From: Keir Fraser <keir.xen@gmail.com>
To: Jan Beulich <JBeulich@suse.com>,
"xen-devel@lists.xensource.com" <xen-devel@lists.xensource.com>
Subject: Re: [PATCH, RFC] Re: x86: gnttab_clear_flag() abusing clear_bit()
Date: Tue, 07 Feb 2012 05:10:48 +0000 [thread overview]
Message-ID: <CB5662D8.2A8BB%keir.xen@gmail.com> (raw)
In-Reply-To: <4F310C5B0200007800071870@nat28.tlf.novell.com>
On 07/02/2012 10:34, "Jan Beulich" <JBeulich@suse.com> wrote:
>>>> On 06.02.12 at 18:06, "Jan Beulich" <JBeulich@suse.com> wrote:
>> Back in c/s 17194:af33f2054f47 bitops got restricted to 4-bytes and
>> larger operands only. gnttab_clear_flag() cheats in casting a uint16_t *
>> to unsigned long * - how is that not a problem in the context of the
>> goal that said c/s set, in particular for v2 of the interface? (Given that
>> this is being implemented as arch-specific routine - so far for reasons
>> that escape me - this should be simple to fix by using inline assembly
>> rather than clear_bit().)
>>
>> Further I just spotted one instance where the or of two bit positions
>> gets passed to this function - that's quite definitely a bug I would say.
>>
>> And, quite the opposite, __fixup_status_for_pin() ands out the
>> negation of bit positions rather than masks... (Which also raises
>> the question whether it really would need to be clear_bit() above
>> instead of the cheaper __clear_bit().)
>
> Below the tentative fix for all of the above problems. In the light
> of the comment at the top of x86's bitops.h I'm awaiting our gcc
> experts' response regarding the safety of using "+m" here.
Looks fine to me, in principle. I would add a comment to the x86
gnttab_clear_flag() explaining why we have to open code something that looks
a lot like clear_bit().
-- Keir
> Jan
>
> --- 2012-02-06.orig/xen/common/grant_table.c
> +++ 2012-02-06/xen/common/grant_table.c
> @@ -397,7 +397,8 @@ static int _set_status_v2(domid_t domid
> (id != domid) ||
> (!readonly && (flags & GTF_readonly)) )
> {
> - gnttab_clear_flag(_GTF_reading | _GTF_writing, status);
> + gnttab_clear_flag(_GTF_writing, status);
> + gnttab_clear_flag(_GTF_reading, status);
> PIN_FAIL(done, GNTST_general_error,
> "Unstable flags (%x) or dom (%d). (expected dom %d) "
> "(r/w: %d)\n",
> @@ -1715,14 +1716,14 @@ __release_grant_for_copy(
> under the domain's grant table lock. */
> /* Only safe on transitive grants. Even then, note that we don't
> attempt to drop any pin on the referent grant. */
> -static void __fixup_status_for_pin(struct active_grant_entry *act,
> +static void __fixup_status_for_pin(const struct active_grant_entry *act,
> uint16_t *status)
> {
> if ( !(act->pin & GNTPIN_hstw_mask) )
> - *status &= ~_GTF_writing;
> + *status &= ~GTF_writing;
>
> if ( !(act->pin & GNTPIN_hstr_mask) )
> - *status &= ~_GTF_reading;
> + *status &= ~GTF_reading;
> }
>
> /* Grab a frame number from a grant entry and update the flags and pin
> --- 2012-02-06.orig/xen/include/asm-ia64/grant_table.h
> +++ 2012-02-06/xen/include/asm-ia64/grant_table.h
> @@ -5,6 +5,8 @@
> #ifndef __ASM_GRANT_TABLE_H__
> #define __ASM_GRANT_TABLE_H__
>
> +#include <asm/intrinsics.h>
> +
> #define INITIAL_NR_GRANT_FRAMES 1
>
> // for grant map/unmap
> @@ -82,9 +84,15 @@ int guest_physmap_add_page(struct domain
>
> #define gnttab_mark_dirty(d, f) ((void)f)
>
> -static inline void gnttab_clear_flag(unsigned long nr, uint16_t *addr)
> +static inline void gnttab_clear_flag(unsigned int nr, volatile uint16_t *st)
> {
> - clear_bit(nr, addr);
> + uint16_t mask = ~(1 << nr), old;
> + CMPXCHG_BUGCHECK_DECL
> +
> + do {
> + CMPXCHG_BUGCHECK(st);
> + old = *st;
> + } while (cmpxchg_rel(st, old, old & mask) != old);
> }
>
> #define gnttab_host_mapping_get_page_type(op, ld, rd) \
> --- 2012-02-06.orig/xen/include/asm-x86/grant_table.h
> +++ 2012-02-06/xen/include/asm-x86/grant_table.h
> @@ -48,9 +48,11 @@ int replace_grant_host_mapping(
>
> #define gnttab_mark_dirty(d, f) paging_mark_dirty((d), (f))
>
> -static inline void gnttab_clear_flag(unsigned long nr, uint16_t *addr)
> +static inline void gnttab_clear_flag(unsigned int nr, uint16_t *addr)
> {
> - clear_bit(nr, (unsigned long *)addr);
> + asm volatile ("lock btrw %1,%0"
> + : "+m" (*addr)
> + : "Ir" (nr));
> }
>
> /* Foreign mappings of HHVM-guest pages do not modify the type count. */
>
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel
next prev parent reply other threads:[~2012-02-07 5:10 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2012-02-06 17:06 x86: gnttab_clear_flag() abusing clear_bit() Jan Beulich
2012-02-07 10:34 ` [PATCH, RFC] " Jan Beulich
2012-02-07 5:10 ` Keir Fraser [this message]
2012-02-09 9:01 ` Jan Beulich
2012-02-09 12:33 ` Keir Fraser
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CB5662D8.2A8BB%keir.xen@gmail.com \
--to=keir.xen@gmail.com \
--cc=JBeulich@suse.com \
--cc=xen-devel@lists.xensource.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).