linux-kernel - Re: [PATCH] mailbox: Allow NULL message sending

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] [day] [month] [year] [list]

Message-ID: <CABb+yY0QFgZ4htYicKTJ-vav-rbkGF-wMbMaRVze5AAeum5WFA@mail.gmail.com>
Date: Sat, 20 Dec 2025 17:38:32 -0600
From: Jassi Brar <jassisinghbrar@...il.com>
To: Joonwon Kang <joonwonkang@...gle.com>
Cc: jonathanh@...dia.com, linux-kernel@...r.kernel.org, sudeep.holla@....com, 
	thierry.reding@...il.com
Subject: Re: [PATCH] mailbox: Allow NULL message sending

On Mon, Dec 8, 2025 at 12:51 AM Joonwon Kang <joonwonkang@...gle.com> wrote:
>
> On Wed, Dec 3, 2025 at 11:57 PM Joonwon Kang <joonwonkang@...gle.com> wrote:
> > >
> > > On Tue, Nov 25, 2025 at 11:00 PM Joonwon Kang <joonwonkang@...gle.com> wrote:
> > > > >
> > > > > Clients may want to send interrupt only without any useful message
> > > > > involved. Since the current mailbox framework does not allow NULL
> > > > > message sending(although it allows receiving it), the clients should
> > > > > allocate a dummy message buffer and pretend sending it. Besides, if
> > > > > the mailbox controller calls `mbox_chan_txdone()` when the client
> > > > > drivers happen to send NULL message anyway, it will result in unexpected
> > > > > results by making the tx status messed up. This commit lifts the
> > > > > limitation and allows the clients to send interrupt only without any
> > > > > message buffer allocated.
> > > > >
> > > > Interrupts without data messages are called 'doorbells' and we already
> > > > support them.
> > > > thanks
> > >
> > > I am not sure if it is already supported. Let me draw two cases which imply
> > > that it is not supported. If the cases make sense, could you reconsider the
> > > patch? If it is supported in another branch, could you refer me to that
> > > branch? I am currently referring to the `for-next` branch of your mailbox
> > > repo.
> > >
> > I believe you are talking about some hypothetical situation?
> > Otherwise, which controller is that?
> > A controller driver is supposed to either expect data or not, but not both.
>
> I did not notice this controller's expectation since I could not find this info
> in the API doc. So, now I believe that Case 2 could be seen as a hypothetical
> situation. However, what about Case 1? If the message to send is NULL,
> `chan->cl->tx_done(chan->cl, mssg, r)` and `complete(&chan->tx_complete)` will
> **never** be called from `tx_tick()`. It also means that there is no way for a
> client to know that the sending is really done or not. Even though a controller
> driver(I mean any typical controller, not a specific controller) calls
> `mbox_chan_txdone()` after receiving the corresponding ACK interrupt from the
> remote to the previously sent NULL message, the client will be blocking not
> knowing the sending is done.
>
For non-data channels (doorbells), the driver may expect the doorbell
info at runtime
via 'mssg'. If it doesn't, the clients send '0' or any arbitrary
non-null value which remains
unused by anyone. Maybe the api can define some MAILBOX_NULL_DATA integer that
can be used instead because the controller driver anyway doesn't need it.
If/when the doorbell is rung is upto the controller - it may be just
setting a bit or observing
a status bit for the remote or whatever. The driver must say tx_done()
appropriately.

Again, it will help if we talk about the controller you have in mind exactly.

Cheers!