Re: [PATCH] xen/tmem: Fix uses of unmatched __map_domain_page()

[Bob Liu <bob.liu@oracle.com>]

On 12/04/2013 05:00 AM, Konrad Rzeszutek Wilk wrote:
> On Wed, Nov 27, 2013 at 02:55:04PM +0000, Andrew Cooper wrote:
>> I noticed this while looking through tmem_xen.h with regards to the
>> recently-discovered Coverity issues.  As the issue was not noticed or
>> referenced in Bob's cleanup series, I figured it was fair game, given its
>> severity.
>>
>> __map_domain_page() *must* be matched with an unmap_domain_page().  These five
>> static inline functions each map a page (or two), then throw away the context
>> needed to unmap it.
> 
> I was trying to figure out how it worked before. I had been running with
> tze enabled (I hope!) and I did not trigger any mapcache exhaustion.
> 
> Ah wait, I had been on my nighly regression system  - which has some
> guests that use tmem but they don't create any load fast enough.
> 
> Let me queue this up and test it. Bob, would appreciate you testing
> it too - just in case.
> 

I've done the testing with this patch on trunk and I didn't see any problem.

Both dedup and tze are enabled in my testing and I'm sure they were in use.

[root@ca-test65 bob.liu]# xm tmem-list --long --all |
/usr/sbin/xen-tmem-list-parse
WARNING: xend/xm is deprecated.
total tmem ops=4139783 (errors=5461) -- tmem pages avail=20918
datastructs: objs=1039 (max=2130) pgps=32422 (max=53149) nodes=1994
(max=2803) pages=26709 (max=46611) pcds=13776 (max=35691) deduped:
avg=6.35% (curr=35.47%) tze savings=4.34%

Thanks,
-Bob