linux-kernel - Re: [PATCH] accel/ivpu: Add handling of VPU_JSM_STATUS_MVNCI_CONTEXT_VIOLATION

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <80f49ba8-caea-47d5-be38-dd1eefd09988@linux.intel.com>
Date: Thu, 24 Apr 2025 12:22:31 +0200
From: Jacek Lawrynowicz <jacek.lawrynowicz@...ux.intel.com>
To: Greg KH <gregkh@...uxfoundation.org>
Cc: linux-kernel@...r.kernel.org, stable@...r.kernel.org,
 Karol Wachowski <karol.wachowski@...el.com>
Subject: Re: [PATCH] accel/ivpu: Add handling of
 VPU_JSM_STATUS_MVNCI_CONTEXT_VIOLATION_HW

Hi,

On 4/22/2025 2:17 PM, Greg KH wrote:
> On Tue, Apr 08, 2025 at 11:57:11AM +0200, Jacek Lawrynowicz wrote:
>> From: Karol Wachowski <karol.wachowski@...el.com>
>>
>> commit dad945c27a42dfadddff1049cf5ae417209a8996 upstream.
>>
>> Trigger recovery of the NPU upon receiving HW context violation from
>> the firmware. The context violation error is a fatal error that prevents
>> any subsequent jobs from being executed. Without this fix it is
>> necessary to reload the driver to restore the NPU operational state.
>>
>> This is simplified version of upstream commit as the full implementation
>> would require all engine reset/resume logic to be backported.
> 
> We REALLY do not like taking patches that are not upstream.  Why not
> backport all of the needed patches instead, how many would that be?
> Taking one-off patches like this just makes it harder/impossible to
> maintain the code over time as further fixes in this same area will NOT
> apply properly at all.
> 
> Think about what you want to be touching 5 years from now, a one-off
> change that doesn't match the rest of the kernel tree, or something that
> is the same?

Sure, I'm totally on board with backporting all required patches.
I thought it was not possible due to 100 line limit.

This would be the minimum set of patches:

Patch 1:
 drivers/accel/ivpu/ivpu_drv.c   | 32 +++-----------
 drivers/accel/ivpu/ivpu_drv.h   |  2 +
 drivers/accel/ivpu/ivpu_job.c   | 78 ++++++++++++++++++++++++++-------
 drivers/accel/ivpu/ivpu_job.h   |  1 +
 drivers/accel/ivpu/ivpu_mmu.c   |  3 +-
 drivers/accel/ivpu/ivpu_sysfs.c |  5 ++-
 6 files changed, 75 insertions(+), 46 deletions(-)

Patch 2:
 drivers/accel/ivpu/ivpu_job.c | 15 ++++++---------
 1 file changed, 6 insertions(+), 9 deletions(-)

Patch 3:
 drivers/accel/ivpu/ivpu_job.c     |   2 +-
 drivers/accel/ivpu/ivpu_jsm_msg.c |   3 +-
 drivers/accel/ivpu/vpu_boot_api.h |  45 +++--
 drivers/accel/ivpu/vpu_jsm_api.h  | 303 +++++++++++++++++++++++++-----
 4 files changed, 293 insertions(+), 60 deletions(-)

Patch 4:
 drivers/accel/ivpu/ivpu_job.c | 27 ++++++++++++++++++++++++++-
 1 file changed, 26 insertions(+), 1 deletion(-)

First patch needs some changes to apply correctly to 6.12 but the rest of them apply pretty cleanly.
Is this acceptable?

Regards,
Jacek