Re: [Qemu-devel] [PATCH 2/2] qemu-ga: Add the guest-suspend command

[Eric Blake <eblake@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1707 bytes --]

On 01/16/2012 01:19 PM, Luiz Capitulino wrote:
>> memset(), strlen, strcpy, and friends in <string.h> are all in the class
>> of functions that I think are unintentional omissions from the list of
>> async-signal-safe functions (they don't read/modify anything but the
>> pointers passed in, so the _only_ reason I can think of why they _might_
>> have been omitted from the list is that there might be some machine
>> state that could be observably different if you were interrupted in the
>> middle of one of these operations, such as a processor flag bit when
>> using a rep prefix on x86 controlling which direction to move, but no
>> one has ever pointed me to a definitive answer to why they were omitted).
> 
> If this is right we shouldn't be using them then...

The _nice_ thing is that the functions in <string.h> are trivially
replaceable by naive variants that _are_ async-signal-safe, since the
algorithms behind them are so trivial.  It's just that it's annoying to
have to tell users that they have to write non-optimized code when doing
string ops in a signal handler or after a fork (C code tends to not be
as nice as the hand-tuned assembly in glibc for all these low-level
functions), for what so far appears to be a theoretical rather than a
confirmed restriction on why the standard does not require async-safety.

I guess it's time for me to follow through with my threat to file a bug
against the POSIX folks to get the string functions added to the list of
async-signal-safe, and/or give me stronger justification why they are
not already there.

-- 
Eric Blake   eblake@redhat.com    +1-919-301-3266
Libvirt virtualization library http://libvirt.org


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 620 bytes --]