linux-kernel - Re: [PATCH 5/7] mmc_spi: allow higher timeouts for SPI mode

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: <20090311154605.GD1475@console-pimps.org>
Date:	Wed, 11 Mar 2009 15:46:05 +0000
From:	Matt Fleming <matt@...sole-pimps.org>
To:	Wolfgang Mües <wolfgang.mues@...rswald.de>
Cc:	Pierre Ossman <drzeus@...eus.cx>,
	Andrew Morton <akpm@...ux-foundation.org>,
	David Brownell <dbrownell@...rs.sourceforge.net>,
	Mike Frysinger <vapier.adi@...il.com>,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH 5/7] mmc_spi: allow higher timeouts for SPI mode

On Wed, Mar 11, 2009 at 03:55:00PM +0100, Wolfgang Mües wrote:
> 
> My patch 6 in mmc_spi_skip() is doing a busy-wait for a short while ( less 
> than 1 jiffie), and starts to call schedule() inside the loop if the card is 
> slower.
> 

OK, but if my machine runs at 100 HZ then a jiffie is 10ms. Previously
(without your patch) we waited for 300ms in the write case and 100ms in
the read case. So, it's not unreasonable to think that a response is
going to take more than 10ms. With your patch we're almost always going
to schedule() with no indication of exactly when we're going to come
back.

> My goal was to avoid the long-lasting busy waiting. I have measured times up 
> to 900ms! With my patch, the longest busy waiting will be 1 jiffie. 
> 

I agree that busy-waiting for 900ms would be a bit mad. Is there a
reason that you didn't implement this with msleep() as was noted in the
comment above the timeout?


                /* REVISIT investigate msleep() to avoid busy-wait I/O
                 * in at least some cases.
                 */


> And yes, if the SD card is sending its response after 5 jiffies, it is 
> recognized only after the scheduler schedules this process, which will incure 
> a delay to the data transfer. The amount of delay is determined by the number 
> of running processes and the number of HZ.
> 

Have you benchmarked this case? Do you know approximately how long it
is before we return from the schedule() under various workloads?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/