[PATCH v2 3/3] bpf: hash: use per-bucket spinlock

ust enough. This patch converts the per-hashtable lock into per-bucket spinlock, so that contention can be decreased a lot. Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 50 -- 1 file changed, 32 insertions(+), 18 deletions(-) diff --git a/

[PATCH v2 0/3] bpf: hash: use per-bucket spinlock

mic counter V2: - fix checking on buckets size V1: - fix the wrong 3/3 patch kernel/bpf/hashtab.c | 64 +++- 1 file changed, 38 insertions(+), 26 deletions(-) Thanks, Ming Lei -- To unsubscribe from this list: send the line "unsubs

[PATCH v2 1/3] bpf: hash: use atomic count

Preparing for removing global per-hashtable lock, so the counter need to be defined as aotmic_t first. Acked-by: Daniel Borkmann Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 12 ++-- 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/kernel/bpf/hashtab.c b/kernel/bpf

[PATCH v2 2/3] bpf: hash: move select_bucket() out of htab's spinlock

The spinlock is just used for protecting the per-bucket hlist, so it isn't needed for selecting bucket. Acked-by: Daniel Borkmann Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 6 ++ 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/kernel/bpf/hashtab.c b/kerne

[PATCH v1 0/3] bpf: hash: use per-bucket spinlock

Hi, This patchset tries to optimize ebpf hash map, and follows the idea: Both htab_map_update_elem() and htab_map_delete_elem() can be called from eBPF program, and they may be in kernel hot path, it isn't efficient to use a per-hashtable lock in this two helpers,

[PATCH v1 2/3] bpf: hash: move select_bucket() out of htab's spinlock

The spinlock is just used for protecting the per-bucket hlist, so it isn't needed for selecting bucket. Acked-by: Daniel Borkmann Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 6 ++ 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/kernel/bpf/hashtab.c b/kerne

[PATCH v1 1/3] bpf: hash: use atomic count

Preparing for removing global per-hashtable lock, so the counter need to be defined as aotmic_t first. Acked-by: Daniel Borkmann Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 12 ++-- 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/kernel/bpf/hashtab.c b/kernel/bpf

[PATCH v1 3/3] bpf: hash: use per-bucket spinlock

ust enough. This patch converts the per-hashtable lock into per-bucket spinlock, so that contention can be decreased a lot. Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 46 ++ 1 file changed, 30 insertions(+), 16 deletions(-) diff --git a/

Re: [PATCH 3/3] bpf: hash: use per-bucket spinlock

On Mon, Dec 28, 2015 at 5:13 PM, Daniel Borkmann wrote: > On 12/26/2015 10:31 AM, Ming Lei wrote: >> >> From: Ming Lei >> >> Both htab_map_update_elem() and htab_map_delete_elem() can be >> called from eBPF program, and they may be in kernel hot path, >

[PATCH 2/3] bpf: hash: move select_bucket() out of htab's spinlock

The spinlock is just used for protecting the per-bucket hlist, so it isn't needed for selecting bucket. Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 6 ++ 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c index 2615388..d8

[PATCH 0/3] bpf: hash: use per-bucket spinlock

Hi, This patchset tries to optimize ebpf hash map, and follows the idea: Both htab_map_update_elem() and htab_map_delete_elem() can be called from eBPF program, and they may be in kernel hot path, it isn't efficient to use a per-hashtable lock in this two helpers,

[PATCH 3/3] bpf: hash: use per-bucket spinlock

From: Ming Lei Both htab_map_update_elem() and htab_map_delete_elem() can be called from eBPF program, and they may be in kernel hot path, so it isn't efficient to use a per-hashtable lock in this two helpers. The per-hashtable spinlock is used just for protecting bucket's hlist, and

[PATCH 1/3] bpf: hash: use atomic count

Preparing for removing global per-hashtable lock, so the counter need to be defined as aotmic_t first. Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 12 ++-- 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c index 34777b3

Re: [PATCH 4/6] bpf: hash: convert per-hashtable lock into per-bucket bit spinlock

On Fri, Dec 18, 2015 at 2:20 PM, Alexei Starovoitov wrote: > On Wed, Dec 16, 2015 at 02:58:08PM +0800, Ming Lei wrote: >> On Wed, Dec 16, 2015 at 1:01 PM, Yang Shi wrote: >> >> > >> > I recalled Steven confirmed raw_spin_lock has the lockdep benefit too i

Re: [PATCH 5/6] bpf: hash: avoid to call kmalloc() in eBPF prog

On Wed, Dec 16, 2015 at 7:10 AM, Alexei Starovoitov wrote: > On Tue, Dec 15, 2015 at 07:21:03PM +0800, Ming Lei wrote: >> kmalloc() is often a bit time-consuming, also >> one atomic counter has to be used to track the total >> allocated elements, which is also not good.

Re: [PATCH 4/6] bpf: hash: convert per-hashtable lock into per-bucket bit spinlock

about the lockdep benifit, :-( But for this lock, I think lockdep isn't such important, because it is the intermost lock, and it can be used just for protecting the bucket list and nothing else need to be covered. Thanks, Ming Lei -- To unsubscribe from this list: send the line "unsubscr

Re: [PATCH 4/6] bpf: hash: convert per-hashtable lock into per-bucket bit spinlock

Hi Alexei, On Wed, Dec 16, 2015 at 6:51 AM, Alexei Starovoitov wrote: > On Tue, Dec 15, 2015 at 07:21:02PM +0800, Ming Lei wrote: >> Both htab_map_update_elem() and htab_map_delete_elem() can be >> called from eBPF program, and they may be in kernel hot path, >> so it isn

[PATCH 5/6] bpf: hash: avoid to call kmalloc() in eBPF prog

cost can be decreased. >From my test, at least 10% fio throughput is improved in block I/O test when tools/biolatency of bcc(iovisor) is running. Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 204 +-- 1 file changed, 167 insertions(+),

[PATCH 4/6] bpf: hash: convert per-hashtable lock into per-bucket bit spinlock

should be enough. This patch converts the per-hashtable lock into per-bucket bit spinlock, so that contention can be decreased a lot, and no extra memory can be consumed for these locks. Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 38 ++ 1 file c

[PATCH 1/6] bpf: hash: use atomic count

Preparing for removing global per-hashtable lock, so the counter need to be defined as aotmic_t first. Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 12 ++-- 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c index 34777b3

[PATCH 0/6] bpf: hash: optimization

Hi, This patchset tries to optimize ebpf hash map, and follows the ideas: 1) Both htab_map_update_elem() and htab_map_delete_elem() can be called from eBPF program, and they may be in kernel hot path, so it isn't efficient to use a per-hashtable lock in this two helpers, so this patch converts th

[PATCH 2/6] hlist: prepare for supporting bit spinlock

ose. bit spinlock isn't efficient as spin lock, but the contention shouldn't be very high for operating per-bucket hlist, so bit spinlock should be OK. Signed-off-by: Ming Lei --- include/linux/rculist.h | 55 + 1 file changed, 55 insert

[PATCH 3/6] bpf: hash: move select_bucket() out of htab's spinlock

The spinlock is just used for protecting the per-bucket hlist, so it isn't needed for selecting bucket. Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c | 6 ++ 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/kernel/bpf/hashtab.c b/kernel/bpf/hashtab.c index 2615388..d8

[PATCH 6/6] bpf: hash: reorganize 'struct htab_elem'

Lifetime for hash fields and liftime for kfree_rcu fields can't be overlapped, so re-organizing them for better readabilty. Also one sizeof(void *) should be saved with this change, and cache footprint can got improved too. Signed-off-by: Ming Lei --- kernel/bpf/hashtab.c

4.2-rc3: APM xgene2: ethernet doesn't work

Hi Guys, When booting from UEFI/ACPI, sometimes there is a crash[1] from rx path, sometimes there isn't any rx packets comming. Firmware version: 2.02.10 Thanks, [1], crash log Call trace: skbuff: skb_over_panic: text:ffc00048094c len:85 put:60 head: data: (null) ta

Re: [RESEND][PATCH] Bluetooth: Make request workqueue freezable

On Wed, May 20, 2015 at 4:40 PM, Oliver Neukum wrote: > On Wed, 2015-05-20 at 08:29 +0200, Takashi Iwai wrote: >> The data is cached in RAM. More specifically, the former loaded >> firmware files are reloaded and saved at suspend for each device >> object. See fw_pm_notify() in firmware_class.c.