mjhjmtu

I have "page allocation failure" reported on my system:

[some_app]: page allocation failure: order:4, mode:0x2040d0

Could someone please explain what that mode exactly stands for? Am I right that this is for the following GFP flags: GFP_NOTRACK | GFP_COMP | GFP_WAIT | GFP_IO | GFP_FS ?

Kernel version is 3.10.0-693.21.1.el7.AV1.x86_64.

The flags seem to be defined in file <kernel source directory>/include/linux/gfp.h, and at least on kernel 4.9.105, mode 0x2040d0 would seem to map to:

GFP_NOTRACK | GFP_COMP | GFP_FS | GFP_IO | GFP_RECLAIMABLE

But if I Google for flag definitions, I see in some sources value 0x10 defined as GFP_WAIT instead of GFP_RECLAIMABLE, which seems to match your source.

This LWN discussion might be useful reading, but the best description I can see is in the comments in include/linux/gfp.h file.

In general, these mode flags modify the working of the page allocator.

• GFP_NOTRACK: avoids tracking with kmemcheck.


• GFP_COMP: address compound page metadata


• GFP_FS: indicates that the allocator can call down the the low-level filesystem to reclaim pages if necessary; if this is cleared, I think would indicate the allocation is for some filesystem code that may be holding locks... which might be important when using a swap file, for example.


• GFP_IO: indicates that the allocator can start physical I/O to reclaim pages to satisfy this request.


• GFP_RECLAIMABLE: "[this] is used for slab allocations that specify SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers." This flag is apparently used by memory allocations for filesystems. Basically it seems to mean that there is a kernel function (a shrinker) that can be called to free or minimize this allocation if necessary.