From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:48908) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1Ws8lT-0005UP-1p for qemu-devel@nongnu.org; Wed, 04 Jun 2014 06:50:00 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1Ws8lK-0001zY-0M for qemu-devel@nongnu.org; Wed, 04 Jun 2014 06:49:50 -0400 Message-ID: <538EF9AC.8030802@gmail.com> Date: Wed, 04 Jun 2014 18:49:16 +0800 From: Chen Gang MIME-Version: 1.0 References: <538C6F98.5070005@gmail.com> <538D79D4.5000900@huawei.com> <538DD187.5050400@gmail.com> <20140604100148.GC2618@work-vm> In-Reply-To: <20140604100148.GC2618@work-vm> Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [PATCH-trivial] arch_init.c: Always be sure that 'encoded_buf' and 'current_buf' are lock protected List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: "Dr. David Alan Gilbert" Cc: ChenLiang , quintela@redhat.com, QEMU Trivial , Michael Tokarev , QEMU Developers , arei.gonglei@huawei.com Firstly, thank you very much for reviewing the related 2 patches. On 06/04/2014 06:01 PM, Dr. David Alan Gilbert wrote: > * Chen Gang (gang.chen.5i5j@gmail.com) wrote: >> >> Firstly, thank you very much for reviewing the related 2 patches. >> >> On 06/03/2014 03:31 PM, ChenLiang wrote: >>> On 2014/6/2 20:35, Chen Gang wrote: >>> >>>> 'encoded_buf' and 'current_buf' are lock protected during using in >>>> save_xbzrle_page() in ram_save_page(), and during freeing in >>>> migration_end(). >>>> >>>> So recommend to let them lock protected during starting, just like we >>>> have done to 'cache'. > > This isn't one for Trivial - it needs some thought. Locks always need thought. > OK, I shall notice about it next time. "Locks always need thought", so prefer some 'templates' to simplify thinking, especially for some simple locks (not nested, either not performance sensitive). >>>> Signed-off-by: Chen Gang >>>> --- >>>> arch_init.c | 4 +++- >>>> 1 file changed, 3 insertions(+), 1 deletion(-) >>>> >>>> diff --git a/arch_init.c b/arch_init.c >>>> index 23044c1..784922c 100644 >>>> --- a/arch_init.c >>>> +++ b/arch_init.c >>>> @@ -783,12 +783,12 @@ static int ram_save_setup(QEMUFile *f, void *opaque) >>>> error_report("Error creating cache"); >>>> return -1; >>>> } >>>> - XBZRLE_cache_unlock(); >>>> >>>> /* We prefer not to abort if there is no memory */ >>>> XBZRLE.encoded_buf = g_try_malloc0(TARGET_PAGE_SIZE); >>>> if (!XBZRLE.encoded_buf) { >>>> error_report("Error allocating encoded_buf"); >>>> + XBZRLE_cache_unlock(); >>>> return -1; >>>> } >>>> >>>> @@ -797,8 +797,10 @@ static int ram_save_setup(QEMUFile *f, void *opaque) >>>> error_report("Error allocating current_buf"); >>>> g_free(XBZRLE.encoded_buf); >>>> XBZRLE.encoded_buf = NULL; >>>> + XBZRLE_cache_unlock(); >>>> return -1; >>>> } >>>> + XBZRLE_cache_unlock(); >>>> >>>> acct_clear(); >>>> } >>> >>> >>> It isn't necessary. >>> >> >> For me, it is still necessary to be improved. >> >> - If this code is not performance sensitive: >> >> "always lock protected" will let source code easy understanding for >> readers, and easy sustainable for maintainers. >> >> - Else (performance sensitive): >> >> Need give related comments to it (e.g. for performance reason, let it >> out of lock protected). So readers easily understand, and the related >> maintainers may notice about it at any time. >> >> It is not good to give comments in a function, except the code is >> not used in normal way. > > I think the locks for the cache were originally there to handle the race > between a resize (on the main thread) against the use (in the migration thread). > That doesn't happen for the encoded_buf so I don't think we have that problem. > > Is there any other case that it could race? > I can't see one - I think all of the other start/end is done within the migration > thread. > What you thought and what Chen Liang though are fine to me. But I guess it is a simple lock, if we let them always lock protected, we need not think them more. One simple lock 'template' can be: "always let it lock protected during its whole life", that will let all related readers and maintainers easy understanding and avoid to think of more. Our lock is not nested, although I am not quit sure whether it is performance sensitive (I guess not). Thanks. -- Chen Gang Open, share, and attitude like air, water, and life which God blessed