From: Andrew Cooper <Andrew.Cooper3@citrix.com>
To: Anthony Perard <anthony.perard@citrix.com>,
"devel@edk2.groups.io" <devel@edk2.groups.io>
Cc: "xen-devel@lists.xenproject.org" <xen-devel@lists.xenproject.org>,
Ard Biesheuvel <ardb+tianocore@kernel.org>,
Jiewen Yao <jiewen.yao@intel.com>,
Jordan Justen <jordan.l.justen@intel.com>,
Gerd Hoffmann <kraxel@redhat.com>, Julien Grall <julien@xen.org>
Subject: Re: [PATCH 1/1] OvmfPkg/XenPvBlkDxe: Fix memory barrier macro
Date: Tue, 19 Jul 2022 14:12:40 +0000 [thread overview]
Message-ID: <73483ebf-d641-e4f2-6e23-b0336e62f359@citrix.com> (raw)
In-Reply-To: <20220719135230.32838-1-anthony.perard@citrix.com>
On 19/07/2022 14:52, Anthony Perard wrote:
> diff --git a/OvmfPkg/XenPvBlkDxe/XenPvBlkDxe.h b/OvmfPkg/XenPvBlkDxe/XenPvBlkDxe.h
> index 350b7bd309c0..67ee1899e9a8 100644
> --- a/OvmfPkg/XenPvBlkDxe/XenPvBlkDxe.h
> +++ b/OvmfPkg/XenPvBlkDxe/XenPvBlkDxe.h
> @@ -11,8 +11,9 @@
> #define __EFI_XEN_PV_BLK_DXE_H__
>
> #include <Uefi.h>
> +#include "FullMemoryFence.h"
>
> -#define xen_mb() MemoryFence()
> +#define xen_mb() FullMemoryFence()
> #define xen_rmb() MemoryFence()
> #define xen_wmb() MemoryFence()
Ok, so the old MemoryFence() is definitely bogus here.
However, it doesn't need to be an mfence instruction. All that is
needed is smp_mb(), which these days is
asm volatile ( "lock addl $0, -4(%%rsp)" ::: "memory" )
because that has the required read/write ordering properties without the
extra serialising property that mfence has.
Furthermore, ...
>
> diff --git a/OvmfPkg/XenPvBlkDxe/X86GccFullMemoryFence.c b/OvmfPkg/XenPvBlkDxe/X86GccFullMemoryFence.c
> new file mode 100644
> index 000000000000..92d107def470
> --- /dev/null
> +++ b/OvmfPkg/XenPvBlkDxe/X86GccFullMemoryFence.c
> @@ -0,0 +1,20 @@
> +/** @file
> + Copyright (C) 2022, Citrix Ltd.
> +
> + SPDX-License-Identifier: BSD-2-Clause-Patent
> +**/
> +
> +#include "FullMemoryFence.h"
> +
> +//
> +// Like MemoryFence() but prevent stores from been reorded with loads by
> +// the CPU on X64.
> +//
> +VOID
> +EFIAPI
> +FullMemoryFence (
> + VOID
> + )
> +{
> + __asm__ __volatile__ ("mfence":::"memory");
> +}
... stuff like this needs to come from a single core location, and not
opencoded for each driver.
~Andrew
prev parent reply other threads:[~2022-07-19 14:12 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-07-19 13:52 [PATCH 1/1] OvmfPkg/XenPvBlkDxe: Fix memory barrier macro Anthony PERARD
2022-07-19 14:12 ` Andrew Cooper [this message]
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-list 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=73483ebf-d641-e4f2-6e23-b0336e62f359@citrix.com \
--to=devel@edk2.groups.io \
/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