From 54313503f9a3c34a9e45aad7654976b7f50b9ae7 Mon Sep 17 00:00:00 2001
From: Dan Carpenter <dan.carpenter@oracle.com>
Date: Fri, 25 Sep 2015 14:36:55 +0300
Subject: drm/amdgpu: signedness bug in amdgpu_cs_parser_init()
The "i" variable should be signed or it leads to a crash in the error
handling code.
Fixes: 1d263474c441 ('drm/amdgpu: unwind properly in amdgpu_cs_parser_init()')
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 749420f1ea6f..cb3c274edb0a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -156,7 +156,8 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
uint64_t *chunk_array_user;
uint64_t *chunk_array;
struct amdgpu_fpriv *fpriv = p->filp->driver_priv;
- unsigned size, i;
+ unsigned size;
+ int i;
int ret;
if (cs->in.num_chunks == 0)
--
cgit
From 028423b0d8fa27c30a46da4af2a5cab230369de6 Mon Sep 17 00:00:00 2001
From: Arnd Bergmann <arnd@arndb.de>
Date: Wed, 7 Oct 2015 09:41:27 +0200
Subject: drm/amdgpu: fix 32-bit compiler warning
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
The new amdgpu driver passes a user space pointer in a 64-bit structure
member, which is the correct way to do it, but it attempts to
directly cast it to a __user pointer in the kernel, which causes
a warning in three places:
drm/amd/amdgpu/amdgpu_cs.c: In function 'amdgpu_cs_parser_init':
drm/amd/amdgpu/amdgpu_cs.c:180:21: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
chunk_array_user = (uint64_t __user *)(cs->in.chunks);
This changes all three to add an intermediate cast to 'unsigned long'
as other drivers do. This avoids the warning and works correctly on
both 32-bit and 64-bit architectures.
Fixes: e60b344f6c0eff ("drm/amdgpu: optimize amdgpu_parser_init")
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index cb3c274edb0a..fd16652aa277 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -177,7 +177,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
/* get chunks */
INIT_LIST_HEAD(&p->validated);
- chunk_array_user = (uint64_t __user *)(cs->in.chunks);
+ chunk_array_user = (uint64_t __user *)(unsigned long)(cs->in.chunks);
if (copy_from_user(chunk_array, chunk_array_user,
sizeof(uint64_t)*cs->in.num_chunks)) {
ret = -EFAULT;
@@ -197,7 +197,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
struct drm_amdgpu_cs_chunk user_chunk;
uint32_t __user *cdata;
- chunk_ptr = (void __user *)chunk_array[i];
+ chunk_ptr = (void __user *)(unsigned long)chunk_array[i];
if (copy_from_user(&user_chunk, chunk_ptr,
sizeof(struct drm_amdgpu_cs_chunk))) {
ret = -EFAULT;
@@ -208,7 +208,7 @@ int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
p->chunks[i].length_dw = user_chunk.length_dw;
size = p->chunks[i].length_dw;
- cdata = (void __user *)user_chunk.chunk_data;
+ cdata = (void __user *)(unsigned long)user_chunk.chunk_data;
p->chunks[i].user_ptr = cdata;
p->chunks[i].kdata = drm_malloc_ab(size, sizeof(uint32_t));
--
cgit
From c113ea1c4f4a7592f17e53d658873b83af0dbb4b Mon Sep 17 00:00:00 2001
From: Alex Deucher <alexander.deucher@amd.com>
Date: Thu, 8 Oct 2015 16:30:37 -0400
Subject: drm/amdgpu: rework sdma structures
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Rework the sdma structures in the driver to
consolidate all of the sdma info into a single
structure and allow for asics that may have
different numbers of sdma instances.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 22 +--
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c | 4 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c | 4 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 7 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c | 10 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c | 4 +-
drivers/gpu/drm/amd/amdgpu/cik_sdma.c | 130 ++++++++---------
drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c | 156 ++++++++++----------
drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c | 166 +++++++++++-----------
9 files changed, 245 insertions(+), 258 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 6647fb26ef25..afc984806c4c 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1708,7 +1708,7 @@ struct amdgpu_vce {
/*
* SDMA
*/
-struct amdgpu_sdma {
+struct amdgpu_sdma_instance {
/* SDMA firmware */
const struct firmware *fw;
uint32_t fw_version;
@@ -1718,6 +1718,13 @@ struct amdgpu_sdma {
bool burst_nop;
};
+struct amdgpu_sdma {
+ struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES];
+ struct amdgpu_irq_src trap_irq;
+ struct amdgpu_irq_src illegal_inst_irq;
+ int num_instances;
+};
+
/*
* Firmware
*/
@@ -2064,9 +2071,7 @@ struct amdgpu_device {
struct amdgpu_gfx gfx;
/* sdma */
- struct amdgpu_sdma sdma[AMDGPU_MAX_SDMA_INSTANCES];
- struct amdgpu_irq_src sdma_trap_irq;
- struct amdgpu_irq_src sdma_illegal_inst_irq;
+ struct amdgpu_sdma sdma;
/* uvd */
bool has_uvd;
@@ -2203,17 +2208,18 @@ static inline void amdgpu_ring_write(struct amdgpu_ring *ring, uint32_t v)
ring->ring_free_dw--;
}
-static inline struct amdgpu_sdma * amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
+static inline struct amdgpu_sdma_instance *
+amdgpu_get_sdma_instance(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
int i;
- for (i = 0; i < AMDGPU_MAX_SDMA_INSTANCES; i++)
- if (&adev->sdma[i].ring == ring)
+ for (i = 0; i < adev->sdma.num_instances; i++)
+ if (&adev->sdma.instance[i].ring == ring)
break;
if (i < AMDGPU_MAX_SDMA_INSTANCES)
- return &adev->sdma[i];
+ return &adev->sdma.instance[i];
else
return NULL;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
index dd2037bc0b4a..0e1376317683 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
@@ -649,12 +649,12 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
case KGD_ENGINE_SDMA1:
hdr = (const union amdgpu_firmware_header *)
- adev->sdma[0].fw->data;
+ adev->sdma.instance[0].fw->data;
break;
case KGD_ENGINE_SDMA2:
hdr = (const union amdgpu_firmware_header *)
- adev->sdma[1].fw->data;
+ adev->sdma.instance[1].fw->data;
break;
default:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
index dfd1d503bccf..79fa5c7de856 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
@@ -523,12 +523,12 @@ static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
case KGD_ENGINE_SDMA1:
hdr = (const union amdgpu_firmware_header *)
- adev->sdma[0].fw->data;
+ adev->sdma.instance[0].fw->data;
break;
case KGD_ENGINE_SDMA2:
hdr = (const union amdgpu_firmware_header *)
- adev->sdma[1].fw->data;
+ adev->sdma.instance[1].fw->data;
break;
default:
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 749420f1ea6f..29fc45ce64dd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -104,10 +104,11 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
}
break;
case AMDGPU_HW_IP_DMA:
- if (ring < 2) {
- *out_ring = &adev->sdma[ring].ring;
+ if (ring < adev->sdma.num_instances) {
+ *out_ring = &adev->sdma.instance[ring].ring;
} else {
- DRM_ERROR("only two SDMA rings are supported\n");
+ DRM_ERROR("only %d SDMA rings are supported\n",
+ adev->sdma.num_instances);
return -EINVAL;
}
break;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
index 78233225dd36..3f5f2d58ad94 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_kms.c
@@ -218,8 +218,8 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
break;
case AMDGPU_HW_IP_DMA:
type = AMD_IP_BLOCK_TYPE_SDMA;
- ring_mask = adev->sdma[0].ring.ready ? 1 : 0;
- ring_mask |= ((adev->sdma[1].ring.ready ? 1 : 0) << 1);
+ for (i = 0; i < adev->sdma.num_instances; i++)
+ ring_mask |= ((adev->sdma.instance[i].ring.ready ? 1 : 0) << i);
ib_start_alignment = AMDGPU_GPU_PAGE_SIZE;
ib_size_alignment = 1;
break;
@@ -341,10 +341,10 @@ static int amdgpu_info_ioctl(struct drm_device *dev, void *data, struct drm_file
fw_info.feature = 0;
break;
case AMDGPU_INFO_FW_SDMA:
- if (info->query_fw.index >= 2)
+ if (info->query_fw.index >= adev->sdma.num_instances)
return -EINVAL;
- fw_info.ver = adev->sdma[info->query_fw.index].fw_version;
- fw_info.feature = adev->sdma[info->query_fw.index].feature_version;
+ fw_info.ver = adev->sdma.instance[info->query_fw.index].fw_version;
+ fw_info.feature = adev->sdma.instance[info->query_fw.index].feature_version;
break;
default:
return -EINVAL;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
index 30dce235ddeb..b13a74b273a6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ring.c
@@ -540,8 +540,8 @@ static int amdgpu_debugfs_ring_info(struct seq_file *m, void *data)
static int amdgpu_gfx_index = offsetof(struct amdgpu_device, gfx.gfx_ring[0]);
static int cayman_cp1_index = offsetof(struct amdgpu_device, gfx.compute_ring[0]);
static int cayman_cp2_index = offsetof(struct amdgpu_device, gfx.compute_ring[1]);
-static int amdgpu_dma1_index = offsetof(struct amdgpu_device, sdma[0].ring);
-static int amdgpu_dma2_index = offsetof(struct amdgpu_device, sdma[1].ring);
+static int amdgpu_dma1_index = offsetof(struct amdgpu_device, sdma.instance[0].ring);
+static int amdgpu_dma2_index = offsetof(struct amdgpu_device, sdma.instance[1].ring);
static int r600_uvd_index = offsetof(struct amdgpu_device, uvd.ring);
static int si_vce1_index = offsetof(struct amdgpu_device, vce.ring[0]);
static int si_vce2_index = offsetof(struct amdgpu_device, vce.ring[1]);
diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
index 9ea9de457da3..814598e76c98 100644
--- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
+++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
@@ -96,7 +96,7 @@ static int cik_sdma_init_microcode(struct amdgpu_device *adev)
{
const char *chip_name;
char fw_name[30];
- int err, i;
+ int err = 0, i;
DRM_DEBUG("\n");
@@ -119,24 +119,24 @@ static int cik_sdma_init_microcode(struct amdgpu_device *adev)
default: BUG();
}
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
if (i == 0)
snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma.bin", chip_name);
else
snprintf(fw_name, sizeof(fw_name), "radeon/%s_sdma1.bin", chip_name);
- err = request_firmware(&adev->sdma[i].fw, fw_name, adev->dev);
+ err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev);
if (err)
goto out;
- err = amdgpu_ucode_validate(adev->sdma[i].fw);
+ err = amdgpu_ucode_validate(adev->sdma.instance[i].fw);
}
out:
if (err) {
printk(KERN_ERR
"cik_sdma: Failed to load firmware \"%s\"\n",
fw_name);
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
- release_firmware(adev->sdma[i].fw);
- adev->sdma[i].fw = NULL;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ release_firmware(adev->sdma.instance[i].fw);
+ adev->sdma.instance[i].fw = NULL;
}
}
return err;
@@ -168,7 +168,7 @@ static uint32_t cik_sdma_ring_get_rptr(struct amdgpu_ring *ring)
static uint32_t cik_sdma_ring_get_wptr(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
- u32 me = (ring == &adev->sdma[0].ring) ? 0 : 1;
+ u32 me = (ring == &adev->sdma.instance[0].ring) ? 0 : 1;
return (RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) & 0x3fffc) >> 2;
}
@@ -183,14 +183,14 @@ static uint32_t cik_sdma_ring_get_wptr(struct amdgpu_ring *ring)
static void cik_sdma_ring_set_wptr(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
- u32 me = (ring == &adev->sdma[0].ring) ? 0 : 1;
+ u32 me = (ring == &adev->sdma.instance[0].ring) ? 0 : 1;
WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], (ring->wptr << 2) & 0x3fffc);
}
static void cik_sdma_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
{
- struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring);
+ struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring);
int i;
for (i = 0; i < count; i++)
@@ -248,7 +248,7 @@ static void cik_sdma_ring_emit_hdp_flush(struct amdgpu_ring *ring)
SDMA_POLL_REG_MEM_EXTRA_FUNC(3)); /* == */
u32 ref_and_mask;
- if (ring == &ring->adev->sdma[0].ring)
+ if (ring == &ring->adev->sdma.instance[0].ring)
ref_and_mask = GPU_HDP_FLUSH_DONE__SDMA0_MASK;
else
ref_and_mask = GPU_HDP_FLUSH_DONE__SDMA1_MASK;
@@ -327,8 +327,8 @@ static bool cik_sdma_ring_emit_semaphore(struct amdgpu_ring *ring,
*/
static void cik_sdma_gfx_stop(struct amdgpu_device *adev)
{
- struct amdgpu_ring *sdma0 = &adev->sdma[0].ring;
- struct amdgpu_ring *sdma1 = &adev->sdma[1].ring;
+ struct amdgpu_ring *sdma0 = &adev->sdma.instance[0].ring;
+ struct amdgpu_ring *sdma1 = &adev->sdma.instance[1].ring;
u32 rb_cntl;
int i;
@@ -336,7 +336,7 @@ static void cik_sdma_gfx_stop(struct amdgpu_device *adev)
(adev->mman.buffer_funcs_ring == sdma1))
amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size);
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]);
rb_cntl &= ~SDMA0_GFX_RB_CNTL__RB_ENABLE_MASK;
WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl);
@@ -376,7 +376,7 @@ static void cik_sdma_enable(struct amdgpu_device *adev, bool enable)
cik_sdma_rlc_stop(adev);
}
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
me_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]);
if (enable)
me_cntl &= ~SDMA0_F32_CNTL__HALT_MASK;
@@ -402,8 +402,8 @@ static int cik_sdma_gfx_resume(struct amdgpu_device *adev)
u32 wb_offset;
int i, j, r;
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
- ring = &adev->sdma[i].ring;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ ring = &adev->sdma.instance[i].ring;
wb_offset = (ring->rptr_offs * 4);
mutex_lock(&adev->srbm_mutex);
@@ -502,26 +502,25 @@ static int cik_sdma_load_microcode(struct amdgpu_device *adev)
u32 fw_size;
int i, j;
- if (!adev->sdma[0].fw || !adev->sdma[1].fw)
- return -EINVAL;
-
/* halt the MEs */
cik_sdma_enable(adev, false);
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
- hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ if (!adev->sdma.instance[i].fw)
+ return -EINVAL;
+ hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
amdgpu_ucode_print_sdma_hdr(&hdr->header);
fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
- adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
- adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
- if (adev->sdma[i].feature_version >= 20)
- adev->sdma[i].burst_nop = true;
+ adev->sdma.instance[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
+ adev->sdma.instance[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
+ if (adev->sdma.instance[i].feature_version >= 20)
+ adev->sdma.instance[i].burst_nop = true;
fw_data = (const __le32 *)
- (adev->sdma[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
+ (adev->sdma.instance[i].fw->data + le32_to_cpu(hdr->header.ucode_array_offset_bytes));
WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0);
for (j = 0; j < fw_size; j++)
WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++));
- WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma[i].fw_version);
+ WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma.instance[i].fw_version);
}
return 0;
@@ -830,7 +829,7 @@ static void cik_sdma_vm_set_pte_pde(struct amdgpu_ib *ib,
*/
static void cik_sdma_vm_pad_ib(struct amdgpu_ib *ib)
{
- struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring);
+ struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ib->ring);
u32 pad_count;
int i;
@@ -934,6 +933,8 @@ static int cik_sdma_early_init(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+ adev->sdma.num_instances = SDMA_MAX_INSTANCE;
+
cik_sdma_set_ring_funcs(adev);
cik_sdma_set_irq_funcs(adev);
cik_sdma_set_buffer_funcs(adev);
@@ -946,7 +947,7 @@ static int cik_sdma_sw_init(void *handle)
{
struct amdgpu_ring *ring;
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
- int r;
+ int r, i;
r = cik_sdma_init_microcode(adev);
if (r) {
@@ -955,43 +956,33 @@ static int cik_sdma_sw_init(void *handle)
}
/* SDMA trap event */
- r = amdgpu_irq_add_id(adev, 224, &adev->sdma_trap_irq);
+ r = amdgpu_irq_add_id(adev, 224, &adev->sdma.trap_irq);
if (r)
return r;
/* SDMA Privileged inst */
- r = amdgpu_irq_add_id(adev, 241, &adev->sdma_illegal_inst_irq);
+ r = amdgpu_irq_add_id(adev, 241, &adev->sdma.illegal_inst_irq);
if (r)
return r;
/* SDMA Privileged inst */
- r = amdgpu_irq_add_id(adev, 247, &adev->sdma_illegal_inst_irq);
- if (r)
- return r;
-
- ring = &adev->sdma[0].ring;
- ring->ring_obj = NULL;
-
- ring = &adev->sdma[1].ring;
- ring->ring_obj = NULL;
-
- ring = &adev->sdma[0].ring;
- sprintf(ring->name, "sdma0");
- r = amdgpu_ring_init(adev, ring, 256 * 1024,
- SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0), 0xf,
- &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP0,
- AMDGPU_RING_TYPE_SDMA);
+ r = amdgpu_irq_add_id(adev, 247, &adev->sdma.illegal_inst_irq);
if (r)
return r;
- ring = &adev->sdma[1].ring;
- sprintf(ring->name, "sdma1");
- r = amdgpu_ring_init(adev, ring, 256 * 1024,
- SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0), 0xf,
- &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP1,
- AMDGPU_RING_TYPE_SDMA);
- if (r)
- return r;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ ring = &adev->sdma.instance[i].ring;
+ ring->ring_obj = NULL;
+ sprintf(ring->name, "sdma%d", i);
+ r = amdgpu_ring_init(adev, ring, 256 * 1024,
+ SDMA_PACKET(SDMA_OPCODE_NOP, 0, 0), 0xf,
+ &adev->sdma.trap_irq,
+ (i == 0) ?
+ AMDGPU_SDMA_IRQ_TRAP0 : AMDGPU_SDMA_IRQ_TRAP1,
+ AMDGPU_RING_TYPE_SDMA);
+ if (r)
+ return r;
+ }
return r;
}
@@ -999,9 +990,10 @@ static int cik_sdma_sw_init(void *handle)
static int cik_sdma_sw_fini(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+ int i;
- amdgpu_ring_fini(&adev->sdma[0].ring);
- amdgpu_ring_fini(&adev->sdma[1].ring);
+ for (i = 0; i < adev->sdma.num_instances; i++)
+ amdgpu_ring_fini(&adev->sdma.instance[i].ring);
return 0;
}
@@ -1078,7 +1070,7 @@ static void cik_sdma_print_status(void *handle)
dev_info(adev->dev, "CIK SDMA registers\n");
dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n",
RREG32(mmSRBM_STATUS2));
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n",
i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i]));
dev_info(adev->dev, " SDMA%d_ME_CNTL=0x%08X\n",
@@ -1223,7 +1215,7 @@ static int cik_sdma_process_trap_irq(struct amdgpu_device *adev,
case 0:
switch (queue_id) {
case 0:
- amdgpu_fence_process(&adev->sdma[0].ring);
+ amdgpu_fence_process(&adev->sdma.instance[0].ring);
break;
case 1:
/* XXX compute */
@@ -1236,7 +1228,7 @@ static int cik_sdma_process_trap_irq(struct amdgpu_device *adev,
case 1:
switch (queue_id) {
case 0:
- amdgpu_fence_process(&adev->sdma[1].ring);
+ amdgpu_fence_process(&adev->sdma.instance[1].ring);
break;
case 1:
/* XXX compute */
@@ -1334,8 +1326,10 @@ static const struct amdgpu_ring_funcs cik_sdma_ring_funcs = {
static void cik_sdma_set_ring_funcs(struct amdgpu_device *adev)
{
- adev->sdma[0].ring.funcs = &cik_sdma_ring_funcs;
- adev->sdma[1].ring.funcs = &cik_sdma_ring_funcs;
+ int i;
+
+ for (i = 0; i < adev->sdma.num_instances; i++)
+ adev->sdma.instance[i].ring.funcs = &cik_sdma_ring_funcs;
}
static const struct amdgpu_irq_src_funcs cik_sdma_trap_irq_funcs = {
@@ -1349,9 +1343,9 @@ static const struct amdgpu_irq_src_funcs cik_sdma_illegal_inst_irq_funcs = {
static void cik_sdma_set_irq_funcs(struct amdgpu_device *adev)
{
- adev->sdma_trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
- adev->sdma_trap_irq.funcs = &cik_sdma_trap_irq_funcs;
- adev->sdma_illegal_inst_irq.funcs = &cik_sdma_illegal_inst_irq_funcs;
+ adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
+ adev->sdma.trap_irq.funcs = &cik_sdma_trap_irq_funcs;
+ adev->sdma.illegal_inst_irq.funcs = &cik_sdma_illegal_inst_irq_funcs;
}
/**
@@ -1416,7 +1410,7 @@ static void cik_sdma_set_buffer_funcs(struct amdgpu_device *adev)
{
if (adev->mman.buffer_funcs == NULL) {
adev->mman.buffer_funcs = &cik_sdma_buffer_funcs;
- adev->mman.buffer_funcs_ring = &adev->sdma[0].ring;
+ adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
}
}
@@ -1431,7 +1425,7 @@ static void cik_sdma_set_vm_pte_funcs(struct amdgpu_device *adev)
{
if (adev->vm_manager.vm_pte_funcs == NULL) {
adev->vm_manager.vm_pte_funcs = &cik_sdma_vm_pte_funcs;
- adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring;
+ adev->vm_manager.vm_pte_funcs_ring = &adev->sdma.instance[0].ring;
adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
}
}
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
index 14e87234171a..f8b868c7c496 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
@@ -118,7 +118,7 @@ static int sdma_v2_4_init_microcode(struct amdgpu_device *adev)
{
const char *chip_name;
char fw_name[30];
- int err, i;
+ int err = 0, i;
struct amdgpu_firmware_info *info = NULL;
const struct common_firmware_header *header = NULL;
const struct sdma_firmware_header_v1_0 *hdr;
@@ -132,27 +132,27 @@ static int sdma_v2_4_init_microcode(struct amdgpu_device *adev)
default: BUG();
}
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
if (i == 0)
snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma.bin", chip_name);
else
snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma1.bin", chip_name);
- err = request_firmware(&adev->sdma[i].fw, fw_name, adev->dev);
+ err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev);
if (err)
goto out;
- err = amdgpu_ucode_validate(adev->sdma[i].fw);
+ err = amdgpu_ucode_validate(adev->sdma.instance[i].fw);
if (err)
goto out;
- hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data;
- adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
- adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
- if (adev->sdma[i].feature_version >= 20)
- adev->sdma[i].burst_nop = true;
+ hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
+ adev->sdma.instance[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
+ adev->sdma.instance[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
+ if (adev->sdma.instance[i].feature_version >= 20)
+ adev->sdma.instance[i].burst_nop = true;
if (adev->firmware.smu_load) {
info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i];
info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i;
- info->fw = adev->sdma[i].fw;
+ info->fw = adev->sdma.instance[i].fw;
header = (const struct common_firmware_header *)info->fw->data;
adev->firmware.fw_size +=
ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
@@ -164,9 +164,9 @@ out:
printk(KERN_ERR
"sdma_v2_4: Failed to load firmware \"%s\"\n",
fw_name);
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
- release_firmware(adev->sdma[i].fw);
- adev->sdma[i].fw = NULL;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ release_firmware(adev->sdma.instance[i].fw);
+ adev->sdma.instance[i].fw = NULL;
}
}
return err;
@@ -199,7 +199,7 @@ static uint32_t sdma_v2_4_ring_get_rptr(struct amdgpu_ring *ring)
static uint32_t sdma_v2_4_ring_get_wptr(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
- int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1;
+ int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1;
u32 wptr = RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) >> 2;
return wptr;
@@ -215,14 +215,14 @@ static uint32_t sdma_v2_4_ring_get_wptr(struct amdgpu_ring *ring)
static void sdma_v2_4_ring_set_wptr(struct amdgpu_ring *ring)
{
struct amdgpu_device *adev = ring->adev;
- int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1;
+ int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1;
WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2);
}
static void sdma_v2_4_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
{
- struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring);
+ struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring);
int i;
for (i = 0; i < count; i++)
@@ -284,7 +284,7 @@ static void sdma_v2_4_ring_emit_hdp_flush(struct amdgpu_ring *ring)
{
u32 ref_and_mask = 0;
- if (ring == &ring->adev->sdma[0].ring)
+ if (ring == &ring->adev->sdma.instance[0].ring)
ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA0, 1);
else
ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA1, 1);
@@ -368,8 +368,8 @@ static bool sdma_v2_4_ring_emit_semaphore(struct amdgpu_ring *ring,
*/
static void sdma_v2_4_gfx_stop(struct amdgpu_device *adev)
{
- struct amdgpu_ring *sdma0 = &adev->sdma[0].ring;
- struct amdgpu_ring *sdma1 = &adev->sdma[1].ring;
+ struct amdgpu_ring *sdma0 = &adev->sdma.instance[0].ring;
+ struct amdgpu_ring *sdma1 = &adev->sdma.instance[1].ring;
u32 rb_cntl, ib_cntl;
int i;
@@ -377,7 +377,7 @@ static void sdma_v2_4_gfx_stop(struct amdgpu_device *adev)
(adev->mman.buffer_funcs_ring == sdma1))
amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size);
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]);
rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0);
WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl);
@@ -419,7 +419,7 @@ static void sdma_v2_4_enable(struct amdgpu_device *adev, bool enable)
sdma_v2_4_rlc_stop(adev);
}
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
f32_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]);
if (enable)
f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 0);
@@ -445,8 +445,8 @@ static int sdma_v2_4_gfx_resume(struct amdgpu_device *adev)
u32 wb_offset;
int i, j, r;
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
- ring = &adev->sdma[i].ring;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ ring = &adev->sdma.instance[i].ring;
wb_offset = (ring->rptr_offs * 4);
mutex_lock(&adev->srbm_mutex);
@@ -545,29 +545,23 @@ static int sdma_v2_4_load_microcode(struct amdgpu_device *adev)
const __le32 *fw_data;
u32 fw_size;
int i, j;
- bool smc_loads_fw = false; /* XXX fix me */
-
- if (!adev->sdma[0].fw || !adev->sdma[1].fw)
- return -EINVAL;
/* halt the MEs */
sdma_v2_4_enable(adev, false);
- if (smc_loads_fw) {
- /* XXX query SMC for fw load complete */
- } else {
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
- hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data;
- amdgpu_ucode_print_sdma_hdr(&hdr->header);
- fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
- fw_data = (const __le32 *)
- (adev->sdma[i].fw->data +
- le32_to_cpu(hdr->header.ucode_array_offset_bytes));
- WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0);
- for (j = 0; j < fw_size; j++)
- WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++));
- WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma[i].fw_version);
- }
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ if (!adev->sdma.instance[i].fw)
+ return -EINVAL;
+ hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
+ amdgpu_ucode_print_sdma_hdr(&hdr->header);
+ fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
+ fw_data = (const __le32 *)
+ (adev->sdma.instance[i].fw->data +
+ le32_to_cpu(hdr->header.ucode_array_offset_bytes));
+ WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0);
+ for (j = 0; j < fw_size; j++)
+ WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++));
+ WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma.instance[i].fw_version);
}
return 0;
@@ -894,7 +888,7 @@ static void sdma_v2_4_vm_set_pte_pde(struct amdgpu_ib *ib,
*/
static void sdma_v2_4_vm_pad_ib(struct amdgpu_ib *ib)
{
- struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring);
+ struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ib->ring);
u32 pad_count;
int i;
@@ -952,6 +946,8 @@ static int sdma_v2_4_early_init(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+ adev->sdma.num_instances = SDMA_MAX_INSTANCE;
+
sdma_v2_4_set_ring_funcs(adev);
sdma_v2_4_set_buffer_funcs(adev);
sdma_v2_4_set_vm_pte_funcs(adev);
@@ -963,21 +959,21 @@ static int sdma_v2_4_early_init(void *handle)
static int sdma_v2_4_sw_init(void *handle)
{
struct amdgpu_ring *ring;
- int r;
+ int r, i;
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
/* SDMA trap event */
- r = amdgpu_irq_add_id(adev, 224, &adev->sdma_trap_irq);
+ r = amdgpu_irq_add_id(adev, 224, &adev->sdma.trap_irq);
if (r)
return r;
/* SDMA Privileged inst */
- r = amdgpu_irq_add_id(adev, 241, &adev->sdma_illegal_inst_irq);
+ r = amdgpu_irq_add_id(adev, 241, &adev->sdma.illegal_inst_irq);
if (r)
return r;
/* SDMA Privileged inst */
- r = amdgpu_irq_add_id(adev, 247, &adev->sdma_illegal_inst_irq);
+ r = amdgpu_irq_add_id(adev, 247, &adev->sdma.illegal_inst_irq);
if (r)
return r;
@@ -987,31 +983,20 @@ static int sdma_v2_4_sw_init(void *handle)
return r;
}
- ring = &adev->sdma[0].ring;
- ring->ring_obj = NULL;
- ring->use_doorbell = false;
-
- ring = &adev->sdma[1].ring;
- ring->ring_obj = NULL;
- ring->use_doorbell = false;
-
- ring = &adev->sdma[0].ring;
- sprintf(ring->name, "sdma0");
- r = amdgpu_ring_init(adev, ring, 256 * 1024,
- SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf,
- &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP0,
- AMDGPU_RING_TYPE_SDMA);
- if (r)
- return r;
-
- ring = &adev->sdma[1].ring;
- sprintf(ring->name, "sdma1");
- r = amdgpu_ring_init(adev, ring, 256 * 1024,
- SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf,
- &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP1,
- AMDGPU_RING_TYPE_SDMA);
- if (r)
- return r;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ ring = &adev->sdma.instance[i].ring;
+ ring->ring_obj = NULL;
+ ring->use_doorbell = false;
+ sprintf(ring->name, "sdma%d", i);
+ r = amdgpu_ring_init(adev, ring, 256 * 1024,
+ SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf,
+ &adev->sdma.trap_irq,
+ (i == 0) ?
+ AMDGPU_SDMA_IRQ_TRAP0 : AMDGPU_SDMA_IRQ_TRAP1,
+ AMDGPU_RING_TYPE_SDMA);
+ if (r)
+ return r;
+ }
return r;
}
@@ -1019,9 +1004,10 @@ static int sdma_v2_4_sw_init(void *handle)
static int sdma_v2_4_sw_fini(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+ int i;
- amdgpu_ring_fini(&adev->sdma[0].ring);
- amdgpu_ring_fini(&adev->sdma[1].ring);
+ for (i = 0; i < adev->sdma.num_instances; i++)
+ amdgpu_ring_fini(&adev->sdma.instance[i].ring);
return 0;
}
@@ -1100,7 +1086,7 @@ static void sdma_v2_4_print_status(void *handle)
dev_info(adev->dev, "VI SDMA registers\n");
dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n",
RREG32(mmSRBM_STATUS2));
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n",
i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i]));
dev_info(adev->dev, " SDMA%d_F32_CNTL=0x%08X\n",
@@ -1243,7 +1229,7 @@ static int sdma_v2_4_process_trap_irq(struct amdgpu_device *adev,
case 0:
switch (queue_id) {
case 0:
- amdgpu_fence_process(&adev->sdma[0].ring);
+ amdgpu_fence_process(&adev->sdma.instance[0].ring);
break;
case 1:
/* XXX compute */
@@ -1256,7 +1242,7 @@ static int sdma_v2_4_process_trap_irq(struct amdgpu_device *adev,
case 1:
switch (queue_id) {
case 0:
- amdgpu_fence_process(&adev->sdma[1].ring);
+ amdgpu_fence_process(&adev->sdma.instance[1].ring);
break;
case 1:
/* XXX compute */
@@ -1345,8 +1331,10 @@ static const struct amdgpu_ring_funcs sdma_v2_4_ring_funcs = {
static void sdma_v2_4_set_ring_funcs(struct amdgpu_device *adev)
{
- adev->sdma[0].ring.funcs = &sdma_v2_4_ring_funcs;
- adev->sdma[1].ring.funcs = &sdma_v2_4_ring_funcs;
+ int i;
+
+ for (i = 0; i < adev->sdma.num_instances; i++)
+ adev->sdma.instance[i].ring.funcs = &sdma_v2_4_ring_funcs;
}
static const struct amdgpu_irq_src_funcs sdma_v2_4_trap_irq_funcs = {
@@ -1360,9 +1348,9 @@ static const struct amdgpu_irq_src_funcs sdma_v2_4_illegal_inst_irq_funcs = {
static void sdma_v2_4_set_irq_funcs(struct amdgpu_device *adev)
{
- adev->sdma_trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
- adev->sdma_trap_irq.funcs = &sdma_v2_4_trap_irq_funcs;
- adev->sdma_illegal_inst_irq.funcs = &sdma_v2_4_illegal_inst_irq_funcs;
+ adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
+ adev->sdma.trap_irq.funcs = &sdma_v2_4_trap_irq_funcs;
+ adev->sdma.illegal_inst_irq.funcs = &sdma_v2_4_illegal_inst_irq_funcs;
}
/**
@@ -1428,7 +1416,7 @@ static void sdma_v2_4_set_buffer_funcs(struct amdgpu_device *adev)
{
if (adev->mman.buffer_funcs == NULL) {
adev->mman.buffer_funcs = &sdma_v2_4_buffer_funcs;
- adev->mman.buffer_funcs_ring = &adev->sdma[0].ring;
+ adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
}
}
@@ -1443,7 +1431,7 @@ static void sdma_v2_4_set_vm_pte_funcs(struct amdgpu_device *adev)
{
if (adev->vm_manager.vm_pte_funcs == NULL) {
adev->vm_manager.vm_pte_funcs = &sdma_v2_4_vm_pte_funcs;
- adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring;
+ adev->vm_manager.vm_pte_funcs_ring = &adev->sdma.instance[0].ring;
adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
}
}
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
index 9bfe92df15f7..670555a45da9 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
@@ -184,7 +184,7 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev)
{
const char *chip_name;
char fw_name[30];
- int err, i;
+ int err = 0, i;
struct amdgpu_firmware_info *info = NULL;
const struct common_firmware_header *header = NULL;
const struct sdma_firmware_header_v1_0 *hdr;
@@ -204,27 +204,27 @@ static int sdma_v3_0_init_microcode(struct amdgpu_device *adev)
default: BUG();
}
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
if (i == 0)
snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma.bin", chip_name);
else
snprintf(fw_name, sizeof(fw_name), "amdgpu/%s_sdma1.bin", chip_name);
- err = request_firmware(&adev->sdma[i].fw, fw_name, adev->dev);
+ err = request_firmware(&adev->sdma.instance[i].fw, fw_name, adev->dev);
if (err)
goto out;
- err = amdgpu_ucode_validate(adev->sdma[i].fw);
+ err = amdgpu_ucode_validate(adev->sdma.instance[i].fw);
if (err)
goto out;
- hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data;
- adev->sdma[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
- adev->sdma[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
- if (adev->sdma[i].feature_version >= 20)
- adev->sdma[i].burst_nop = true;
+ hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
+ adev->sdma.instance[i].fw_version = le32_to_cpu(hdr->header.ucode_version);
+ adev->sdma.instance[i].feature_version = le32_to_cpu(hdr->ucode_feature_version);
+ if (adev->sdma.instance[i].feature_version >= 20)
+ adev->sdma.instance[i].burst_nop = true;
if (adev->firmware.smu_load) {
info = &adev->firmware.ucode[AMDGPU_UCODE_ID_SDMA0 + i];
info->ucode_id = AMDGPU_UCODE_ID_SDMA0 + i;
- info->fw = adev->sdma[i].fw;
+ info->fw = adev->sdma.instance[i].fw;
header = (const struct common_firmware_header *)info->fw->data;
adev->firmware.fw_size +=
ALIGN(le32_to_cpu(header->ucode_size_bytes), PAGE_SIZE);
@@ -235,9 +235,9 @@ out:
printk(KERN_ERR
"sdma_v3_0: Failed to load firmware \"%s\"\n",
fw_name);
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
- release_firmware(adev->sdma[i].fw);
- adev->sdma[i].fw = NULL;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ release_firmware(adev->sdma.instance[i].fw);
+ adev->sdma.instance[i].fw = NULL;
}
}
return err;
@@ -276,7 +276,7 @@ static uint32_t sdma_v3_0_ring_get_wptr(struct amdgpu_ring *ring)
/* XXX check if swapping is necessary on BE */
wptr = ring->adev->wb.wb[ring->wptr_offs] >> 2;
} else {
- int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1;
+ int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1;
wptr = RREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me]) >> 2;
}
@@ -300,7 +300,7 @@ static void sdma_v3_0_ring_set_wptr(struct amdgpu_ring *ring)
adev->wb.wb[ring->wptr_offs] = ring->wptr << 2;
WDOORBELL32(ring->doorbell_index, ring->wptr << 2);
} else {
- int me = (ring == &ring->adev->sdma[0].ring) ? 0 : 1;
+ int me = (ring == &ring->adev->sdma.instance[0].ring) ? 0 : 1;
WREG32(mmSDMA0_GFX_RB_WPTR + sdma_offsets[me], ring->wptr << 2);
}
@@ -308,7 +308,7 @@ static void sdma_v3_0_ring_set_wptr(struct amdgpu_ring *ring)
static void sdma_v3_0_ring_insert_nop(struct amdgpu_ring *ring, uint32_t count)
{
- struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ring);
+ struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ring);
int i;
for (i = 0; i < count; i++)
@@ -369,7 +369,7 @@ static void sdma_v3_0_ring_emit_hdp_flush(struct amdgpu_ring *ring)
{
u32 ref_and_mask = 0;
- if (ring == &ring->adev->sdma[0].ring)
+ if (ring == &ring->adev->sdma.instance[0].ring)
ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA0, 1);
else
ref_and_mask = REG_SET_FIELD(ref_and_mask, GPU_HDP_FLUSH_DONE, SDMA1, 1);
@@ -454,8 +454,8 @@ static bool sdma_v3_0_ring_emit_semaphore(struct amdgpu_ring *ring,
*/
static void sdma_v3_0_gfx_stop(struct amdgpu_device *adev)
{
- struct amdgpu_ring *sdma0 = &adev->sdma[0].ring;
- struct amdgpu_ring *sdma1 = &adev->sdma[1].ring;
+ struct amdgpu_ring *sdma0 = &adev->sdma.instance[0].ring;
+ struct amdgpu_ring *sdma1 = &adev->sdma.instance[1].ring;
u32 rb_cntl, ib_cntl;
int i;
@@ -463,7 +463,7 @@ static void sdma_v3_0_gfx_stop(struct amdgpu_device *adev)
(adev->mman.buffer_funcs_ring == sdma1))
amdgpu_ttm_set_active_vram_size(adev, adev->mc.visible_vram_size);
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
rb_cntl = RREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i]);
rb_cntl = REG_SET_FIELD(rb_cntl, SDMA0_GFX_RB_CNTL, RB_ENABLE, 0);
WREG32(mmSDMA0_GFX_RB_CNTL + sdma_offsets[i], rb_cntl);
@@ -500,7 +500,7 @@ static void sdma_v3_0_ctx_switch_enable(struct amdgpu_device *adev, bool enable)
u32 f32_cntl;
int i;
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
f32_cntl = RREG32(mmSDMA0_CNTL + sdma_offsets[i]);
if (enable)
f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_CNTL,
@@ -530,7 +530,7 @@ static void sdma_v3_0_enable(struct amdgpu_device *adev, bool enable)
sdma_v3_0_rlc_stop(adev);
}
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
f32_cntl = RREG32(mmSDMA0_F32_CNTL + sdma_offsets[i]);
if (enable)
f32_cntl = REG_SET_FIELD(f32_cntl, SDMA0_F32_CNTL, HALT, 0);
@@ -557,8 +557,8 @@ static int sdma_v3_0_gfx_resume(struct amdgpu_device *adev)
u32 doorbell;
int i, j, r;
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
- ring = &adev->sdma[i].ring;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ ring = &adev->sdma.instance[i].ring;
wb_offset = (ring->rptr_offs * 4);
mutex_lock(&adev->srbm_mutex);
@@ -669,23 +669,22 @@ static int sdma_v3_0_load_microcode(struct amdgpu_device *adev)
u32 fw_size;
int i, j;
- if (!adev->sdma[0].fw || !adev->sdma[1].fw)
- return -EINVAL;
-
/* halt the MEs */
sdma_v3_0_enable(adev, false);
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
- hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma[i].fw->data;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ if (!adev->sdma.instance[i].fw)
+ return -EINVAL;
+ hdr = (const struct sdma_firmware_header_v1_0 *)adev->sdma.instance[i].fw->data;
amdgpu_ucode_print_sdma_hdr(&hdr->header);
fw_size = le32_to_cpu(hdr->header.ucode_size_bytes) / 4;
fw_data = (const __le32 *)
- (adev->sdma[i].fw->data +
+ (adev->sdma.instance[i].fw->data +
le32_to_cpu(hdr->header.ucode_array_offset_bytes));
WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], 0);
for (j = 0; j < fw_size; j++)
WREG32(mmSDMA0_UCODE_DATA + sdma_offsets[i], le32_to_cpup(fw_data++));
- WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma[i].fw_version);
+ WREG32(mmSDMA0_UCODE_ADDR + sdma_offsets[i], adev->sdma.instance[i].fw_version);
}
return 0;
@@ -701,21 +700,21 @@ static int sdma_v3_0_load_microcode(struct amdgpu_device *adev)
*/
static int sdma_v3_0_start(struct amdgpu_device *adev)
{
- int r;
+ int r, i;
if (!adev->firmware.smu_load) {
r = sdma_v3_0_load_microcode(adev);
if (r)
return r;
} else {
- r = adev->smu.smumgr_funcs->check_fw_load_finish(adev,
- AMDGPU_UCODE_ID_SDMA0);
- if (r)
- return -EINVAL;
- r = adev->smu.smumgr_funcs->check_fw_load_finish(adev,
- AMDGPU_UCODE_ID_SDMA1);
- if (r)
- return -EINVAL;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ r = adev->smu.smumgr_funcs->check_fw_load_finish(adev,
+ (i == 0) ?
+ AMDGPU_UCODE_ID_SDMA0 :
+ AMDGPU_UCODE_ID_SDMA1);
+ if (r)
+ return -EINVAL;
+ }
}
/* unhalt the MEs */
@@ -1013,7 +1012,7 @@ static void sdma_v3_0_vm_set_pte_pde(struct amdgpu_ib *ib,
*/
static void sdma_v3_0_vm_pad_ib(struct amdgpu_ib *ib)
{
- struct amdgpu_sdma *sdma = amdgpu_get_sdma_instance(ib->ring);
+ struct amdgpu_sdma_instance *sdma = amdgpu_get_sdma_instance(ib->ring);
u32 pad_count;
int i;
@@ -1071,6 +1070,12 @@ static int sdma_v3_0_early_init(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+ switch (adev->asic_type) {
+ default:
+ adev->sdma.num_instances = SDMA_MAX_INSTANCE;
+ break;
+ }
+
sdma_v3_0_set_ring_funcs(adev);
sdma_v3_0_set_buffer_funcs(adev);
sdma_v3_0_set_vm_pte_funcs(adev);
@@ -1082,21 +1087,21 @@ static int sdma_v3_0_early_init(void *handle)
static int sdma_v3_0_sw_init(void *handle)
{
struct amdgpu_ring *ring;
- int r;
+ int r, i;
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
/* SDMA trap event */
- r = amdgpu_irq_add_id(adev, 224, &adev->sdma_trap_irq);
+ r = amdgpu_irq_add_id(adev, 224, &adev->sdma.trap_irq);
if (r)
return r;
/* SDMA Privileged inst */
- r = amdgpu_irq_add_id(adev, 241, &adev->sdma_illegal_inst_irq);
+ r = amdgpu_irq_add_id(adev, 241, &adev->sdma.illegal_inst_irq);
if (r)
return r;
/* SDMA Privileged inst */
- r = amdgpu_irq_add_id(adev, 247, &adev->sdma_illegal_inst_irq);
+ r = amdgpu_irq_add_id(adev, 247, &adev->sdma.illegal_inst_irq);
if (r)
return r;
@@ -1106,33 +1111,23 @@ static int sdma_v3_0_sw_init(void *handle)
return r;
}
- ring = &adev->sdma[0].ring;
- ring->ring_obj = NULL;
- ring->use_doorbell = true;
- ring->doorbell_index = AMDGPU_DOORBELL_sDMA_ENGINE0;
-
- ring = &adev->sdma[1].ring;
- ring->ring_obj = NULL;
- ring->use_doorbell = true;
- ring->doorbell_index = AMDGPU_DOORBELL_sDMA_ENGINE1;
-
- ring = &adev->sdma[0].ring;
- sprintf(ring->name, "sdma0");
- r = amdgpu_ring_init(adev, ring, 256 * 1024,
- SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf,
- &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP0,
- AMDGPU_RING_TYPE_SDMA);
- if (r)
- return r;
-
- ring = &adev->sdma[1].ring;
- sprintf(ring->name, "sdma1");
- r = amdgpu_ring_init(adev, ring, 256 * 1024,
- SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf,
- &adev->sdma_trap_irq, AMDGPU_SDMA_IRQ_TRAP1,
- AMDGPU_RING_TYPE_SDMA);
- if (r)
- return r;
+ for (i = 0; i < adev->sdma.num_instances; i++) {
+ ring = &adev->sdma.instance[i].ring;
+ ring->ring_obj = NULL;
+ ring->use_doorbell = true;
+ ring->doorbell_index = (i == 0) ?
+ AMDGPU_DOORBELL_sDMA_ENGINE0 : AMDGPU_DOORBELL_sDMA_ENGINE1;
+
+ sprintf(ring->name, "sdma%d", i);
+ r = amdgpu_ring_init(adev, ring, 256 * 1024,
+ SDMA_PKT_NOP_HEADER_OP(SDMA_OP_NOP), 0xf,
+ &adev->sdma.trap_irq,
+ (i == 0) ?
+ AMDGPU_SDMA_IRQ_TRAP0 : AMDGPU_SDMA_IRQ_TRAP1,
+ AMDGPU_RING_TYPE_SDMA);
+ if (r)
+ return r;
+ }
return r;
}
@@ -1140,9 +1135,10 @@ static int sdma_v3_0_sw_init(void *handle)
static int sdma_v3_0_sw_fini(void *handle)
{
struct amdgpu_device *adev = (struct amdgpu_device *)handle;
+ int i;
- amdgpu_ring_fini(&adev->sdma[0].ring);
- amdgpu_ring_fini(&adev->sdma[1].ring);
+ for (i = 0; i < adev->sdma.num_instances; i++)
+ amdgpu_ring_fini(&adev->sdma.instance[i].ring);
return 0;
}
@@ -1222,7 +1218,7 @@ static void sdma_v3_0_print_status(void *handle)
dev_info(adev->dev, "VI SDMA registers\n");
dev_info(adev->dev, " SRBM_STATUS2=0x%08X\n",
RREG32(mmSRBM_STATUS2));
- for (i = 0; i < SDMA_MAX_INSTANCE; i++) {
+ for (i = 0; i < adev->sdma.num_instances; i++) {
dev_info(adev->dev, " SDMA%d_STATUS_REG=0x%08X\n",
i, RREG32(mmSDMA0_STATUS_REG + sdma_offsets[i]));
dev_info(adev->dev, " SDMA%d_F32_CNTL=0x%08X\n",
@@ -1367,7 +1363,7 @@ static int sdma_v3_0_process_trap_irq(struct amdgpu_device *adev,
case 0:
switch (queue_id) {
case 0:
- amdgpu_fence_process(&adev->sdma[0].ring);
+ amdgpu_fence_process(&adev->sdma.instance[0].ring);
break;
case 1:
/* XXX compute */
@@ -1380,7 +1376,7 @@ static int sdma_v3_0_process_trap_irq(struct amdgpu_device *adev,
case 1:
switch (queue_id) {
case 0:
- amdgpu_fence_process(&adev->sdma[1].ring);
+ amdgpu_fence_process(&adev->sdma.instance[1].ring);
break;
case 1:
/* XXX compute */
@@ -1468,8 +1464,10 @@ static const struct amdgpu_ring_funcs sdma_v3_0_ring_funcs = {
static void sdma_v3_0_set_ring_funcs(struct amdgpu_device *adev)
{
- adev->sdma[0].ring.funcs = &sdma_v3_0_ring_funcs;
- adev->sdma[1].ring.funcs = &sdma_v3_0_ring_funcs;
+ int i;
+
+ for (i = 0; i < adev->sdma.num_instances; i++)
+ adev->sdma.instance[i].ring.funcs = &sdma_v3_0_ring_funcs;
}
static const struct amdgpu_irq_src_funcs sdma_v3_0_trap_irq_funcs = {
@@ -1483,9 +1481,9 @@ static const struct amdgpu_irq_src_funcs sdma_v3_0_illegal_inst_irq_funcs = {
static void sdma_v3_0_set_irq_funcs(struct amdgpu_device *adev)
{
- adev->sdma_trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
- adev->sdma_trap_irq.funcs = &sdma_v3_0_trap_irq_funcs;
- adev->sdma_illegal_inst_irq.funcs = &sdma_v3_0_illegal_inst_irq_funcs;
+ adev->sdma.trap_irq.num_types = AMDGPU_SDMA_IRQ_LAST;
+ adev->sdma.trap_irq.funcs = &sdma_v3_0_trap_irq_funcs;
+ adev->sdma.illegal_inst_irq.funcs = &sdma_v3_0_illegal_inst_irq_funcs;
}
/**
@@ -1551,7 +1549,7 @@ static void sdma_v3_0_set_buffer_funcs(struct amdgpu_device *adev)
{
if (adev->mman.buffer_funcs == NULL) {
adev->mman.buffer_funcs = &sdma_v3_0_buffer_funcs;
- adev->mman.buffer_funcs_ring = &adev->sdma[0].ring;
+ adev->mman.buffer_funcs_ring = &adev->sdma.instance[0].ring;
}
}
@@ -1566,7 +1564,7 @@ static void sdma_v3_0_set_vm_pte_funcs(struct amdgpu_device *adev)
{
if (adev->vm_manager.vm_pte_funcs == NULL) {
adev->vm_manager.vm_pte_funcs = &sdma_v3_0_vm_pte_funcs;
- adev->vm_manager.vm_pte_funcs_ring = &adev->sdma[0].ring;
+ adev->vm_manager.vm_pte_funcs_ring = &adev->sdma.instance[0].ring;
adev->vm_manager.vm_pte_funcs_ring->is_pte_ring = true;
}
}
--
cgit
From b495bd3a54e732e846bfde49ba2ea93b62b1de91 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Thu, 10 Sep 2015 14:00:35 +0200
Subject: drm/amdgpu: add option to clear VM page tables after every submit
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
This makes it much easier to find when userspace misses to send some buffers.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 17 ++++++++++++++++-
drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 4 ++++
3 files changed, 21 insertions(+), 1 deletion(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index a3dbbd1fc060..736eb4560fd5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -80,6 +80,7 @@ extern int amdgpu_deep_color;
extern int amdgpu_vm_size;
extern int amdgpu_vm_block_size;
extern int amdgpu_vm_fault_stop;
+extern int amdgpu_vm_debug;
extern int amdgpu_enable_scheduler;
extern int amdgpu_sched_jobs;
extern int amdgpu_sched_hw_submission;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 29fc45ce64dd..25012c790f8f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -567,9 +567,24 @@ static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
if (r)
return r;
}
+
+ }
+
+ r = amdgpu_vm_clear_invalids(adev, vm, &p->ibs[0].sync);
+
+ if (amdgpu_vm_debug && p->bo_list) {
+ /* Invalidate all BOs to test for userspace bugs */
+ for (i = 0; i < p->bo_list->num_entries; i++) {
+ /* ignore duplicates */
+ bo = p->bo_list->array[i].robj;
+ if (!bo)
+ continue;
+
+ amdgpu_vm_bo_invalidate(adev, bo);
+ }
}
- return amdgpu_vm_clear_invalids(adev, vm, &p->ibs[0].sync);
+ return r;
}
static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
index bec091632194..6134b1ece17f 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
@@ -76,6 +76,7 @@ int amdgpu_deep_color = 0;
int amdgpu_vm_size = 8;
int amdgpu_vm_block_size = -1;
int amdgpu_vm_fault_stop = 0;
+int amdgpu_vm_debug = 0;
int amdgpu_exp_hw_support = 0;
int amdgpu_enable_scheduler = 1;
int amdgpu_sched_jobs = 16;
@@ -145,6 +146,9 @@ module_param_named(vm_block_size, amdgpu_vm_block_size, int, 0444);
MODULE_PARM_DESC(vm_fault_stop, "Stop on VM fault (0 = never (default), 1 = print first, 2 = always)");
module_param_named(vm_fault_stop, amdgpu_vm_fault_stop, int, 0444);
+MODULE_PARM_DESC(vm_debug, "Debug VM handling (0 = disabled (default), 1 = enabled)");
+module_param_named(vm_debug, amdgpu_vm_debug, int, 0644);
+
MODULE_PARM_DESC(exp_hw_support, "experimental hw support (1 = enable, 0 = disable (default))");
module_param_named(exp_hw_support, amdgpu_exp_hw_support, int, 0444);
--
cgit
From 0c418f10104d4aa1d6b83698790898dc9ef1c12d Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Tue, 1 Sep 2015 15:13:53 +0200
Subject: drm/amdgpu: remove the exclusive lock
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Finally getting rid of it.
Signed-off-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 -
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 10 ++--------
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 4 ----
drivers/gpu/drm/amd/amdgpu/amdgpu_display.c | 7 +------
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 21 +++++----------------
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 7 -------
6 files changed, 8 insertions(+), 42 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 911d67032d57..39d96088f7b4 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1955,7 +1955,6 @@ struct amdgpu_device {
struct device *dev;
struct drm_device *ddev;
struct pci_dev *pdev;
- struct rw_semaphore exclusive_lock;
/* ASIC */
enum amd_asic_type asic_type;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 25012c790f8f..5fdc0394561e 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -831,11 +831,8 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
bool reserved_buffers = false;
int i, r;
- down_read(&adev->exclusive_lock);
- if (!adev->accel_working) {
- up_read(&adev->exclusive_lock);
+ if (!adev->accel_working)
return -EBUSY;
- }
parser = amdgpu_cs_parser_create(adev, filp, NULL, NULL, 0);
if (!parser)
@@ -843,8 +840,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
r = amdgpu_cs_parser_init(parser, data);
if (r) {
DRM_ERROR("Failed to initialize parser !\n");
- kfree(parser);
- up_read(&adev->exclusive_lock);
+ amdgpu_cs_parser_fini(parser, r, false);
r = amdgpu_cs_handle_lockup(adev, r);
return r;
}
@@ -915,14 +911,12 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
mutex_unlock(&job->job_lock);
amdgpu_cs_parser_fini_late(parser);
- up_read(&adev->exclusive_lock);
return 0;
}
cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence;
out:
amdgpu_cs_parser_fini(parser, r, reserved_buffers);
- up_read(&adev->exclusive_lock);
r = amdgpu_cs_handle_lockup(adev, r);
return r;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index 28c9ee334dd7..19e185ec3707 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -1418,7 +1418,6 @@ int amdgpu_device_init(struct amdgpu_device *adev,
mutex_init(&adev->gfx.gpu_clock_mutex);
mutex_init(&adev->srbm_mutex);
mutex_init(&adev->grbm_idx_mutex);
- init_rwsem(&adev->exclusive_lock);
mutex_init(&adev->mn_lock);
hash_init(adev->mn_hash);
@@ -1814,8 +1813,6 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
int i, r;
int resched;
- down_write(&adev->exclusive_lock);
-
atomic_inc(&adev->gpu_reset_counter);
/* block TTM */
@@ -1879,7 +1876,6 @@ retry:
dev_info(adev->dev, "GPU reset failed\n");
}
- up_write(&adev->exclusive_lock);
return r;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
index 9b34a3410c32..fdf7525cf120 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_display.c
@@ -47,11 +47,8 @@ static void amdgpu_flip_wait_fence(struct amdgpu_device *adev,
fence = to_amdgpu_fence(*f);
if (fence) {
r = fence_wait(&fence->base, false);
- if (r == -EDEADLK) {
- up_read(&adev->exclusive_lock);
+ if (r == -EDEADLK)
r = amdgpu_gpu_reset(adev);
- down_read(&adev->exclusive_lock);
- }
} else
r = fence_wait(*f, false);
@@ -77,7 +74,6 @@ static void amdgpu_flip_work_func(struct work_struct *__work)
unsigned long flags;
unsigned i;
- down_read(&adev->exclusive_lock);
amdgpu_flip_wait_fence(adev, &work->excl);
for (i = 0; i < work->shared_count; ++i)
amdgpu_flip_wait_fence(adev, &work->shared[i]);
@@ -93,7 +89,6 @@ static void amdgpu_flip_work_func(struct work_struct *__work)
amdgpuCrtc->pflip_status = AMDGPU_FLIP_SUBMITTED;
spin_unlock_irqrestore(&crtc->dev->event_lock, flags);
- up_read(&adev->exclusive_lock);
}
/*
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
index 4010aa6b4e53..1fadc15e64ae 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c
@@ -260,16 +260,8 @@ static void amdgpu_fence_check_lockup(struct work_struct *work)
lockup_work.work);
ring = fence_drv->ring;
- if (!down_read_trylock(&ring->adev->exclusive_lock)) {
- /* just reschedule the check if a reset is going on */
- amdgpu_fence_schedule_check(ring);
- return;
- }
-
- if (amdgpu_fence_activity(ring)) {
+ if (amdgpu_fence_activity(ring))
wake_up_all(&ring->fence_drv.fence_queue);
- }
- up_read(&ring->adev->exclusive_lock);
}
/**
@@ -317,18 +309,15 @@ static bool amdgpu_fence_is_signaled(struct fence *f)
{
struct amdgpu_fence *fence = to_amdgpu_fence(f);
struct amdgpu_ring *ring = fence->ring;
- struct amdgpu_device *adev = ring->adev;
if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq)
return true;
- if (down_read_trylock(&adev->exclusive_lock)) {
- amdgpu_fence_process(ring);
- up_read(&adev->exclusive_lock);
+ amdgpu_fence_process(ring);
+
+ if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq)
+ return true;
- if (atomic64_read(&ring->fence_drv.last_seq) >= fence->seq)
- return true;
- }
return false;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index 7297ca3a0ba7..d81ab785368a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -181,7 +181,6 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
bool kernel = false;
int r;
- down_read(&adev->exclusive_lock);
/* create a gem object to contain this object in */
if (args->in.domains & (AMDGPU_GEM_DOMAIN_GDS |
AMDGPU_GEM_DOMAIN_GWS | AMDGPU_GEM_DOMAIN_OA)) {
@@ -214,11 +213,9 @@ int amdgpu_gem_create_ioctl(struct drm_device *dev, void *data,
memset(args, 0, sizeof(*args));
args->out.handle = handle;
- up_read(&adev->exclusive_lock);
return 0;
error_unlock:
- up_read(&adev->exclusive_lock);
r = amdgpu_gem_handle_lockup(adev, r);
return r;
}
@@ -250,8 +247,6 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
return -EACCES;
}
- down_read(&adev->exclusive_lock);
-
/* create a gem object to contain this object in */
r = amdgpu_gem_object_create(adev, args->size, 0,
AMDGPU_GEM_DOMAIN_CPU, 0,
@@ -293,14 +288,12 @@ int amdgpu_gem_userptr_ioctl(struct drm_device *dev, void *data,
goto handle_lockup;
args->handle = handle;
- up_read(&adev->exclusive_lock);
return 0;
release_object:
drm_gem_object_unreference_unlocked(gobj);
handle_lockup:
- up_read(&adev->exclusive_lock);
r = amdgpu_gem_handle_lockup(adev, r);
return r;
--
cgit
From f48b2659f521301753f9b3b67e308a79c6110346 Mon Sep 17 00:00:00 2001
From: Chunming Zhou <david1.zhou@amd.com>
Date: Fri, 16 Oct 2015 14:06:19 +0800
Subject: drm/amdgpu: fix the broken vm->mutex V2
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
fix the vm->mutex and ww_mutex confilcts.
vm->mutex is always token first, then ww_mutex.
V2: remove unneccessary checking for pt bo.
Change-Id: Iea56e183752c02831126d06d2f5b7a474a6e4743
Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 8 +++++---
drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c | 22 ++++++++++-----------
drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 35 +++------------------------------
3 files changed, 19 insertions(+), 46 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 5fdc0394561e..8727c3099554 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -608,7 +608,6 @@ static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
}
}
- mutex_lock(&vm->mutex);
r = amdgpu_bo_vm_update_pte(parser, vm);
if (r) {
goto out;
@@ -619,7 +618,6 @@ static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
parser->filp);
out:
- mutex_unlock(&vm->mutex);
return r;
}
@@ -827,6 +825,8 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
{
struct amdgpu_device *adev = dev->dev_private;
union drm_amdgpu_cs *cs = data;
+ struct amdgpu_fpriv *fpriv = filp->driver_priv;
+ struct amdgpu_vm *vm = &fpriv->vm;
struct amdgpu_cs_parser *parser;
bool reserved_buffers = false;
int i, r;
@@ -844,7 +844,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
r = amdgpu_cs_handle_lockup(adev, r);
return r;
}
-
+ mutex_lock(&vm->mutex);
r = amdgpu_cs_parser_relocs(parser);
if (r == -ENOMEM)
DRM_ERROR("Not enough memory for command submission!\n");
@@ -911,12 +911,14 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
mutex_unlock(&job->job_lock);
amdgpu_cs_parser_fini_late(parser);
+ mutex_unlock(&vm->mutex);
return 0;
}
cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence;
out:
amdgpu_cs_parser_fini(parser, r, reserved_buffers);
+ mutex_unlock(&vm->mutex);
r = amdgpu_cs_handle_lockup(adev, r);
return r;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
index d81ab785368a..087332858853 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c
@@ -115,9 +115,10 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_pri
struct amdgpu_vm *vm = &fpriv->vm;
struct amdgpu_bo_va *bo_va;
int r;
-
+ mutex_lock(&vm->mutex);
r = amdgpu_bo_reserve(rbo, false);
if (r) {
+ mutex_unlock(&vm->mutex);
return r;
}
@@ -128,7 +129,7 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj, struct drm_file *file_pri
++bo_va->ref_count;
}
amdgpu_bo_unreserve(rbo);
-
+ mutex_unlock(&vm->mutex);
return 0;
}
@@ -141,9 +142,10 @@ void amdgpu_gem_object_close(struct drm_gem_object *obj,
struct amdgpu_vm *vm = &fpriv->vm;
struct amdgpu_bo_va *bo_va;
int r;
-
+ mutex_lock(&vm->mutex);
r = amdgpu_bo_reserve(rbo, true);
if (r) {
+ mutex_unlock(&vm->mutex);
dev_err(adev->dev, "leaking bo va because "
"we fail to reserve bo (%d)\n", r);
return;
@@ -155,6 +157,7 @@ void amdgpu_gem_object_close(struct drm_gem_object *obj,
}
}
amdgpu_bo_unreserve(rbo);
+ mutex_unlock(&vm->mutex);
}
static int amdgpu_gem_handle_lockup(struct amdgpu_device *adev, int r)
@@ -481,18 +484,13 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
goto error_unreserve;
}
- mutex_lock(&bo_va->vm->mutex);
r = amdgpu_vm_clear_freed(adev, bo_va->vm);
if (r)
- goto error_unlock;
-
+ goto error_unreserve;
if (operation == AMDGPU_VA_OP_MAP)
r = amdgpu_vm_bo_update(adev, bo_va, &bo_va->bo->tbo.mem);
-error_unlock:
- mutex_unlock(&bo_va->vm->mutex);
-
error_unreserve:
ttm_eu_backoff_reservation(&ticket, &list);
@@ -549,10 +547,11 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
gobj = drm_gem_object_lookup(dev, filp, args->handle);
if (gobj == NULL)
return -ENOENT;
-
+ mutex_lock(&fpriv->vm.mutex);
rbo = gem_to_amdgpu_bo(gobj);
r = amdgpu_bo_reserve(rbo, false);
if (r) {
+ mutex_unlock(&fpriv->vm.mutex);
drm_gem_object_unreference_unlocked(gobj);
return r;
}
@@ -560,6 +559,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
bo_va = amdgpu_vm_bo_find(&fpriv->vm, rbo);
if (!bo_va) {
amdgpu_bo_unreserve(rbo);
+ mutex_unlock(&fpriv->vm.mutex);
return -ENOENT;
}
@@ -584,7 +584,7 @@ int amdgpu_gem_va_ioctl(struct drm_device *dev, void *data,
if (!r && !(args->flags & AMDGPU_VM_DELAY_UPDATE))
amdgpu_gem_va_update_vm(adev, bo_va, args->operation);
-
+ mutex_unlock(&fpriv->vm.mutex);
drm_gem_object_unreference_unlocked(gobj);
return r;
}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 0675524eb9de..06e207fd007b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -90,11 +90,9 @@ struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
struct amdgpu_bo_list_entry *list;
unsigned i, idx;
- mutex_lock(&vm->mutex);
list = drm_malloc_ab(vm->max_pde_used + 2,
sizeof(struct amdgpu_bo_list_entry));
if (!list) {
- mutex_unlock(&vm->mutex);
return NULL;
}
@@ -119,7 +117,6 @@ struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
list[idx].tv.shared = true;
list_add(&list[idx++].tv.head, head);
}
- mutex_unlock(&vm->mutex);
return list;
}
@@ -970,9 +967,7 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
INIT_LIST_HEAD(&bo_va->invalids);
INIT_LIST_HEAD(&bo_va->vm_status);
- mutex_lock(&vm->mutex);
list_add_tail(&bo_va->bo_list, &bo->va);
- mutex_unlock(&vm->mutex);
return bo_va;
}
@@ -1025,8 +1020,6 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
return -EINVAL;
}
- mutex_lock(&vm->mutex);
-
saddr /= AMDGPU_GPU_PAGE_SIZE;
eaddr /= AMDGPU_GPU_PAGE_SIZE;
@@ -1040,14 +1033,14 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
tmp->it.start, tmp->it.last + 1);
amdgpu_bo_unreserve(bo_va->bo);
r = -EINVAL;
- goto error_unlock;
+ goto error;
}
mapping = kmalloc(sizeof(*mapping), GFP_KERNEL);
if (!mapping) {
amdgpu_bo_unreserve(bo_va->bo);
r = -ENOMEM;
- goto error_unlock;
+ goto error;
}
INIT_LIST_HEAD(&mapping->list);
@@ -1079,9 +1072,6 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
if (vm->page_tables[pt_idx].bo)
continue;
- /* drop mutex to allocate and clear page table */
- mutex_unlock(&vm->mutex);
-
ww_mutex_lock(&resv->lock, NULL);
r = amdgpu_bo_create(adev, AMDGPU_VM_PTE_COUNT * 8,
AMDGPU_GPU_PAGE_SIZE, true,
@@ -1098,32 +1088,19 @@ int amdgpu_vm_bo_map(struct amdgpu_device *adev,
goto error_free;
}
- /* aquire mutex again */
- mutex_lock(&vm->mutex);
- if (vm->page_tables[pt_idx].bo) {
- /* someone else allocated the pt in the meantime */
- mutex_unlock(&vm->mutex);
- amdgpu_bo_unref(&pt);
- mutex_lock(&vm->mutex);
- continue;
- }
-
vm->page_tables[pt_idx].addr = 0;
vm->page_tables[pt_idx].bo = pt;
}
- mutex_unlock(&vm->mutex);
return 0;
error_free:
- mutex_lock(&vm->mutex);
list_del(&mapping->list);
interval_tree_remove(&mapping->it, &vm->va);
trace_amdgpu_vm_bo_unmap(bo_va, mapping);
kfree(mapping);
-error_unlock:
- mutex_unlock(&vm->mutex);
+error:
return r;
}
@@ -1168,7 +1145,6 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
}
}
- mutex_lock(&vm->mutex);
list_del(&mapping->list);
interval_tree_remove(&mapping->it, &vm->va);
trace_amdgpu_vm_bo_unmap(bo_va, mapping);
@@ -1177,7 +1153,6 @@ int amdgpu_vm_bo_unmap(struct amdgpu_device *adev,
list_add(&mapping->list, &vm->freed);
else
kfree(mapping);
- mutex_unlock(&vm->mutex);
amdgpu_bo_unreserve(bo_va->bo);
return 0;
@@ -1201,8 +1176,6 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
list_del(&bo_va->bo_list);
- mutex_lock(&vm->mutex);
-
spin_lock(&vm->status_lock);
list_del(&bo_va->vm_status);
spin_unlock(&vm->status_lock);
@@ -1221,8 +1194,6 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
fence_put(bo_va->last_pt_update);
kfree(bo_va);
-
- mutex_unlock(&vm->mutex);
}
/**
--
cgit
From 4cfdcd9cecc05646b08805ba449811a645b3f4a8 Mon Sep 17 00:00:00 2001
From: Dan Carpenter <dan.carpenter@oracle.com>
Date: Wed, 4 Nov 2015 16:25:09 +0300
Subject: drm/amdgpu: cleanup on error in amdgpu_cs_ioctl()
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
We recently changed the locking in this function and now there is a
missing unlock on error. Also there are some other resources that we
should probably release as well...
Fixes: f48b2659f521 ('drm/amdgpu: fix the broken vm->mutex V2')
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 8727c3099554..233ac9c64eb3 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -875,8 +875,10 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
struct amdgpu_job *job;
struct amdgpu_ring * ring = parser->ibs->ring;
job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
- if (!job)
- return -ENOMEM;
+ if (!job) {
+ r = -ENOMEM;
+ goto out;
+ }
job->base.sched = &ring->sched;
job->base.s_entity = &parser->ctx->rings[ring->idx].entity;
job->adev = parser->adev;
--
cgit
From e4a58a28b50f30e72292b6659d94410cbf7355ad Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Thu, 5 Nov 2015 17:00:25 +0100
Subject: drm/amdgpu: fix leaking the IBs on error
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Fixing a memory leak when the scheduler is enabled.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 25 ++++++++++++-------------
1 file changed, 12 insertions(+), 13 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index dfc4d02c7a38..ecc82dfe83f8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -499,16 +499,12 @@ static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser)
for (i = 0; i < parser->nchunks; i++)
drm_free_large(parser->chunks[i].kdata);
kfree(parser->chunks);
- if (!amdgpu_enable_scheduler)
- {
- if (parser->ibs)
- for (i = 0; i < parser->num_ibs; i++)
- amdgpu_ib_free(parser->adev, &parser->ibs[i]);
- kfree(parser->ibs);
- if (parser->uf.bo)
- drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
- }
-
+ if (parser->ibs)
+ for (i = 0; i < parser->num_ibs; i++)
+ amdgpu_ib_free(parser->adev, &parser->ibs[i]);
+ kfree(parser->ibs);
+ if (parser->uf.bo)
+ drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
kfree(parser);
}
@@ -888,11 +884,14 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
job->base.owner = parser->filp;
mutex_init(&job->job_lock);
if (job->ibs[job->num_ibs - 1].user) {
- memcpy(&job->uf, &parser->uf,
- sizeof(struct amdgpu_user_fence));
+ job->uf = parser->uf;
job->ibs[job->num_ibs - 1].user = &job->uf;
+ parser->uf.bo = NULL;
}
+ parser->ibs = NULL;
+ parser->num_ibs = 0;
+
job->free_job = amdgpu_cs_free_job;
mutex_lock(&job->job_lock);
r = amd_sched_entity_push_job(&job->base);
@@ -905,7 +904,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
cs->out.handle =
amdgpu_ctx_add_fence(parser->ctx, ring,
&job->base.s_fence->base);
- parser->ibs[parser->num_ibs - 1].sequence = cs->out.handle;
+ job->ibs[job->num_ibs - 1].sequence = cs->out.handle;
list_sort(NULL, &parser->validated, cmp_size_smaller_first);
ttm_eu_fence_buffer_objects(&parser->ticket,
--
cgit
From 7e52a81c2f0326a85d3ebc005829bcd604731c6d Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Wed, 4 Nov 2015 15:44:39 +0100
Subject: drm/amdgpu: cleanup amdgpu_cs_parser handling
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
No need any more to allocate that structure dynamically, just put it on the
stack. This is a start to cleanup some of the scheduler fallouts.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 5 --
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 88 +++++++++++++---------------------
2 files changed, 33 insertions(+), 60 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index b801b6702812..7b02e3455172 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -2256,11 +2256,6 @@ void amdgpu_pci_config_reset(struct amdgpu_device *adev);
bool amdgpu_card_posted(struct amdgpu_device *adev);
void amdgpu_update_display_priority(struct amdgpu_device *adev);
bool amdgpu_boot_test_post_card(struct amdgpu_device *adev);
-struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev,
- struct drm_file *filp,
- struct amdgpu_ctx *ctx,
- struct amdgpu_ib *ibs,
- uint32_t num_ibs);
int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data);
int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index ecc82dfe83f8..bf32096b8eb7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -127,30 +127,6 @@ int amdgpu_cs_get_ring(struct amdgpu_device *adev, u32 ip_type,
return 0;
}
-struct amdgpu_cs_parser *amdgpu_cs_parser_create(struct amdgpu_device *adev,
- struct drm_file *filp,
- struct amdgpu_ctx *ctx,
- struct amdgpu_ib *ibs,
- uint32_t num_ibs)
-{
- struct amdgpu_cs_parser *parser;
- int i;
-
- parser = kzalloc(sizeof(struct amdgpu_cs_parser), GFP_KERNEL);
- if (!parser)
- return NULL;
-
- parser->adev = adev;
- parser->filp = filp;
- parser->ctx = ctx;
- parser->ibs = ibs;
- parser->num_ibs = num_ibs;
- for (i = 0; i < num_ibs; i++)
- ibs[i].ctx = ctx;
-
- return parser;
-}
-
int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
{
union drm_amdgpu_cs *cs = data;
@@ -490,6 +466,7 @@ static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int err
static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser)
{
unsigned i;
+
if (parser->ctx)
amdgpu_ctx_put(parser->ctx);
if (parser->bo_list)
@@ -505,7 +482,6 @@ static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser)
kfree(parser->ibs);
if (parser->uf.bo)
drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
- kfree(parser);
}
/**
@@ -824,36 +800,36 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
union drm_amdgpu_cs *cs = data;
struct amdgpu_fpriv *fpriv = filp->driver_priv;
struct amdgpu_vm *vm = &fpriv->vm;
- struct amdgpu_cs_parser *parser;
+ struct amdgpu_cs_parser parser = {};
bool reserved_buffers = false;
int i, r;
if (!adev->accel_working)
return -EBUSY;
- parser = amdgpu_cs_parser_create(adev, filp, NULL, NULL, 0);
- if (!parser)
- return -ENOMEM;
- r = amdgpu_cs_parser_init(parser, data);
+ parser.adev = adev;
+ parser.filp = filp;
+
+ r = amdgpu_cs_parser_init(&parser, data);
if (r) {
DRM_ERROR("Failed to initialize parser !\n");
- amdgpu_cs_parser_fini(parser, r, false);
+ amdgpu_cs_parser_fini(&parser, r, false);
r = amdgpu_cs_handle_lockup(adev, r);
return r;
}
mutex_lock(&vm->mutex);
- r = amdgpu_cs_parser_relocs(parser);
+ r = amdgpu_cs_parser_relocs(&parser);
if (r == -ENOMEM)
DRM_ERROR("Not enough memory for command submission!\n");
else if (r && r != -ERESTARTSYS)
DRM_ERROR("Failed to process the buffer list %d!\n", r);
else if (!r) {
reserved_buffers = true;
- r = amdgpu_cs_ib_fill(adev, parser);
+ r = amdgpu_cs_ib_fill(adev, &parser);
}
if (!r) {
- r = amdgpu_cs_dependencies(adev, parser);
+ r = amdgpu_cs_dependencies(adev, &parser);
if (r)
DRM_ERROR("Failed in the dependencies handling %d!\n", r);
}
@@ -861,36 +837,38 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
if (r)
goto out;
- for (i = 0; i < parser->num_ibs; i++)
- trace_amdgpu_cs(parser, i);
+ for (i = 0; i < parser.num_ibs; i++)
+ trace_amdgpu_cs(&parser, i);
- r = amdgpu_cs_ib_vm_chunk(adev, parser);
+ r = amdgpu_cs_ib_vm_chunk(adev, &parser);
if (r)
goto out;
- if (amdgpu_enable_scheduler && parser->num_ibs) {
+ if (amdgpu_enable_scheduler && parser.num_ibs) {
struct amdgpu_job *job;
- struct amdgpu_ring * ring = parser->ibs->ring;
+ struct amdgpu_ring * ring = parser.ibs->ring;
+
job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
if (!job) {
r = -ENOMEM;
goto out;
}
+
job->base.sched = &ring->sched;
- job->base.s_entity = &parser->ctx->rings[ring->idx].entity;
- job->adev = parser->adev;
- job->ibs = parser->ibs;
- job->num_ibs = parser->num_ibs;
- job->base.owner = parser->filp;
+ job->base.s_entity = &parser.ctx->rings[ring->idx].entity;
+ job->adev = parser.adev;
+ job->ibs = parser.ibs;
+ job->num_ibs = parser.num_ibs;
+ job->base.owner = parser.filp;
mutex_init(&job->job_lock);
if (job->ibs[job->num_ibs - 1].user) {
- job->uf = parser->uf;
+ job->uf = parser.uf;
job->ibs[job->num_ibs - 1].user = &job->uf;
- parser->uf.bo = NULL;
+ parser.uf.bo = NULL;
}
- parser->ibs = NULL;
- parser->num_ibs = 0;
+ parser.ibs = NULL;
+ parser.num_ibs = 0;
job->free_job = amdgpu_cs_free_job;
mutex_lock(&job->job_lock);
@@ -902,24 +880,24 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
goto out;
}
cs->out.handle =
- amdgpu_ctx_add_fence(parser->ctx, ring,
+ amdgpu_ctx_add_fence(parser.ctx, ring,
&job->base.s_fence->base);
job->ibs[job->num_ibs - 1].sequence = cs->out.handle;
- list_sort(NULL, &parser->validated, cmp_size_smaller_first);
- ttm_eu_fence_buffer_objects(&parser->ticket,
- &parser->validated,
+ list_sort(NULL, &parser.validated, cmp_size_smaller_first);
+ ttm_eu_fence_buffer_objects(&parser.ticket,
+ &parser.validated,
&job->base.s_fence->base);
mutex_unlock(&job->job_lock);
- amdgpu_cs_parser_fini_late(parser);
+ amdgpu_cs_parser_fini_late(&parser);
mutex_unlock(&vm->mutex);
return 0;
}
- cs->out.handle = parser->ibs[parser->num_ibs - 1].sequence;
+ cs->out.handle = parser.ibs[parser.num_ibs - 1].sequence;
out:
- amdgpu_cs_parser_fini(parser, r, reserved_buffers);
+ amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
mutex_unlock(&vm->mutex);
r = amdgpu_cs_handle_lockup(adev, r);
return r;
--
cgit
From 7034decf6a5b1ff778d83ff9d7ce1f0b404804e4 Mon Sep 17 00:00:00 2001
From: Chunming Zhou <David1.Zhou@amd.com>
Date: Wed, 11 Nov 2015 14:56:00 +0800
Subject: drm/amdgpu: add command submission workflow tracepoint
OGL needs these tracepoints to investigate performance issue.
Change-Id: I5e58187d061253f7d665dfce8e4e163ba91d3e2b
Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 2 +
drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h | 51 +++++++++++++++++++++++++
drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h | 24 ++++++++++--
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 1 +
5 files changed, 76 insertions(+), 4 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index bf32096b8eb7..2ae73d5232dd 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -888,7 +888,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
ttm_eu_fence_buffer_objects(&parser.ticket,
&parser.validated,
&job->base.s_fence->base);
-
+ trace_amdgpu_cs_ioctl(job);
mutex_unlock(&job->job_lock);
amdgpu_cs_parser_fini_late(&parser);
mutex_unlock(&vm->mutex);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
index dcf4a8aca680..67f778f6eedb 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
@@ -26,6 +26,7 @@
#include <linux/sched.h>
#include <drm/drmP.h>
#include "amdgpu.h"
+#include "amdgpu_trace.h"
static struct fence *amdgpu_sched_dependency(struct amd_sched_job *sched_job)
{
@@ -45,6 +46,7 @@ static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job)
}
job = to_amdgpu_job(sched_job);
mutex_lock(&job->job_lock);
+ trace_amdgpu_sched_run_job(job);
r = amdgpu_ib_schedule(job->adev,
job->num_ibs,
job->ibs,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h
index 26e2d50a6f64..8f9834ab1bd5 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h
@@ -48,6 +48,57 @@ TRACE_EVENT(amdgpu_cs,
__entry->fences)
);
+TRACE_EVENT(amdgpu_cs_ioctl,
+ TP_PROTO(struct amdgpu_job *job),
+ TP_ARGS(job),
+ TP_STRUCT__entry(
+ __field(struct amdgpu_device *, adev)
+ __field(struct amd_sched_job *, sched_job)
+ __field(struct amdgpu_ib *, ib)
+ __field(struct fence *, fence)
+ __field(char *, ring_name)
+ __field(u32, num_ibs)
+ ),
+
+ TP_fast_assign(
+ __entry->adev = job->adev;
+ __entry->sched_job = &job->base;
+ __entry->ib = job->ibs;
+ __entry->fence = &job->base.s_fence->base;
+ __entry->ring_name = job->ibs[0].ring->name;
+ __entry->num_ibs = job->num_ibs;
+ ),
+ TP_printk("adev=%p, sched_job=%p, first ib=%p, sched fence=%p, ring name:%s, num_ibs:%u",
+ __entry->adev, __entry->sched_job, __entry->ib,
+ __entry->fence, __entry->ring_name, __entry->num_ibs)
+);
+
+TRACE_EVENT(amdgpu_sched_run_job,
+ TP_PROTO(struct amdgpu_job *job),
+ TP_ARGS(job),
+ TP_STRUCT__entry(
+ __field(struct amdgpu_device *, adev)
+ __field(struct amd_sched_job *, sched_job)
+ __field(struct amdgpu_ib *, ib)
+ __field(struct fence *, fence)
+ __field(char *, ring_name)
+ __field(u32, num_ibs)
+ ),
+
+ TP_fast_assign(
+ __entry->adev = job->adev;
+ __entry->sched_job = &job->base;
+ __entry->ib = job->ibs;
+ __entry->fence = &job->base.s_fence->base;
+ __entry->ring_name = job->ibs[0].ring->name;
+ __entry->num_ibs = job->num_ibs;
+ ),
+ TP_printk("adev=%p, sched_job=%p, first ib=%p, sched fence=%p, ring name:%s, num_ibs:%u",
+ __entry->adev, __entry->sched_job, __entry->ib,
+ __entry->fence, __entry->ring_name, __entry->num_ibs)
+);
+
+
TRACE_EVENT(amdgpu_vm_grab_id,
TP_PROTO(unsigned vmid, int ring),
TP_ARGS(vmid, ring),
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h b/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h
index 144f50acc971..c89dc777768f 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h
@@ -16,6 +16,8 @@ TRACE_EVENT(amd_sched_job,
TP_ARGS(sched_job),
TP_STRUCT__entry(
__field(struct amd_sched_entity *, entity)
+ __field(struct amd_sched_job *, sched_job)
+ __field(struct fence *, fence)
__field(const char *, name)
__field(u32, job_count)
__field(int, hw_job_count)
@@ -23,16 +25,32 @@ TRACE_EVENT(amd_sched_job,
TP_fast_assign(
__entry->entity = sched_job->s_entity;
+ __entry->sched_job = sched_job;
+ __entry->fence = &sched_job->s_fence->base;
__entry->name = sched_job->sched->name;
__entry->job_count = kfifo_len(
&sched_job->s_entity->job_queue) / sizeof(sched_job);
__entry->hw_job_count = atomic_read(
&sched_job->sched->hw_rq_count);
),
- TP_printk("entity=%p, ring=%s, job count:%u, hw job count:%d",
- __entry->entity, __entry->name, __entry->job_count,
- __entry->hw_job_count)
+ TP_printk("entity=%p, sched job=%p, fence=%p, ring=%s, job count:%u, hw job count:%d",
+ __entry->entity, __entry->sched_job, __entry->fence, __entry->name,
+ __entry->job_count, __entry->hw_job_count)
);
+
+TRACE_EVENT(amd_sched_process_job,
+ TP_PROTO(struct amd_sched_fence *fence),
+ TP_ARGS(fence),
+ TP_STRUCT__entry(
+ __field(struct fence *, fence)
+ ),
+
+ TP_fast_assign(
+ __entry->fence = &fence->base;
+ ),
+ TP_printk("fence=%p signaled", __entry->fence)
+);
+
#endif
/* This part must be outside protection */
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
index fe5b3c415f18..b8925fea577c 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
@@ -346,6 +346,7 @@ static void amd_sched_process_job(struct fence *f, struct fence_cb *cb)
list_del_init(&s_fence->list);
spin_unlock_irqrestore(&sched->fence_list_lock, flags);
}
+ trace_amd_sched_process_job(s_fence);
fence_put(&s_fence->base);
wake_up_interruptible(&sched->wake_up_worker);
}
--
cgit
From e284022163716ecf11c37fd1057c35d689ef2c11 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Thu, 5 Nov 2015 19:49:48 +0100
Subject: drm/amdgpu: fix incorrect mutex usage v3
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Before this patch the scheduler fence was created when we push the job
into the queue, so we could only get the fence after pushing it.
The mutex now was necessary to prevent the thread pushing the jobs to
the hardware from running faster than the thread pushing the jobs into
the queue.
Otherwise the thread pushing jobs into the queue would have accessed
possible freed up memory when it tries to get a reference to the fence.
So what you get in the end is thread A:
mutex_lock(&job->lock);
...
Kick of thread B.
...
mutex_unlock(&job->lock);
And thread B:
mutex_lock(&job->lock);
....
mutex_unlock(&job->lock);
kfree(job);
I'm actually not sure if I'm still up to date on this, but this usage
pattern used to be not allowed with mutexes. See here as well
https://lwn.net/Articles/575460/.
v2: remove unrelated changes, fix missing owner
v3: rebased, add more commit message
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 43 +++++++++++++++------------
drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c | 27 +++++++----------
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 10 +------
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 3 +-
5 files changed, 37 insertions(+), 48 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 7b02e3455172..0f187027c753 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1225,7 +1225,7 @@ struct amdgpu_job {
struct amdgpu_device *adev;
struct amdgpu_ib *ibs;
uint32_t num_ibs;
- struct mutex job_lock;
+ void *owner;
struct amdgpu_user_fence uf;
int (*free_job)(struct amdgpu_job *job);
};
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 2ae73d5232dd..44cf977ae4f6 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -845,8 +845,9 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
goto out;
if (amdgpu_enable_scheduler && parser.num_ibs) {
- struct amdgpu_job *job;
struct amdgpu_ring * ring = parser.ibs->ring;
+ struct amd_sched_fence *fence;
+ struct amdgpu_job *job;
job = kzalloc(sizeof(struct amdgpu_job), GFP_KERNEL);
if (!job) {
@@ -859,37 +860,41 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
job->adev = parser.adev;
job->ibs = parser.ibs;
job->num_ibs = parser.num_ibs;
- job->base.owner = parser.filp;
- mutex_init(&job->job_lock);
+ job->owner = parser.filp;
+ job->free_job = amdgpu_cs_free_job;
+
if (job->ibs[job->num_ibs - 1].user) {
job->uf = parser.uf;
job->ibs[job->num_ibs - 1].user = &job->uf;
parser.uf.bo = NULL;
}
- parser.ibs = NULL;
- parser.num_ibs = 0;
-
- job->free_job = amdgpu_cs_free_job;
- mutex_lock(&job->job_lock);
- r = amd_sched_entity_push_job(&job->base);
- if (r) {
- mutex_unlock(&job->job_lock);
+ fence = amd_sched_fence_create(job->base.s_entity,
+ parser.filp);
+ if (!fence) {
+ r = -ENOMEM;
amdgpu_cs_free_job(job);
kfree(job);
goto out;
}
- cs->out.handle =
- amdgpu_ctx_add_fence(parser.ctx, ring,
- &job->base.s_fence->base);
+ job->base.s_fence = fence;
+ fence_get(&fence->base);
+
+ cs->out.handle = amdgpu_ctx_add_fence(parser.ctx, ring,
+ &fence->base);
job->ibs[job->num_ibs - 1].sequence = cs->out.handle;
- list_sort(NULL, &parser.validated, cmp_size_smaller_first);
- ttm_eu_fence_buffer_objects(&parser.ticket,
- &parser.validated,
- &job->base.s_fence->base);
+ parser.ibs = NULL;
+ parser.num_ibs = 0;
+
trace_amdgpu_cs_ioctl(job);
- mutex_unlock(&job->job_lock);
+ amd_sched_entity_push_job(&job->base);
+
+ list_sort(NULL, &parser.validated, cmp_size_smaller_first);
+ ttm_eu_fence_buffer_objects(&parser.ticket, &parser.validated,
+ &fence->base);
+ fence_put(&fence->base);
+
amdgpu_cs_parser_fini_late(&parser);
mutex_unlock(&vm->mutex);
return 0;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
index 8ef9e4415fcc..438c05254695 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sched.c
@@ -45,12 +45,8 @@ static struct fence *amdgpu_sched_run_job(struct amd_sched_job *sched_job)
return NULL;
}
job = to_amdgpu_job(sched_job);
- mutex_lock(&job->job_lock);
trace_amdgpu_sched_run_job(job);
- r = amdgpu_ib_schedule(job->adev,
- job->num_ibs,
- job->ibs,
- job->base.owner);
+ r = amdgpu_ib_schedule(job->adev, job->num_ibs, job->ibs, job->owner);
if (r) {
DRM_ERROR("Error scheduling IBs (%d)\n", r);
goto err;
@@ -63,7 +59,6 @@ err:
if (job->free_job)
job->free_job(job);
- mutex_unlock(&job->job_lock);
kfree(job);
return fence ? &fence->base : NULL;
}
@@ -89,21 +84,19 @@ int amdgpu_sched_ib_submit_kernel_helper(struct amdgpu_device *adev,
return -ENOMEM;
job->base.sched = &ring->sched;
job->base.s_entity = &adev->kernel_ctx.rings[ring->idx].entity;
+ job->base.s_fence = amd_sched_fence_create(job->base.s_entity, owner);
+ if (!job->base.s_fence) {
+ kfree(job);
+ return -ENOMEM;
+ }
+ *f = fence_get(&job->base.s_fence->base);
+
job->adev = adev;
job->ibs = ibs;
job->num_ibs = num_ibs;
- job->base.owner = owner;
- mutex_init(&job->job_lock);
+ job->owner = owner;
job->free_job = free_job;
- mutex_lock(&job->job_lock);
- r = amd_sched_entity_push_job(&job->base);
- if (r) {
- mutex_unlock(&job->job_lock);
- kfree(job);
- return r;
- }
- *f = fence_get(&job->base.s_fence->base);
- mutex_unlock(&job->job_lock);
+ amd_sched_entity_push_job(&job->base);
} else {
r = amdgpu_ib_schedule(adev, num_ibs, ibs, owner);
if (r)
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
index ccb7c1554f5e..ea30d6ad4c13 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c
@@ -276,21 +276,13 @@ static bool amd_sched_entity_in(struct amd_sched_job *sched_job)
*
* Returns 0 for success, negative error code otherwise.
*/
-int amd_sched_entity_push_job(struct amd_sched_job *sched_job)
+void amd_sched_entity_push_job(struct amd_sched_job *sched_job)
{
struct amd_sched_entity *entity = sched_job->s_entity;
- struct amd_sched_fence *fence = amd_sched_fence_create(
- entity, sched_job->owner);
-
- if (!fence)
- return -ENOMEM;
-
- sched_job->s_fence = fence;
wait_event(entity->sched->job_scheduled,
amd_sched_entity_in(sched_job));
trace_amd_sched_job(sched_job);
- return 0;
}
/**
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index 4d05ca6fb057..939692b14f4b 100644
--- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
+++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
@@ -79,7 +79,6 @@ struct amd_sched_job {
struct amd_gpu_scheduler *sched;
struct amd_sched_entity *s_entity;
struct amd_sched_fence *s_fence;
- void *owner;
};
extern const struct fence_ops amd_sched_fence_ops;
@@ -131,7 +130,7 @@ int amd_sched_entity_init(struct amd_gpu_scheduler *sched,
uint32_t jobs);
void amd_sched_entity_fini(struct amd_gpu_scheduler *sched,
struct amd_sched_entity *entity);
-int amd_sched_entity_push_job(struct amd_sched_job *sched_job);
+void amd_sched_entity_push_job(struct amd_sched_job *sched_job);
struct amd_sched_fence *amd_sched_fence_create(
struct amd_sched_entity *s_entity, void *owner);
--
cgit
From 5d82730af746abca2aa74e00de6370d338df7e95 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Fri, 13 Nov 2015 13:04:50 +0100
Subject: drm/amdgpu: fix handling order in scheduler CS
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
We need to clear parser.ibs and num_ibs before amd_sched_fence_create,
otherwise the IB could be freed twice if fence creates fails.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 10 +++++-----
1 file changed, 5 insertions(+), 5 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 44cf977ae4f6..6096effd6a56 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -858,11 +858,14 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
job->base.sched = &ring->sched;
job->base.s_entity = &parser.ctx->rings[ring->idx].entity;
job->adev = parser.adev;
- job->ibs = parser.ibs;
- job->num_ibs = parser.num_ibs;
job->owner = parser.filp;
job->free_job = amdgpu_cs_free_job;
+ job->ibs = parser.ibs;
+ job->num_ibs = parser.num_ibs;
+ parser.ibs = NULL;
+ parser.num_ibs = 0;
+
if (job->ibs[job->num_ibs - 1].user) {
job->uf = parser.uf;
job->ibs[job->num_ibs - 1].user = &job->uf;
@@ -884,9 +887,6 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
&fence->base);
job->ibs[job->num_ibs - 1].sequence = cs->out.handle;
- parser.ibs = NULL;
- parser.num_ibs = 0;
-
trace_amdgpu_cs_ioctl(job);
amd_sched_entity_push_job(&job->base);
--
cgit
From 984810fc45389c545719fbb4219e8a12b27032a4 Mon Sep 17 00:00:00 2001
From: Christian König <christian.koenig@amd.com>
Date: Sat, 14 Nov 2015 21:05:35 +0100
Subject: drm/amdgpu: cleanup scheduler command submission
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Unify the two code path again, cause they do pretty much the same thing.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <davdi1.zhou@amd.com>
Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
---
drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 63 +++++++++++++---------------------
2 files changed, 25 insertions(+), 39 deletions(-)
(limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c')
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
index 0f187027c753..0ce5c1e45ec8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
@@ -1210,6 +1210,7 @@ struct amdgpu_cs_parser {
/* relocations */
struct amdgpu_bo_list_entry *vm_bos;
struct list_head validated;
+ struct fence *fence;
struct amdgpu_ib *ibs;
uint32_t num_ibs;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index 6096effd6a56..3afcf0237c25 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -439,8 +439,18 @@ static int cmp_size_smaller_first(void *priv, struct list_head *a,
return (int)la->robj->tbo.num_pages - (int)lb->robj->tbo.num_pages;
}
-static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int error, bool backoff)
+/**
+ * cs_parser_fini() - clean parser states
+ * @parser: parser structure holding parsing context.
+ * @error: error number
+ *
+ * If error is set than unvalidate buffer, otherwise just free memory
+ * used by parsing context.
+ **/
+static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
{
+ unsigned i;
+
if (!error) {
/* Sort the buffer list from the smallest to largest buffer,
* which affects the order of buffers in the LRU list.
@@ -455,17 +465,13 @@ static void amdgpu_cs_parser_fini_early(struct amdgpu_cs_parser *parser, int err
list_sort(NULL, &parser->validated, cmp_size_smaller_first);
ttm_eu_fence_buffer_objects(&parser->ticket,
- &parser->validated,
- &parser->ibs[parser->num_ibs-1].fence->base);
+ &parser->validated,
+ parser->fence);
} else if (backoff) {
ttm_eu_backoff_reservation(&parser->ticket,
&parser->validated);
}
-}
-
-static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser)
-{
- unsigned i;
+ fence_put(parser->fence);
if (parser->ctx)
amdgpu_ctx_put(parser->ctx);
@@ -484,20 +490,6 @@ static void amdgpu_cs_parser_fini_late(struct amdgpu_cs_parser *parser)
drm_gem_object_unreference_unlocked(&parser->uf.bo->gem_base);
}
-/**
- * cs_parser_fini() - clean parser states
- * @parser: parser structure holding parsing context.
- * @error: error number
- *
- * If error is set than unvalidate buffer, otherwise just free memory
- * used by parsing context.
- **/
-static void amdgpu_cs_parser_fini(struct amdgpu_cs_parser *parser, int error, bool backoff)
-{
- amdgpu_cs_parser_fini_early(parser, error, backoff);
- amdgpu_cs_parser_fini_late(parser);
-}
-
static int amdgpu_bo_vm_update_pte(struct amdgpu_cs_parser *p,
struct amdgpu_vm *vm)
{
@@ -582,15 +574,9 @@ static int amdgpu_cs_ib_vm_chunk(struct amdgpu_device *adev,
}
r = amdgpu_bo_vm_update_pte(parser, vm);
- if (r) {
- goto out;
- }
- amdgpu_cs_sync_rings(parser);
- if (!amdgpu_enable_scheduler)
- r = amdgpu_ib_schedule(adev, parser->num_ibs, parser->ibs,
- parser->filp);
+ if (!r)
+ amdgpu_cs_sync_rings(parser);
-out:
return r;
}
@@ -881,7 +867,7 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
goto out;
}
job->base.s_fence = fence;
- fence_get(&fence->base);
+ parser.fence = fence_get(&fence->base);
cs->out.handle = amdgpu_ctx_add_fence(parser.ctx, ring,
&fence->base);
@@ -890,17 +876,16 @@ int amdgpu_cs_ioctl(struct drm_device *dev, void *data, struct drm_file *filp)
trace_amdgpu_cs_ioctl(job);
amd_sched_entity_push_job(&job->base);
- list_sort(NULL, &parser.validated, cmp_size_smaller_first);
- ttm_eu_fence_buffer_objects(&parser.ticket, &parser.validated,
- &fence->base);
- fence_put(&fence->base);
+ } else {
+ struct amdgpu_fence *fence;
- amdgpu_cs_parser_fini_late(&parser);
- mutex_unlock(&vm->mutex);
- return 0;
+ r = amdgpu_ib_schedule(adev, parser.num_ibs, parser.ibs,
+ parser.filp);
+ fence = parser.ibs[parser.num_ibs - 1].fence;
+ parser.fence = fence_get(&fence->base);
+ cs->out.handle = parser.ibs[parser.num_ibs - 1].sequence;
}
- cs->out.handle = parser.ibs[parser.num_ibs - 1].sequence;
out:
amdgpu_cs_parser_fini(&parser, r, reserved_buffers);
mutex_unlock(&vm->mutex);
--
cgit