lkml.org 
[lkml]   [2012]   [Aug]   [14]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: efi_pstore: question about how to remove create_sysfs_entry() from a write callback.
On Tue, Aug 14, 2012 at 9:05 AM, Seiji Aguchi <seiji.aguchi@hds.com> wrote:
> Hi,
>
> I'm sending an email to discuss how to remove create_sysfs_entry() from a write callback.
>
> [Problem]
>
> Current efi_pstore creates sysfs entries ,which enable users to access to NVRAM, in a write callback.
> If a kernel panic happens in interrupt contexts, pstore may fail because it could sleep due to dynamic
> memory allocations during creating sysfs entries.
>
> To resolve the problem above, my goal here is removing create_sysfs_entry() from a write callback.
>
> [Ideas]
>
> (1) Introduce a workqueue updating sysfs entries
>
> To remove create_sysfs_entry() from a write callback,
> It seems to be possible if efi_pstore updates its sysfs files
> by scanning existing entries in NVRAM with a GetNextVariable()
> in a workqueue.
>
>
> I created a prototype patch based on an idea above but can't avoid a race
> between SetVariable() in a write callback and GetNextVariable() in a workqueue.
> It is not guaranteed by EFI specification.
>
> EFI 2.3.1 specification, page 217.
> <snip>
> Calls to SetVariable() between calls to
> GetNextVariableName() may produce unpredictable results.
> <snip>

Can we not serialize this with &efivars->lock if it is updated to
disable interrupts? A loop in the workqueue that locks, iterates
through ->get_next_variable, and compares against searches in
efivars->list should work, no?

>
>
> (2) Don't support sysfs entries in efi_pstore.
>
> Another idea is _not_ updating sysfs entries at all in efi_pstore.
> This can avoid a race SetVariable() and GetNextVariable().
>
> write callback
> - simply write a new entry with SetVariable().
> - This fits a discussion about holding multiple logs in a thread below.
> http://marc.info/?l=linux-kernel&m=134316268011854&w=2
>
> erase callback
> - simply erase an existing entry with SetVariable().
>
> read callback
> - Scaning existing entries with GetNextVariable().
> We can avoid a race between GetNextVariable() in a read callback
> and SetVariable() in a write/erase callback by protecting them with efi_lock.
>
> IMO, idea (2) is reasonable because we already have an interface, /dev/pstore, which users can access
> to NVRAM and we don't need to support multiple user interfaces.
>
> Any comments are welcome.
>
> Seiji


\
 
 \ /
  Last update: 2012-08-14 20:42    [W:0.045 / U:0.472 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site