Messages in this thread |  | | Date | Tue, 26 Sep 2000 19:56:12 +0200 | From | Andrea Arcangeli <> | Subject | Re: refill_inactive() |
| |
On Tue, Sep 26, 2000 at 10:12:44AM -0700, Linus Torvalds wrote: > It should probably just be a GFP_USER (ie not the GFP_KERNEL "try very > hard").
GFP_KERNEL and GFP_USER have to try equally very hard until the machine runs _truly_ out of memory.
When the machine runs truly out of memory I think it would be better in 2.4.x not to cause a shortage of the atomic pool, but to make GFP_KERNEL work like GFP_USER (so not giving it access to the atomic pool). This is risky if some GFP_KERNEL allocation is deadlock prone (like raid1 before 2.4.x) and that's why it's better not to do that in 2.2.x to decrease the probability to hit those deadlocks.
GFP_BUFFER (getblk) should still access the whole atomic pool instead (because it's deadlock prone and fixing it would impact all fs).
Andrea - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org Please read the FAQ at http://www.tux.org/lkml/
|  |