Re: [PATCH 0/2] iommu/iova: Solve longterm IOVA issue

On Fri, Sep 25, 2020 at 2:56 AM John Garry wrote: > > This series contains a patch to solve the longterm IOVA issue which > leizhen originally tried to address at [0]. > > I also included the small optimisation from Cong Wang, which never seems > to be have been accepted [

Re: [Patch v3 1/3] iommu: avoid unnecessary magazine allocations

On Wed, Jan 22, 2020 at 9:07 AM Robin Murphy wrote: > > On 21/01/2020 5:21 pm, Cong Wang wrote: > > On Tue, Jan 21, 2020 at 3:11 AM Robin Murphy wrote: > >> > >> On 18/12/2019 4:39 am, Cong Wang wrote: > >>> The IOVA cache algorithm implemented

Re: [Patch v3 2/3] iommu: optimize iova_magazine_free_pfns()

On Wed, Jan 22, 2020 at 9:34 AM Robin Murphy wrote: > Sorry, but without convincing evidence, this change just looks like > churn for the sake of it. The time I wasted on arguing with you isn't worth anything than the value this patch brings. So let's just drop it to save some time. Thanks.

Re: [Patch v3 2/3] iommu: optimize iova_magazine_free_pfns()

On Tue, Jan 21, 2020 at 1:52 AM Robin Murphy wrote: > > On 18/12/2019 4:39 am, Cong Wang wrote: > > If the magazine is empty, iova_magazine_free_pfns() should > > be a nop, however it misses the case of mag->size==0. So we > > should just call iova_magazine_empty().

Re: [Patch v3 1/3] iommu: avoid unnecessary magazine allocations

On Tue, Jan 21, 2020 at 3:11 AM Robin Murphy wrote: > > On 18/12/2019 4:39 am, Cong Wang wrote: > > The IOVA cache algorithm implemented in IOMMU code does not > > exactly match the original algorithm described in the paper > > "Magazines and Vmem: Extending the Sl

Re: [Patch v3 0/3] iommu: reduce spinlock contention on fast path

On Tue, Dec 17, 2019 at 8:40 PM Cong Wang wrote: > > This patchset contains three small optimizations for the global spinlock > contention in IOVA cache. Our memcache perf test shows this reduced its > p999 latency down by 45% on AMD when IOMMU is enabled. > > (Resending v3

[Patch v3 3/3] iommu: avoid taking iova_rbtree_lock twice

Both find_iova() and __free_iova() take iova_rbtree_lock, there is no reason to take and release it twice inside free_iova(). Fold them into one critical section by calling the unlock versions instead. Cc: Joerg Roedel Cc: John Garry Signed-off-by: Cong Wang --- drivers/iommu/iova.c | 8

[Patch v3 0/3] iommu: reduce spinlock contention on fast path

This patchset contains three small optimizations for the global spinlock contention in IOVA cache. Our memcache perf test shows this reduced its p999 latency down by 45% on AMD when IOMMU is enabled. (Resending v3 on Joerg's request.) Cong Wang (3): iommu: avoid unnecessary magazine

[Patch v3 2/3] iommu: optimize iova_magazine_free_pfns()

ed-off-by: Cong Wang --- drivers/iommu/iova.c | 22 +++--- 1 file changed, 11 insertions(+), 11 deletions(-) diff --git a/drivers/iommu/iova.c b/drivers/iommu/iova.c index cb473ddce4cf..184d4c0e20b5 100644 --- a/drivers/iommu/iova.c +++ b/drivers/iommu/iova.c @@ -797,13 +797,23 @@

[Patch v3 1/3] iommu: avoid unnecessary magazine allocations

es. Together with a few other changes to make it exactly match the pseudo code in the paper. Cc: Joerg Roedel Cc: John Garry Signed-off-by: Cong Wang --- drivers/iommu/iova.c | 45 +++- 1 file changed, 28 insertions(+), 17 deletions(-) diff --git a/drivers/i

[Patch v3 0/3] iommu: reduce spinlock contention on fast path

This patchset contains three small optimizations for the global spinlock contention in IOVA cache. Our memcache perf test shows this reduced its p999 latency down by 45% on AMD when IOMMU is enabled. Cong Wang (3): iommu: avoid unnecessary magazine allocations iommu: optimize

[Patch v3 2/3] iommu: optimize iova_magazine_free_pfns()

ed-off-by: Cong Wang --- drivers/iommu/iova.c | 22 +++--- 1 file changed, 11 insertions(+), 11 deletions(-) diff --git a/drivers/iommu/iova.c b/drivers/iommu/iova.c index cb473ddce4cf..184d4c0e20b5 100644 --- a/drivers/iommu/iova.c +++ b/drivers/iommu/iova.c @@ -797,13 +797,23 @@

[Patch v3 1/3] iommu: avoid unnecessary magazine allocations

es. Together with a few other changes to make it exactly match the pseudo code in the paper. Cc: Joerg Roedel Cc: John Garry Signed-off-by: Cong Wang --- drivers/iommu/iova.c | 45 +++- 1 file changed, 28 insertions(+), 17 deletions(-) diff --git a/drivers/i

[Patch v3 3/3] iommu: avoid taking iova_rbtree_lock twice

Both find_iova() and __free_iova() take iova_rbtree_lock, there is no reason to take and release it twice inside free_iova(). Fold them into one critical section by calling the unlock versions instead. Cc: Joerg Roedel Cc: John Garry Signed-off-by: Cong Wang --- drivers/iommu/iova.c | 8

Re: [Patch v2 2/3] iommu: optimize iova_magazine_free_pfns()

On Mon, Dec 2, 2019 at 2:02 AM John Garry wrote: > > On 30/11/2019 06:02, Cong Wang wrote: > > On Fri, Nov 29, 2019 at 5:24 AM John Garry wrote: > >> > >> On 29/11/2019 00:48, Cong Wang wrote: > >>> If the maganize is empty, iova_magazine_free_pfns() sh

Re: [Patch v2 2/3] iommu: optimize iova_magazine_free_pfns()

On Mon, Dec 2, 2019 at 8:59 AM Christoph Hellwig wrote: > > > + return (mag && mag->size == IOVA_MAG_SIZE); > > > + return (!mag || mag->size == 0); > > No need for the braces in both cases. The current code is already this, I don't want to mix coding style changes with a

Re: [Patch v2 1/3] iommu: match the original algorithm

On Mon, Dec 2, 2019 at 2:55 AM John Garry wrote: > Apart from this change, did anyone ever consider kmem cache for the > magazines? You can always make any changes you want after this patch, I can't do all optimizations in one single patch. :) So, I will leave this to you. Thanks.

Re: [Patch v2 1/3] iommu: match the original algorithm

On Mon, Dec 2, 2019 at 8:58 AM Christoph Hellwig wrote: > > I think a subject line better describes what you change, no that > it matches an original algorithm. The fact that the fix matches > the original algorithm can go somewhere towards the commit log, > preferably with a reference to the

Re: [Patch v2 3/3] iommu: avoid taking iova_rbtree_lock twice

On Fri, Nov 29, 2019 at 5:34 AM John Garry wrote: > > On 29/11/2019 00:48, Cong Wang wrote: > > Both find_iova() and __free_iova() take iova_rbtree_lock, > > there is no reason to take and release it twice inside > > free_iova(). > > > > Fold them into the cr

Re: [Patch v2 2/3] iommu: optimize iova_magazine_free_pfns()

On Fri, Nov 29, 2019 at 5:24 AM John Garry wrote: > > On 29/11/2019 00:48, Cong Wang wrote: > > If the maganize is empty, iova_magazine_free_pfns() should > > magazine Good catch! > > > be a nop, however it misses the case of mag->size==0. So we > >

Re: [Patch v2 1/3] iommu: match the original algorithm

On Fri, Nov 29, 2019 at 6:43 AM John Garry wrote: > > On 29/11/2019 00:48, Cong Wang wrote: > > The IOVA cache algorithm implemented in IOMMU code does not > > exactly match the original algorithm described in the paper. > > > > which paper? It's in drivers/iomm

[Patch v2 0/3] iommu: reduce spinlock contention on fast path

This patchset contains three small optimizations for the global spinlock contention in IOVA cache. Our memcache perf test shows this reduced its p999 latency down by 45% on AMD when IOMMU is enabled. Cong Wang (3): iommu: match the original algorithm iommu: optimize iova_magazine_free_pfns

[Patch v2 3/3] iommu: avoid taking iova_rbtree_lock twice

Both find_iova() and __free_iova() take iova_rbtree_lock, there is no reason to take and release it twice inside free_iova(). Fold them into the critical section by calling the unlock versions instead. Cc: Joerg Roedel Signed-off-by: Cong Wang --- drivers/iommu/iova.c | 8 ++-- 1 file

[Patch v2 2/3] iommu: optimize iova_magazine_free_pfns()

If the maganize is empty, iova_magazine_free_pfns() should be a nop, however it misses the case of mag->size==0. So we should just call iova_magazine_empty(). This should reduce the contention on iovad->iova_rbtree_lock a little bit. Cc: Joerg Roedel Signed-off-by: Cong Wang --- drivers

[Patch v2 1/3] iommu: match the original algorithm

and only recycle them when all of them are full. Before this patch, rcache->depot[] contains either full or freed entries, after this patch, it contains either full or empty (but allocated) entries. Cc: Joerg Roedel Signed-off-by: Cong Wang --- drivers/iommu/iova.c |

Re: [PATCH 1/3] iommu: match the original algorithm

On Wed, Nov 27, 2019 at 10:01 AM John Garry wrote: > > On 21/11/2019 00:13, Cong Wang wrote: > > The IOVA cache algorithm implemented in IOMMU code does not > > exactly match the original algorithm described in the paper. > > > > Particularly, it doesn't need to

[PATCH 3/3] iommu: avoid taking iova_rbtree_lock twice

Both find_iova() and __free_iova() take iova_rbtree_lock, there is no reason to take and release it twice inside free_iova(). Fold them into the critical section by calling the unlock versions instead. Cc: Joerg Roedel Signed-off-by: Cong Wang --- drivers/iommu/iova.c | 8 ++-- 1 file

[PATCH 0/3] iommu: reduce spinlock contention on fast path

This patchset contains three small optimizations for the global spinlock contention in IOVA cache. Our memcache perf test shows this reduced its p999 latency down by 45% on AMD when IOMMU is enabled. Cong Wang (3): iommu: match the original algorithm iommu: optimize iova_magazine_free_pfns

[PATCH 2/3] iommu: optimize iova_magazine_free_pfns()

If the maganize is empty, iova_magazine_free_pfns() should be a nop, however it misses the case of mag->size==0. So we should just call iova_magazine_empty(). This should reduce the contention on iovad->iova_rbtree_lock a little bit. Cc: Joerg Roedel Signed-off-by: Cong Wang --- drivers

[PATCH 1/3] iommu: match the original algorithm

: Joerg Roedel Signed-off-by: Cong Wang --- drivers/iommu/iova.c | 14 -- 1 file changed, 8 insertions(+), 6 deletions(-) diff --git a/drivers/iommu/iova.c b/drivers/iommu/iova.c index 41c605b0058f..92f72a85e62a 100644 --- a/drivers/iommu/iova.c +++ b/drivers/iommu/iova.c @@ -900,7 +900,7