Messages in this thread | | | Date | Fri, 22 Jul 2011 08:13:02 -0400 | From | Don Zickus <> | Subject | Re: [PATCH] pstore: change mutex locking to spin_locks |
| |
On Fri, Jul 22, 2011 at 08:33:22AM +0800, Huang Ying wrote: > On 07/22/2011 01:57 AM, Luck, Tony wrote: > >>> Is it safe to call pstore_mkfile with IRQ disabled? > >>> > >>> pstore_mkfile -> d_alloc_name -> d_alloc -> kmem_cache_alloc(, GFP_KERNEL). > >> > >> Don't know. But would that mean we would have to put the pstore_mkfile > >> on a workqueue then or something similar? > > > > That might be a good idea anyway. In the "oops" case we'd like the file > > to appear in the pstore filesystem if the system stayed healthy despite > > the oops[1]. There isn't any reason why the pstore entry must appear instantly. > > Delaying the creation would avoid running into problems related to the > > oops. > > For oops, it may be better to delay writing into something like > workqueue. But for panic, I think we should write the record to backend > (such as ERST) as soon as possible. So maybe it is better to write to > backend as soon as possible and delay writing to pstore filesystem.
In the panic case do we care if the pstore fs is mounted (which leads us to run pstore_mkfile)?
Actually it seems like most of the entry points into pstore_dumper would not require the fs to create a new file. I think the exception is an oops.
Cheers, Don
| |