On Mon, 20 May 2019, Christoph Hellwig wrote: > On Mon, May 20, 2019 at 11:09:25AM +0200, Oliver Neukum wrote: > > we actually do. It is just higher up in the calling path: > > Perfect! > > > So, do we need to audit the mem_flags again? > > What are we supposed to use? GFP_KERNEL? > > GFP_KERNEL if you can block, GFP_ATOMIC if you can't for a good reason, > that is the allocation is from irq context or under a spinlock. If you > think you have a case where you think you don't want to block, but it > is not because of the above reasons we need to have a chat about the > details. What if the allocation requires the kernel to swap some old pages out to the backing store, but the backing store is on the device that the driver is managing? The swap can't take place until the current I/O operation is complete (assuming the driver can handle only one I/O operation at a time), and the current operation can't complete until the old pages are swapped out. Result: deadlock. Isn't that the whole reason for using GFP_NOIO in the first place? Alan Stern
