All of lore.kernel.org
 help / color / mirror / Atom feed
From: Jocelyn Falempe <jfalempe@redhat.com>
To: Nicolas Baranger <nicolas.baranger@3xo.fr>,
	Thomas Zimmermann <tzimmermann@suse.de>
Cc: dri-devel@lists.freedesktop.org, airlied@redhat.com,
	Maarten Lankhorst <maarten.lankhorst@linux.intel.com>,
	Maxime Ripard <mripard@kernel.org>,
	David Airlie <airlied@gmail.com>, Simona Vetter <simona@ffwll.ch>,
	linux-kernel@vger.kernel.org
Subject: Re: Include ASPEED ast-drm 1.15.1 video driver in kernel tree
Date: Fri, 14 Feb 2025 10:11:22 +0100	[thread overview]
Message-ID: <cd7a9908-d4ba-45ca-a5cb-de8ac7ef72d0@redhat.com> (raw)
In-Reply-To: <984c317de1027f5886390a65f1f66126@3xo.fr>

On 13/02/2025 10:27, Nicolas Baranger wrote:
> Dear Thomas
> 
> Thanks for answer and help.
> 
> Yes, due to .date total removal in linux 6.14 (https://github.com/ 
> torvalds/linux/commit/cb2e1c2136f71618142557ceca3a8802e87a44cd <https:// 
> github.com/torvalds/linux/commit/ 
> cb2e1c2136f71618142557ceca3a8802e87a44cd>) the last DKMS sources are :
> https://xba.soartist.net/ast-drm_nba_20250211/nba-dkms/ 
> nba_last_src_20250212/src/ <https://xba.soartist.net/ast- 
> drm_nba_20250211/nba-dkms/nba_last_src_20250212/src/>
> 
> You can also find this sources in directory drivers/gpu/drm/ast_new of 
> the tarball https://xba.soartist.net/ast-drm_nba_20250211/nba-kernel/ 
> linux-6.14.0.1-ast1.15.1-rc2_nba0_20250212.tar.gz <https:// 
> xba.soartist.net/ast-drm_nba_20250211/nba-kernel/linux-6.14.0.1- 
> ast1.15.1-rc2_nba0_20250212.tar.gz>
> 
> I'm surprised by the fact the in-kernel driver 0.1.0 is more advanced 
> than Aspeed version 1.15.1 because on my system it has very poor 
> rendering and is very slow, twinkle is high and had poor colors.
> The screen flickering is high and it's like if I was using a very old 
> cathode ray tube monitor (In fact I'm using a SAMSUNG LCD monitor which 
> is perfectly functionnal and which display a nice and eyes confortable 
> picture when using ast 1.15.1 driver or the video output of the Nvidia 
> GPU ).
> 
> 
> My testing system is a test Xeon server with an AST2400 BMC with its AST 
> VGA card as the main video output (to be able to have a screen on the 
> BMC KVM) +a discrete NVIDIA GPU I'm using for GPGPU and 3D rendering 
> with Nvidia prime render offload.
> What I constat with embed kernel driver 0.1.0 is that the Xeon processor 
> is doing the video job for example when watching a video, and it's not 
> the case with version 1.15.1 even when displaying on the AST VGA card a 
> vulkan rotating cube (compute by nvidia GPU with nvidia prime but 
> display by the AST VGA card of the AST2400).
> Note that with in-kernel version 0.1.0 it's nearly impossible to make 
> round the vulkan cube at more than half a round by  second where it's 
> working (very) fine for a 32MB video memory card with version 1.15.1 as 
> you can see in the video present in the online directory
> 
> I'm not developer or kernel developer so be sure that I wouldn't have 
> done all this work if the in-kernel ast version 0.1.0 was usable out-of- 
> the-box
> 
> Sure you can give me a patch I will test on this server (building 
> mainline+ast_new yesterday tooks 19 minutes on this server)
> 
> PS:
> here is a 'git diff linux-6.14.0.1-ast-rc2/drivers/gpu/drm/ast 
> linux-6.14.0.1-ast-rc2/drivers/gpu/drm/ast_new'
> https://xba.soartist.net/ast-drm_nba_20250211/nba-dump/ast- 
> fullpatch.patch <https://xba.soartist.net/ast-drm_nba_20250211/nba-dump/ 
> ast-fullpatch.patch>
> Diff is about 250+ kb so the 2 drivers seems to have nothing to do with 
> each others...
> 
> Thanks again for help
> 
> Kind regards
> Nicolas
> 
> 
> Le 2025-02-13 08:57, Thomas Zimmermann a écrit :
> 
>> Hi Nicolas
>>
>> Am 12.02.25 um 19:58 schrieb Nicolas Baranger:
>>> Dear maintener
>>
>> That's mostly me and Jocelyn.
>>
>>>
>>> I did include ast-drm driver version 1.15.1 (in replacement of 
>>> version 0.1.0) on the new mainline kernel too (6.14.0-rc2) and I 
>>> issue a new dkms patch
>>>
>>> Last DKMS patch had been sucessfully tested on mainline.
>>> And last ast.ko version 1.15.1 included in linux tree had also been 
>>> sucessfully tested
>>>
>>> Online directory is updated with :
>>> - new DKMS patch
>>> - new DKMS srouces
>>> - new DKMS debian package
>>> - new tarball of mainline included ast_new ported in kernel tree
>>> - new kernel debian package (mainline with ast_new)
>>>
>>>
>>> NB: online directory is here: https://xba.soartist.net/ast- 
>>> drm_nba_20250211/ <https://xba.soartist.net/ast-drm_nba_20250211/>
>>>
>>> Please let me know what I should do to see this change in linux-next
>>
>> I'm having a little trouble with figuring out which of the many driver 
>> sources is the relevant one. Am I correct to assume it's the one at
>>
>> https://xba.soartist.net/ast-drm_nba_20250211/nba-dkms/ 
>> nba_last_src_20250212/src/ <https://xba.soartist.net/ast- 
>> drm_nba_20250211/nba-dkms/nba_last_src_20250212/src/>
>>
>>
>> About that driver: Although the official driver reports an ancient 
>> version number, it is an up-to-date driver. It is actually more up-to- 
>> date than Aspeed's package. Both drivers share source code and a few 
>> years ago there was an effort to bring the kernel's driver up to the 
>> same feature set. Since then, the kernel's driver has been updated, 
>> reworked and improved.
>>
>> About the performance: From what I can tell, the only significant 
>> difference in these drivers is memory management. Your ast_new driver 
>> uses an older algorithm that we replaced quite a few releases ago. The 
>> old version was unreliable on systems with little video memory, so we 
>> had to replace it.  I don't know why the new code should be slower though.

Regarding the performances of ast driver, I remember doing profiling 
some times ago, and when running glxgears (with llvmpipe), 65% of the 
CPU time was wasted in page fault 
(https://elixir.bootlin.com/linux/v6.13.2/source/drivers/gpu/drm/drm_gem_shmem_helper.c#L534)
But as this driver is mostly used for console/basic desktop usage, I 
didn't investigate more.

If I remember correctly, the switch to shmem, is because some devices 
have only 16MB of memory, and 1920x1200x32bits takes ~9MB, so it's not 
possible to have double buffering in this case. (And this is required by 
most desktop environment).

The switch to shmem was done with "f2fa5a99ca81c drm/ast: Convert ast to 
SHMEM", and introduced in v6.2. So maybe if you can try with a v6.1 
kernel, using the built-in ast driver and report if it has better 
performances.

Best regards,

-- 

Jocelyn

>>
>> If I give you a patch against a recent Linux kernel, are you capable 
>> of building the patched kernel and testing that change on your system?
>>
>> Best regards
>> Thomas
>>
>>
>>>
>>> Thanks for help
>>>
>>> Kind regards
>>> Nicolas Baranger
>>>
>>>
>>> Le 2025-02-11 19:15, Nicolas Baranger a écrit :
>>>
>>>> Dear maintener
>>>>
>>>> For my own usage, I did make work the ASPEED ast-drm 1.15.1 video 
>>>> driver on mainline kernel (6.13.0 + 6.13.1).
>>>>
>>>> ASPEED video driver is availiable here:
>>>> https://www.aspeedtech.com/file/support/Linux_DRM_1.15.1_4.tar.gz 
>>>> <https://www.aspeedtech.com/file/support/Linux_DRM_1.15.1_4.tar.gz>
>>>>
>>>> But it only work for LTS kernel
>>>> So I modify the DKMS package and I build a new Debian DKMS package 
>>>> with the adapted  source.
>>>> My patch can be find here :
>>>> https://xba.soartist.net/ast-drm_nba_20250211/nba-dkms/astdiff.patch 
>>>> <https://xba.soartist.net/ast-drm_nba_20250211/nba-dkms/astdiff.patch>
>>>> See the README:
>>>> https://xba.soartist.net/ast-drm_nba_20250211/nba-dkms/README 
>>>> <https://xba.soartist.net/ast-drm_nba_20250211/nba-dkms/README>
>>>>
>>>> Using this new 'ast 1.15.1' driver, performance are amazing compared 
>>>> to the 'ast' driver include in kernel tree, specially when using a 
>>>> discrete GPU and offloading VULKAN / 3D on it but using AST VGA card 
>>>> as the main video card and as the main and only video output (the 
>>>> discrete GPU is used only for offloading 3D or for cuda/opencl)
>>>>
>>>> So to make things easier, I include the new 'ast 1.15.1' driver in 
>>>> kernel tree as AST_NEW : linux-6.13.1-ast/drivers/gpu/drm/ast_new'
>>>> It's working fine as you can see on this video :
>>>> https://xba.soartist.net/ast-drm_nba_20250211/ 
>>>> vulcan_nvidia_prime_render_offload_on_ast_vga_card.webm <https:// 
>>>> xba.soartist.net/ast-drm_nba_20250211/ 
>>>> vulcan_nvidia_prime_render_offload_on_ast_vga_card.webm> I upload 
>>>> all the work I've done here :
>>>> https://xba.soartist.net/ast-drm_nba_20250211/ <https:// 
>>>> xba.soartist.net/ast-drm_nba_20250211/>
>>>>
>>>> See the global README :
>>>> https://xba.soartist.net/ast-drm_nba_20250211/README <https:// 
>>>> xba.soartist.net/ast-drm_nba_20250211/README>
>>>>
>>>> and the README in nba-kernel sub-directory :
>>>> https://xba.soartist.net/ast-drm_nba_20250211/nba-kernel/README 
>>>> <https://xba.soartist.net/ast-drm_nba_20250211/nba-kernel/README>
>>>>
>>>> I'm not a developer so please let me know if I made the things the 
>>>> right way and if this new 'ast 1.15.1' driver can be ported to 
>>>> linux-next or linux-? ?
>>>> If you need more explanations, do not hesitate to contact me, I 
>>>> would be happy to help
>>>>
>>>> Kind regards
>>>> Nicolas Baranger


  parent reply	other threads:[~2025-02-14  9:11 UTC|newest]

Thread overview: 26+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-02-11 18:15 Include ASPEED ast-drm 1.15.1 video driver in kernel tree Nicolas Baranger
2025-02-12 18:58 ` Nicolas Baranger
2025-02-12 19:14   ` Maarten Lankhorst
2025-02-13  7:57   ` Thomas Zimmermann
2025-02-13  9:27     ` Nicolas Baranger
2025-02-13  9:36       ` Nicolas Baranger
2025-02-14  9:11       ` Jocelyn Falempe [this message]
2025-02-14 12:09         ` Nicolas Baranger
2025-02-14 12:36         ` Thomas Zimmermann
2025-02-14 15:01           ` Nicolas Baranger
2025-02-14 17:03             ` Nicolas Baranger
2025-02-14 17:52               ` Nicolas Baranger
2025-02-17  8:11                 ` Thomas Zimmermann
2025-02-17  8:37                   ` Nicolas Baranger
2025-02-21 11:57                     ` Nicolas Baranger
2025-02-24  8:53       ` Jani Nikula
2025-02-24  9:57         ` Nicolas Baranger
2025-02-28 10:06 ` Thomas Zimmermann
2025-03-03  8:07   ` Nicolas Baranger
2025-03-03  8:19     ` Thomas Zimmermann
2025-03-03  8:44       ` Nicolas Baranger
2025-03-03 10:16         ` Thomas Zimmermann
2025-03-03 15:48           ` Nicolas Baranger
2025-03-15 10:33           ` Nicolas Baranger
  -- strict thread matches above, loose matches on Subject: below --
2025-02-14 12:46 Thomas Zimmermann
2025-02-14 15:10 ` Nicolas Baranger

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=cd7a9908-d4ba-45ca-a5cb-de8ac7ef72d0@redhat.com \
    --to=jfalempe@redhat.com \
    --cc=airlied@gmail.com \
    --cc=airlied@redhat.com \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=maarten.lankhorst@linux.intel.com \
    --cc=mripard@kernel.org \
    --cc=nicolas.baranger@3xo.fr \
    --cc=simona@ffwll.ch \
    --cc=tzimmermann@suse.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.