Date:	Fri, 8 Jun 2007 13:44:09 -0700
From:	Andrew Morton <akpm@...ux-foundation.org>
To:	"Keshavamurthy, Anil S" <anil.s.keshavamurthy@...el.com>
Cc:	linux-kernel@...r.kernel.org, ak@...e.de, gregkh@...e.de,
	muli@...ibm.com, asit.k.mallick@...el.com,
	suresh.b.siddha@...el.com, arjan@...ux.intel.com,
	ashok.raj@...el.com, shaohua.li@...el.com, davem@...emloft.net
Subject: Re: [Intel-IOMMU 02/10] Library routine for pre-allocat pool
 handling

On Fri, 8 Jun 2007 13:12:00 -0700
"Keshavamurthy, Anil S" <anil.s.keshavamurthy@...el.com> wrote:

> On Fri, Jun 08, 2007 at 12:01:07PM -0700, Andrew Morton wrote:
> > On Fri, 8 Jun 2007 11:21:57 -0700
> > "Keshavamurthy, Anil S" <anil.s.keshavamurthy@...el.com> wrote:
> > 
> > > On Thu, Jun 07, 2007 at 04:27:26PM -0700, Andrew Morton wrote:
> > > > On Wed, 06 Jun 2007 11:57:00 -0700
> > > > anil.s.keshavamurthy@...el.com wrote:
> > > > 
> > > > > Signed-off-by: Anil S Keshavamurthy <anil.s.keshavamurthy@...el.com>
> > > > 
> > > > That was a terse changelog.
> > > > 
> > > > Obvious question: how does this differ from mempools, and would it be
> > > > better to fill in any gaps in mempool functionality instead of
> > > > implementing something similar-looking?
> > > 
> > > Very good question. Mempool pre-allocates the elements
> > > to the required minimum count size during its initilization time.
> > > However when mempool_alloc() is called it tries to obtain the
> > > element from OS and if that fails then it looks for the element in 
> > > its pool. If there are no elements in its pool and if the gpf_t 
> > > flags says it can wait then it waits untill someone puts the element 
> > > back to pool, else if gpf_t flag say it can;t wait then it returns NULL. 
> > > In other words, mempool acts as *emergency* pool, i.e only if the OS fails 
> > > to allocate the required memory, then the pool object is used.
> > > 
> > > 
> > > In the IOMMU case, we need exactly opposite of what mempool provides,
> > > i.e we always want to look for the element in the pool and if the pool
> > > has no element then go to OS as a worst case. This resource pool
> > > library routines do the same. Again, this resource pools 
> > > grows and shrinks automatically to maintain the minimum pool 
> > > elements in the background. I am not sure whether this totally
> > > opposite functionality of mempools and resource pools can be 
> > > merged.
> > 
> > Confused.
> > 
> > If resource pools are not designed to provide extra robustness via an
> > emergency pool, then what _are_ they designed for?  (Boy this is a hard way
> > to write a changelog!)
> 
> The resource pool indeed provide extra robustness, the initial pool size will
> be equal to min_count + grow_count. If the pool object count goes below
> min_count, then pool grows in the background while serving as emergency
> pool with min_count of objects in it. If we run out of emergency pool objects
> before the pool grow in the background, then we go to OS for allocation.

This wholly duplicates kswapd functionality.

> Similary, if the pool objects grows above the max threshold,
> the objects are freed to OS in the background thread maintaining
> the pool objects close to min_count + grow_count size.

That problem was _introduced_ by resource-pools, so yes, it also needs to
be solved there.

> 
> > 
> > > In fact the very first version of this IOMMU patch used mempools
> > > and the performance was worse because mempool did not help as
> > > IOMMU did a very frequent alloc and free of pool objects and
> > > every call to alloc/free used to go to os. Andi Kleen, 
> > > noticied and told us that mempool usage for IOMMU is wrong and
> > > hence we came up with resource pool concept.
> > 
> > You _seem_ to be saying that the resource pools are there purely for
> > alloc/free performance reasons.  If so, I'd be skeptical: slab is pretty
> > darned fast.
> We need several objects of size say( 4 * sizeof(u64)) and reuse
> them in dma map/unmap api calls for managing io virtual allocation address that
> this driver has dished out. Hence having pool of objects where we put 
> the element in the linked list and and get it from the linked list is pretty
> fast compared to slab.

slab is fast (and IO is slow!).  Do you have benchmark results?

> > 
> > > > 
> > > > The changelog very much should describe all this, as well as explaining
> > > > what the dynamic behaviour of this new thing is, and what applications are
> > > > envisaged, what problems it solves, etc, etc.
> > > 
> > > I can gladly update the changelog if the resource pool concept is 
> > > approved. I will fix all the below minor comments.
> > > 
> > > I envision that this might be useful for all vendor's (IBM, AMD, Intel, etc) IOMMU driver
> > > and for any kernel component which does lots of dynamic alloc/free an object of same size.
> > > 
> > 
> > That's what kmem_cache_alloc() is for?!?!
> We had this kmem_cache_alloc() with mempool concept earlier and Andi suggest to 
> come up with something pre-allocated pool.  
> Andi, Can you chime in please.


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists