public-mirrors/linux
1
0
Fork 0
mirror of git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git synced 2025-09-18 22:14:16 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

drm/amdgpu: move queue_bitmap to an independent structure (v3)

Browse source 
To allocate independent queue_bitmap for each XCD,
then the old bitmap policy can be continued to use
with a clear logic.

Use mec_bitmap[0] as default for all non-GC 9.4.3 IPs.

v2: squash commits to avoid breaking the build
v3: unify naming style

Signed-off-by: Le Ma <le.ma@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit is contained in:
Le Ma 2022-07-27 14:35:49 +08:00 • committed by Alex Deucher
parent 277bd3371f
commit be697aa3a7
9 changed files with 47 additions and 32 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
View file

@ -162,7 +162,7 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *adev)
* clear
*/
bitmap_complement(gpu_resources.cp_queue_bitmap,
adev->gfx.mec.queue_bitmap,
adev->gfx.mec_bitmap[0].queue_bitmap,
KGD_MAX_QUEUES);
/* According to linux/bitmap.h we shouldn't use bitmap_clear if

2
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v9.c
View file

@ -778,7 +778,7 @@ void kgd_gfx_v9_get_cu_occupancy(struct amdgpu_device *adev, int pasid,
* Iterate through the shader engines and arrays of the device
* to get number of waves in flight
*/
bitmap_complement(cp_queue_bitmap, adev->gfx.mec.queue_bitmap,
bitmap_complement(cp_queue_bitmap, adev->gfx.mec_bitmap[0].queue_bitmap,
KGD_MAX_QUEUES);
max_queue_cnt = adev->gfx.mec.num_pipe_per_mec *
adev->gfx.mec.num_queue_per_pipe;

39
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
View file

@ -63,10 +63,10 @@ void amdgpu_queue_mask_bit_to_mec_queue(struct amdgpu_device *adev, int bit,
}
bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev,
int mec, int pipe, int queue)
int xcc_id, int mec, int pipe, int queue)
{
return test_bit(amdgpu_gfx_mec_queue_to_bit(adev, mec, pipe, queue),
adev->gfx.mec.queue_bitmap);
adev->gfx.mec_bitmap[xcc_id].queue_bitmap);
}
int amdgpu_gfx_me_queue_to_bit(struct amdgpu_device *adev,
@ -204,29 +204,38 @@ bool amdgpu_gfx_is_high_priority_compute_queue(struct amdgpu_device *adev,
void amdgpu_gfx_compute_queue_acquire(struct amdgpu_device *adev)
{
int i, queue, pipe;
int i, j, queue, pipe;
bool multipipe_policy = amdgpu_gfx_is_compute_multipipe_capable(adev);
int max_queues_per_mec = min(adev->gfx.mec.num_pipe_per_mec *
adev->gfx.mec.num_queue_per_pipe,
adev->gfx.num_compute_rings);
int num_xcd = (adev->gfx.num_xcd > 1) ? adev->gfx.num_xcd : 1;
if (multipipe_policy) {
/* policy: make queues evenly cross all pipes on MEC1 only */
for (i = 0; i < max_queues_per_mec; i++) {
pipe = i % adev->gfx.mec.num_pipe_per_mec;
queue = (i / adev->gfx.mec.num_pipe_per_mec) %
adev->gfx.mec.num_queue_per_pipe;
/* policy: make queues evenly cross all pipes on MEC1 only
* for multiple xcc, just use the original policy for simplicity */
for (j = 0; j < num_xcd; j++) {
for (i = 0; i < max_queues_per_mec; i++) {
pipe = i % adev->gfx.mec.num_pipe_per_mec;
queue = (i / adev->gfx.mec.num_pipe_per_mec) %
adev->gfx.mec.num_queue_per_pipe;
set_bit(pipe * adev->gfx.mec.num_queue_per_pipe + queue,
adev->gfx.mec.queue_bitmap);
set_bit(pipe * adev->gfx.mec.num_queue_per_pipe + queue,
adev->gfx.mec_bitmap[j].queue_bitmap);
}
}
} else {
/* policy: amdgpu owns all queues in the given pipe */
for (i = 0; i < max_queues_per_mec; ++i)
set_bit(i, adev->gfx.mec.queue_bitmap);
for (j = 0; j < num_xcd; j++) {
for (i = 0; i < max_queues_per_mec; ++i)
set_bit(i, adev->gfx.mec_bitmap[j].queue_bitmap);
}
}
dev_dbg(adev->dev, "mec queue bitmap weight=%d\n", bitmap_weight(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES));
for (j = 0; j < num_xcd; j++) {
dev_dbg(adev->dev, "mec queue bitmap weight=%d\n",
bitmap_weight(adev->gfx.mec_bitmap[j].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES));
}
}
void amdgpu_gfx_graphics_queue_acquire(struct amdgpu_device *adev)
@ -268,7 +277,7 @@ static int amdgpu_gfx_kiq_acquire(struct amdgpu_device *adev,
* adev->gfx.mec.num_queue_per_pipe;
while (--queue_bit >= 0) {
if (test_bit(queue_bit, adev->gfx.mec.queue_bitmap))
if (test_bit(queue_bit, adev->gfx.mec_bitmap[0].queue_bitmap))
continue;
amdgpu_queue_mask_bit_to_mec_queue(adev, queue_bit, &mec, &pipe, &queue);
@ -516,7 +525,7 @@ int amdgpu_gfx_enable_kcq(struct amdgpu_device *adev)
return -EINVAL;
for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
if (!test_bit(i, adev->gfx.mec.queue_bitmap))
if (!test_bit(i, adev->gfx.mec_bitmap[0].queue_bitmap))
continue;
/* This situation may be hit in the future if a new HW

7
drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
View file

@ -76,7 +76,9 @@ struct amdgpu_mec {
u32 num_pipe_per_mec;
u32 num_queue_per_pipe;
void *mqd_backup[AMDGPU_MAX_COMPUTE_RINGS + 1];
};
struct amdgpu_mec_bitmap {
/* These are the resources for which amdgpu takes ownership */
DECLARE_BITMAP(queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
};
@ -296,6 +298,7 @@ struct amdgpu_gfx {
struct amdgpu_ce ce;
struct amdgpu_me me;
struct amdgpu_mec mec;
struct amdgpu_mec_bitmap mec_bitmap[AMDGPU_MAX_GC_INSTANCES];
struct amdgpu_kiq kiq[AMDGPU_MAX_GC_INSTANCES];
struct amdgpu_imu imu;
bool rs64_enable; /* firmware format */
@ -425,8 +428,8 @@ int amdgpu_gfx_mec_queue_to_bit(struct amdgpu_device *adev, int mec,
int pipe, int queue);
void amdgpu_queue_mask_bit_to_mec_queue(struct amdgpu_device *adev, int bit,
int *mec, int *pipe, int *queue);
bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int mec,
int pipe, int queue);
bool amdgpu_gfx_is_mec_queue_enabled(struct amdgpu_device *adev, int inst,
int mec, int pipe, int queue);
bool amdgpu_gfx_is_high_priority_compute_queue(struct amdgpu_device *adev,
struct amdgpu_ring *ring);
bool amdgpu_gfx_is_high_priority_graphics_queue(struct amdgpu_device *adev,

6
drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c
View file

@ -4219,7 +4219,7 @@ static int gfx_v10_0_mec_init(struct amdgpu_device *adev)
const struct gfx_firmware_header_v1_0 *mec_hdr = NULL;
bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
/* take ownership of the relevant compute queues */
amdgpu_gfx_compute_queue_acquire(adev);
@ -4614,8 +4614,8 @@ static int gfx_v10_0_sw_init(void *handle)
for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k,
j))
if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
k, j))
continue;
r = gfx_v10_0_compute_ring_init(adev, ring_id,

6
drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c
View file

@ -699,7 +699,7 @@ static int gfx_v11_0_mec_init(struct amdgpu_device *adev)
u32 *hpd;
size_t mec_hpd_size;
bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
/* take ownership of the relevant compute queues */
amdgpu_gfx_compute_queue_acquire(adev);
@ -1374,8 +1374,8 @@ static int gfx_v11_0_sw_init(void *handle)
for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k,
j))
if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
k, j))
continue;
r = gfx_v11_0_compute_ring_init(adev, ring_id,

5
drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c
View file

@ -2728,7 +2728,7 @@ static int gfx_v7_0_mec_init(struct amdgpu_device *adev)
u32 *hpd;
size_t mec_hpd_size;
bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
/* take ownership of the relevant compute queues */
amdgpu_gfx_compute_queue_acquire(adev);
@ -4456,7 +4456,8 @@ static int gfx_v7_0_sw_init(void *handle)
for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
k, j))
continue;
r = gfx_v7_0_compute_ring_init(adev,

7
drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
View file

@ -1304,7 +1304,7 @@ static int gfx_v8_0_mec_init(struct amdgpu_device *adev)
u32 *hpd;
size_t mec_hpd_size;
bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
/* take ownership of the relevant compute queues */
amdgpu_gfx_compute_queue_acquire(adev);
@ -2001,7 +2001,8 @@ static int gfx_v8_0_sw_init(void *handle)
for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
k, j))
continue;
r = gfx_v8_0_compute_ring_init(adev,
@ -4319,7 +4320,7 @@ static int gfx_v8_0_kiq_kcq_enable(struct amdgpu_device *adev)
int r, i;
for (i = 0; i < AMDGPU_MAX_COMPUTE_QUEUES; ++i) {
if (!test_bit(i, adev->gfx.mec.queue_bitmap))
if (!test_bit(i, adev->gfx.mec_bitmap[0].queue_bitmap))
continue;
/* This situation may be hit in the future if a new HW

5
drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
View file

@ -1713,7 +1713,7 @@ static int gfx_v9_0_mec_init(struct amdgpu_device *adev)
const struct gfx_firmware_header_v1_0 *mec_hdr;
bitmap_zero(adev->gfx.mec.queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
bitmap_zero(adev->gfx.mec_bitmap[0].queue_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
/* take ownership of the relevant compute queues */
amdgpu_gfx_compute_queue_acquire(adev);
@ -2154,7 +2154,8 @@ static int gfx_v9_0_sw_init(void *handle)
for (i = 0; i < adev->gfx.mec.num_mec; ++i) {
for (j = 0; j < adev->gfx.mec.num_queue_per_pipe; j++) {
for (k = 0; k < adev->gfx.mec.num_pipe_per_mec; k++) {
if (!amdgpu_gfx_is_mec_queue_enabled(adev, i, k, j))
if (!amdgpu_gfx_is_mec_queue_enabled(adev, 0, i,
k, j))
continue;
r = gfx_v9_0_compute_ring_init(adev,