Olaf Hering
2012-Aug-28 08:32 UTC
[PATCH] xenpaging: use poll timeout if no paging target exists
# HG changeset patch
# User Olaf Hering <olaf@aepfle.de>
# Date 1346142745 -7200
# Node ID b088e473c7fb1a47b9578693c1750cd27767cc25
# Parent 03ef29089830b119b6efd87db8ab5c38b7428938
xenpaging: use poll timeout if no paging target exists
Currently xenpaging will use 100% cpu time if a paging target is not yet
set via xenstore. The reason is that ->use_poll_timeout is initialized
to zero. Another case is when the paging target is set to zero. In this
case ->use_poll_timeout will not be set to 1.
Fix the first case by initializing use_poll_timeout to 1 in
xenpaging_init. The second case is fixed by setting use_poll_timeout in
case the target is zero and there are no more pages left to resume.
Signed-off-by: Olaf Hering <olaf@aepfle.de>
diff -r 03ef29089830 -r b088e473c7fb tools/xenpaging/xenpaging.c
--- a/tools/xenpaging/xenpaging.c
+++ b/tools/xenpaging/xenpaging.c
@@ -482,6 +482,9 @@ static struct xenpaging *xenpaging_init(
goto err;
}
+ /* Use a poll timeout until xenpaging got a target from xenstore */
+ paging->use_poll_timeout = 1;
+
return paging;
err:
@@ -1019,6 +1022,8 @@ int main(int argc, char *argv[])
{
if ( paging->num_paged_out )
resume_pages(paging, paging->num_paged_out);
+ else
+ paging->use_poll_timeout = 1;
}
/* Evict more pages if target not reached */
else if ( tot_pages > paging->target_tot_pages )
Ian Campbell
2012-Sep-07 16:05 UTC
Re: [PATCH] xenpaging: use poll timeout if no paging target exists
On Tue, 2012-08-28 at 09:32 +0100, Olaf Hering wrote:> # HG changeset patch > # User Olaf Hering <olaf@aepfle.de> > # Date 1346142745 -7200 > # Node ID b088e473c7fb1a47b9578693c1750cd27767cc25 > # Parent 03ef29089830b119b6efd87db8ab5c38b7428938 > xenpaging: use poll timeout if no paging target exists > > Currently xenpaging will use 100% cpu time if a paging target is not yet > set via xenstore. The reason is that ->use_poll_timeout is initialized > to zero. Another case is when the paging target is set to zero. In this > case ->use_poll_timeout will not be set to 1. > > Fix the first case by initializing use_poll_timeout to 1 in > xenpaging_init. The second case is fixed by setting use_poll_timeout in > case the target is zero and there are no more pages left to resume. > > Signed-off-by: Olaf Hering <olaf@aepfle.de> > > diff -r 03ef29089830 -r b088e473c7fb tools/xenpaging/xenpaging.c > --- a/tools/xenpaging/xenpaging.c > +++ b/tools/xenpaging/xenpaging.c > @@ -482,6 +482,9 @@ static struct xenpaging *xenpaging_init( > goto err; > } > > + /* Use a poll timeout until xenpaging got a target from xenstore */ > + paging->use_poll_timeout = 1;Don''t you want a negative (== infinite) timeout in this case?> + > return paging; > > err: > @@ -1019,6 +1022,8 @@ int main(int argc, char *argv[]) > { > if ( paging->num_paged_out ) > resume_pages(paging, paging->num_paged_out); > + else > + paging->use_poll_timeout = 1; > } > /* Evict more pages if target not reached */ > else if ( tot_pages > paging->target_tot_pages ) > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xen.org > http://lists.xen.org/xen-devel
Olaf Hering
2012-Sep-07 16:14 UTC
Re: [PATCH] xenpaging: use poll timeout if no paging target exists
On Fri, Sep 07, Ian Campbell wrote:> > + /* Use a poll timeout until xenpaging got a target from xenstore */ > > + paging->use_poll_timeout = 1; > > Don''t you want a negative (== infinite) timeout in this case?Maybe, but this change is the simplest fix for the isse. Waiting forever may have bad side effects, this is something to consider for 4.3. Olaf
Ian Campbell
2012-Sep-08 05:44 UTC
Re: [PATCH] xenpaging: use poll timeout if no paging target exists
On Fri, 2012-09-07 at 17:14 +0100, Olaf Hering wrote:> On Fri, Sep 07, Ian Campbell wrote: > > > > + /* Use a poll timeout until xenpaging got a target from xenstore */ > > > + paging->use_poll_timeout = 1; > > > > Don''t you want a negative (== infinite) timeout in this case? > > Maybe, but this change is the simplest fix for the isse. > Waiting forever may have bad side effects, this is something to consider > for 4.3.We are in the 4.3 development window now, and this patch didn''t say anything about being for 4.2. I think this fix (or the larger one) are 4.2.1 material at this point. Ian