Re: urcu lfht and free listing

Re: urcu lfht and free listing

[Mathieu Desnoyers]

----- On Jun 10, 2016, at 6:58 AM, D'Alessandro, Luke K ldalessa@indiana.edu wrote:

> Hi List,
> 
> Is it safe to free list and reuse hash table nodes (cds_lfht_node structures
> used in cds_lfht_add_replace, cds_lfht_iter_get_node/cds_lfht_del) without
> going through a synchronize_rcu() barrier?

Hi,

No, it is not safe.

From rculfhash.h:

cds_lfht_del:

 * After successful removal, a grace period must be waited for before
 * freeing the memory reserved for old node (which can be accessed with
 * cds_lfht_iter_get_node).

cds_lfht_add_replace:

 * After successful replacement, a grace period must be waited for before
 * freeing the memory reserved for the returned node.

cds_lfht_replace:

 * After successful replacement, a grace period must be waited for before
 * freeing the memory reserved for the old node (which can be accessed
 * with cds_lfht_iter_get_node).

You can either invoke synchronize_rcu() before re-using or freeing
the removed node, or use the call_rcu() or defer_rcu() mechanisms to
defer reclaim after a grace period (handled by worker threads).

Thanks,

Mathieu

> 
> Thanks,
> Luke
> _______________________________________________
> lttng-dev mailing list
> lttng-dev@lists.lttng.org
> https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
_______________________________________________
lttng-dev mailing list
lttng-dev@lists.lttng.org
https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev

Re: urcu lfht and free listing

[D'Alessandro, Luke K]

[-- Attachment #1.1: Type: text/plain, Size: 1973 bytes --]


> On Jun 10, 2016, at 2:22 PM, Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
> 
> ----- On Jun 10, 2016, at 6:58 AM, D'Alessandro, Luke K ldalessa@indiana.edu wrote:
> 
>> Hi List,
>> 
>> Is it safe to free list and reuse hash table nodes (cds_lfht_node structures
>> used in cds_lfht_add_replace, cds_lfht_iter_get_node/cds_lfht_del) without
>> going through a synchronize_rcu() barrier?
> 
> Hi,
> 
> No, it is not safe.
> 
> From rculfhash.h:
> 
> cds_lfht_del:
> 
> * After successful removal, a grace period must be waited for before
> * freeing the memory reserved for old node (which can be accessed with
> * cds_lfht_iter_get_node).
> 
> cds_lfht_add_replace:
> 
> * After successful replacement, a grace period must be waited for before
> * freeing the memory reserved for the returned node.
> 
> cds_lfht_replace:
> 
> * After successful replacement, a grace period must be waited for before
> * freeing the memory reserved for the old node (which can be accessed
> * with cds_lfht_iter_get_node).

Hi Mathieu,

Thanks for the response.

I did read that documentation but it only references “freeing” the memory which I read as “returning the memory to the allocator via free()” which can sometimes be an issue as it can cascade into an munmap that is concurrent with some reader. I wasn’t sure if simply reusing the memory was a problem.

> You can either invoke synchronize_rcu() before re-using or freeing
> the removed node, or use the call_rcu() or defer_rcu() mechanisms to
> defer reclaim after a grace period (handled by worker threads).

Great, thank you.

Thanks,

Luke

> 
> Thanks,
> 
> Mathieu
> 
>> 
>> Thanks,
>> Luke
>> _______________________________________________
>> lttng-dev mailing list
>> lttng-dev@lists.lttng.org
>> https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
> 
> --
> Mathieu Desnoyers
> EfficiOS Inc.
> http://www.efficios.com


[-- Attachment #1.2: Message signed with OpenPGP using GPGMail --]
[-- Type: application/pgp-signature, Size: 842 bytes --]

[-- Attachment #2: Type: text/plain, Size: 156 bytes --]

_______________________________________________
lttng-dev mailing list
lttng-dev@lists.lttng.org
https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev

Re: urcu lfht and free listing

[Mathieu Desnoyers]

----- On Jun 10, 2016, at 2:36 PM, D'Alessandro, Luke K ldalessa@indiana.edu wrote:

>> On Jun 10, 2016, at 2:22 PM, Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
>> wrote:
>> 
>> ----- On Jun 10, 2016, at 6:58 AM, D'Alessandro, Luke K ldalessa@indiana.edu
>> wrote:
>> 
>>> Hi List,
>>> 
>>> Is it safe to free list and reuse hash table nodes (cds_lfht_node structures
>>> used in cds_lfht_add_replace, cds_lfht_iter_get_node/cds_lfht_del) without
>>> going through a synchronize_rcu() barrier?
>> 
>> Hi,
>> 
>> No, it is not safe.
>> 
>> From rculfhash.h:
>> 
>> cds_lfht_del:
>> 
>> * After successful removal, a grace period must be waited for before
>> * freeing the memory reserved for old node (which can be accessed with
>> * cds_lfht_iter_get_node).
>> 
>> cds_lfht_add_replace:
>> 
>> * After successful replacement, a grace period must be waited for before
>> * freeing the memory reserved for the returned node.
>> 
>> cds_lfht_replace:
>> 
>> * After successful replacement, a grace period must be waited for before
>> * freeing the memory reserved for the old node (which can be accessed
>> * with cds_lfht_iter_get_node).
> 
> Hi Mathieu,
> 
> Thanks for the response.
> 
> I did read that documentation but it only references “freeing” the memory which
> I read as “returning the memory to the allocator via free()” which can
> sometimes be an issue as it can cascade into an munmap that is concurrent with
> some reader. I wasn’t sure if simply reusing the memory was a problem.

Indeed, it could be clearer. I just updated the comments in the master branch.

commit 5002434419b458f1de889b3ed1110b440d3bf5b0
Author: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Date:   Fri Jun 10 14:43:40 2016 -0400

    rculfhash: Documentation: clarify need for grace period before "re-using"
    
    Grace period must be waited for in case a node removed from the hash
    table is re-used, similarly to the reclaim use-case.
    
    Reported-by: Luke K D'Alessandro <ldalessa@indiana.edu>
    Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>

Thanks!

Mathieu

> 
>> You can either invoke synchronize_rcu() before re-using or freeing
>> the removed node, or use the call_rcu() or defer_rcu() mechanisms to
>> defer reclaim after a grace period (handled by worker threads).
> 
> Great, thank you.
> 
> Thanks,
> 
> Luke
> 
>> 
>> Thanks,
>> 
>> Mathieu
>> 
>>> 
>>> Thanks,
>>> Luke
>>> _______________________________________________
>>> lttng-dev mailing list
>>> lttng-dev@lists.lttng.org
>>> https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
>> 
>> --
>> Mathieu Desnoyers
>> EfficiOS Inc.
> > http://www.efficios.com

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
_______________________________________________
lttng-dev mailing list
lttng-dev@lists.lttng.org
https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev

urcu lfht and free listing

[D'Alessandro, Luke K]

Hi List,

Is it safe to free list and reuse hash table nodes (cds_lfht_node structures used in cds_lfht_add_replace, cds_lfht_iter_get_node/cds_lfht_del) without going through a synchronize_rcu() barrier?

Thanks,
Luke
_______________________________________________
lttng-dev mailing list
lttng-dev@lists.lttng.org
https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev