Message-Id: <EB8EC5FD-AB6C-48C3-8980-65E8CB444BDF@linux.microsoft.com>
Date: Tue, 3 Sep 2024 11:06:53 -0700
From: Allen Pais <apais@...ux.microsoft.com>
To: Takashi Sakamoto <o-takashi@...amocchi.jp>
Cc: linux1394-devel@...ts.sourceforge.net,
 Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
 linux-sound@...r.kernel.org,
 edmund.raile@...tonmail.com,
 linux-media@...r.kernel.org,
 netdev@...r.kernel.org
Subject: Re: [RFT][PATCH 0/5] firewire: use sleepable workqueue to handle 1394
 OHCI IT/IR context events



>This series of change is inspired by BH workqueue available in recent
>kernel.

>In Linux FireWire subsystem, tasklet softIRQ context has been utilized to
>operate 1394 OHCI Isochronous Transmit (IT) and Isochronous Receive (IR)
>contexts. The tasklet context is not preferable, as you know.

>I have already received a proposal[1][2] to replace the usage of tasklet
>with BH workqueue. However, the proposal includes bare replacement for 1394
>OHCI IT, IR, Asynchronous Transmit (AT), and Asynchronous Receive (AR)
>contexts with neither any care of actual usage for each context nor
>practical test reports. In theory, this kind of change should be done by
>step by step with enough amount of evaluation over software design to avoid
>any disorder.

>In this series of changes, the usage of tasklet for 1394 OHCI IT/IR
>contexts is just replaced, as a first step. In 1394 OHCI IR/IT events,
>software is expected to process the content of page dedicated to DMA
>transmission for each isochronous context. It means that the content can be
>processed concurrently per isochronous context. Additionally, the content
>of page is immutable as long as the software schedules the transmission
>again for the page. It means that the task to process the content can sleep
>or be preempted. Due to the characteristics, BH workqueue is _not_ used.

>At present, 1394 OHCI driver is responsible for the maintenance of tasklet
>context, while in this series of change the core function is responsible
>for the maintenance of workqueue and work items. This change is an attempt
>to let each implementation focus on own task.

>The change affects the following implementations of unit protocols which
>operate isochronous contexts:

>- firewire-net for IP over 1394 (RFC 2734/3146)
>- firedtv
>- drivers in ALSA firewire stack for IEC 61883-1/6
>- user space applications

>As long as reading their codes, the first two drivers look to require no
>change. While the drivers in ALSA firewire stack require change to switch
>the type of context in which callback is executed. The series of change
>includes a patch for them to adapt to work process context.

>Finally, these changes are tested by devices supported by ALSA firewire
>stack with/without no-period-wakeup runtime of PCM substream. I also tested
>examples in libhinoko[3] as samples of user space applications. Currently I
>face no issue.

>On the other hand, I have neither tested for firewire-net nor firedtv,
>since I have never used these functions. If someone has any experience to
>use them, I would request to test the change.

>[1] https://lore.kernel.org/lkml/20240403144558.13398-1-apais@linux.microsoft.com/
>[2] https://github.com/allenpais/for-6.9-bh-conversions/issues/1
>[3] https://git.kernel.org/pub/scm/libs/ieee1394/libhinoko.git/


>Regards

Thank you for doing this work. You will probably need to send out a v2
As most of you patches have single line comment instead of Block style
Commnents (/* ..*/). Please have it fixed.

- Allen

>Takashi Sakamoto (5):
> firewire: core: allocate workqueue to handle isochronous contexts in
>   card
> firewire: core: add local API for work items scheduled to workqueue
>   specific to isochronous contexts
> firewire: ohci: process IT/IR events in sleepable work process to
>   obsolete tasklet softIRQ
> firewire: core: non-atomic memory allocation for isochronous event to
>   user client
> ALSA: firewire: use nonatomic PCM operation

Powered by blists - more mailing lists