[Qemu-devel] Re: [PATCH v6] Introduce qemu_madvise()

From: Blue Swirl <blauwirbel@gmail.com>
To: "Andreas Färber" <andreas.faerber@web.de>
Cc: Andrea Arcangeli <aarcange@redhat.com>,
	qemu-devel@nongnu.org, Alexander Graf <agraf@suse.de>
Subject: [Qemu-devel] Re: [PATCH v6] Introduce qemu_madvise()
Date: Wed, 15 Sep 2010 19:50:53 +0000	[thread overview]
Message-ID: <AANLkTin4xT97iw2Bo+xSVvgcx=JoeoYh62DifzQPw4Ka@mail.gmail.com> (raw)
In-Reply-To: <15F942FE-97D7-4747-A80E-219FF54DDB04@web.de>

On Wed, Sep 15, 2010 at 7:35 PM, Andreas Färber <andreas.faerber@web.de> wrote:
> Am 15.09.2010 um 21:00 schrieb Blue Swirl:
>
>> madvise() actually
>> returns 0 or -1 with error code in errno. Should we try to match that?
>
> posix_madvise() doesn't seem to... "otherwise, an error number shall be
> returned to indicate the error"

If we match posix_madvise(), then the wrapper function probably should
become qemu_posix_madvise() and we should prefer posix_madvise() over
madvise().

> It documents EINVAL for invalid advice, and so does madvise. Are we using
> ENOTSUP knowingly?

Let's use EINVAL.

> Maybe return -1 and set errno to the posix_madvise() return code? (but is it
> errno = -retval or errno = retval then?)

An advantage of posix_madvise() semantics is that no mucking with
errno is needed. Though existing code should be checked if they rely
on madvise() semantics instead.