);
if (user_addr) {
@@ -1784,7 +1775,6 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
allocate_init_user_pages_failed:
err_pin_bo:
remove_kgd_mem_from_kfd_bo_list(*mem, avm->process_info);
-err_set_mem_partition:
drm_vma_node_revoke(&gobj->vma_node, drm_priv);
err_node_allow:
/* Don't unreserve system mem limit twice */
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 09d1a98bd11e..129c593cb2bd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1060,7 +1060,7 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
return NULL;
}
gtt->gobj = &bo->base;
- gtt->pool_id = NUMA_NO_NODE;
+ gtt->pool_id = abo->mem_id;
if (abo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
caching = ttm_write_combined;
@@ -1155,24 +1155,6 @@ static void amdgpu_ttm_tt_unpopulate(struct ttm_device *bdev,
return ttm_pool_free(pool, ttm);
}
-/**
- * amdgpu_ttm_tt_set_mem_pool - Set the TTM memory pool for the TTM BO
- * @tbo: The ttm_buffer_object that backs the VRAM bo
- * @mem_id: to select the initialized ttm pool corresponding to the memory partition
- */
-int amdgpu_ttm_tt_set_mem_pool(struct ttm_buffer_object *tbo, int mem_id)
-{
- struct ttm_tt *ttm = tbo->ttm;
- struct amdgpu_ttm_tt *gtt;
-
- if (!ttm && !ttm_tt_is_populated(ttm))
- return -EINVAL;
-
- gtt = ttm_to_amdgpu_ttm_tt(ttm);
- gtt->pool_id = mem_id;
- return 0;
-}
-
/**
* amdgpu_ttm_tt_get_userptr - Return the userptr GTT ttm_tt for the current
* task
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index fe32de1bf4d5..8ef048a0a33e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -192,7 +192,6 @@ bool amdgpu_ttm_tt_has_userptr(struct ttm_tt *ttm);
struct mm_struct *amdgpu_ttm_tt_get_usermm(struct ttm_tt *ttm);
bool amdgpu_ttm_tt_affect_userptr(struct ttm_tt *ttm, unsigned long start,
unsigned long end, unsigned long *userptr);
-int amdgpu_ttm_tt_set_mem_pool(struct ttm_buffer_object *tbo, int mem_id);
bool amdgpu_ttm_tt_userptr_invalidated(struct ttm_tt *ttm,
int *last_invalidated);
bool amdgpu_ttm_tt_is_userptr(struct ttm_tt *ttm);
--
cgit v1.2.3
From 3ebfd221c1a83e5f0edadb87d173d8fd93d1d125 Mon Sep 17 00:00:00 2001
From: Philip Yang <Philip.Yang@amd.com>
Date: Wed, 8 Mar 2023 11:57:00 -0500
Subject: drm/amdkfd: Store xcp partition id to amdgpu bo
For memory accounting per compute partition and export drm amdgpu bo and
then import to KFD, we need the xcp id to account the memory usage or
find the KFD node of the original amdgpu bo to create the KFD bo on the
correct adev KFD node.
Set xcp_id_plus1 of amdgpu_bo_param to create bo and store xcp_id to
amddgpu bo. Add helper macro to get the mem_id from adev and xcp_id.
v2: squash in fix ("drm/amdgpu: Fix BO creation failure on GFX 9.4.3 dGPU")
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 4 ++++
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 11 ++++++-----
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 4 ++--
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 15 ++++++++++-----
drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 12 ++++++++----
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 6 +++++-
drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c | 5 +++--
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 2 +-
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 4 ++--
10 files changed, 42 insertions(+), 23 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index 324cb566ca2f..05c54776951b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -330,6 +330,10 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
uint64_t size, u32 alloc_flag);
+#define KFD_XCP_MEM_ID(adev, xcp_id) \
+ ((adev)->xcp_mgr && (xcp_id) >= 0 ?\
+ (adev)->xcp_mgr->xcp[(xcp_id)].mem_id : -1)
+
#define KFD_XCP_MEMORY_SIZE(n) ((n)->adev->gmc.num_mem_partitions ?\
(n)->adev->gmc.mem_partitions[(n)->xcp->mem_id].size /\
(n)->adev->xcp_mgr->num_xcp_per_mem_partition :\
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index c234dc0db799..8724a0be31b8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -1634,6 +1634,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
uint64_t *offset, uint32_t flags, bool criu_resume)
{
struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
+ struct amdgpu_fpriv *fpriv = container_of(avm, struct amdgpu_fpriv, vm);
enum ttm_bo_type bo_type = ttm_bo_type_device;
struct sg_table *sg = NULL;
uint64_t user_addr = 0;
@@ -1641,7 +1642,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
struct drm_gem_object *gobj = NULL;
u32 domain, alloc_domain;
uint64_t aligned_size;
- int8_t mem_id = -1;
+ int8_t xcp_id = -1;
u64 alloc_flags;
int ret;
@@ -1660,7 +1661,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
alloc_flags |= (flags & KFD_IOC_ALLOC_MEM_FLAGS_PUBLIC) ?
AMDGPU_GEM_CREATE_CPU_ACCESS_REQUIRED : 0;
}
- mem_id = avm->mem_id;
+ xcp_id = fpriv->xcp_id == ~0 ? 0 : fpriv->xcp_id;
} else if (flags & KFD_IOC_ALLOC_MEM_FLAGS_GTT) {
domain = alloc_domain = AMDGPU_GEM_DOMAIN_GTT;
alloc_flags = 0;
@@ -1718,12 +1719,12 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
goto err_reserve_limit;
}
- pr_debug("\tcreate BO VA 0x%llx size 0x%llx domain %s mem_id %d\n",
+ pr_debug("\tcreate BO VA 0x%llx size 0x%llx domain %s xcp_id %d\n",
va, (*mem)->aql_queue ? size << 1 : size,
- domain_string(alloc_domain), mem_id);
+ domain_string(alloc_domain), xcp_id);
ret = amdgpu_gem_object_create(adev, aligned_size, 1, alloc_domain, alloc_flags,
- bo_type, NULL, &gobj, mem_id + 1);
+ bo_type, NULL, &gobj, xcp_id + 1);
if (ret) {
pr_debug("Failed to create BO on domain %s. ret %d\n",
domain_string(alloc_domain), ret);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 33ebee18b80d..7e8839cc6f58 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -98,7 +98,7 @@ int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size,
int alignment, u32 initial_domain,
u64 flags, enum ttm_bo_type type,
struct dma_resv *resv,
- struct drm_gem_object **obj, int8_t mem_id_plus1)
+ struct drm_gem_object **obj, int8_t xcp_id_plus1)
{
struct amdgpu_bo *bo;
struct amdgpu_bo_user *ubo;
@@ -116,7 +116,7 @@ int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size,
bp.flags = flags;
bp.domain = initial_domain;
bp.bo_ptr_size = sizeof(struct amdgpu_bo);
- bp.mem_id_plus1 = mem_id_plus1;
+ bp.xcp_id_plus1 = xcp_id_plus1;
r = amdgpu_bo_create_user(adev, &bp, &ubo);
if (r)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h
index 646c4fcc8e40..f30264782ba2 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.h
@@ -43,7 +43,7 @@ int amdgpu_gem_object_create(struct amdgpu_device *adev, unsigned long size,
int alignment, u32 initial_domain,
u64 flags, enum ttm_bo_type type,
struct dma_resv *resv,
- struct drm_gem_object **obj, int8_t mem_id_plus1);
+ struct drm_gem_object **obj, int8_t xcp_id_plus1);
int amdgpu_mode_dumb_create(struct drm_file *file_priv,
struct drm_device *dev,
struct drm_mode_create_dumb *args);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index b2d11c4f39b0..42c02f48c3a8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -131,14 +131,15 @@ void amdgpu_bo_placement_from_domain(struct amdgpu_bo *abo, u32 domain)
if (domain & AMDGPU_GEM_DOMAIN_VRAM) {
unsigned int visible_pfn = adev->gmc.visible_vram_size >> PAGE_SHIFT;
+ int8_t mem_id = KFD_XCP_MEM_ID(adev, abo->xcp_id);
- if (adev->gmc.mem_partitions && abo->mem_id >= 0) {
- places[c].fpfn = adev->gmc.mem_partitions[abo->mem_id].range.fpfn;
+ if (adev->gmc.mem_partitions && mem_id >= 0) {
+ places[c].fpfn = adev->gmc.mem_partitions[mem_id].range.fpfn;
/*
* memory partition range lpfn is inclusive start + size - 1
* TTM place lpfn is exclusive start + size
*/
- places[c].lpfn = adev->gmc.mem_partitions[abo->mem_id].range.lpfn + 1;
+ places[c].lpfn = adev->gmc.mem_partitions[mem_id].range.lpfn + 1;
} else {
places[c].fpfn = 0;
places[c].lpfn = 0;
@@ -583,8 +584,12 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
bo->flags = bp->flags;
- /* bo->mem_id -1 means any partition */
- bo->mem_id = bp->mem_id_plus1 - 1;
+ if (adev->gmc.mem_partitions)
+ /* For GPUs with spatial partitioning, bo->xcp_id=-1 means any partition */
+ bo->xcp_id = bp->xcp_id_plus1 - 1;
+ else
+ /* For GPUs without spatial partitioning */
+ bo->xcp_id = 0;
if (!amdgpu_bo_support_uswc(bo->flags))
bo->flags &= ~AMDGPU_GEM_CREATE_CPU_GTT_USWC;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
index eb24a66ccee5..05496b97ef93 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.h
@@ -56,8 +56,8 @@ struct amdgpu_bo_param {
bool no_wait_gpu;
struct dma_resv *resv;
void (*destroy)(struct ttm_buffer_object *bo);
- /* memory partition number plus 1, 0 means any partition */
- int8_t mem_id_plus1;
+ /* xcp partition number plus 1, 0 means any partition */
+ int8_t xcp_id_plus1;
};
/* bo virtual addresses in a vm */
@@ -111,8 +111,12 @@ struct amdgpu_bo {
#endif
struct kgd_mem *kfd_bo;
- /* memory partition number, -1 means any partition */
- int8_t mem_id;
+ /*
+ * For GPUs with spatial partitioning, xcp partition number, -1 means
+ * any partition. For other ASICs without spatial partition, always 0
+ * for memory accounting.
+ */
+ int8_t xcp_id;
};
struct amdgpu_bo_user {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 129c593cb2bd..23101c82519a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1051,6 +1051,7 @@ static void amdgpu_ttm_backend_destroy(struct ttm_device *bdev,
static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
uint32_t page_flags)
{
+ struct amdgpu_device *adev = amdgpu_ttm_adev(bo->bdev);
struct amdgpu_bo *abo = ttm_to_amdgpu_bo(bo);
struct amdgpu_ttm_tt *gtt;
enum ttm_caching caching;
@@ -1060,7 +1061,10 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
return NULL;
}
gtt->gobj = &bo->base;
- gtt->pool_id = abo->mem_id;
+ if (adev->gmc.mem_partitions && abo->xcp_id >= 0)
+ gtt->pool_id = KFD_XCP_MEM_ID(adev, abo->xcp_id);
+ else
+ gtt->pool_id = abo->xcp_id;
if (abo->flags & AMDGPU_GEM_CREATE_CPU_GTT_USWC)
caching = ttm_write_combined;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
index 62fc7e8d326e..cc3b1b596e56 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c
@@ -502,6 +502,7 @@ exit:
int amdgpu_vm_pt_create(struct amdgpu_device *adev, struct amdgpu_vm *vm,
int level, bool immediate, struct amdgpu_bo_vm **vmbo)
{
+ struct amdgpu_fpriv *fpriv = container_of(vm, struct amdgpu_fpriv, vm);
struct amdgpu_bo_param bp;
struct amdgpu_bo *bo;
struct dma_resv *resv;
@@ -534,7 +535,7 @@ int amdgpu_vm_pt_create(struct amdgpu_device *adev, struct amdgpu_vm *vm,
bp.type = ttm_bo_type_kernel;
bp.no_wait_gpu = immediate;
- bp.mem_id_plus1 = vm->mem_id + 1;
+ bp.xcp_id_plus1 = fpriv->xcp_id == ~0 ? 0 : fpriv->xcp_id + 1;
if (vm->root.bo)
bp.resv = vm->root.bo->tbo.base.resv;
@@ -560,7 +561,7 @@ int amdgpu_vm_pt_create(struct amdgpu_device *adev, struct amdgpu_vm *vm,
bp.type = ttm_bo_type_kernel;
bp.resv = bo->tbo.base.resv;
bp.bo_ptr_size = sizeof(struct amdgpu_bo);
- bp.mem_id_plus1 = vm->mem_id + 1;
+ bp.xcp_id_plus1 = fpriv->xcp_id == ~0 ? 0 : fpriv->xcp_id + 1;
r = amdgpu_bo_create(adev, &bp, &(*vmbo)->shadow);
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index 263d17a8b433..7ea80bdf8e1e 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -1248,7 +1248,7 @@ static void gmc_v9_0_get_coherence_flags(struct amdgpu_device *adev,
is_local = (!is_vram && (adev->flags & AMD_IS_APU) &&
num_possible_nodes() <= 1) ||
(is_vram && adev == bo_adev &&
- bo->mem_id == vm->mem_id);
+ KFD_XCP_MEM_ID(adev, bo->xcp_id) == vm->mem_id);
snoop = true;
if (uncached) {
mtype = MTYPE_UC;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index a700d9ccd054..45959892bc0f 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -556,7 +556,7 @@ svm_range_vram_node_new(struct kfd_node *node, struct svm_range *prange,
bp.type = ttm_bo_type_device;
bp.resv = NULL;
if (node->xcp)
- bp.mem_id_plus1 = node->xcp->mem_id + 1;
+ bp.xcp_id_plus1 = node->xcp->id + 1;
r = amdgpu_bo_create_user(node->adev, &bp, &ubo);
if (r) {
@@ -567,7 +567,7 @@ svm_range_vram_node_new(struct kfd_node *node, struct svm_range *prange,
pr_debug("alloc bo at offset 0x%lx size 0x%lx on partition %d\n",
bo->tbo.resource->start << PAGE_SHIFT, bp.size,
- bp.mem_id_plus1 - 1);
+ bp.xcp_id_plus1 - 1);
r = amdgpu_bo_reserve(bo, true);
if (r) {
--
cgit v1.2.3
From 45b3a914d40e63d2c9e3a3e02fb2014be975b9b0 Mon Sep 17 00:00:00 2001
From: Alex Deucher <alexander.deucher@amd.com>
Date: Tue, 16 May 2023 17:16:30 -0400
Subject: drm/amdgpu/gmc9: fix 64 bit division in partition code
Rework logic or use do_div() to avoid problems on 32 bit.
v2: add a missing case for XCP macro
v3: fix out of bounds array access
v4: fix xcp handling harder
Acked-by: Guchun Chen <guchun.chen@amd.com> (v1)
Reviewed-by: Mukul Joshi <mukul.joshi@amd.com> (v3)
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 15 +++++++++++++++
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 9 ++++-----
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 5 ++++-
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 11 ++++++-----
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 8 ++++++--
5 files changed, 35 insertions(+), 13 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
index 739eb7c0d133..5de92c9ab18f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
@@ -794,3 +794,18 @@ void amdgpu_amdkfd_unlock_kfd(struct amdgpu_device *adev)
{
kgd2kfd_unlock_kfd();
}
+
+
+u64 amdgpu_amdkfd_xcp_memory_size(struct amdgpu_device *adev, int xcp_id)
+{
+ u64 tmp;
+ s8 mem_id = KFD_XCP_MEM_ID(adev, xcp_id);
+
+ if (adev->gmc.num_mem_partitions && xcp_id >= 0 && mem_id >= 0) {
+ tmp = adev->gmc.mem_partitions[mem_id].size;
+ do_div(tmp, adev->xcp_mgr->num_xcp_per_mem_partition);
+ return tmp;
+ } else {
+ return adev->gmc.real_vram_size;
+ }
+}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
index be43d71ba7ef..94cc456761e5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
@@ -333,15 +333,14 @@ int amdgpu_amdkfd_reserve_mem_limit(struct amdgpu_device *adev,
void amdgpu_amdkfd_unreserve_mem_limit(struct amdgpu_device *adev,
uint64_t size, u32 alloc_flag, int8_t xcp_id);
+u64 amdgpu_amdkfd_xcp_memory_size(struct amdgpu_device *adev, int xcp_id);
+
#define KFD_XCP_MEM_ID(adev, xcp_id) \
((adev)->xcp_mgr && (xcp_id) >= 0 ?\
(adev)->xcp_mgr->xcp[(xcp_id)].mem_id : -1)
-#define KFD_XCP_MEMORY_SIZE(adev, xcp_id)\
- ((adev)->gmc.num_mem_partitions && (xcp_id) >= 0 ?\
- (adev)->gmc.mem_partitions[KFD_XCP_MEM_ID((adev), (xcp_id))].size /\
- (adev)->xcp_mgr->num_xcp_per_mem_partition :\
- (adev)->gmc.real_vram_size)
+#define KFD_XCP_MEMORY_SIZE(adev, xcp_id) amdgpu_amdkfd_xcp_memory_size((adev), (xcp_id))
+
#if IS_ENABLED(CONFIG_HSA_AMD)
void amdgpu_amdkfd_gpuvm_init_mem_limits(void);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 23101c82519a..902773ce41b5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -814,11 +814,14 @@ static void amdgpu_ttm_gart_bind_gfx9_mqd(struct amdgpu_device *adev,
struct amdgpu_ttm_tt *gtt = (void *)ttm;
uint64_t total_pages = ttm->num_pages;
int num_xcc = max(1U, adev->gfx.num_xcc_per_xcp);
- uint64_t page_idx, pages_per_xcc = total_pages / num_xcc;
+ uint64_t page_idx, pages_per_xcc;
int i;
uint64_t ctrl_flags = (flags & ~AMDGPU_PTE_MTYPE_VG10_MASK) |
AMDGPU_PTE_MTYPE_VG10(AMDGPU_MTYPE_NC);
+ pages_per_xcc = total_pages;
+ do_div(pages_per_xcc, num_xcc);
+
for (i = 0, page_idx = 0; i < num_xcc; i++, page_idx += pages_per_xcc) {
/* MQD page: use default flags */
amdgpu_gart_bind(adev,
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index 7ea80bdf8e1e..f70e666cecf2 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -1914,9 +1914,10 @@ gmc_v9_0_init_acpi_mem_ranges(struct amdgpu_device *adev,
adev->gmc.num_mem_partitions = num_ranges;
/* If there is only partition, don't use entire size */
- if (adev->gmc.num_mem_partitions == 1)
- mem_ranges[0].size =
- (mem_ranges[0].size * (mem_groups - 1) / mem_groups);
+ if (adev->gmc.num_mem_partitions == 1) {
+ mem_ranges[0].size = mem_ranges[0].size * (mem_groups - 1);
+ do_div(mem_ranges[0].size, mem_groups);
+ }
}
static void
@@ -1948,8 +1949,8 @@ gmc_v9_0_init_sw_mem_ranges(struct amdgpu_device *adev,
break;
}
- size = (adev->gmc.real_vram_size >> AMDGPU_GPU_PAGE_SHIFT) /
- adev->gmc.num_mem_partitions;
+ size = adev->gmc.real_vram_size >> AMDGPU_GPU_PAGE_SHIFT;
+ size /= adev->gmc.num_mem_partitions;
for (i = 0; i < adev->gmc.num_mem_partitions; ++i) {
mem_ranges[i].range.fpfn = start_addr;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index 206851c9e642..b0f0d31bf3e6 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
@@ -1939,10 +1939,14 @@ void svm_range_set_max_pages(struct amdgpu_device *adev)
uint64_t max_pages;
uint64_t pages, _pages;
uint64_t min_pages = 0;
- int i;
+ int i, id;
for (i = 0; i < adev->kfd.dev->num_nodes; i++) {
- pages = KFD_XCP_MEMORY_SIZE(adev, adev->kfd.dev->nodes[i]->xcp->id) >> 17;
+ if (adev->kfd.dev->nodes[i]->xcp)
+ id = adev->kfd.dev->nodes[i]->xcp->id;
+ else
+ id = -1;
+ pages = KFD_XCP_MEMORY_SIZE(adev, id) >> 17;
pages = clamp(pages, 1ULL << 9, 1ULL << 18);
pages = rounddown_pow_of_two(pages);
min_pages = min_not_zero(min_pages, pages);
--
cgit v1.2.3
From 01c3f464743b64e6e65cb9bad951458986819a42 Mon Sep 17 00:00:00 2001
From: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Date: Wed, 17 May 2023 20:10:48 +0530
Subject: drm/amd/amdgpu: Fix errors & warnings in amdgpu_ttm.c
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Fix below checkpatch insisted error & warnings:
ERROR: Macros with complex values should be enclosed in parentheses
WARNING: Prefer 'unsigned int' to bare use of 'unsigned'
WARNING: braces {} are not necessary for single statement blocks
WARNING: Block comments use a trailing */ on a separate line
WARNING: Missing a blank line after declarations
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Acked-by: Luben Tuikov <luben.tuikov@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 25 +++++++++++++------------
1 file changed, 13 insertions(+), 12 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 902773ce41b5..b8adcebb0d89 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -65,7 +65,7 @@
MODULE_IMPORT_NS(DMA_BUF);
-#define AMDGPU_TTM_VRAM_MAX_DW_READ (size_t)128
+#define AMDGPU_TTM_VRAM_MAX_DW_READ ((size_t)128)
static int amdgpu_ttm_backend_bind(struct ttm_device *bdev,
struct ttm_tt *ttm,
@@ -184,11 +184,11 @@ static void amdgpu_evict_flags(struct ttm_buffer_object *bo,
static int amdgpu_ttm_map_buffer(struct ttm_buffer_object *bo,
struct ttm_resource *mem,
struct amdgpu_res_cursor *mm_cur,
- unsigned window, struct amdgpu_ring *ring,
+ unsigned int window, struct amdgpu_ring *ring,
bool tmz, uint64_t *size, uint64_t *addr)
{
struct amdgpu_device *adev = ring->adev;
- unsigned offset, num_pages, num_dw, num_bytes;
+ unsigned int offset, num_pages, num_dw, num_bytes;
uint64_t src_addr, dst_addr;
struct amdgpu_job *job;
void *cpu_addr;
@@ -1060,9 +1060,9 @@ static struct ttm_tt *amdgpu_ttm_tt_create(struct ttm_buffer_object *bo,
enum ttm_caching caching;
gtt = kzalloc(sizeof(struct amdgpu_ttm_tt), GFP_KERNEL);
- if (gtt == NULL) {
+ if (!gtt)
return NULL;
- }
+
gtt->gobj = &bo->base;
if (adev->gmc.mem_partitions && abo->xcp_id >= 0)
gtt->pool_id = KFD_XCP_MEM_ID(adev, abo->xcp_id);
@@ -1847,9 +1847,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
*place on the VRAM, so reserve it early.
*/
r = amdgpu_ttm_fw_reserve_vram_init(adev);
- if (r) {
+ if (r)
return r;
- }
/*
*The reserved vram for driver must be pinned to the specified
@@ -1873,7 +1872,8 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
/* allocate memory as required for VGA
* This is used for VGA emulation and pre-OS scanout buffers to
* avoid display artifacts while transitioning between pre-OS
- * and driver. */
+ * and driver.
+ */
if (!adev->gmc.is_app_apu) {
r = amdgpu_bo_create_kernel_at(adev, 0,
adev->mman.stolen_vga_size,
@@ -1902,7 +1902,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
}
DRM_INFO("amdgpu: %uM of VRAM memory ready\n",
- (unsigned) (adev->gmc.real_vram_size / (1024 * 1024)));
+ (unsigned int)(adev->gmc.real_vram_size / (1024 * 1024)));
/* Compute GTT size, either based on TTM limit
* or whatever the user passed on module init.
@@ -1919,7 +1919,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
return r;
}
DRM_INFO("amdgpu: %uM of GTT memory ready.\n",
- (unsigned)(gtt_size / (1024 * 1024)));
+ (unsigned int)(gtt_size / (1024 * 1024)));
/* Initialize preemptible memory pool */
r = amdgpu_preempt_mgr_init(adev);
@@ -1961,6 +1961,7 @@ int amdgpu_ttm_init(struct amdgpu_device *adev)
void amdgpu_ttm_fini(struct amdgpu_device *adev)
{
int idx;
+
if (!adev->mman.initialized)
return;
@@ -2089,10 +2090,10 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,
bool vm_needs_flush, bool tmz)
{
struct amdgpu_device *adev = ring->adev;
- unsigned num_loops, num_dw;
+ unsigned int num_loops, num_dw;
struct amdgpu_job *job;
uint32_t max_bytes;
- unsigned i;
+ unsigned int i;
int r;
if (!direct_submit && !ring->sched.ready) {
--
cgit v1.2.3
From 109b4d8cfe4279da1cbcbcd99ae54cb2b2aee521 Mon Sep 17 00:00:00 2001
From: Su Hui <suhui@nfschina.com>
Date: Mon, 15 May 2023 09:34:28 +0800
Subject: drm/amdgpu: remove unnecessary (void*) conversions
No need cast (void*) to (struct amdgpu_device *).
Signed-off-by: Su Hui <suhui@nfschina.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 4 ++--
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 2 +-
6 files changed, 7 insertions(+), 7 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
index e94d0cf3f793..c657bed350ac 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c
@@ -1638,7 +1638,7 @@ int amdgpu_debugfs_regs_init(struct amdgpu_device *adev)
static int amdgpu_debugfs_test_ib_show(struct seq_file *m, void *unused)
{
- struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
+ struct amdgpu_device *adev = m->private;
struct drm_device *dev = adev_to_drm(adev);
int r = 0, i;
@@ -1749,7 +1749,7 @@ static int amdgpu_debugfs_benchmark(void *data, u64 val)
static int amdgpu_debugfs_vm_info_show(struct seq_file *m, void *unused)
{
- struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
+ struct amdgpu_device *adev = m->private;
struct drm_device *dev = adev_to_drm(adev);
struct drm_file *file;
int r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
index 928814492d5b..876ec35b8f83 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
@@ -833,7 +833,7 @@ static const struct dma_fence_ops amdgpu_job_fence_ops = {
#if defined(CONFIG_DEBUG_FS)
static int amdgpu_debugfs_fence_info_show(struct seq_file *m, void *unused)
{
- struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
+ struct amdgpu_device *adev = m->private;
int i;
for (i = 0; i < AMDGPU_MAX_RINGS; ++i) {
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index fec9df354652..74055cba3dc9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -951,7 +951,7 @@ int amdgpu_mode_dumb_create(struct drm_file *file_priv,
#if defined(CONFIG_DEBUG_FS)
static int amdgpu_debugfs_gem_info_show(struct seq_file *m, void *unused)
{
- struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
+ struct amdgpu_device *adev = m->private;
struct drm_device *dev = adev_to_drm(adev);
struct drm_file *file;
int r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
index b27ac48ca123..ebeddc9a37e9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ib.c
@@ -463,7 +463,7 @@ int amdgpu_ib_ring_tests(struct amdgpu_device *adev)
static int amdgpu_debugfs_sa_info_show(struct seq_file *m, void *unused)
{
- struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
+ struct amdgpu_device *adev = m->private;
seq_printf(m, "--------------------- DELAYED --------------------- \n");
amdgpu_sa_bo_dump_debug_info(&adev->ib_pools[AMDGPU_IB_POOL_DELAYED],
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 815098be4c2f..41d047e5de69 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -1468,7 +1468,7 @@ void amdgpu_disable_vblank_kms(struct drm_crtc *crtc)
static int amdgpu_debugfs_firmware_info_show(struct seq_file *m, void *unused)
{
- struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
+ struct amdgpu_device *adev = m->private;
struct drm_amdgpu_info_firmware fw_info;
struct drm_amdgpu_query_fw query_fw;
struct atom_context *ctx = adev->mode_info.atom_context;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index b8adcebb0d89..509b26e9d0c4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -2257,7 +2257,7 @@ int amdgpu_ttm_evict_resources(struct amdgpu_device *adev, int mem_type)
static int amdgpu_ttm_page_pool_show(struct seq_file *m, void *unused)
{
- struct amdgpu_device *adev = (struct amdgpu_device *)m->private;
+ struct amdgpu_device *adev = m->private;
return ttm_pool_debugfs(&adev->mman.bdev.pool, m);
}
--
cgit v1.2.3
From 73ade646c545feda7c5df9b9c78c5d011ce76463 Mon Sep 17 00:00:00 2001
From: Christoph Hellwig <hch@lst.de>
Date: Thu, 18 May 2023 15:51:38 +0200
Subject: drm/amdgpu: stop including swiotlb.h
amdgpu does not need swiotlb.h, so stop including it.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 1 -
1 file changed, 1 deletion(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 509b26e9d0c4..0abad5f89421 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -38,7 +38,6 @@
#include <linux/seq_file.h>
#include <linux/slab.h>
#include <linux/swap.h>
-#include <linux/swiotlb.h>
#include <linux/dma-buf.h>
#include <linux/sizes.h>
#include <linux/module.h>
--
cgit v1.2.3
From c3aaca43fb07ce05f3a3bd85288eb3d500469be5 Mon Sep 17 00:00:00 2001
From: Mukul Joshi <mukul.joshi@amd.com>
Date: Wed, 17 May 2023 15:53:50 -0400
Subject: drm/amdgpu: Add a low priority scheduler for VRAM clearing
Add a low priority DRM scheduler for VRAM clearing instead of using
the exisiting high priority scheduler. Use the high priority scheduler
for migrations and evictions.
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 4 ++--
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 37 ++++++++++++++++++++++--------
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 5 +++-
3 files changed, 34 insertions(+), 12 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
index 42c02f48c3a8..d9e331508389 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_object.c
@@ -627,7 +627,7 @@ int amdgpu_bo_create(struct amdgpu_device *adev,
bo->tbo.resource->mem_type == TTM_PL_VRAM) {
struct dma_fence *fence;
- r = amdgpu_fill_buffer(bo, 0, bo->tbo.base.resv, &fence);
+ r = amdgpu_fill_buffer(bo, 0, bo->tbo.base.resv, &fence, true);
if (unlikely(r))
goto fail_unreserve;
@@ -1354,7 +1354,7 @@ void amdgpu_bo_release_notify(struct ttm_buffer_object *bo)
if (WARN_ON_ONCE(!dma_resv_trylock(bo->base.resv)))
return;
- r = amdgpu_fill_buffer(abo, AMDGPU_POISON, bo->base.resv, &fence);
+ r = amdgpu_fill_buffer(abo, AMDGPU_POISON, bo->base.resv, &fence, true);
if (!WARN_ON(r)) {
amdgpu_bo_fence(abo, fence, false);
dma_fence_put(fence);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 0abad5f89421..473eeac1f03b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -383,7 +383,8 @@ static int amdgpu_move_blit(struct ttm_buffer_object *bo,
(abo->flags & AMDGPU_GEM_CREATE_VRAM_WIPE_ON_RELEASE)) {
struct dma_fence *wipe_fence = NULL;
- r = amdgpu_fill_buffer(abo, AMDGPU_POISON, NULL, &wipe_fence);
+ r = amdgpu_fill_buffer(abo, AMDGPU_POISON, NULL, &wipe_fence,
+ false);
if (r) {
goto error;
} else if (wipe_fence) {
@@ -2036,8 +2037,18 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
r);
return;
}
+
+ r = drm_sched_entity_init(&adev->mman.delayed,
+ DRM_SCHED_PRIORITY_NORMAL, &sched,
+ 1, NULL);
+ if (r) {
+ DRM_ERROR("Failed setting up TTM BO move entity (%d)\n",
+ r);
+ goto error_free_entity;
+ }
} else {
drm_sched_entity_destroy(&adev->mman.entity);
+ drm_sched_entity_destroy(&adev->mman.delayed);
dma_fence_put(man->move);
man->move = NULL;
}
@@ -2049,6 +2060,11 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
size = adev->gmc.visible_vram_size;
man->size = size;
adev->mman.buffer_funcs_enabled = enable;
+
+ return;
+
+error_free_entity:
+ drm_sched_entity_destroy(&adev->mman.entity);
}
static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev,
@@ -2056,14 +2072,16 @@ static int amdgpu_ttm_prepare_job(struct amdgpu_device *adev,
unsigned int num_dw,
struct dma_resv *resv,
bool vm_needs_flush,
- struct amdgpu_job **job)
+ struct amdgpu_job **job,
+ bool delayed)
{
enum amdgpu_ib_pool_type pool = direct_submit ?
AMDGPU_IB_POOL_DIRECT :
AMDGPU_IB_POOL_DELAYED;
int r;
-
- r = amdgpu_job_alloc_with_ib(adev, &adev->mman.entity,
+ struct drm_sched_entity *entity = delayed ? &adev->mman.delayed :
+ &adev->mman.entity;
+ r = amdgpu_job_alloc_with_ib(adev, entity,
AMDGPU_FENCE_OWNER_UNDEFINED,
num_dw * 4, pool, job);
if (r)
@@ -2104,7 +2122,7 @@ int amdgpu_copy_buffer(struct amdgpu_ring *ring, uint64_t src_offset,
num_loops = DIV_ROUND_UP(byte_count, max_bytes);
num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->copy_num_dw, 8);
r = amdgpu_ttm_prepare_job(adev, direct_submit, num_dw,
- resv, vm_needs_flush, &job);
+ resv, vm_needs_flush, &job, false);
if (r)
return r;
@@ -2140,7 +2158,7 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring, uint32_t src_data,
uint64_t dst_addr, uint32_t byte_count,
struct dma_resv *resv,
struct dma_fence **fence,
- bool vm_needs_flush)
+ bool vm_needs_flush, bool delayed)
{
struct amdgpu_device *adev = ring->adev;
unsigned int num_loops, num_dw;
@@ -2153,7 +2171,7 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring, uint32_t src_data,
num_loops = DIV_ROUND_UP_ULL(byte_count, max_bytes);
num_dw = ALIGN(num_loops * adev->mman.buffer_funcs->fill_num_dw, 8);
r = amdgpu_ttm_prepare_job(adev, false, num_dw, resv, vm_needs_flush,
- &job);
+ &job, delayed);
if (r)
return r;
@@ -2176,7 +2194,8 @@ static int amdgpu_ttm_fill_mem(struct amdgpu_ring *ring, uint32_t src_data,
int amdgpu_fill_buffer(struct amdgpu_bo *bo,
uint32_t src_data,
struct dma_resv *resv,
- struct dma_fence **f)
+ struct dma_fence **f,
+ bool delayed)
{
struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev);
struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
@@ -2205,7 +2224,7 @@ int amdgpu_fill_buffer(struct amdgpu_bo *bo,
goto error;
r = amdgpu_ttm_fill_mem(ring, src_data, to, cur_size, resv,
- &next, true);
+ &next, true, delayed);
if (r)
goto error;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
index 8ef048a0a33e..e82b1edee7a4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h
@@ -61,6 +61,8 @@ struct amdgpu_mman {
struct mutex gtt_window_lock;
/* Scheduler entity for buffer moves */
struct drm_sched_entity entity;
+ /* Scheduler entity for VRAM clearing */
+ struct drm_sched_entity delayed;
struct amdgpu_vram_mgr vram_mgr;
struct amdgpu_gtt_mgr gtt_mgr;
@@ -152,7 +154,8 @@ int amdgpu_ttm_copy_mem_to_mem(struct amdgpu_device *adev,
int amdgpu_fill_buffer(struct amdgpu_bo *bo,
uint32_t src_data,
struct dma_resv *resv,
- struct dma_fence **fence);
+ struct dma_fence **fence,
+ bool delayed);
int amdgpu_ttm_alloc_gart(struct ttm_buffer_object *bo);
void amdgpu_ttm_recover_gart(struct ttm_buffer_object *tbo);
--
cgit v1.2.3
From 9535a86a4072babc37dc6bdadae52bdbb88166f5 Mon Sep 17 00:00:00 2001
From: Shiwu Zhang <shiwu.zhang@amd.com>
Date: Wed, 17 May 2023 14:15:05 +0800
Subject: drm/amdgpu: bypass bios dependent operations
Since bios reading does not work currently so just bypass all operations
related to bios
v2: hardcode the vram info for APP_APU case (hawking)
v3: correct the vram_width with channel number * channel size (lijo)
Signed-off-by: Shiwu Zhang <shiwu.zhang@amd.com>
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 66 +++++++++++++++++++-----------
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 7 +++-
drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 63 +++++++++++++++++-----------
3 files changed, 85 insertions(+), 51 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index ff9ca0dbeb5a..8f50ca2bee97 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -1389,6 +1389,15 @@ int amdgpu_device_resize_fb_bar(struct amdgpu_device *adev)
return 0;
}
+static bool amdgpu_device_read_bios(struct amdgpu_device *adev)
+{
+ if (hweight32(adev->aid_mask) && (adev->flags & AMD_IS_APU)) {
+ return false;
+ }
+
+ return true;
+}
+
/*
* GPU helpers function.
*/
@@ -1408,6 +1417,9 @@ bool amdgpu_device_need_post(struct amdgpu_device *adev)
if (amdgpu_sriov_vf(adev))
return false;
+ if (!amdgpu_device_read_bios(adev))
+ return false;
+
if (amdgpu_passthrough(adev)) {
/* for FIJI: In whole GPU pass-through virtualization case, after VM reboot
* some old smc fw still need driver do vPost otherwise gpu hang, while
@@ -2318,14 +2330,16 @@ static int amdgpu_device_ip_early_init(struct amdgpu_device *adev)
return r;
/* Read BIOS */
- if (!amdgpu_get_bios(adev))
- return -EINVAL;
+ if (amdgpu_device_read_bios(adev)) {
+ if (!amdgpu_get_bios(adev))
+ return -EINVAL;
- r = amdgpu_atombios_init(adev);
- if (r) {
- dev_err(adev->dev, "amdgpu_atombios_init failed\n");
- amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0);
- return r;
+ r = amdgpu_atombios_init(adev);
+ if (r) {
+ dev_err(adev->dev, "amdgpu_atombios_init failed\n");
+ amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_INIT_FAIL, 0, 0);
+ return r;
+ }
}
/*get pf2vf msg info at it's earliest time*/
@@ -3945,25 +3959,27 @@ int amdgpu_device_init(struct amdgpu_device *adev,
}
}
- if (adev->is_atom_fw) {
- /* Initialize clocks */
- r = amdgpu_atomfirmware_get_clock_info(adev);
- if (r) {
- dev_err(adev->dev, "amdgpu_atomfirmware_get_clock_info failed\n");
- amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
- goto failed;
- }
- } else {
- /* Initialize clocks */
- r = amdgpu_atombios_get_clock_info(adev);
- if (r) {
- dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n");
- amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
- goto failed;
+ if (adev->bios) {
+ if (adev->is_atom_fw) {
+ /* Initialize clocks */
+ r = amdgpu_atomfirmware_get_clock_info(adev);
+ if (r) {
+ dev_err(adev->dev, "amdgpu_atomfirmware_get_clock_info failed\n");
+ amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
+ goto failed;
+ }
+ } else {
+ /* Initialize clocks */
+ r = amdgpu_atombios_get_clock_info(adev);
+ if (r) {
+ dev_err(adev->dev, "amdgpu_atombios_get_clock_info failed\n");
+ amdgpu_vf_error_put(adev, AMDGIM_ERROR_VF_ATOMBIOS_GET_CLOCK_FAIL, 0, 0);
+ goto failed;
+ }
+ /* init i2c buses */
+ if (!amdgpu_device_has_dc_support(adev))
+ amdgpu_atombios_i2c_init(adev);
}
- /* init i2c buses */
- if (!amdgpu_device_has_dc_support(adev))
- amdgpu_atombios_i2c_init(adev);
}
fence_driver_init:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 473eeac1f03b..d2d0d27f9053 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1696,7 +1696,7 @@ static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev)
uint32_t reserve_size = 0;
int ret;
- if (!amdgpu_sriov_vf(adev)) {
+ if (adev->bios && !amdgpu_sriov_vf(adev)) {
if (amdgpu_atomfirmware_mem_training_supported(adev))
mem_train_support = true;
else
@@ -1713,7 +1713,10 @@ static int amdgpu_ttm_reserve_tmr(struct amdgpu_device *adev)
if (adev->bios)
reserve_size =
amdgpu_atomfirmware_get_fw_reserved_fb_size(adev);
- if (!reserve_size)
+
+ if (!adev->bios && adev->ip_versions[GC_HWIP][0] == IP_VERSION(9, 4, 3))
+ reserve_size = max(reserve_size, (uint32_t)280 << 20);
+ else if (!reserve_size)
reserve_size = DISCOVERY_TMR_OFFSET;
if (mem_train_support) {
diff --git a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
index 1e8b2aaa48c1..be7823d82150 100644
--- a/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c
@@ -2010,34 +2010,49 @@ static int gmc_v9_0_sw_init(void *handle)
spin_lock_init(&adev->gmc.invalidate_lock);
- r = amdgpu_atomfirmware_get_vram_info(adev,
- &vram_width, &vram_type, &vram_vendor);
- if (amdgpu_sriov_vf(adev))
- /* For Vega10 SR-IOV, vram_width can't be read from ATOM as RAVEN,
- * and DF related registers is not readable, seems hardcord is the
- * only way to set the correct vram_width
- */
- adev->gmc.vram_width = 2048;
- else if (amdgpu_emu_mode != 1)
- adev->gmc.vram_width = vram_width;
+ if (!(adev->bios) || adev->gmc.is_app_apu) {
+ if (adev->flags & AMD_IS_APU) {
+ if (adev->gmc.is_app_apu) {
+ adev->gmc.vram_type = AMDGPU_VRAM_TYPE_HBM;
+ adev->gmc.vram_width = 128 * 64;
+ } else {
+ adev->gmc.vram_type = AMDGPU_VRAM_TYPE_DDR4;
+ adev->gmc.vram_width = 64 * 64;
+ }
+ } else {
+ adev->gmc.vram_type = AMDGPU_VRAM_TYPE_HBM;
+ adev->gmc.vram_width = 128 * 64;
+ }
+ } else {
+ r = amdgpu_atomfirmware_get_vram_info(adev,
+ &vram_width, &vram_type, &vram_vendor);
+ if (amdgpu_sriov_vf(adev))
+ /* For Vega10 SR-IOV, vram_width can't be read from ATOM as RAVEN,
+ * and DF related registers is not readable, seems hardcord is the
+ * only way to set the correct vram_width
+ */
+ adev->gmc.vram_width = 2048;
+ else if (amdgpu_emu_mode != 1)
+ adev->gmc.vram_width = vram_width;
- if (!adev->gmc.vram_width) {
- int chansize, numchan;
+ if (!adev->gmc.vram_width) {
+ int chansize, numchan;
- /* hbm memory channel size */
- if (adev->flags & AMD_IS_APU)
- chansize = 64;
- else
- chansize = 128;
- if (adev->df.funcs &&
- adev->df.funcs->get_hbm_channel_number) {
- numchan = adev->df.funcs->get_hbm_channel_number(adev);
- adev->gmc.vram_width = numchan * chansize;
+ /* hbm memory channel size */
+ if (adev->flags & AMD_IS_APU)
+ chansize = 64;
+ else
+ chansize = 128;
+ if (adev->df.funcs &&
+ adev->df.funcs->get_hbm_channel_number) {
+ numchan = adev->df.funcs->get_hbm_channel_number(adev);
+ adev->gmc.vram_width = numchan * chansize;
+ }
}
- }
- adev->gmc.vram_type = vram_type;
- adev->gmc.vram_vendor = vram_vendor;
+ adev->gmc.vram_type = vram_type;
+ adev->gmc.vram_vendor = vram_vendor;
+ }
switch (adev->ip_versions[GC_HWIP][0]) {
case IP_VERSION(9, 1, 0):
case IP_VERSION(9, 2, 2):
--
cgit v1.2.3