[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [PATCH RESEND 2/2] memory: Add tracepoint for dirty sync
From: |
Peter Xu |
Subject: |
Re: [PATCH RESEND 2/2] memory: Add tracepoint for dirty sync |
Date: |
Tue, 17 Aug 2021 12:05:50 -0400 |
On Tue, Aug 17, 2021 at 09:25:56AM +0200, David Hildenbrand wrote:
> On 17.08.21 03:37, Peter Xu wrote:
> > Trace at memory_region_sync_dirty_bitmap() for log_sync() or
> > global_log_sync()
> > on memory regions. One trace line should suffice when it finishes, so as to
> > estimate the time used for each log sync process.
>
> I wonder if a start/finish would be even nicer. At least it wouldn't really
> result in significantly more code changes :)
Note that the "name"s I added is not only for not using start/end, it's about
knowing which memory listener is slow. Start/end won't achieve that if we
don't have a name for them. So far I just wanted to identify majorly kvm,
vhost and kvm-smram, however it'll always be good when some log_sync is missed
when tracing.
I'm also wondering whether kvm-smram needs a whole bitmap as I don't know what
RAM would be touched within system manager mode (as I thought it should only
touch a very limited range and should be defined somewhere), but that's
off-topic.
If we want to make it start/end pair, I can do that too. But the 1st patch
will still be wanted.
Thanks,
--
Peter Xu