Re: LRU_GEN v2 & v5.12 build issue

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hi Yu Zhao

CONFIG_MAXSMP=n worked!

The main distros are with CONFIG_MAXSMP=y. It may not make sense on desktops, even for an amd threadripper, but in datacenter / HPC, for example, it is interesting to add maxsmp compatibility that virtually removes the n cpus limit.

Thanks,
Alexandre Frade




---- Ativado Sáb, 01 mai 2021 23:45:11 -0300 Yu Zhao <yuzhao@xxxxxxxxxx> escreveu ----

On Sat, May 1, 2021 at 8:39 PM Yu Zhao <yuzhao@xxxxxxxxxx> wrote:
>
> On Sat, May 1, 2021 at 8:31 PM Alexandre Frade <kernel@xxxxxxxxxx> wrote:
> >
> > Dear Yu Zhao,
> >
> > I don't know if it's a known issue with the 5.12 kernel released [9f4ad9e]:
> >
> > LRU_GEN v2 & v5.12 final:
>
> Hi Alexandre,
>
> I think I hit this problem with Ubuntu too. Does your config include
> the following?
> CONFIG_MAXSMP=y
> CONFIG_NODES_SHIFT=10
> CONFIG_NR_CPUS=8192
> If so, then it's the same problem: there are no spare bits left in
> page->flags. Disabling
> CONFIG_MAXSMP should fix it, unless you really want to use 8192 CPUs
> and 1024 NUMA nodes :)
>
> Please let me know. Thanks!

Sorry I forgot to clarify that this option should be disabled
regardless whether you want to use the multigenerational lru or not.

This option is for debugging and will bring unnecessary memory
overhead to production systems.



[Index of Archives]     [Linux ARM Kernel]     [Linux ARM]     [Linux Omap]     [Fedora ARM]     [IETF Annouce]     [Bugtraq]     [Linux OMAP]     [Linux MIPS]     [eCos]     [Asterisk Internet PBX]     [Linux API]

  Powered by Linux