linux-kernel - Re: [PATCH v4] introduce macro spin_event

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: <1236876609.5090.934.camel@laptop>
Date:	Thu, 12 Mar 2009 17:50:09 +0100
From:	Peter Zijlstra <peterz@...radead.org>
To:	Timur Tabi <timur@...escale.com>
Cc:	Grant Likely <grant.likely@...retlab.ca>,
	Benjamin Herrenschmidt <benh@...nel.crashing.org>,
	Alan Cox <alan@...rguk.ukuu.org.uk>,
	linux-kernel@...r.kernel.org, rdreier@...co.com,
	jirislaby@...il.com, will.newton@...il.com, hancockrwd@...il.com,
	jeremy@...p.org
Subject: Re: [PATCH v4] introduce macro spin_event_timeout()

On Thu, 2009-03-12 at 11:19 -0500, Timur Tabi wrote:
> Grant Likely wrote:
> 
> > #define spin_until_timeout(condition, timeout, rc)          \
> >       for (unsigned long __timeout = jiffies + (timeout);     \
> >               (!(rc = (condition)) && time_after(jiffies, __timeout)); )
> 
> Ooo, that's good.
> 
> I'm still not crazy about using jiffies, since it doesn't get
> incremented when interrupts are disabled, and I'd like this function to
> work in those cases.  How about get_cycles()?  I know it's not supported
> on all platforms, but I'm willing to live with that.
> 
> The other problem with get_cycles() is that there doesn't appear to be a
> num_cycles_per_usec() function, so there's no way for me to scale the
> count to a fixed time period.

sched_clock() does that, but:
  - it falls back to jiffies on poor platforms
  - it requires to be called with IRQs disabled
  - it can basically jump any random way on funky hardware

then there is cpu_clock(int cpu):
  - still falls back to jiffies on poor platforms
  - is monotonic when used on the same cpu
  - can drift up to a few jiffies when used between cpus

But something that seems to always work, is simply count loops and rely
on whatever delay is in the specified loop.

#define spin_until_timeout(condition, timeout, rc) \
	for (unsigned long __timeout = 0; \
	     !(rc = (condition)) && __timeout < (timeout); \
	     __timeout++)

--
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/