Move the code that put one idle sh(hot in cache, but happens to be
zero referenced) back to active stage to __find_stripe(). Because
that's what need to do every time you invoke __find_stripe().
Moving it there avoids duplicate code, as well as makes a bit more
sense, IMO, as it tells a whole story now.
Signed-off-by: Yuanhan Liu <[email protected]>
---
drivers/md/raid5.c | 50 ++++++++++++++++++--------------------------------
1 file changed, 18 insertions(+), 32 deletions(-)
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index 77dfd72..e7fa818 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -567,8 +567,25 @@ static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
- if (sh->sector == sector && sh->generation == generation)
+ if (sh->sector == sector && sh->generation == generation) {
+ if (!atomic_inc_not_zero(&sh->count)) {
+ spin_lock(&conf->device_lock);
+ if (!atomic_read(&sh->count)) {
+ if (!test_bit(STRIPE_HANDLE, &sh->state))
+ atomic_inc(&conf->active_stripes);
+ BUG_ON(list_empty(&sh->lru) &&
+ !test_bit(STRIPE_EXPANDING, &sh->state));
+ list_del_init(&sh->lru);
+ if (sh->group) {
+ sh->group->stripes_cnt--;
+ sh->group = NULL;
+ }
+ }
+ atomic_inc(&sh->count);
+ spin_unlock(&conf->device_lock);
+ }
return sh;
+ }
pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
return NULL;
}
@@ -698,21 +715,6 @@ get_active_stripe(struct r5conf *conf, sector_t sector,
init_stripe(sh, sector, previous);
atomic_inc(&sh->count);
}
- } else if (!atomic_inc_not_zero(&sh->count)) {
- spin_lock(&conf->device_lock);
- if (!atomic_read(&sh->count)) {
- if (!test_bit(STRIPE_HANDLE, &sh->state))
- atomic_inc(&conf->active_stripes);
- BUG_ON(list_empty(&sh->lru) &&
- !test_bit(STRIPE_EXPANDING, &sh->state));
- list_del_init(&sh->lru);
- if (sh->group) {
- sh->group->stripes_cnt--;
- sh->group = NULL;
- }
- }
- atomic_inc(&sh->count);
- spin_unlock(&conf->device_lock);
}
} while (sh == NULL);
@@ -771,22 +773,6 @@ static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh
hash = stripe_hash_locks_hash(head_sector);
spin_lock_irq(conf->hash_locks + hash);
head = __find_stripe(conf, head_sector, conf->generation);
- if (head && !atomic_inc_not_zero(&head->count)) {
- spin_lock(&conf->device_lock);
- if (!atomic_read(&head->count)) {
- if (!test_bit(STRIPE_HANDLE, &head->state))
- atomic_inc(&conf->active_stripes);
- BUG_ON(list_empty(&head->lru) &&
- !test_bit(STRIPE_EXPANDING, &head->state));
- list_del_init(&head->lru);
- if (head->group) {
- head->group->stripes_cnt--;
- head->group = NULL;
- }
- }
- atomic_inc(&head->count);
- spin_unlock(&conf->device_lock);
- }
spin_unlock_irq(conf->hash_locks + hash);
if (!head)
--
1.9.0
Remove the unnecessary "!atomic_read(&sh->count)" check, as the previous
"atomic_inc_not_zero(&sh->count)" check assures sh->count to be 0.
The only reason I can think of that we need such check is to consider
the lock race issue.
First of all, I doubt there is another process could modify an in-cache
but zero referenced sh while it's being protected by a hash lock. Hence,
I would say sh->count will be consistent to 0 in that "if !atomic_inc_not_zero"
block.
Secondly, just assume there is a chance that someone outside the lock
modifies sh->count(by atomic_inc?). It could lead to some problem.
To make it clear, here I paste few lines of key code:
if (!atomic_inc_not_zero(&sh->count)) {
spin_lock(&conf->device_lock);
if (!atomic_read(&sh->count)) {
....
}
...
}
At the time we enter the first if block, sh->count is zero. And just assume
someone increases sh->count from somewhere while acquiring the lock,
the following if block will not be executed then, leaving some fileds,
such as conf->active_stripes, not being set properly.
So, we should execute the second if block whenever we entered the first
if block no matter sh->count stays with 0 or not.
Signed-off-by: Yuanhan Liu <[email protected]>
---
Neil, I'm a bit concerned that I missed something in this patch. Please
kindly correct me if I'm wrong :)
---
drivers/md/raid5.c | 18 ++++++++----------
1 file changed, 8 insertions(+), 10 deletions(-)
diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
index e7fa818..17ece2a 100644
--- a/drivers/md/raid5.c
+++ b/drivers/md/raid5.c
@@ -570,16 +570,14 @@ static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
if (sh->sector == sector && sh->generation == generation) {
if (!atomic_inc_not_zero(&sh->count)) {
spin_lock(&conf->device_lock);
- if (!atomic_read(&sh->count)) {
- if (!test_bit(STRIPE_HANDLE, &sh->state))
- atomic_inc(&conf->active_stripes);
- BUG_ON(list_empty(&sh->lru) &&
- !test_bit(STRIPE_EXPANDING, &sh->state));
- list_del_init(&sh->lru);
- if (sh->group) {
- sh->group->stripes_cnt--;
- sh->group = NULL;
- }
+ if (!test_bit(STRIPE_HANDLE, &sh->state))
+ atomic_inc(&conf->active_stripes);
+ BUG_ON(list_empty(&sh->lru) &&
+ !test_bit(STRIPE_EXPANDING, &sh->state));
+ list_del_init(&sh->lru);
+ if (sh->group) {
+ sh->group->stripes_cnt--;
+ sh->group = NULL;
}
atomic_inc(&sh->count);
spin_unlock(&conf->device_lock);
--
1.9.0
On Wed, 6 May 2015 17:45:49 +0800 Yuanhan Liu <[email protected]>
wrote:
> Move the code that put one idle sh(hot in cache, but happens to be
> zero referenced) back to active stage to __find_stripe(). Because
> that's what need to do every time you invoke __find_stripe().
>
> Moving it there avoids duplicate code, as well as makes a bit more
> sense, IMO, as it tells a whole story now.
Thanks for this. It is a good cleanup.
However I don't want to make any new changes to the RAID5 code until I find a
couple of bugs that I'm hunting. So I won't apply it just yet.
Remind me in a couple of weeks if I seem to have forgotten.
Thanks,
NeilBrown
>
> Signed-off-by: Yuanhan Liu <[email protected]>
> ---
> drivers/md/raid5.c | 50 ++++++++++++++++++--------------------------------
> 1 file changed, 18 insertions(+), 32 deletions(-)
>
> diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
> index 77dfd72..e7fa818 100644
> --- a/drivers/md/raid5.c
> +++ b/drivers/md/raid5.c
> @@ -567,8 +567,25 @@ static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
>
> pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
> hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
> - if (sh->sector == sector && sh->generation == generation)
> + if (sh->sector == sector && sh->generation == generation) {
> + if (!atomic_inc_not_zero(&sh->count)) {
> + spin_lock(&conf->device_lock);
> + if (!atomic_read(&sh->count)) {
> + if (!test_bit(STRIPE_HANDLE, &sh->state))
> + atomic_inc(&conf->active_stripes);
> + BUG_ON(list_empty(&sh->lru) &&
> + !test_bit(STRIPE_EXPANDING, &sh->state));
> + list_del_init(&sh->lru);
> + if (sh->group) {
> + sh->group->stripes_cnt--;
> + sh->group = NULL;
> + }
> + }
> + atomic_inc(&sh->count);
> + spin_unlock(&conf->device_lock);
> + }
> return sh;
> + }
> pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
> return NULL;
> }
> @@ -698,21 +715,6 @@ get_active_stripe(struct r5conf *conf, sector_t sector,
> init_stripe(sh, sector, previous);
> atomic_inc(&sh->count);
> }
> - } else if (!atomic_inc_not_zero(&sh->count)) {
> - spin_lock(&conf->device_lock);
> - if (!atomic_read(&sh->count)) {
> - if (!test_bit(STRIPE_HANDLE, &sh->state))
> - atomic_inc(&conf->active_stripes);
> - BUG_ON(list_empty(&sh->lru) &&
> - !test_bit(STRIPE_EXPANDING, &sh->state));
> - list_del_init(&sh->lru);
> - if (sh->group) {
> - sh->group->stripes_cnt--;
> - sh->group = NULL;
> - }
> - }
> - atomic_inc(&sh->count);
> - spin_unlock(&conf->device_lock);
> }
> } while (sh == NULL);
>
> @@ -771,22 +773,6 @@ static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh
> hash = stripe_hash_locks_hash(head_sector);
> spin_lock_irq(conf->hash_locks + hash);
> head = __find_stripe(conf, head_sector, conf->generation);
> - if (head && !atomic_inc_not_zero(&head->count)) {
> - spin_lock(&conf->device_lock);
> - if (!atomic_read(&head->count)) {
> - if (!test_bit(STRIPE_HANDLE, &head->state))
> - atomic_inc(&conf->active_stripes);
> - BUG_ON(list_empty(&head->lru) &&
> - !test_bit(STRIPE_EXPANDING, &head->state));
> - list_del_init(&head->lru);
> - if (head->group) {
> - head->group->stripes_cnt--;
> - head->group = NULL;
> - }
> - }
> - atomic_inc(&head->count);
> - spin_unlock(&conf->device_lock);
> - }
> spin_unlock_irq(conf->hash_locks + hash);
>
> if (!head)
On Fri, May 08, 2015 at 03:28:00PM +1000, NeilBrown wrote:
> On Wed, 6 May 2015 17:45:49 +0800 Yuanhan Liu <[email protected]>
> wrote:
>
> > Move the code that put one idle sh(hot in cache, but happens to be
> > zero referenced) back to active stage to __find_stripe(). Because
> > that's what need to do every time you invoke __find_stripe().
> >
> > Moving it there avoids duplicate code, as well as makes a bit more
> > sense, IMO, as it tells a whole story now.
>
> Thanks for this. It is a good cleanup.
>
> However I don't want to make any new changes to the RAID5 code until I find a
> couple of bugs that I'm hunting. So I won't apply it just yet.
> Remind me in a couple of weeks if I seem to have forgotten.
Got it. Thanks.
--yliu
>
> >
> > Signed-off-by: Yuanhan Liu <[email protected]>
> > ---
> > drivers/md/raid5.c | 50 ++++++++++++++++++--------------------------------
> > 1 file changed, 18 insertions(+), 32 deletions(-)
> >
> > diff --git a/drivers/md/raid5.c b/drivers/md/raid5.c
> > index 77dfd72..e7fa818 100644
> > --- a/drivers/md/raid5.c
> > +++ b/drivers/md/raid5.c
> > @@ -567,8 +567,25 @@ static struct stripe_head *__find_stripe(struct r5conf *conf, sector_t sector,
> >
> > pr_debug("__find_stripe, sector %llu\n", (unsigned long long)sector);
> > hlist_for_each_entry(sh, stripe_hash(conf, sector), hash)
> > - if (sh->sector == sector && sh->generation == generation)
> > + if (sh->sector == sector && sh->generation == generation) {
> > + if (!atomic_inc_not_zero(&sh->count)) {
> > + spin_lock(&conf->device_lock);
> > + if (!atomic_read(&sh->count)) {
> > + if (!test_bit(STRIPE_HANDLE, &sh->state))
> > + atomic_inc(&conf->active_stripes);
> > + BUG_ON(list_empty(&sh->lru) &&
> > + !test_bit(STRIPE_EXPANDING, &sh->state));
> > + list_del_init(&sh->lru);
> > + if (sh->group) {
> > + sh->group->stripes_cnt--;
> > + sh->group = NULL;
> > + }
> > + }
> > + atomic_inc(&sh->count);
> > + spin_unlock(&conf->device_lock);
> > + }
> > return sh;
> > + }
> > pr_debug("__stripe %llu not in cache\n", (unsigned long long)sector);
> > return NULL;
> > }
> > @@ -698,21 +715,6 @@ get_active_stripe(struct r5conf *conf, sector_t sector,
> > init_stripe(sh, sector, previous);
> > atomic_inc(&sh->count);
> > }
> > - } else if (!atomic_inc_not_zero(&sh->count)) {
> > - spin_lock(&conf->device_lock);
> > - if (!atomic_read(&sh->count)) {
> > - if (!test_bit(STRIPE_HANDLE, &sh->state))
> > - atomic_inc(&conf->active_stripes);
> > - BUG_ON(list_empty(&sh->lru) &&
> > - !test_bit(STRIPE_EXPANDING, &sh->state));
> > - list_del_init(&sh->lru);
> > - if (sh->group) {
> > - sh->group->stripes_cnt--;
> > - sh->group = NULL;
> > - }
> > - }
> > - atomic_inc(&sh->count);
> > - spin_unlock(&conf->device_lock);
> > }
> > } while (sh == NULL);
> >
> > @@ -771,22 +773,6 @@ static void stripe_add_to_batch_list(struct r5conf *conf, struct stripe_head *sh
> > hash = stripe_hash_locks_hash(head_sector);
> > spin_lock_irq(conf->hash_locks + hash);
> > head = __find_stripe(conf, head_sector, conf->generation);
> > - if (head && !atomic_inc_not_zero(&head->count)) {
> > - spin_lock(&conf->device_lock);
> > - if (!atomic_read(&head->count)) {
> > - if (!test_bit(STRIPE_HANDLE, &head->state))
> > - atomic_inc(&conf->active_stripes);
> > - BUG_ON(list_empty(&head->lru) &&
> > - !test_bit(STRIPE_EXPANDING, &head->state));
> > - list_del_init(&head->lru);
> > - if (head->group) {
> > - head->group->stripes_cnt--;
> > - head->group = NULL;
> > - }
> > - }
> > - atomic_inc(&head->count);
> > - spin_unlock(&conf->device_lock);
> > - }
> > spin_unlock_irq(conf->hash_locks + hash);
> >
> > if (!head)
>