Re: [Libvirt-cim] [PATCH V3 04/11] CSI, add lock to protect shared data
Since I don't have the email to reply-to, here is a link: https://www.redhat.com/archives/libvirt-cim/2012-December/msg00026.html Why was the decision to reuse the lifecycle_mutex rather than generating a data specific mutex? Seems like that mutex is used for other operations to protect filters and indications. Thus unless the two are somehow tied, I really think it'd be better to create/use a data specific mutex. Furthermore, there's a few mutex adds around thread mgmt that probably should be a separate patch. That is - the non "dom_list" protection additions appear to be different. John
于 2013-3-11 22:41, John Ferlan 写道:
Since I don't have the email to reply-to, here is a link:
https://www.redhat.com/archives/libvirt-cim/2012-December/msg00026.html
Why was the decision to reuse the lifecycle_mutex rather than generating a data specific mutex? Seems like that mutex is used for other operations to protect filters and indications. Thus unless the two are somehow tied, I really think it'd be better to create/use a data specific mutex.
Furthermore, there's a few mutex adds around thread mgmt that probably should be a separate patch. That is - the non "dom_list" protection additions appear to be different.
John
It is just the quickest way to fix bug, it should be a separate lock just as mentioned in the commit message for performance, you can form a patch to fix it above or replace this patch.
_______________________________________________ Libvirt-cim mailing list Libvirt-cim@redhat.com https://www.redhat.com/mailman/listinfo/libvirt-cim
-- Best Regards Wenchao Xia
于 2013-3-12 13:10, Wenchao Xia 写道:
于 2013-3-11 22:41, John Ferlan 写道:
Since I don't have the email to reply-to, here is a link:
https://www.redhat.com/archives/libvirt-cim/2012-December/msg00026.html
Why was the decision to reuse the lifecycle_mutex rather than generating a data specific mutex? Seems like that mutex is used for other operations to protect filters and indications. Thus unless the two are somehow tied, I really think it'd be better to create/use a data specific mutex.
Furthermore, there's a few mutex adds around thread mgmt that probably should be a separate patch. That is - the non "dom_list" protection additions appear to be different.
John
It is just the quickest way to fix bug, it should be a separate lock just as mentioned in the commit message for performance, you can form a patch to fix it above or replace this patch.
_______________________________________________ Libvirt-cim mailing list Libvirt-cim@redhat.com https://www.redhat.com/mailman/listinfo/libvirt-cim
Rechecked the code, the lifecycle_mutex is original meant to protect global variables, actually it should be declared as: struct CSI_gloable_data { pthread_mutex_t lifecycle_mutex, bool lifecycle_enabled, csi_thread_data_t csi_thread_data[CSI_NUM_PLATFORMS], }; Since above is a pure code structure improve and need quite a change, I hope to leave it untouched with a commit message, and rebase become easier. So this patch will only fix lock missing problem, if worthy, later we can send a patch fix the code struture problem as "CSI: folder global lock and structure ", and a patch improves lock as "CSI: use smaller lock". -- Best Regards Wenchao Xia
participants (2)
-
John Ferlan -
Wenchao Xia