LKML Archive on lore.kernel.org help / color / mirror / Atom feed
* iova RB tree setup tweak.
@ 2008-02-11 21:56 mark gross
2008-02-11 22:29 ` Andrew Morton
0 siblings, 1 reply; 3+ messages in thread
From: mark gross @ 2008-02-11 21:56 UTC (permalink / raw)
To: Andrew Morton; +Cc: linux-kernel, linux-mm
The following patch merges two functions into one allowing for a 3%
reduction in overhead in locating, allocating and inserting pages for
use in IOMMU operations.
Its a bit of a eye-crosser so I welcome any RB-tree / MM experts to take
a look. It works by re-using some of the information gathered in the
search for the pages to use in setting up the IOTLB's in the insertion
of the iova structure into the RB tree.
--mgross
Singed-off-by: <mgross@linux.intel.com>
Index: linux-2.6.24-mm1/drivers/pci/iova.c
===================================================================
--- linux-2.6.24-mm1.orig/drivers/pci/iova.c 2008-02-07 11:05:44.000000000 -0800
+++ linux-2.6.24-mm1/drivers/pci/iova.c 2008-02-07 13:05:03.000000000 -0800
@@ -72,20 +72,25 @@
return pad_size;
}
-static int __alloc_iova_range(struct iova_domain *iovad, unsigned long size,
- unsigned long limit_pfn, struct iova *new, bool size_aligned)
+static int __alloc_and_insert_iova_range(struct iova_domain *iovad,
+ unsigned long size, unsigned long limit_pfn,
+ struct iova *new, bool size_aligned)
{
- struct rb_node *curr = NULL;
+ struct rb_node *prev, *curr = NULL;
unsigned long flags;
unsigned long saved_pfn;
unsigned int pad_size = 0;
/* Walk the tree backwards */
spin_lock_irqsave(&iovad->iova_rbtree_lock, flags);
saved_pfn = limit_pfn;
curr = __get_cached_rbnode(iovad, &limit_pfn);
+ prev = curr;
while (curr) {
struct iova *curr_iova = container_of(curr, struct iova, node);
+
if (limit_pfn < curr_iova->pfn_lo)
goto move_left;
else if (limit_pfn < curr_iova->pfn_hi)
@@ -99,6 +104,7 @@
adjust_limit_pfn:
limit_pfn = curr_iova->pfn_lo - 1;
move_left:
+ prev = curr;
curr = rb_prev(curr);
}
@@ -115,7 +121,33 @@
new->pfn_lo = limit_pfn - (size + pad_size) + 1;
new->pfn_hi = new->pfn_lo + size - 1;
+ /* Insert the new_iova into domain rbtree by holding writer lock */
+ /* Add new node and rebalance tree. */
+ {
+ struct rb_node **entry = &((prev)), *parent = NULL;
+ /* Figure out where to put new node */
+ while (*entry) {
+ struct iova *this = container_of(*entry,
+ struct iova, node);
+ parent = *entry;
+
+ if (new->pfn_lo < this->pfn_lo)
+ entry = &((*entry)->rb_left);
+ else if (new->pfn_lo > this->pfn_lo)
+ entry = &((*entry)->rb_right);
+ else
+ BUG(); /* this should not happen */
+ }
+
+ /* Add new node and rebalance tree. */
+ rb_link_node(&new->node, parent, entry);
+ rb_insert_color(&new->node, &iovad->rbroot);
+ }
+ __cached_rbnode_insert_update(iovad, saved_pfn, new);
+
spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags);
+
+
return 0;
}
@@ -171,23 +203,15 @@
size = __roundup_pow_of_two(size);
spin_lock_irqsave(&iovad->iova_alloc_lock, flags);
- ret = __alloc_iova_range(iovad, size, limit_pfn, new_iova,
- size_aligned);
+ ret = __alloc_and_insert_iova_range(iovad, size, limit_pfn,
+ new_iova, size_aligned);
+ spin_unlock_irqrestore(&iovad->iova_alloc_lock, flags);
if (ret) {
- spin_unlock_irqrestore(&iovad->iova_alloc_lock, flags);
free_iova_mem(new_iova);
return NULL;
}
- /* Insert the new_iova into domain rbtree by holding writer lock */
- spin_lock(&iovad->iova_rbtree_lock);
- iova_insert_rbtree(&iovad->rbroot, new_iova);
- __cached_rbnode_insert_update(iovad, limit_pfn, new_iova);
- spin_unlock(&iovad->iova_rbtree_lock);
-
- spin_unlock_irqrestore(&iovad->iova_alloc_lock, flags);
-
return new_iova;
}
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: iova RB tree setup tweak. 2008-02-11 21:56 iova RB tree setup tweak mark gross @ 2008-02-11 22:29 ` Andrew Morton 2008-02-12 15:33 ` mark gross 0 siblings, 1 reply; 3+ messages in thread From: Andrew Morton @ 2008-02-11 22:29 UTC (permalink / raw) To: mgross; +Cc: linux-kernel, linux-mm, Greg KH On Mon, 11 Feb 2008 13:56:51 -0800 mark gross <mgross@linux.intel.com> wrote: > The following patch merges two functions into one allowing for a 3% > reduction in overhead in locating, allocating and inserting pages for > use in IOMMU operations. > > Its a bit of a eye-crosser so I welcome any RB-tree / MM experts to take > a look. It works by re-using some of the information gathered in the > search for the pages to use in setting up the IOTLB's in the insertion > of the iova structure into the RB tree. > I guess this is a PCI patch hence I'd be tagging is as to-be-merged-via-greg's-pci-tree. > > Singed-off-by: <mgross@linux.intel.com> > ow, that musta hurt. > > Index: linux-2.6.24-mm1/drivers/pci/iova.c > =================================================================== > --- linux-2.6.24-mm1.orig/drivers/pci/iova.c 2008-02-07 11:05:44.000000000 -0800 > +++ linux-2.6.24-mm1/drivers/pci/iova.c 2008-02-07 13:05:03.000000000 -0800 > @@ -72,20 +72,25 @@ > return pad_size; > } > > -static int __alloc_iova_range(struct iova_domain *iovad, unsigned long size, > - unsigned long limit_pfn, struct iova *new, bool size_aligned) > +static int __alloc_and_insert_iova_range(struct iova_domain *iovad, > + unsigned long size, unsigned long limit_pfn, > + struct iova *new, bool size_aligned) > { > - struct rb_node *curr = NULL; > + struct rb_node *prev, *curr = NULL; > unsigned long flags; > unsigned long saved_pfn; > unsigned int pad_size = 0; > > /* Walk the tree backwards */ > spin_lock_irqsave(&iovad->iova_rbtree_lock, flags); > saved_pfn = limit_pfn; > curr = __get_cached_rbnode(iovad, &limit_pfn); > + prev = curr; > while (curr) { > struct iova *curr_iova = container_of(curr, struct iova, node); > + > if (limit_pfn < curr_iova->pfn_lo) > goto move_left; > else if (limit_pfn < curr_iova->pfn_hi) > @@ -99,6 +104,7 @@ For some reason patch(1) claims that the patch is corrupted at this line. ^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: iova RB tree setup tweak. 2008-02-11 22:29 ` Andrew Morton @ 2008-02-12 15:33 ` mark gross 0 siblings, 0 replies; 3+ messages in thread From: mark gross @ 2008-02-12 15:33 UTC (permalink / raw) To: Andrew Morton; +Cc: linux-kernel, linux-mm, Greg KH On Mon, Feb 11, 2008 at 02:29:46PM -0800, Andrew Morton wrote: > On Mon, 11 Feb 2008 13:56:51 -0800 > mark gross <mgross@linux.intel.com> wrote: > > > The following patch merges two functions into one allowing for a 3% > > reduction in overhead in locating, allocating and inserting pages for > > use in IOMMU operations. > > > > Its a bit of a eye-crosser so I welcome any RB-tree / MM experts to take > > a look. It works by re-using some of the information gathered in the > > search for the pages to use in setting up the IOTLB's in the insertion > > of the iova structure into the RB tree. > > > > I guess this is a PCI patch hence I'd be tagging is as > to-be-merged-via-greg's-pci-tree. > > > > > Singed-off-by: <mgross@linux.intel.com> > > > > ow, that musta hurt. Not at first but ever since I posted the patch its been a bit. > > > > > Index: linux-2.6.24-mm1/drivers/pci/iova.c > > =================================================================== > > --- linux-2.6.24-mm1.orig/drivers/pci/iova.c 2008-02-07 11:05:44.000000000 -0800 > > +++ linux-2.6.24-mm1/drivers/pci/iova.c 2008-02-07 13:05:03.000000000 -0800 > > @@ -72,20 +72,25 @@ > > return pad_size; > > } > > > > -static int __alloc_iova_range(struct iova_domain *iovad, unsigned long size, > > - unsigned long limit_pfn, struct iova *new, bool size_aligned) > > +static int __alloc_and_insert_iova_range(struct iova_domain *iovad, > > + unsigned long size, unsigned long limit_pfn, > > + struct iova *new, bool size_aligned) > > { > > - struct rb_node *curr = NULL; > > + struct rb_node *prev, *curr = NULL; > > unsigned long flags; > > unsigned long saved_pfn; > > unsigned int pad_size = 0; > > > > /* Walk the tree backwards */ > > spin_lock_irqsave(&iovad->iova_rbtree_lock, flags); > > saved_pfn = limit_pfn; > > curr = __get_cached_rbnode(iovad, &limit_pfn); > > + prev = curr; > > while (curr) { > > struct iova *curr_iova = container_of(curr, struct iova, node); > > + > > if (limit_pfn < curr_iova->pfn_lo) > > goto move_left; > > else if (limit_pfn < curr_iova->pfn_hi) > > @@ -99,6 +104,7 @@ > > For some reason patch(1) claims that the patch is corrupted at this line. > My bad. Try the following: --mgross Signed-off-by : <mgross@linux.intel.com> fa-la-la Index: linux-2.6.24-mm1/drivers/pci/iova.c =================================================================== --- linux-2.6.24-mm1.orig/drivers/pci/iova.c 2008-02-07 11:05:44.000000000 -0800 +++ linux-2.6.24-mm1/drivers/pci/iova.c 2008-02-12 07:12:47.000000000 -0800 @@ -72,10 +72,11 @@ return pad_size; } -static int __alloc_iova_range(struct iova_domain *iovad, unsigned long size, - unsigned long limit_pfn, struct iova *new, bool size_aligned) +static int __alloc_and_insert_iova_range(struct iova_domain *iovad, + unsigned long size, unsigned long limit_pfn, + struct iova *new, bool size_aligned) { - struct rb_node *curr = NULL; + struct rb_node *prev, *curr = NULL; unsigned long flags; unsigned long saved_pfn; unsigned int pad_size = 0; @@ -84,8 +85,10 @@ spin_lock_irqsave(&iovad->iova_rbtree_lock, flags); saved_pfn = limit_pfn; curr = __get_cached_rbnode(iovad, &limit_pfn); + prev = curr; while (curr) { struct iova *curr_iova = container_of(curr, struct iova, node); + if (limit_pfn < curr_iova->pfn_lo) goto move_left; else if (limit_pfn < curr_iova->pfn_hi) @@ -99,6 +102,7 @@ adjust_limit_pfn: limit_pfn = curr_iova->pfn_lo - 1; move_left: + prev = curr; curr = rb_prev(curr); } @@ -115,7 +119,33 @@ new->pfn_lo = limit_pfn - (size + pad_size) + 1; new->pfn_hi = new->pfn_lo + size - 1; + /* Insert the new_iova into domain rbtree by holding writer lock */ + /* Add new node and rebalance tree. */ + { + struct rb_node **entry = &((prev)), *parent = NULL; + /* Figure out where to put new node */ + while (*entry) { + struct iova *this = container_of(*entry, + struct iova, node); + parent = *entry; + + if (new->pfn_lo < this->pfn_lo) + entry = &((*entry)->rb_left); + else if (new->pfn_lo > this->pfn_lo) + entry = &((*entry)->rb_right); + else + BUG(); /* this should not happen */ + } + + /* Add new node and rebalance tree. */ + rb_link_node(&new->node, parent, entry); + rb_insert_color(&new->node, &iovad->rbroot); + } + __cached_rbnode_insert_update(iovad, saved_pfn, new); + spin_unlock_irqrestore(&iovad->iova_rbtree_lock, flags); + + return 0; } @@ -171,23 +201,15 @@ size = __roundup_pow_of_two(size); spin_lock_irqsave(&iovad->iova_alloc_lock, flags); - ret = __alloc_iova_range(iovad, size, limit_pfn, new_iova, - size_aligned); + ret = __alloc_and_insert_iova_range(iovad, size, limit_pfn, + new_iova, size_aligned); + spin_unlock_irqrestore(&iovad->iova_alloc_lock, flags); if (ret) { - spin_unlock_irqrestore(&iovad->iova_alloc_lock, flags); free_iova_mem(new_iova); return NULL; } - /* Insert the new_iova into domain rbtree by holding writer lock */ - spin_lock(&iovad->iova_rbtree_lock); - iova_insert_rbtree(&iovad->rbroot, new_iova); - __cached_rbnode_insert_update(iovad, limit_pfn, new_iova); - spin_unlock(&iovad->iova_rbtree_lock); - - spin_unlock_irqrestore(&iovad->iova_alloc_lock, flags); - return new_iova; } ^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2008-02-12 15:30 UTC | newest] Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2008-02-11 21:56 iova RB tree setup tweak mark gross 2008-02-11 22:29 ` Andrew Morton 2008-02-12 15:33 ` mark gross
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).