Re: [PATCH] MMIO: Make coalesced mmio use a device per zone

[Jan Kiszka <jan.kiszka@siemens.com>]

On 2011-07-19 12:32, Avi Kivity wrote:
> On 07/19/2011 01:17 PM, Sasha Levin wrote:
>> On Tue, 2011-07-19 at 12:59 +0300, Avi Kivity wrote:
>> >  On 07/19/2011 12:53 PM, Sasha Levin wrote:
>> >  >  >   Make these per-guest instead of global.  The lock may be
>> contended, and
>> >  >  >   the list shouldn't hold items from different guests (why is
>> it needed,
>> >  >  >   anyway?)
>> >  >  >
>> >  >
>> >  >  We only need the list for removal, since we only have the range
>> we want
>> >  >  to remove, and we want to find all devices which contain this
>> range.
>> >  >
>> >
>> >  All devices in the same guest which contain this range.  Your patch
>> >  removes devices from all guests.
>> >
>>
>> Yup. I've messed up guest-locality. Will fix.
>>
>> Also, I found this comment when increasing NR_IOBUS_DEVS:
>>
>> /*
>>   * It would be nice to use something smarter than a linear search,
>> TBD...
>>   * Thankfully we dont expect many devices to register (famous last
>> words :),
>>   * so until then it will suffice.  At least its abstracted so we can
>> change
>>   * in one place.
>>   */
>>
>> Since so far we've registered 5-6 devices, and now it may increase
>> significantly (since we may want to do the same change to ioeventfds,
>> which work the same way) - how would you feel if we make devices
>> register range(s) and do a rbtree lookup instead of a linear search?
>>
> 
> It makes sense.  In fact your change is a good first step - so far it
> was impossible to to a clever search since the seaching code was not
> aware of the ranges (and could not be, since the single coalesced mmio
> device had multiple ranges).
> 
> Rather than an rbtree, I suggest putting all ranges in an array and
> sorting it, then using binary search.

Another improvement - unfortunately less transparent for user space -
would be to overcome the single ring buffer that forces us to hold a
central lock in user space while processing the entries. We rather need
per-device rings. While waiting for coalesced VGA MMIO being processed,
way too many kittens are killed.

I have this on our agenda, but I wouldn't be disappointed as well if
someone else is faster.

Jan

-- 
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux