On Mon, Mar 24, 2025 at 5:06 AM Sergey Kovalenko
wrote:
>
> Predict an optimal delay to enable GFXOFF for the next interval
> based on the request count:
> - less than 15 requests per second - zero delay
> - less than 25 requests per second - default delay
> - 25 and more requests per second - don
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
struct _ddebug_info already has almost all dyndbg's info for a module,
so finish the encapsulation. This puts the datum closer to where its
needed, improving the chance that we can obsolete the _ddebug.modame
field with a desc_modname(dp) accessor f
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Since commit
85f7f6c0edb8 ("dynamic_debug: process multiple debug-queries on a line")
Multi-query commands have been allowed:
modprobe drm dyndbg="class DRM_UT_CORE +p; class DRM_UT_KMS +p"
modprobe drm dyndbg=<
[ 203.902703] dyndbg: query p
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
New fn validates parsing and effect of queries using combinations of
commas and spaces to delimit the tokens.
It manipulates pr-debugs in builtin module/params, so might have deps
I havent foreseen on odd configurations.
Signed-off-by: Jim Cromie
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
This new test-fn runs 3 module/submodule modprobe scenarios, variously
using both the generic dyndbg= modprobe arg, and the
test-module's classmap-params to manipulate the test-mod*'s pr_debugs.
In all cases, the current flag-settings are counted and
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
This does basic testing of classmaps using '%' separated
multi-queries. It modprobes test_dynamic_debug with several classes
enabled, and counts to verify that the expected sites show the
enablement in the control file.
Signed-off-by: Jim Cromie
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Current classmap code protects class'd pr_debugs from unintended
changes by "legacy" unclassed queries:
# this doesn't disable all of DRM_UT_* categories
echo "-p" > /proc/dynamic_debug/control
# name the class to change it - protective bu
ping?
On Thu, Mar 20, 2025 at 12:22 PM Alex Deucher wrote:
>
> GC12 only has 1 mec.
>
> Fixes: 52cb80c12e8a ("drm/amdgpu: Add gfx v12_0 ip block support (v6)")
> Signed-off-by: Alex Deucher
> ---
> drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
ping on this series?
Alex
On Thu, Mar 20, 2025 at 12:57 PM Alex Deucher wrote:
>
> In dev core dump, dump the full header fifo for
> each queue. Each FIFO has 8 entries.
>
> Signed-off-by: Alex Deucher
> ---
> drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 62 +--
> 1 file cha
ping on this series?
On Thu, Mar 20, 2025 at 4:36 PM Alex Deucher wrote:
>
> It's not used outside of amdgpu_gfx.c.
>
> Signed-off-by: Alex Deucher
> ---
> drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c | 4 ++--
> drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h | 2 --
> 2 files changed, 2 insertions(+), 4 d
ping on this series?
Alex
On Wed, Mar 19, 2025 at 12:23 PM Alex Deucher wrote:
>
> We shouldn't return after the last section.
> We need to update the rest of the CSIB.
>
> Signed-off-by: Alex Deucher
> ---
> drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c | 2 --
> 1 file changed, 2 deletions(-)
>
> di
We should only increase the deferred errors in UMC block.
Signed-off-by: Xiang Liu
---
drivers/gpu/drm/amd/amdgpu/amdgpu_aca.c | 4
drivers/gpu/drm/amd/amdgpu/amdgpu_aca.h | 8
drivers/gpu/drm/amd/amdgpu/amdgpu_xgmi.c | 2 +-
drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c | 8 +++-
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
dyndbg's CLASSMAP-v1 api was broken; DECLARE_DYNDBG_CLASSMAP tried to
do too much. Its replaced by DRM_CLASSMAP_DEFINE, which creates &
EXPORTs a classmap (in DRM core), and DRM_CLASSMAP_USE which refers to
the classmap defined elsewhere.
The drive
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 KFD_MAX_NUM_OF_QUEUES_PER_PROCESS is not enough
and test fails at function find_available_queue_slot().
So increasing the nub
Applied 1-4 with some minor fixes.
Alex
On Sat, Mar 22, 2025 at 3:03 PM Alexandre Demers
wrote:
>
> gmc_v7_0_is_idle() does exactly what we need, so use it.
>
> Signed-off-by: Alexandre Demers
> ---
> drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 9 +
> 1 file changed, 1 insertion(+), 8 dele
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 gp
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 KFD_MAX_NUM_OF_QUEUES_PER_PROCESS is not enough
> and test fails
On 3/24/2025 13:45, Saleemkhan Jamadar wrote:
Hi Mario,
Made the suggested change.
Regards,
Saleem
I'm not sure what happened to the below text (maybe email client failure?).
But in any case the change looks good to me now. You can add:
Reviewed-by: Mario Limonciello
When committing it.
Ping?
On Fri, Feb 28, 2025 at 3:39 PM Alex Deucher wrote:
>
> Add a new parameter to disable BAR resizing. Note that this
> only disables the driver from attempting to resize the BAR,
> The BIOS may have resized the BAR at boot.
>
> Some teams have found this useful in debugging P2P DMA
> issues
Le 20/03/2025 à 19:51, Jim Cromie a écrit :
Remove the DD_CLASS_TYPE_*_NAMES classmap types and code.
These 2 classmap types accept class names at the PARAM interface, for
example:
echo +DRM_UT_CORE,-DRM_UT_KMS > /sys/module/drm/parameters/debug_names
The code works, but its only used by
On 3/24/2025 14:57, Alex Deucher wrote:
Ping?
On Fri, Feb 28, 2025 at 3:39 PM Alex Deucher wrote:
Add a new parameter to disable BAR resizing. Note that this
only disables the driver from attempting to resize the BAR,
The BIOS may have resized the BAR at boot.
Some teams have found this use
Hi Mario,
Made the suggested change.
Regards,
Saleem
--2.34.1
Content-Type: text/plain; charset=UTF-8; format=fixed
Content-Transfer-Encoding: 8bit
v1 - DRM_ERROR to drm_err (Mario)
Update message to identifiy the vblank initialization fail case
Signed-off-by: Saleemkhan Jamadar
On Fri, Mar 21, 2025 at 9:58 PM Alexandre Demers
wrote:
>
> The following series is intented to remove duplicated defines, shifts and
> masks or
> to classify them where they belong. si_enums.h has been used as a garbage can
> for anything and everything when moving SI code from radeon to amdgpu.
On Fri, Mar 21, 2025 at 9:48 PM Alexandre Demers
wrote:
>
> They will be used later when switching away from sid.h/si_enums.h.
>
> To prevent redefinition clashes, comment out the ones in sid.h. They will be
> removed later.
>
> Signed-off-by: Alexandre Demers
> ---
> .../drm/amd/include/asic_re
Add an INFO query to check if user queues are supported.
v2: switch to a mask of IPs (Marek)
v3: move to drm_amdgpu_info_device (Marek)
Cc: marek.ol...@amd.com
Cc: prike.li...@amd.com
Cc: sunil.kha...@amd.com
Cc: yogesh.mohanmarimu...@amd.com
Signed-off-by: Alex Deucher
---
drivers/gpu/drm/amd/
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 gpu
> >> nodes, the queues crea
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
tiny/bochs has 5 DRM_UT_* debugs, make them controllable when
CONFIG_DRM_USE_DYNAMIC_DEBUG=y by telling dyndbg that the module has
class'd debugs.
Signed-off-by: Jim Cromie
Reviewed-by: Louis Chauvet
---
drivers/gpu/drm/tiny/bochs.c | 2 ++
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Add new drm_dyndbg_user.c with a single call to
DYNDBG_CLASSMAP_USE(drm_debug_classes). This creates a _class_user
record (and a linkage dependency).
I agree, this could be a very nice thing to automagically have the _USE
call included. But if th
Le 16/03/2025 à 20:46, jim.cro...@gmail.com a écrit :
hi Louis,
On Tue, Feb 25, 2025 at 7:16 AM Louis Chauvet wrote:
Le 25/01/2025 à 07:45, Jim Cromie a écrit :
DECLARE_DYNDBG_CLASSMAP() has a design error; its usage fails a basic
K&R rule: "define once, refer many times".
It is used a
Le 20/03/2025 à 19:51, Jim Cromie a écrit :
DECLARE_DYNDBG_CLASSMAP() has a design error; its usage fails a basic
K&R rule: "define once, refer many times".
When DRM_USE_DYNAMIC_DEBUG=y, it is used across DRM core & drivers;
each invocation allocates/inits the classmap understood by that
modu
Le 20/03/2025 à 19:51, Jim Cromie a écrit :
Add __DYNDBG_CLASSMAP_CHECK to implement these arg-checks at compile:
0 <= _base < 63
class_names is not empty
class_names[0] is a string
(class_names.length + _base) < 63
These compile-time checks will prevent severa
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Treat comma as a token terminator, just like a space. This allows a
user to avoid quoting hassles when spaces are otherwise needed:
:#> modprobe drm dyndbg=class,DRM_UT_CORE,+p\;class,DRM_UT_KMS,+p
or as a boot arg:
drm.dyndbg=class,DRM_UT_CO
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Following the dyndbg-api-fix, replace DECLARE_DYNDBG_CLASSMAP with
DRM_CLASSMAP_USE. This refs the defined & exported classmap, rather
than re-declaring it redundantly, and error-prone-ly.
This resolves the appearance of "class:_UNKNOWN_" in the co
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Following the dyndbg-api-fix, replace DECLARE_DYNDBG_CLASSMAP with
DRM_CLASSMAP_USE. This refs the defined & exported classmap, rather
than re-declaring it redundantly, and error-prone-ly.
This resolves the appearance of "class:_UNKNOWN_" in the co
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Add mention of comma and percent delimiters into the respective
paragraphs describing their equivalents: space and newline.
cc: linux-...@vger.kernel.org
Signed-off-by: Jim Cromie
I think this should go with the previous patches introducing the
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
When writing queries to >control, flags are parsed 1st, since they are
the only required field, and they require specific compositions. So
if the flags draw an error (on those specifics), then keyword errors
aren't reported. This can be mildly conf
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
With CONFIG_DRM_USE_DYNAMIC_DEBUG=y, __drm_printfn_dbg() gets an
unused variable warning/error on 'category', even though the usage
follows immediately, in drm_debug_enabled(category).
For static-key optimized dyndbg, the macro doesn't actually chec
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
we currently get:
WARNING: Argument 'name' is not used in function-like macro
on:
#define DRM_CLASSMAP_USE(name) /* nothing here */
Following this advice is wrong here, and shouldn't be fixed by
ignoring args altogether; the macro should prop
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Invoke DYNAMIC_DEBUG_CLASSMAP_PARAM to hook drm.debug (__drm_debug) to the
DRM_UT_* classmap, replacing the ad-hoc wiring previously doing it.
Add DRM_CLASSMAP_* adapter macros to selectively use
DYNAMIC_DEBUG_CLASSMAP_* when DRM_USE_DYNAMIC_DEBUG=y
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Invoke DRM_CLASSMAP_USE from xe_drm_client.c. When built with
CONFIG_DRM_USE_DYNAMIC_DEBUG=y, this tells dydnbg that Xe uses
has drm.debug calls.
Signed-off-by: Jim Cromie
---
drivers/gpu/drm/xe/xe_drm_client.c | 2 ++
1 file changed, 2 inserti
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Following the dyndbg-api-fix, replace DECLARE_DYNDBG_CLASSMAP with
DRM_CLASSMAP_USE. This refs the defined & exported classmap, rather
than re-declaring it redundantly, and error-prone-ly.
This resolves the appearance of "class:_UNKNOWN_" in the co
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Describe the 3 API macros providing dynamic_debug's classmaps
DYNDBG_CLASSMAP_DEFINE - create & export a classmap
DYNDBG_CLASSMAP_USE- refer to exported map
DYNDBG_CLASSMAP_PARAM - bind control param to the classmap
DYNDBG_CLASSMAP_PARAM_REF +
Le 20/03/2025 à 19:51, Jim Cromie a écrit :
Split dynamic_emit_prefix() to separate out _INCL_LOOKUPs:
1. keep dynamic_emit_prefix() static inline
check _INCL_ANY flags before calling 2
2. __dynamic_emit_prefix()
prints [TID] or and trailing space if +t flag
check _INCL_LOOKUP f
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
move the DYNAMIC_DEBUG_CLASSMAP_PARAM macro from test-dynamic-debug.c into
the header, and refine it, by distinguishing the 2 use cases:
1.DYNAMIC_DEBUG_CLASSMAP_PARAM_REF
for DRM, to pass in extern __drm_debug by name.
dyndbg keeps bits i
Le 20/03/2025 à 19:51, Jim Cromie a écrit :
The body of ddebug_attach_module_classes() is dominated by a
code-block that finds the contiguous subrange of classmaps matching on
modname, and saves it into the ddebug_table's info record.
Implement this block in a macro to accommodate different c
On Mon, Mar 24, 2025 at 11:14 AM Sergey Kovalenko
wrote:
>
> "Moreover the work only gets scheduled/cancelled
> around a ref count of 0" - and this happens more than 600 times
> per second under load, as you can see from the table.
> Moreover these 600 GFXOFF requests are executed with a very
> l
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
Following the dyndbg-api-fix, replace DECLARE_DYNDBG_CLASSMAP with
DRM_CLASSMAP_USE. This refs the defined & exported classmap, rather
than re-declaring it redundantly, and error-prone-ly.
This resolves the appearance of "class:_UNKNOWN_" in the co
From: Dmitry Baryshkov
Switch drm_dp_mst_topology.c to use new set of DPCD read / write helpers.
Reviewed-by: Lyude Paul
Acked-by: Jani Nikula
Signed-off-by: Dmitry Baryshkov
---
drivers/gpu/drm/display/drm_dp_mst_topology.c | 105 +-
1 file changed, 51 insertions(+),
Am 21.03.25 um 19:05 schrieb Philipp Stanner:
> On Fri, 2025-03-21 at 16:58 +0100, Christian König wrote:
>> Sometimes drivers need to be able to submit multiple jobs which
>> depend on
>> each other to different schedulers at the same time, but using
>> drm_sched_job_add_dependency() can't fail an
Am 24.03.25 um 09:35 schrieb Tvrtko Ursulin:
>
> On 21/03/2025 15:58, Christian König wrote:
>> Sometimes drivers need to be able to submit multiple jobs which depend on
>> each other to different schedulers at the same time, but using
>> drm_sched_job_add_dependency() can't fail any more after the
[Public]
Hi all,
This week this patchset was tested on 4 systems, two dGPU and two APU based,
and tested across multiple display and connection types.
APU
* Single Display eDP -> 1080p 60hz, 2560x1600 120hz, 1920x1200 165hz
* Single Display DP (SST DSC) -> 4k144hz, 4k240hz
On Mon, Mar 3, 2025 at 9:46 AM Alex Deucher wrote:
>
> On Fri, Feb 21, 2025 at 8:38 AM Prike Liang wrote:
> >
> > Implement the kernel graphics queue pipe reset,and the driver
> > will fallback to pipe reset when the queue reset fails. However,
> > the ME FW hasn't fully supported pipe reset yet
On Mon, Mar 24, 2025 at 8:01 AM Liang, Prike wrote:
>
> [Public]
>
> It seems there still requires implementing the gfx11/gfx12 pipe reset from
> the project perspective, so please help review the patch series.
See my latest comments on patch 1. With those addressed the series is:
Acked-by: Ale
Le 20/03/2025 à 19:52, Jim Cromie a écrit :
The vkms driver has a number of DRM_UT_* debugs, make them
controllable when CONFIG_DRM_USE_DYNAMIC_DEBUG=y by telling dyndbg
that the module uses them.
Signed-off-by: Jim Cromie
Reviewed-by: Louis Chauvet
---
drivers/gpu/drm/vkms/vkms_drv.c
Le 20/03/2025 à 19:51, Jim Cromie a écrit :
recompose struct _ddebug_info, inserting proper sub-structs.
The struct currently has 2 pairs of fields: descs, num_descs and
classes, num_classes. Several for-loops operate on these field pairs,
soon many more will be added.
Looping over these bl
Le 20/03/2025 à 19:51, Jim Cromie a écrit :
Add _INCL_LOOKUP condition to separate +mfsl flags from +t, allowing
(after refactoring) to avoid a needless call-return.
Add a PREFIX_CACHED bit to remember that a pr-debug callsite is:
- enabled, with +p
- wants a dynamic-prefix, with _INCL_LOOKU
Le 20/03/2025 à 19:51, Jim Cromie a écrit :
dynamic-debug has several __sections, each with ,
num_, and it iterates over these with a 2-index for-loop.
These loops are fiddly with the 2 names.
We have only 2 such loops now, but are getting more soon; lets
embed/abstract the fiddlyness in the
Le 20/03/2025 à 19:51, Jim Cromie a écrit :
Classmaps are stored in an elf section/array, but currently are
individually list-linked onto dyndbg's per-module ddebug_table for
operation. This is unnecessary.
Just like dyndbg's descriptors, classes are packed in compile order;
so even with many
On Mon, Mar 24, 2025 at 10:34 AM Sergey Kovalenko
wrote:
>
> Hello Alex!
>
> "If there are a lot of requests to toggle gfxoff, the worker thread to
> allow it again gets
> cancelled and scheduled again, extending the time it's disallowed." -
> That's true except one thing:
> cancelling and schedul
Applied. Thanks.
Alex
On Sat, Mar 22, 2025 at 3:03 PM Alexandre Demers
wrote:
>
> While I don't think it is being used anywhere, if it were used, it would
> be wrong. We can base this assumption on MAX_POWER_MASK, where the shift is
> by 16 bits.
>
> Signed-off-by: Alexandre Demers
> ---
> dr
[AMD Official Use Only - AMD Internal Distribution Only]
Reviewed-by: Hawking Zhang
Regards,
Hawking
-Original Message-
From: Liu, Xiang(Dean)
Sent: Monday, March 24, 2025 22:14
To: amd-gfx@lists.freedesktop.org
Cc: Zhang, Hawking ; Wang, Yang(Kevin)
; Zhou1, Tao ; Chai, Thomas
; Yang
virtio_gpu has 10 DRM_UT_CORE debugs, make them controllable when
CONFIG_DRM_USE_DYNAMIC_DEBUG=y by telling dyndbg that the module has
class'd debugs.
Signed-off-by: Jim Cromie
---
drivers/gpu/drm/virtio/virtgpu_drv.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/gpu/drm/virtio/v
The gud driver has a number of DRM_UT_* debugs, make them
controllable when CONFIG_DRM_USE_DYNAMIC_DEBUG=y by telling dyndbg
that the module uses them.
Signed-off-by: Jim Cromie
---
drivers/gpu/drm/gud/gud_drv.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/drivers/gpu/drm/gud/gud_drv.c
[AMD Official Use Only - AMD Internal Distribution Only]
From: Likun Gao
Transfer to use function amdgpu_ip_map_init to map ip instance for
aqua_vanjaram instead of operation on different ASIC.
Signed-off-by: Likun Gao
Reviewed-by: Hawking Zhang
---
drivers/gpu/drm/amd/amdgpu/aqua_vanjaram.
[AMD Official Use Only - AMD Internal Distribution Only]
Hi Tomasz,
We can only have zero rpm on/off setting.
After confirmation, the fan stop temperature setting is not exposed to algin
windows driver.
So the fw always pick the default value. I know it may be not reasonable but it
is the truth.
Update ta ra block to keep sync with RAS TA.
Signed-off-by: Stanley.Yang
---
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 1 +
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 7 +++
drivers/gpu/drm/amd/amdgpu/ta_ras_if.h | 3 +++
3 files changed, 11 insertions(+)
diff --git a/drivers/gpu/drm/amd/am
[AMD Official Use Only - AMD Internal Distribution Only]
From: Likun Gao
IP instance map init function can be an common function instead of operation on
different ASIC.
V2: Create amdgpu_ip.[ch] file for ip related functions.
Signed-off-by: Likun Gao
Reviewed-by: Hawking Zhang
Reviewed-by: L
Ping?
On Tue, Mar 18, 2025 at 9:16 AM Alex Deucher wrote:
>
> Ping?
>
> On Thu, Mar 6, 2025 at 10:54 AM Alex Deucher
> wrote:
> >
> > Describes what debugfs files are available and what
> > they are used for.
> >
> > v2: fix some typos (Mark Glines)
> > v3: Address comments from Siqueira and Ke
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 KFD_MAX_NUM_OF_QUEUES_PER_PR
Compute partition DPX is possible in NPS2 mode. Update the compatible
modes for DPX.
Signed-off-by: Lijo Lazar
---
drivers/gpu/drm/amd/amdgpu/aqua_vanjaram.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/aqua_vanjaram.c
b/drivers/gpu/drm/amd/am
[AMD Official Use Only - AMD Internal Distribution Only]
Reviewed-by: Tao Zhou
> -Original Message-
> From: amd-gfx On Behalf Of Lijo Lazar
> Sent: Monday, March 24, 2025 3:59 PM
> To: amd-gfx@lists.freedesktop.org
> Cc: Zhang, Hawking ; Deucher, Alexander
>
> Subject: [PATCH] drm/amdg
[AMD Official Use Only - AMD Internal Distribution Only]
Reviewed-by: Asad Kamal
Thanks & Regards
Asad
-Original Message-
From: amd-gfx On Behalf Of Lijo Lazar
Sent: Monday, March 24, 2025 1:29 PM
To: amd-gfx@lists.freedesktop.org
Cc: Zhang, Hawking ; Deucher, Alexander
Subject: [PAT
Hello,
I want to prepend that I've never done kernel programming and am not familiar
with the amdgpu driver. That said, I read through the issues on Mesa regarding
its behavior when RADV runs out of VRAM [1] and the related information.
Thereby I stumbled upon Christian König mentioning that it
There is a reference couting / lifecycle issue with drm_connector when used
with a USB-C dock. The problem has been previously reproduced on both Intel and
AMD GPUs.
On both Intel and AMD, the symptoms are:
- multiple connectors being listed in sysfs `sys/class/drm/cardX/` (because
the old
On Fri, 2025-03-21 at 16:58 +0100, Christian König wrote:
> Sometimes drivers need to be able to submit multiple jobs which
> depend on
> each other to different schedulers at the same time, but using
> drm_sched_job_add_dependency() can't fail any more after the first
> job is
> initialized.
>
>
Predict an optimal delay to enable GFXOFF for the next interval
based on the request count:
- less than 15 requests per second - zero delay
- less than 25 requests per second - default delay
- 25 and more requests per second - don't enable GFXOFF
The algorithm allows maintaining low power consump
The problem occurs in this part of ttm_tt_populate(), in the nokaslr case
the loop is entered and repeatedly run because ttm_dma32_pages allocated exceeds
the ttm_dma32_pages_limit which leads to lots of calls to ttm_global_swapout().
if (!strcmp(get_current()->comm, "stellaris"))
printk(K
On 21/03/2025 15:58, Christian König wrote:
Sometimes drivers need to be able to submit multiple jobs which depend on
each other to different schedulers at the same time, but using
drm_sched_job_add_dependency() can't fail any more after the first job is
initialized.
This function preallocate
On 21/03/2025 15:58, Christian König wrote:
For the unlikely case that we ran into an ENOMEM while fixing up the gang
submission dependencies we can't clean up any more since the gang
members are already armed.
Fix this by using pre-allocated dependency slots and re-ordering the
code, also fix
From: Dmitry Baryshkov
Switch drm_dp_helper.c to use new set of DPCD read / write helpers.
Reviewed-by: Lyude Paul
Acked-by: Jani Nikula
Signed-off-by: Dmitry Baryshkov
---
drivers/gpu/drm/display/drm_dp_helper.c | 296 +---
1 file changed, 116 insertions(+), 180
From: Dmitry Baryshkov
Switch drm_dp_cec.c to use new set of DPCD read / write helpers.
Reviewed-by: Lyude Paul
Acked-by: Jani Nikula
Signed-off-by: Dmitry Baryshkov
---
drivers/gpu/drm/display/drm_dp_cec.c | 37 ++--
1 file changed, 18 insertions(+), 19 delet
Existing DPCD access functions return an error code or the number of
bytes being read / write in case of partial access. However a lot of
drivers either (incorrectly) ignore partial access or mishandle error
codes. In other cases this results in a boilerplate code which compares
returned value with
From: Dmitry Baryshkov
Switch drm_dp_tunnel.c to use new set of DPCD read / write helpers.
Reviewed-by: Lyude Paul
Acked-by: Jani Nikula
Signed-off-by: Dmitry Baryshkov
---
drivers/gpu/drm/display/drm_dp_tunnel.c | 20 ++--
1 file changed, 10 insertions(+), 10 deletions(-)
d
[Public]
It seems there still requires implementing the gfx11/gfx12 pipe reset from the
project perspective, so please help review the patch series.
Regards,
Prike
> -Original Message-
> From: amd-gfx On Behalf Of Liang,
> Prike
> Sent: Monday, March 3, 2025 7:23 PM
> To: Alex De
From: Dmitry Baryshkov
drm_dp_dpcd_read_link_status() follows the "return error code or number
of bytes read" protocol, with the code returning less bytes than
requested in case of some errors. However most of the drivers
interpreted that as "return error code in case of any error". Switch
drm_dp
[AMD Official Use Only - AMD Internal Distribution Only]
Reviewed-by: Tao Zhou
> -Original Message-
> From: amd-gfx On Behalf Of
> Stanley.Yang
> Sent: Tuesday, March 25, 2025 12:06 PM
> To: amd-gfx@lists.freedesktop.org
> Cc: Yang, Stanley
> Subject: [PATCH Review 1/1] drm/amdgpu: Upd
[AMD Official Use Only - AMD Internal Distribution Only]
Series is
Reviewed-by: Hawking Zhang
Regards,
Hawking
-Original Message-
From: amd-gfx On Behalf Of Gao, Likun
Sent: Tuesday, March 25, 2025 12:33
To: amd-gfx list
Cc: Gao, Likun
Subject: [PATCH 1/2] drm/amdgpu: make ip map ini
87 matches
Mail list logo
