Re: [RFC PATCH 0/3] Rework Power handlers

[Owen Smith <owen.smith@xxxxxxxxxx>]

The problem is relatively rare - it tends to hit occasionally during our automated tests, and is seen more often with xenfilt on other device stacks (vGPU in particular).

These patches have been applied to the XS queue for some time. I refactored and simplified the patches in Dec 2021 - these have been applied since then and we have released at least 2 versions since with no reported power failure issues. Previously this was 1 big messy patch that, I think, has been released at least once. No power related issues detected in all our automated testing or through our support.

Owen

On Thu, Feb 1, 2024 at 5:35 PM Paul Durrant <xadimgnik@xxxxxxxxx> wrote:

On 10/01/2024 13:42, Owen Smith wrote:
> Current power handler code is passed to a single threaded work thread which
> can be delayed enough to trigger 0x9F (DRIVER_POWER_STATE_FAILURE) bugchecks.
> Rework power handlers to math standard practive, using pended IRPs,
> IoCompletionRoutines and deferred IoWorkItems where required.
>
> Martin Harvey (2):
>    [XenFilt] Asynchronous power handling.
>    [XenBus] Asynchronous power handling
>

I advised Martin against these changes over a year ago. How prevalent is
the problem? It took *years* to get this, and the PnP code stable. How
much testing have these changes had?

   Paul

> Owen Smith (1):
>    ASSERT(Frame->Mdl != NULL) before dereference
>
>   src/xenbus/fdo.c    | 792 ++++++++++++++++++++++++--------------------
>   src/xenbus/gnttab.c |   1 +
>   src/xenbus/pdo.c    | 250 ++++++--------
>   src/xenfilt/fdo.c   | 498 +++++++---------------------
>   src/xenfilt/pdo.c   | 503 +++++++---------------------
>   5 files changed, 777 insertions(+), 1267 deletions(-)
>