This is the mail archive of the
libc-ports@sources.redhat.com
mailing list for the libc-ports project.
Re: [Patch, mips] Faster strcmp for mips
- From: Steve Ellcey <sellcey at mips dot com>
- To: Carlos O'Donell <carlos at systemhalted dot org>
- Cc: "libc-ports at sourceware dot org" <libc-ports at sourceware dot org>
- Date: Fri, 15 Nov 2013 10:52:55 -0800
- Subject: Re: [Patch, mips] Faster strcmp for mips
- Authentication-results: sourceware.org; auth=none
- References: <1384464221 dot 2484 dot 86 dot camel at ubuntu-sellcey> <CAE2sS1jDnPP2hKn6u-nnvAnKgwQrrVUsDUrGWjyUu8+EBXZjgA at mail dot gmail dot com>
On Fri, 2013-11-15 at 13:47 -0500, Carlos O'Donell wrote:
> On Thu, Nov 14, 2013 at 4:23 PM, Steve Ellcey <sellcey@mips.com> wrote:
> > This means it could be loading bytes beyond the end of the strings being
> > compared but it looks like other architecture specific strcmp functions
> > are also doing this optimization and the newlib version of strcmp also does
> > this.
>
> I thought that doing so was dangerous? I'm pretty sure we've been trying
> to fix such "load bytes beyond the end of the string" issues because you
> could have a string that straddles a page boundary with the next page
> unmapped and such an optimized routine would fault on a read from the
> unmapped page.
>
> How do you plan to fix that?
>
> Cheers,
> Carlos.
I think the assumption has been that if an aligned word contains at
least one byte of the string, it is OK to read that entire word. I
don't think a single aligned word (4 or 8 bytes depending on the
architecture) can straddle a page boundary. If the word was unaligned,
then I think it could straddle a page boundary and you could get into
trouble.
Steve Ellcey
sellcey@mips.com