On 02/22/2020 08:34 PM, Metztli Information Technology wrote:
On Sat, Feb 22, 2020 at 3:56 AM Edward Shishkin <edward.shishkin@xxxxxxxxx> wrote:
On 02/22/2020 05:51 AM, Metztli Information Technology wrote:
Niltze, Ed-
Starting with kernel 5.4.19, linux/fs/fs-writeback.c , at line 2066, changes
from
set_worker_desc("flush-%s", dev_name(wb->bdi->dev));
to
set_worker_desc("flush-%s", bdi_dev_name(wb->bdi));
I realized that when I was hacking 5.4.19 and, subsequently, my currently running kernel --as there was no RC3:
uname -a
Linux huitzilopochtli 5.4.0-4+reiser4.0.2-amd64 #1 SMP Debian 5.4.20-1+reiser4.0.2 (2020-02-14) x86_64 GNU/Linux
< https://sourceforge.net/projects/metztli-reiser4/files/Reiser4-SFRN-4.0.2_Linux-5.4.20-1-RC2_for-Debian_Buster/ >
I have just built 5.5.5-1 and realized, upon applying reiser4-for-5.5.1.patch.gz, that your patch should be modified
 accordingly; else, the patch will fail and will necessitate manual intervention, i.e., similar to 5.4.18 vs 5.4.19:
< https://metztli.it/readOnlyEphemeral/fs-writeback_c.png >
so, everything works fine after that "manual intervention"?
Indeed, Ed. No issues thus far. I upgraded my local dev environment to 5.4.20-1+reiser4 on February 15, 2020; in such environment I have built several PHP-7.3 iterations and I also built kernel 5.5.5-1+reiser4 yesterday. Additionally, replaced 5.4.20-1+reiser4 cloud flavour in a Google Compute Engine (GCE) customized Zstd transparent compression instance -- which hosts multiple PHP and Ruby applications and functions as a media / web server, as well -- without issues thus far.
All the problems comes from upstream. Specifically from VFS and the
block layer. Not from Debian.
I'll try to prepare instructions on how to find problematic commit by
bisecting.
The bad news is that the changes coming from there are clearly destructive.
Consider, for example, this "optimization":
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=aa65c29ce1b6e1990cd2c7d8004bbea7ff3aff38
What an idiot one need to be in order to lock up a lot of pages at once
at the VFS level? :((( Sigh..
