[lkml]   [2012]   [Apr]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [RFC] propagate gfp_t to page table alloc functions
On 04/24/2012 03:13 PM, Nick Piggin wrote:

> 2012/4/24 Minchan Kim <>:
>> On 04/24/2012 02:16 PM, KAMEZAWA Hiroyuki wrote:
>>> (2012/04/23 17:55), Minchan Kim wrote:
>>>> As I test some code, I found a problem about deadlock by lockdep.
>>>> The reason I saw the message is __vmalloc calls map_vm_area which calls
>>>> pud/pmd_alloc without gfp_t. so although we call __vmalloc with
>>>> GFP_ATOMIC or GFP_NOIO, it ends up allocating pages with GFP_KERNEL.
>>>> The should be a BUG. This patch fixes it by passing gfp_to to low page
>>>> table allocate functions.
>>>> Signed-off-by: Minchan Kim <>
>>> Hmm ? vmalloc should support GFP_ATOMIC ?
>> I'm not sure but alloc_large_system_hash already has used.
>> And it's not specific on GFP_ATOMIC.
>> We have to care of GFP_NOFS and GFP_NOIO to prevent deadlock on reclaim
>> context.
>> There are some places to use GFP_NOFS and we don't emit any warning
>> message in case of that.
> What's the lockdep warning?

It's just some private-test code, not-mainlined and lockdep warning is like this.

[ INFO: inconsistent lock state ]
3.4.0-rc3-next-20120417+ #80 Not tainted
inconsistent {RECLAIM_FS-ON-W} -> {IN-RECLAIM_FS-R} usage.

It seems test code calls vmalloc inside reclaim context so that it enters
reclaim context, again by map_vm_area which allocates pages with GFP_KERNEL.

Of course, I can avoid this problem by fixing the caller but during I look into
this problem, found other places to use gfp_t with "context restriction".


> vmalloc was never supposed to use gfp flags for allocation "context"
> restriction. I.e., it
> was always supposed to have blocking, fs, and io capable allocation
> context. The flags
> were supposed to be a memory type modifier.

You mean "zone modifiers"?

> These different classes of flags is a bit of a problem and source of
> confusion we have.
> We should be doing more checks for them, of course.

It might need some warning in __vmalloc and family which use gfp_t
if the caller use context flags.

> I suspect you need to fix the caller?

Hmm, there are several places to use GFP_NOIO and GFP_NOFS even, GFP_ATOMIC.
I believe it's not trivial now.

> Thanks,
> Nick
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to For more info on Linux MM,
> see: .
> Fight unfair telecom internet charges in Canada: sign
> Don't email: <a href=mailto:""> </a>

Kind regards,
Minchan Kim

 \ /
  Last update: 2012-04-24 09:21    [W:0.111 / U:2.888 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site