diff mbox series

xen/blkfront: correct purging of persistent grants

Message ID 20180928072827.14229-1-jgross@suse.com (mailing list archive)
State New, archived
Headers show
Series xen/blkfront: correct purging of persistent grants | expand

Commit Message

Jürgen Groß Sept. 28, 2018, 7:28 a.m. UTC
Commit a46b53672b2c2e3770b38a4abf90d16364d2584b ("xen/blkfront: cleanup
stale persistent grants") introduced a regression as purged persistent
grants were not pu into the list of free grants again. Correct that.

Signed-off-by: Juergen Gross <jgross@suse.com>
---
 drivers/block/xen-blkfront.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Comments

Alan Robinson Sept. 28, 2018, 7:58 a.m. UTC | #1
On Fri, Sep 28, 2018 at 09:28:27AM +0200, Juergen Gross wrote:
> Commit a46b53672b2c2e3770b38a4abf90d16364d2584b ("xen/blkfront: cleanup
> stale persistent grants") introduced a regression as purged persistent
> grants were not pu into the list of free grants again. Correct that.
s/pu /put /

Alan
Boris Ostrovsky Sept. 28, 2018, 12:45 p.m. UTC | #2
On 9/28/18 3:28 AM, Juergen Gross wrote:
> Commit a46b53672b2c2e3770b38a4abf90d16364d2584b ("xen/blkfront: cleanup
> stale persistent grants") introduced a regression as purged persistent
> grants were not pu into the list of free grants again. Correct that.
>
> Signed-off-by: Juergen Gross <jgross@suse.com>
> ---
>  drivers/block/xen-blkfront.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
> index a71d817e900d..429d20131c7e 100644
> --- a/drivers/block/xen-blkfront.c
> +++ b/drivers/block/xen-blkfront.c
> @@ -2670,8 +2670,8 @@ static void purge_persistent_grants(struct blkfront_info *info)
>  			list_del(&gnt_list_entry->node);
>  			gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
>  			rinfo->persistent_gnts_c--;
> -			__free_page(gnt_list_entry->page);
> -			kfree(gnt_list_entry);
> +			gnt_list_entry->gref = GRANT_INVALID_REF;
> +			list_add_tail(&gnt_list_entry->node, &rinfo->grants);

Sorry, I don't follow this. What is the purpose of removing the grant
from rinfo->grants list with list_del() and then adding it back with
list_add_tail()?

-boris


>  		}
>  
>  		spin_unlock_irqrestore(&rinfo->ring_lock, flags);
Jürgen Groß Sept. 28, 2018, 1:13 p.m. UTC | #3
On 28/09/2018 14:45, Boris Ostrovsky wrote:
> On 9/28/18 3:28 AM, Juergen Gross wrote:
>> Commit a46b53672b2c2e3770b38a4abf90d16364d2584b ("xen/blkfront: cleanup
>> stale persistent grants") introduced a regression as purged persistent
>> grants were not pu into the list of free grants again. Correct that.
>>
>> Signed-off-by: Juergen Gross <jgross@suse.com>
>> ---
>>  drivers/block/xen-blkfront.c | 4 ++--
>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
>> index a71d817e900d..429d20131c7e 100644
>> --- a/drivers/block/xen-blkfront.c
>> +++ b/drivers/block/xen-blkfront.c
>> @@ -2670,8 +2670,8 @@ static void purge_persistent_grants(struct blkfront_info *info)
>>  			list_del(&gnt_list_entry->node);
>>  			gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
>>  			rinfo->persistent_gnts_c--;
>> -			__free_page(gnt_list_entry->page);
>> -			kfree(gnt_list_entry);
>> +			gnt_list_entry->gref = GRANT_INVALID_REF;
>> +			list_add_tail(&gnt_list_entry->node, &rinfo->grants);
> 
> Sorry, I don't follow this. What is the purpose of removing the grant
> from rinfo->grants list with list_del() and then adding it back with
> list_add_tail()?

The persistent grants are at the list head and the non-persistent ones
at the tail.


Juergen
Boris Ostrovsky Sept. 28, 2018, 1:33 p.m. UTC | #4
On 9/28/18 9:13 AM, Juergen Gross wrote:
> On 28/09/2018 14:45, Boris Ostrovsky wrote:
>> On 9/28/18 3:28 AM, Juergen Gross wrote:
>>> Commit a46b53672b2c2e3770b38a4abf90d16364d2584b ("xen/blkfront: cleanup
>>> stale persistent grants") introduced a regression as purged persistent
>>> grants were not pu into the list of free grants again. Correct that.
>>>
>>> Signed-off-by: Juergen Gross <jgross@suse.com>
>>> ---
>>>  drivers/block/xen-blkfront.c | 4 ++--
>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
>>> index a71d817e900d..429d20131c7e 100644
>>> --- a/drivers/block/xen-blkfront.c
>>> +++ b/drivers/block/xen-blkfront.c
>>> @@ -2670,8 +2670,8 @@ static void purge_persistent_grants(struct blkfront_info *info)
>>>  			list_del(&gnt_list_entry->node);
>>>  			gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
>>>  			rinfo->persistent_gnts_c--;
>>> -			__free_page(gnt_list_entry->page);
>>> -			kfree(gnt_list_entry);
>>> +			gnt_list_entry->gref = GRANT_INVALID_REF;
>>> +			list_add_tail(&gnt_list_entry->node, &rinfo->grants);
>> Sorry, I don't follow this. What is the purpose of removing the grant
>> from rinfo->grants list with list_del() and then adding it back with
>> list_add_tail()?
> The persistent grants are at the list head and the non-persistent ones
> at the tail.

Oh, I didn't realize that. But isn't that an optimization (and so not
following this rule should not cause errors)?

-boris
Jürgen Groß Sept. 28, 2018, 1:52 p.m. UTC | #5
On 28/09/2018 15:33, Boris Ostrovsky wrote:
> On 9/28/18 9:13 AM, Juergen Gross wrote:
>> On 28/09/2018 14:45, Boris Ostrovsky wrote:
>>> On 9/28/18 3:28 AM, Juergen Gross wrote:
>>>> Commit a46b53672b2c2e3770b38a4abf90d16364d2584b ("xen/blkfront: cleanup
>>>> stale persistent grants") introduced a regression as purged persistent
>>>> grants were not pu into the list of free grants again. Correct that.
>>>>
>>>> Signed-off-by: Juergen Gross <jgross@suse.com>
>>>> ---
>>>>  drivers/block/xen-blkfront.c | 4 ++--
>>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
>>>> index a71d817e900d..429d20131c7e 100644
>>>> --- a/drivers/block/xen-blkfront.c
>>>> +++ b/drivers/block/xen-blkfront.c
>>>> @@ -2670,8 +2670,8 @@ static void purge_persistent_grants(struct blkfront_info *info)
>>>>  			list_del(&gnt_list_entry->node);
>>>>  			gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
>>>>  			rinfo->persistent_gnts_c--;
>>>> -			__free_page(gnt_list_entry->page);
>>>> -			kfree(gnt_list_entry);
>>>> +			gnt_list_entry->gref = GRANT_INVALID_REF;
>>>> +			list_add_tail(&gnt_list_entry->node, &rinfo->grants);
>>> Sorry, I don't follow this. What is the purpose of removing the grant
>>> from rinfo->grants list with list_del() and then adding it back with
>>> list_add_tail()?
>> The persistent grants are at the list head and the non-persistent ones
>> at the tail.
> 
> Oh, I didn't realize that. But isn't that an optimization (and so not
> following this rule should not cause errors)?

In theory: yes.

The persistent grant handling is rather complicated so I'd like to make
sure not to deviate from the common standards.

When I find some time I want to modify the persistent grant handling to
be more explicit and controlled completely by the frontend (within the
backend defined limits, of course). Until then we should try to modify
persistent grants not too much IMO.


Juergen
Boris Ostrovsky Sept. 28, 2018, 2:03 p.m. UTC | #6
On 9/28/18 9:52 AM, Juergen Gross wrote:
> On 28/09/2018 15:33, Boris Ostrovsky wrote:
>> On 9/28/18 9:13 AM, Juergen Gross wrote:
>>> On 28/09/2018 14:45, Boris Ostrovsky wrote:
>>>> On 9/28/18 3:28 AM, Juergen Gross wrote:
>>>>> Commit a46b53672b2c2e3770b38a4abf90d16364d2584b ("xen/blkfront: cleanup
>>>>> stale persistent grants") introduced a regression as purged persistent
>>>>> grants were not pu into the list of free grants again. Correct that.
>>>>>
>>>>> Signed-off-by: Juergen Gross <jgross@suse.com>
>>>>> ---
>>>>>  drivers/block/xen-blkfront.c | 4 ++--
>>>>>  1 file changed, 2 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
>>>>> index a71d817e900d..429d20131c7e 100644
>>>>> --- a/drivers/block/xen-blkfront.c
>>>>> +++ b/drivers/block/xen-blkfront.c
>>>>> @@ -2670,8 +2670,8 @@ static void purge_persistent_grants(struct blkfront_info *info)
>>>>>  			list_del(&gnt_list_entry->node);
>>>>>  			gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
>>>>>  			rinfo->persistent_gnts_c--;
>>>>> -			__free_page(gnt_list_entry->page);
>>>>> -			kfree(gnt_list_entry);
>>>>> +			gnt_list_entry->gref = GRANT_INVALID_REF;
>>>>> +			list_add_tail(&gnt_list_entry->node, &rinfo->grants);
>>>> Sorry, I don't follow this. What is the purpose of removing the grant
>>>> from rinfo->grants list with list_del() and then adding it back with
>>>> list_add_tail()?
>>> The persistent grants are at the list head and the non-persistent ones
>>> at the tail.
>> Oh, I didn't realize that. But isn't that an optimization (and so not
>> following this rule should not cause errors)?
> In theory: yes.
>
> The persistent grant handling is rather complicated so I'd like to make
> sure not to deviate from the common standards.

I am not arguing with correctness of your patch, so

Reviewed-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>

but I am a little surprised that it fixes Sander's problem.


-boris


>
> When I find some time I want to modify the persistent grant handling to
> be more explicit and controlled completely by the frontend (within the
> backend defined limits, of course). Until then we should try to modify
> persistent grants not too much IMO.
>
>
> Juergen
Jens Axboe Sept. 28, 2018, 3:41 p.m. UTC | #7
On 9/28/18 1:28 AM, Juergen Gross wrote:
> Commit a46b53672b2c2e3770b38a4abf90d16364d2584b ("xen/blkfront: cleanup
> stale persistent grants") introduced a regression as purged persistent
> grants were not pu into the list of free grants again. Correct that.

I'll apply this for 4.19, and if things unexpectedly don't work out,
we can always revert the series next week.
diff mbox series

Patch

diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index a71d817e900d..429d20131c7e 100644
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -2670,8 +2670,8 @@  static void purge_persistent_grants(struct blkfront_info *info)
 			list_del(&gnt_list_entry->node);
 			gnttab_end_foreign_access(gnt_list_entry->gref, 0, 0UL);
 			rinfo->persistent_gnts_c--;
-			__free_page(gnt_list_entry->page);
-			kfree(gnt_list_entry);
+			gnt_list_entry->gref = GRANT_INVALID_REF;
+			list_add_tail(&gnt_list_entry->node, &rinfo->grants);
 		}
 
 		spin_unlock_irqrestore(&rinfo->ring_lock, flags);