This is the mail archive of the
mailing list for the glibc project.
Re: [PATCH] hppa: Optimize atomic_compare_and_exchange_val_acq
- From: Helge Deller <deller at gmx dot de>
- To: Carlos O'Donell <carlos at redhat dot com>, John David Anglin <dave dot anglin at bell dot net>, GNU C Library <libc-alpha at sourceware dot org>
- Cc: "deller at kernel dot org" <deller at kernel dot org>, Mike Frysinger <vapier at gentoo dot org>, Aurelien Jarno <aurelien at aurel32 dot net>
- Date: Thu, 22 Sep 2016 23:38:06 +0200
- Subject: Re: [PATCH] hppa: Optimize atomic_compare_and_exchange_val_acq
- Authentication-results: sourceware.org; auth=none
- References: <58B70052-B987-4C41-B603-F3AAB2FDE34B@bell.net> <email@example.com>
On 22.09.2016 22:37, Carlos O'Donell wrote:
> On 09/22/2016 10:14 AM, John David Anglin wrote:
>> The attached patch replaces the conditional branch tests in
>> atomic_compare_and_exchange_val_acq with conditional instruction
>> nullification. This avoids the stalls associated with conditional
>> branches and the resulting code is shorter. There are no branches in
>> the fast path when the operation is successful.
> Does this really make a measurable difference? The light-weight-syscall
> is probably the most costly part of this entire operation.
> If you can show there is a measurable difference I would be willing
> to accept the removal of the deadlock looping (it becomes a SIGILL
> and you have to look at the core file).
I'm with Dave that the removal of the deadlock looping is OK.
There is no production kernel in real-life (e.g. Debian kernels) which
return EDEADLOCK, because in the Linux kernel we have completely disabled
the LWS debugging by default (ENABLE_LWS_DEBUG=0) and as such I see
no kernel code path where EDEADLOCK can be returned.