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: <20230324142820.61e4f0b6@kernel.org>
Date:   Fri, 24 Mar 2023 14:28:20 -0700
From:   Jakub Kicinski <kuba@...nel.org>
To:     Alexander Duyck <alexander.duyck@...il.com>
Cc:     davem@...emloft.net, netdev@...r.kernel.org, edumazet@...gle.com,
        pabeni@...hat.com, willemb@...gle.com
Subject: Re: [PATCH net-next 1/3] net: provide macros for commonly copied
 lockless queue stop/wake code

On Fri, 24 Mar 2023 08:45:23 -0700 Alexander Duyck wrote:
> On Thu, Mar 23, 2023 at 8:09 PM Jakub Kicinski <kuba@...nel.org> wrote:
> > > We may want to change the values here. The most likely case is "left
> > > enabled" with that being the case we probably want to make that the 0
> > > case. I would then probably make 1 the re-enabled case and -1 the
> > > stopped case.  
> >
> > I chose the return values this way because the important information is
> > whether the queue was in fact stopped (in case the macro is used at the
> > start of .xmit as a safety check). If stopped is zero caller can check
> > !ret vs !!ret.
> >
> > Seems pretty normal for the kernel function called stop() to return 0
> > if it did stop.  
> 
> Except this isn't "stop", this is "maybe stop".

So the return value from try_stop and maybe_stop would be different?
try_stop needs to return 0 if it stopped - the same semantics as
trylock(), AFAIR. Not that I love those semantics, but it's a fairly
strong precedent.

> Maybe we should just
> do away with the stop/wake messaging and go with something such as a
> RTS/CTS type setup. Basically this function is acting as a RTS to
> verify that we have room on the ring to place the frame. If we don't
> we are stopped. The "wake" function is on what is essentially the
> receiving end on the other side of the hardware after it has DMAed the
> frames and is providing the CTS signal back.

I'm definitely open to different naming but wouldn't calling RTS _after_
send be even more confusing than maybe_stop?

> > > With that the decision tree becomes more straightforward as we would do
> > > something like:
> > >       if (result) {
> > >               if (result < 0)
> > >                       Increment stopped stat
> > >                       return
> > >               else
> > >                       Increment restarted stat
> > >       }  
> >
> > Do you see a driver where it matters? ixgbe and co. use
> > netif_tx_queue_try_stop() and again they just test stopped vs not stopped.  
> 
> The thing is in order to make this work for the ixgbe patch you didn't
> use the maybe_stop instead you went with the try_stop. If you replaced
> the ixgbe_maybe_stop_tx with your maybe stop would have to do
> something such as the code above to make it work. That is what I am
> getting at. From what I can tell the only real difference between
> ixgbe_maybe_stop_tx and your maybe_stop is that you avoided having to
> move the restart_queue stat increment out.

I can convert ixgbe further, true, but I needed the try_stop, anyway,
because bnxt does:

if (/* need to stop */) {
	if (xmit_more())
		flush_db_write();
	netif_tx_queue_try_stop();
}

which seems reasonable.

> The general thought is I would prefer to keep it so that 0 is the
> default most likely case in both where the queue is enabled and is
> still enabled. By moving the "take action" items into the 1/-1 values
> then it becomes much easier to sort them out with 1 being a stat
> increment and -1 being an indication to stop transmitting and prep for
> watchdog hang if we don't clear this in the next watchdog period.

Maybe worth taking a step back - the restart stat which ixgbe
maintains made perfect sense when you pioneered this approach but
I think we had a decade of use, and have kprobes now, so we don't
really need to maintain a statistic for a condition with no impact 
to the user? New driver should not care 1 vs -1..

> Also in general it makes it easier to understand if these all work
> with the same logic.
> 
> > > In addition for readability we may want consider adding an enum simliar
> > > to the netdev_tx enum as then we have the return types locked and
> > > usable should we want to specifically pick out one.  
> >
> > Hm, I thought people generally dislike the netdev_tx enum.
> > Maybe it's just me.  
> 
> The thought I had with the enum is to more easily connect the outcomes
> with the sources. It would also help to prevent any confusion on what
> is what. Having the two stop/wake functions return different values is
> a potential source for errors since 0/1 means different things in the
> different functions. Basically since we have 3 possible outcomes using
> the enum would make it very clear what the mapping is between the two.

IMO only two outcomes matter in practice (as mentioned above).
I really like the ability to treat the return value as a bool, if only
we had negative zero we would have a perfect compromise :(

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ