Currently each kfd compute partition/node is sharing
the same uuid of AID, which doen't meet the CUDA spec
for visible device, so corresponding XCD id for each
partition in smu has been assigned to xcp, and exposed
to kfd topology.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/a
SDMA doesn't support oversubsciption, it is the user matter to create
queues over HW limit, but not supposed to be a KFD error.
Signed-off-by: Eric Huang
---
.../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 14 --
.../gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
the entries will be appearing at
/sys/kernel/debug/kfd/proc//pasid_.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_debugfs.c | 77
drivers/gpu/drm/amd/amdkfd/kfd_priv.h| 5 ++
drivers/gpu/drm/amd/amdkfd/kfd_process.c | 3 +
3 files changed, 85
Thanks for the fix, I had the same patch submitted yesterday.
Regards,
Eric
On 2025-04-15 06:44, Dan Carpenter wrote:
The "pdd->drm_priv" NULL check is reversed so it will lead to a NULL
dereference on the next line.
Fixes: 4172b556fd5b ("drm/amdkfd: add smi events for process start and end")
ill
fix the issue.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c | 3 +--
1 file changed, 1 insertion(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
b/drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
index c27fd7aec1c3..83d9384ac815 100644
The mistake will lead to NULL kernel oops, so fix it.
Fixes: 56ed4241e9fe ("drm/amdkfd: add smi events for process start and end")
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/g
Ping ...
On 2025-04-07 16:52, Eric Huang wrote:
rocm-smi will be able to show the events for KFD process
start/end, it is the implementation of this feature.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_process.c| 4
drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c
rocm-smi will be able to show the events for KFD process
start/end, it is the implementation of this feature.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_process.c| 4
drivers/gpu/drm/amd/amdkfd/kfd_smi_events.c | 21 +
drivers/gpu/drm/amd/amdkfd
On 2025-03-24 17:21, Alex Deucher wrote:
On Mon, Mar 24, 2025 at 5:07 PM Eric Huang wrote:
On 2025-03-24 15:32, Alex Deucher wrote:
On Mon, Mar 24, 2025 at 1:26 PM Eric Huang wrote:
kfdtest KFDQMTest.OverSubscribeCpQueues with multiple
gpu mode fails on gfx v9.4.3+NPS4+CPX which has 64
On 2025-03-24 15:32, Alex Deucher wrote:
On Mon, Mar 24, 2025 at 1:26 PM Eric Huang wrote:
kfdtest KFDQMTest.OverSubscribeCpQueues with multiple
gpu mode fails on gfx v9.4.3+NPS4+CPX which has 64 gpu
nodes, the queues created are 65x64=4160, but the number
1024 0f
nubmer will make the test passed.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index f6aedf69c644..054a78207ffe 100644
--- a
In some ASICs L2 cache info may miss in kfd topology,
because the first bitmap may be empty, that means
the first cu may be inactive, so to find the first
active cu will solve the issue.
v2: Only find the first active cu in the first xcc
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd
On 2025-02-06 22:41, Lazar, Lijo wrote:
On 2/6/2025 10:18 PM, Eric Huang wrote:
I understand your concern. KFD currently only reports one L2 instance,
but not every L2 instance. If customers want to have more detail in all
available L2 info, we probably can change the logic in this function
, Lijo ; amd-gfx@lists.freedesktop.org
*Subject:* Re: [PATCH] drm/amdkfd: fix missing L2 cache info in topology
On 2025-02-06 10:14, Lazar, Lijo wrote:
>
> On 1/29/2025 8:50 PM, Eric Huang wrote:
>> In some ASICs L2 cache info may miss in kfd topology,
>> because the first b
On 2025-02-06 10:14, Lazar, Lijo wrote:
On 1/29/2025 8:50 PM, Eric Huang wrote:
In some ASICs L2 cache info may miss in kfd topology,
because the first bitmap may be empty, that means
the first cu may be inactive, so to find the first
active cu will solve the issue.
Signed-off-by: Eric
Ping ..
On 2025-01-29 10:20, Eric Huang wrote:
In some ASICs L2 cache info may miss in kfd topology,
because the first bitmap may be empty, that means
the first cu may be inactive, so to find the first
active cu will solve the issue.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd
In some ASICs L2 cache info may miss in kfd topology,
because the first bitmap may be empty, that means
the first cu may be inactive, so to find the first
active cu will solve the issue.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 18 --
1 file
amdgpu_job_ring may return NULL, which causes kernel NULL
pointer error, using another way to print ring name instead
of ring->name.
Suggested-by: Lijo Lazar
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c | 6 ++
1 file changed, 2 insertions(+), 4 deleti
To fullfill the reset event description.
Suggested-by: Lijo Lazar
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 1 +
3 files changed, 3 insertions(+)
diff --git a
Thanks for your review Lijo, I will send a patch with reset source in
another places.
Regards,
Eric
On 2024-06-04 03:26, Lazar, Lijo wrote:
On 6/3/2024 11:42 PM, Eric Huang wrote:
reset cause is requested by customer as additional
info for gpu reset smi event.
v2: integerate reset sources
reset cause is requested by customer as additional
info for gpu reset smi event.
v2: integerate reset sources suggested by Lijo Lazar
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 3 +++
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 10 +++---
drivers/gpu/drm
reset source or reset cause is very useful info
for reset context, it will be used by events API.
Suggested-by: Lijo Lazar
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_reset.c | 34 +++
drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h | 13 +
2 files
-
From: amd-gfx On Behalf Of Eric Huang
Sent: Friday, May 31, 2024 8:38 PM
To: amd-gfx@lists.freedesktop.org
Cc: Kasiviswanathan, Harish ; Huang, JinHuiEric
Subject: [PATCH] drm/amdkfd: add reset cause in gpu pre-reset smi event
reset cause is requested by customer as additional info for gpu
reset cause is requested by customer as additional
info for gpu reset smi event.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c| 34 +
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h| 17 +++--
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c| 9 ++-
drivers
TLB flush after unmap accidentially was removed on
gfx9.4.2. It is to add it back.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
b/drivers/gpu/drm/amd/amdkfd
The adev can be found from bo by amdgpu_ttm_adev(bo->tbo.bdev),
and adev is also not used in the function
amdgpu_amdkfd_map_gtt_bo_to_gart().
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h | 2 +-
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 3 +--
driv
Signed-off-by: Jonathan Kim
Reviewed-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c
b/drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manag
The field adev->mes.funcs is NULL in function amdgpu_mes_flush_shader_debugger
on non-mes asics, add mes enabling check for call this func to
resolve the error.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_process_queue_manager.c | 3 ++-
1 file changed, 2 insertions(+)
MES on process termination.
Note that the flush call and the MES debugger calls use the same MES
interface but are separated as KFD calls to avoid conflicting with each
other.
Signed-off-by: Jonathan Kim
Tested-by: Alice Wong
Reviewed-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_mes.c
On 2023-11-17 00:20, David Yat Sin wrote:
Fixes issue where user events of type KFD_EVENT_TYPE_HW_EXCEPTION do not
have valid data
Signed-off-by: David Yat Sin
---
drivers/gpu/drm/amd/amdkfd/kfd_events.c | 4
1 file changed, 4 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_e
On 2023-09-26 23:00, Xiaogang.Chen wrote:
From: Xiaogang Chen
prange->svm_bo unref can happen in both mmu callback and a callback after
migrate to system ram. Both are async call in different tasks. Sync svm_bo
unref operation to avoid random "use-after-free".
Signed-off-by: Xiaogang.Chen
-
that do not support the
current exception handling and running KFD tests.
The only time ADD_QUEUE.skip_process_ctx_clear is required is for
debugger use cases where a debugged process is always runtime enabled
when adding a queue.
Tested-by: Shikai Guo
Signed-off-by: Jonathan Kim
Reviewed-by: Eric
On 2023-08-11 09:26, Felix Kuehling wrote:
Am 2023-08-10 um 18:27 schrieb Eric Huang:
There is not UNMAP_QUEUES command sending for queue preemption
because the queue is suspended and test is closed to the end.
Function unmap_queue_cpsch will do nothing after that.
How do you suspend queues
7;s debugger upstreaming patch series? Or did
this come later? This patch only enables the workaround for v9.4.2.
Regards,
Felix
On 2023-08-10 17:52, Eric Huang wrote:
The problem is the queue is suspended before clearing address watch
call in KFD, there is not queue preemption and queue resu
rt of Jon's debugger upstreaming patch series? Or did
this come later? This patch only enables the workaround for v9.4.2.
Regards,
Felix
On 2023-08-10 17:52, Eric Huang wrote:
The problem is the queue is suspended before clearing address watch
call in KFD, there is not queue preemption
lowing apps.
So the solution is to clear the register as gfx v9 in KFD.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c | 8 +---
1 file changed, 1 insertion(+), 7 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
b/driv
Kuehling
On 2023-08-10 16:47, Eric Huang wrote:
KFD currently relies on MEC FW to clear tcp watch control
register by sending MAP_PROCESS packet with 0 of field
tcp_watch_cntl to HWS, but if the queue is suspended, the
packet will not be sent and the previous value will be
left on the register
solution is to clear the register as gfx v9 in KFD.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c | 8 +---
1 file changed, 1 insertion(+), 7 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c
b/drivers/gpu/drm/amd/amdgpu
in a safe manner.
Signed-off-by: Jonathan Kim
Reviewed-by: Eric Huang
Regards,
Eric
---
.../drm/amd/amdgpu/amdgpu_amdkfd_gfx_v11.c| 2 +-
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 13 -
drivers/gpu/drm/amd/amdkfd/kfd_debug.c| 19 +--
driver
To setup ttmp on as default for gfx v9.4.3 in IP hw init.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c | 6 ++
1 file changed, 6 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
b/drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c
index 86a84a0970f0
On 2023-07-11 14:38, Felix Kuehling wrote:
On 2023-07-11 10:28, Eric Huang wrote:
Read/write grace period from/to first xcc instance of
xcp in kfd node.
Signed-off-by: Eric Huang
---
.../drm/amd/amdkfd/kfd_device_queue_manager.c | 21 ---
.../drm/amd/amdkfd
Read/write grace period from/to first xcc instance of
xcp in kfd node.
Signed-off-by: Eric Huang
---
.../drm/amd/amdkfd/kfd_device_queue_manager.c | 21 ---
.../drm/amd/amdkfd/kfd_device_queue_manager.h | 2 +-
.../drm/amd/amdkfd/kfd_packet_manager_v9.c| 8 ---
3 files
OK. Mukul, I will resend this patch based on top of yours.
Regards,
Eric
On 2023-07-10 18:24, Joshi, Mukul wrote:
[AMD Official Use Only - General]
-Original Message-
From: amd-gfx On Behalf Of Eric
Huang
Sent: Monday, July 10, 2023 3:46 PM
To: amd-gfx@lists.freedesktop.org
Cc
Read/write grace period from/to first xcc instance of
xcp in kfd node.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 11 ---
drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h | 2 +-
drivers/gpu/drm/amd/amdkfd/kfd_packet_manager_v9.c| 10
d the difference
for GC v9.4.3 HW spec, i.e. xcc instance.
Signed-off-by: Jonathan Kim
Signed-off-by: Eric Huang
---
.../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c | 8 +-
.../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.h | 27 +++
.../drm/amd/amdgpu/amdgpu_amdkfd_gc_9_4_3.c | 166 +-
..
From: Jonathan Kim
Set watch points for all xcc instances on GFX943.
Signed-off-by: Jonathan Kim
Reviewed-by: Felix Kuehling
Signed-off-by: Eric Huang
Reviewed-by: Jonathan Kim
---
drivers/gpu/drm/amd/amdkfd/kfd_debug.c | 8 +---
1 file changed, 5 insertions(+), 3 deletions(-)
diff
device initialization.
Signed-off-by: Jonathan Kim
Reviewed-by: Felix Kuehling
Signed-off-by: Eric Huang
Reviewed-by: Jonathan Kim
---
drivers/gpu/drm/amd/amdkfd/kfd_debug.h | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
b/drivers
From: Jonathan Kim
Implement the similarities as GC v9.4.2, and the difference
for GC v9.4.3 HW spec, i.e. xcc instance.
Signed-off-by: Jonathan Kim
Signed-off-by: Eric Huang
---
.../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c | 8 +-
.../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.h | 27
Acked-by: Amber Lin
Signed-off-by: Eric Huang
Reviewed-by: Jonathan Kim
---
drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 10 --
drivers/gpu/drm/amd/amdkfd/kfd_topology.h | 3 +++
2 files changed, 11 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
b/d
Jonathan Kim (4):
drm/amdkfd: add kfd2kgd debugger callbacks for GC v9.4.3
drm/amdkfd: restore debugger additional info for gfx v9_4_3
drm/amdkfd: enable watch points globally for gfx943
drm/amdkfd: add multi-process debugging support for GC v9.4.3
.../drm/amd/amdgpu/amdgpu_amdkfd_aldebar
time and set
grace period accordingly.
Signed-off-by: Eric Huang
---
.../drm/amd/amdkfd/kfd_device_queue_manager.c | 9 --
.../drm/amd/amdkfd/kfd_device_queue_manager.h | 2 +-
.../gpu/drm/amd/amdkfd/kfd_packet_manager.c | 32 +++---
-
.../drm/amd/amdkfd
instance needs to get iq wait time and set
grace period accordingly.
Signed-off-by: Eric Huang
---
.../drm/amd/amdkfd/kfd_device_queue_manager.c | 9 --
.../drm/amd/amdkfd/kfd_device_queue_manager.h | 2 +-
.../gpu/drm/amd/amdkfd/kfd_packet_manager.c | 32 +++
.../drm
each xcc instance needs to get iq wait time and set
grace period accordingly.
Signed-off-by: Eric Huang
---
.../drm/amd/amdkfd/kfd_device_queue_manager.c | 9 --
.../drm/amd/amdkfd/kfd_device_queue_manager.h | 2 +-
.../gpu/drm/amd/amdkfd/kfd_packet_manager.c | 32
From: Jonathan Kim
Set watch points for all xcc instances on GFX943.
Signed-off-by: Jonathan Kim
Reviewed-by: Felix Kuehling
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_debug.c | 8 +---
1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/amd
Acked-by: Amber Lin
Signed-off-by: Eric Huang
Reviewed-by: Jonathan Kim
---
drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 10 --
drivers/gpu/drm/amd/amdkfd/kfd_topology.h | 3 +++
2 files changed, 11 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
b/d
To set TTMP setup on by default.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 2 +-
drivers/gpu/drm/amd/amdkfd/kfd_debug.c | 3 ++-
drivers/gpu/drm/amd/amdkfd/kfd_process.c | 6 +++---
3 files changed, 6 insertions(+), 5 deletions(-)
diff --git a/drivers/gpu/drm
device initialization.
Signed-off-by: Jonathan Kim
Reviewed-by: Felix Kuehling
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_debug.h | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
b/drivers/gpu/drm/amd/amdkfd
From: Jonathan Kim
Implement the similarities as GC v9.4.2, and the difference
for GC v9.4.3 HW spec, i.e. xcc instance.
Signed-off-by: Jonathan Kim
Signed-off-by: Eric Huang
---
.../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c | 10 +-
.../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.h | 30
Eric Huang (2):
drm/amdkfd: enable grace period for xcc instance
drm/amdkfd: always keep trap enabled for GC v9.4.3
Jonathan Kim (4):
drm/amdkfd: add kfd2kgd debugger callbacks for GC v9.4.3
drm/amdkfd: restore debugger additional info for gfx v9_4_3
drm/amdkfd: enable watch points
Since GFX9 GPU has multiple xcc instances, this is to
implement this change in KFD for debugger APIs.
Signed-off-by: Eric Huang
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c| 6 --
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gc_9_4_3.c | 6 --
drivers/gpu/drm/amd/amdgpu
Eric Huang (1):
drm/amdkfd: add xcc instance for debugger APIs
Jonathan Kim (4):
drm/amdgpu: add kfd2kgd debugger callbacks for GC v9.4.3
drm/amdkfd: restore debugger additional info for gfx v9_4_3
drm/amdkfd: enable watch points globally for gfx943
drm/amdkfd: add multi-process
device initialization.
Signed-off-by: Jonathan Kim
Reviewed-by: Felix Kuehling
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_debug.h | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
b/drivers/gpu/drm/amd/amdkfd
From: Jonathan Kim
Set watch points for all xcc instances on GFX943.
Signed-off-by: Jonathan Kim
Reviewed-by: Felix Kuehling
Signed-off-by: Eric Huang
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gc_9_4_3.c | 6 --
drivers/gpu/drm/amd/amdkfd/kfd_debug.c | 16 ++--
2
Acked-by: Amber Lin
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 10 --
drivers/gpu/drm/amd/amdkfd/kfd_topology.h | 3 +++
2 files changed, 11 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
b/drivers/gpu/drm/
From: Jonathan Kim
Implement the similarities as GC v9.4.2, and the difference
for GC v9.4.3 HW spec.
Signed-off-by: Jonathan Kim
Signed-off-by: Eric Huang
---
.../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c | 7 +-
.../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.h | 30
.../drm/amd/amdgpu
From: Jonathan Kim
Set watch points for all xcc instances on GFX943.
Signed-off-by: Jonathan Kim
Reviewed-by: Felix Kuehling
Signed-off-by: Eric Huang
---
.../gpu/drm/amd/amdgpu/amdgpu_amdkfd_gc_9_4_3.c | 6 --
drivers/gpu/drm/amd/amdkfd/kfd_debug.c | 16 ++--
2
From: Jonathan Kim
Implement the similarities as GC v9.4.2, and the difference
for GC v9.4.3 HW spec.
Signed-off-by: Jonathan Kim
Signed-off-by: Eric Huang
---
.../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c | 7 +-
.../drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.h | 30
.../drm/amd/amdgpu
Acked-by: Amber Lin
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 10 --
drivers/gpu/drm/amd/amdkfd/kfd_topology.h | 3 +++
2 files changed, 11 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
b/drivers/gpu/drm/
device initialization.
Signed-off-by: Jonathan Kim
Reviewed-by: Felix Kuehling
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_debug.h | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_debug.h
b/drivers/gpu/drm/amd/amdkfd
Since GFX9 GPU has multiple xcc instances, this is to
implement this change in KFD for debugger APIs.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_aldebaran.c | 6 --
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gc_9_4_3.c | 6 --
drivers/gpu/drm/amd/amdgpu
Eric Huang (1):
drm/amdkfd: add xcc instance for debugger APIs
Jonathan Kim (4):
drm/amdgpu: add debugger support for GC v9.4.3
drm/amdkfd: add multi-process debugging support for GC v9.4.3
drm/amdkfd: restore debugger additional info for gfx v9_4_3
drm/amdkfd: enable watch points
Reviewed-by: Eric Huang
Regards,
Eric
On 2023-05-01 16:52, Felix Kuehling wrote:
Don't move DMABuf attachments for PCIe P2P mappings to the SYSTEM domain
when unmapping. This avoids triggering eviction fences unnecessarily.
Instead do the move to SYSTEM and back to GTT when mapping
On 2023-04-28 15:42, Felix Kuehling wrote:
On 2023-04-28 14:09, Eric Huang wrote:
On 2023-04-28 12:41, Felix Kuehling wrote:
On 2023-04-28 10:17, Eric Huang wrote:
On 2023-04-27 23:46, Kuehling, Felix wrote:
[AMD Official Use Only - General]
Re-mapping typically happens after evictions
On 2023-04-28 12:41, Felix Kuehling wrote:
On 2023-04-28 10:17, Eric Huang wrote:
On 2023-04-27 23:46, Kuehling, Felix wrote:
[AMD Official Use Only - General]
Re-mapping typically happens after evictions, before a new eviction
fence gets attached. At that time the old eviction fence
ian König wrote:
Am 26.04.23 um 18:58 schrieb Felix Kuehling:
On 2023-04-26 9:03, Christian König wrote:
Am 25.04.23 um 16:11 schrieb Eric Huang:
Hi Christian,
What do you think about Felix's explanation?
That's unfortunately not something we can do here.
Regards,
Eric
On 2023-0
-26 9:03, Christian König wrote:
Am 25.04.23 um 16:11 schrieb Eric Huang:
Hi Christian,
What do you think about Felix's explanation?
That's unfortunately not something we can do here.
Regards,
Eric
On 2023-04-13 09:28, Felix Kuehling wrote:
Am 2023-04-13 um 07:35 schrieb Chr
Hi Christian,
What do you think about Felix's explanation?
Regards,
Eric
On 2023-04-13 09:28, Felix Kuehling wrote:
Am 2023-04-13 um 07:35 schrieb Christian König:
Am 13.04.23 um 03:01 schrieb Felix Kuehling:
Am 2023-04-12 um 18:25 schrieb Eric Huang:
It is to avoid redundant evictio
It is to avoid redundant eviction for KFD's DMAbuf import
bo when dmaunmapping DMAbuf. The DMAbuf import bo has
been set as AMDGPU_PL_PREEMPT in KFD when mapping.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 7 ++-
1 file changed, 6 insertions(+), 1 del
It is to avoid redundant eviction for KFD's DMAbuf import
bo when dmaunmapping DMAbuf. The DMAbuf import bo has
been set as AMDGPU_PL_PREEMPT in KFD when mapping.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 5 -
1 file changed, 4 insertions(+), 1 del
LE in the imported DMABuf BO.
On 2023-04-10 14:28, Eric Huang wrote:
Hi Felix,
Thanks for your review and suggestion, but unfortunately the
AMDGPU_GEM_DOMAIN_PREEMPTIBLE is not defined in amdgpu_drm.h. I
understand we need the memory eviction on either
kfd_mem_dmamap_dmabuf
pecial case in the above if-block for old_mem->mem_type ==
AMDGPU_PL_PREEMPT: use amdgpu_bo_sync_wait with
owner=AMDGPU_FENCE_OWNER_KFD so that it doesn't wait for eviction
fences
Regards,
Felix
Am 2023-04-04 um 10:36 schrieb Eric Huang:
Here is the backtrace from Jira:
Thu Nov 10 13:10:23 2022]
odule param.
Regards,
Felix
Am 2023-04-03 um 13:59 schrieb Eric Huang:
dmabuf is allocated/mapped as GTT domain, when dma-unmapping dmabuf
changing placement to CPU will trigger memory eviction after calling
ttm_bo_validate, and the eviction will cause performance drop.
Keeping the correct
dmabuf is allocated/mapped as GTT domain, when dma-unmapping dmabuf
changing placement to CPU will trigger memory eviction after calling
ttm_bo_validate, and the eviction will cause performance drop.
Keeping the correct domain will solve the issue.
Signed-off-by: Eric Huang
---
drivers/gpu/drm
Ping.
On 2023-01-05 14:28, Eric Huang wrote:
The point bo->kfd_bo is NULL for queue's write pointer BO
when creating queue on mGPU. To avoid using the pointer
fixes the error.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 2 +-
drivers/gpu
There will be data corruption on vram allocated by svm
if initialization is not being done. Adding sync is to
resolve this issue.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 7 +++
1 file changed, 7 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
b
The point bo->kfd_bo is NULL for queue's write pointer BO
when creating queue on mGPU. To avoid using the pointer
fixes the error.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 2 +-
drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 2 +-
.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c | 4 +---
1 file changed, 1 insertion(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index db772942f7a6..fb1bb593312e 100644
--- a
*From:* amd-gfx on behalf of
Eric Huang
*Sent:* Monday, July 11, 2022 2:41 PM
*To:* amd-gfx@lists.freedesktop.org
*Cc:* Huang, JinHuiEric ; Kuehling, Felix
*Subject:* [PATCH] drm/amdkfd: bump KFD version for unified ctx
save/restore memory
To
To improve performance on queue preemption, allocate ctx s/r
area in VRAM instead of system memory, and migrate it back
to system memory when VRAM is full.
Signed-off-by: Eric Huang
Change-Id: If775782027188dbe84b6868260e429373675434c
---
include/hsakmttypes.h | 1 +
src/queues.c
It is to add new option for always keeping gpu mapping
and bump KFD version for the feature of unified save
restore memory.
Signed-off-by: Eric Huang
Change-Id: Iebee35e6de4d52fa29f82dd19f6bbf5640249492
---
include/linux/kfd_ioctl.h | 6 +-
1 file changed, 5 insertions(+), 1 deletion
To expose unified memory for ctx save/resotre area feature
availablity to libhsakmt.
Signed-off-by: Eric Huang
---
include/uapi/linux/kfd_ioctl.h | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index
To improve performance on queue preemption, allocate ctx s/r
area in VRAM instead of system memory, and migrate it back
to system memory when VRAM is full.
Signed-off-by: Eric Huang
Change-Id: If775782027188dbe84b6868260e429373675434c
---
include/hsakmttypes.h | 1 +
src/queues.c
It is to add new option for always keeping gpu mapping.
Signed-off-by: Eric Huang
Change-Id: Iebee35e6de4d52fa29f82dd19f6bbf5640249492
---
include/linux/kfd_ioctl.h | 2 ++
1 file changed, 2 insertions(+)
diff --git a/include/linux/kfd_ioctl.h b/include/linux/kfd_ioctl.h
index 8a0ed49..5c45f58
It is to add new option for always keeping gpu mapping.
Signed-off-by: Eric Huang
---
include/uapi/linux/kfd_ioctl.h | 2 ++
1 file changed, 2 insertions(+)
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index fd49dde4d5f4..eba04ebfd9a8 100644
--- a/include/uapi
It is to avoid unnecessary queue eviction when range
is not mapped to gpu.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 7 +++
1 file changed, 7 insertions(+)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
b/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
index
Adding always evict queues when flag is set to
KFD_IOCTL_SVM_FLAG_GPU_ALWAYS_MAPPED as if XNACK off.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 6 --
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_svm.c
b/drivers/gpu
amdkfd changes:
Eric Huang (3):
drm/amdkfd: add new flag for svm
drm/amdkfd: change svm range evict
drm/amdkfd: optimize svm range evict
drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 13 +++--
include/uapi/linux/kfd_ioctl.h | 2 ++
2 files changed, 13 insertions(+), 2 deletions
On 2022-06-29 19:29, Felix Kuehling wrote:
On 2022-06-29 18:53, Eric Huang wrote:
On 2022-06-29 18:20, Felix Kuehling wrote:
On 2022-06-28 17:43, Eric Huang wrote:
Two changes:
1. reducing unnecessary evict/unmap when range is not mapped to gpu.
2. adding always evict when flags is set to
On 2022-06-29 18:20, Felix Kuehling wrote:
On 2022-06-28 17:43, Eric Huang wrote:
Two changes:
1. reducing unnecessary evict/unmap when range is not mapped to gpu.
2. adding always evict when flags is set to always_mapped.
Signed-off-by: Eric Huang
---
drivers/gpu/drm/amd/amdkfd
To improve performance on queue preemption, allocate ctx s/r
area in VRAM instead of system memory, and migrate it back
to system memory when VRAM is full.
Signed-off-by: Eric Huang
Change-Id: If775782027188dbe84b6868260e429373675434c
---
include/hsakmttypes.h | 1 +
src/queues.c
1 - 100 of 238 matches
Mail list logo
