On 2019/9/19 0:38, Christoph Hellwig wrote:
On Mon, Sep 16, 2019 at 10:52:33AM -0600, Jens Axboe wrote:
On 9/16/19 3:06 AM, Christoph Hellwig wrote:
On Mon, Sep 16, 2019 at 07:46:29AM +0530, xiubli@xxxxxxxxxx wrote:
From: Xiubo Li <xiubli@xxxxxxxxxx>
To make the patch more readable and cleaner I just split them into 2
small ones to address the issue from @Ming Lei, thanks very much.
I'd be much happier to just see memalloc_noio_save +
memalloc_noio_restore calls in the right places over sprinkling even
more magic GFP_NOIO arguments.
Ugh, I always thought those were kind of lame and band aiding around
places where people are too lazy to fix the path to the gfp args.
Or maybe areas where it's just feasible.
The way I understood the discussion around the introduction of these
flags is that we want to phase out GFP_NOIO and GFP_NOFS in the long
run, given that ammending all calls is basically impossibly, while
marking contexts is pretty easy.
This was also my understanding about the memalloc_xx stuff. By going
through the code and the usage history of it, the best practice of using
it seems as Bart mentioned for the possibly deep call chains.
Thanks
