[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1eac57a5-eae6-4e2b-99d1-2b06c8628b1e@lunn.ch>
Date: Fri, 21 Mar 2025 14:18:53 +0100
From: Andrew Lunn <andrew@...n.ch>
To: Tobias Waldekranz <tobias@...dekranz.com>
Cc: davem@...emloft.net, kuba@...nel.org, maxime.chevallier@...tlin.com,
marcin.s.wojtas@...il.com, linux@...linux.org.uk,
edumazet@...gle.com, pabeni@...hat.com, netdev@...r.kernel.org
Subject: Re: [PATCH v2 net] net: mvpp2: Prevent parser TCAM memory corruption
On Fri, Mar 21, 2025 at 01:41:38PM +0100, Tobias Waldekranz wrote:
> On fre, mar 21, 2025 at 13:12, Andrew Lunn <andrew@...n.ch> wrote:
> >> +static int mvpp2_prs_init_from_hw_unlocked(struct mvpp2 *priv,
> >> + struct mvpp2_prs_entry *pe, int tid)
> >> {
> >> int i;
> >>
> >
> > This is called from quite a few places, and the locking is not always
> > obvious. Maybe add
>
> Agreed, that was why i chose the _unlocked suffix vs. just prefixing
> with _ or something. For sure I can add it, I just want to run something
> by you first:
>
> Originally, my idea was to just protect mvpp2_prs_init_from_hw() and
> mvpp2_prs_hw_write(). Then I realized that the software shadow of the
> SRAM table must also be protected, which is why locking had to be
> hoisted up to the current scope.
>
> > __must_hold(&priv->prs_spinlock)
> >
> > so sparse can verify the call paths ?
>
> So if we add these asserts only to the hardware access leaf functions,
> do we risk inadvertently signaling to future readers that the lock is
> only there to protect the hardware tables?
You can scatter __must_hold() anywhere you want, to indicate the lock
must be held. It has no runtime overhead.
And you can expand the comment where the mutex is defined to say what
it is expected to cover.
FYI: i've never personally used __must_hold(), but i reviewed a patch
recently using it, which made me think it might be useful here. I
don't know if you need additional markup, __acquires() & __releases()
?? You might want to deliberately break the locking and see if sparse
reports it.
Andrew
Powered by blists - more mailing lists