Re: [PATCH v7 00/15] zram: redesign zcomp and rework backends

[zhangbo56]

I reported performance numbers earlier; this is a follow-up focused on
stability.

I ran an 8-hour stress test of this series using the following workload:
16 memcgs each looping a kernel build with tight memory limits to push
anon memory out to the zram swap (zstd), plus continuous CPU hotplug
(~113ms per cycle, 254k cycles). Repeated contained memcg OOM kills kept
the per-CPU streams constantly torn down and re-created under load.

Result: ran to completion, no issues.
  * dmesg clean - no BUG/Oops/stall/lockup/hung task/warning, no zram
    driver messages.
  * zram io_stat: failed_reads=0, failed_writes=0, invalid_io=0 across
    ~3.56 billion operations.
  * All CPUs back online cleanly.

A shorter lz4 (stateless) run was also clean.

Environment: 7.3.0-rc5-mmnew+, x86_64, PREEMPT, 28 CPUs, 8 GB RAM.

For the whole series:
Tested-by: zhangbo56 <zhangbo56@xxxxxxxxxx>

Thanks,
Bo