Re: [PATCH] vmalloc: Removing incorrect logs when vmalloc failed

[tiantao (H)]

在 2020/7/8 21:48, Uladzislau Rezki 写道:
> > > > On 07/07/2020 02:43 PM, Tian Tao wrote:
> > > > > It is not possible to increase size with vmalloc=<size> in arm64
> > > > > architecture and it will mislead.however vmalloc return failure
> > > > > is a rare occurrence in 'many architectures including arm64'.
> > > >
> > > > But there is a chance that vmalloc() might work on architectures
> > > > that support 'vmalloc=' command line i.e after a change and this
> > > > information here might be helpful in those cases.
> > > Agree. At least i see a few users of it:
> > >
> > > <snip>
> > > urezki@pc638:~/data/coding/linux-next.git$ grep -rn early_param ./arch/ | grep vmalloc
> > > ./arch/arm/mm/mmu.c:1152:early_param("vmalloc", early_vmalloc);
> > > ./arch/unicore32/mm/mmu.c:276:early_param("vmalloc", early_vmalloc);
> > > ./arch/x86/mm/pgtable_32.c:86:early_param("vmalloc", parse_vmalloc);
> > > urezki@pc638:~/data/coding/linux-next.git$
> > > <snip>
> > I'm actually having this problem with the arm64 architecture at centos 7.6
> > and pagesize is 64K.
> > I followed the prompts and added vmalloc=<size> to the command to increase
> > the size of the vmalloc.and found out it's not worked.
> > It took me some time to find out that this doesn't work for the arm64
> > architecture, so this log is misleading on arm64.
> Agree, it can take time to understand some code or logic behind of it.
> So in that case having good documentation or comments always help.
>
> > I think it's better not to be prompted than to be prompted incorrectly.
> > I'm sure there will be others with similar problems.
> > So I'd like to solve this problem this time, Please help me with your
> > suggestions.
> > If I change the PATCH to the following, will you accept it?
> Actually it is not up to me to decide what to take or not. Andrew Morton
> is the key person here :) I can just review or make some comments same
> as others.
>
> >        if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
> > +#ifdef CONFIG_ARM64 && CONFIG_XXX
> > +        pr_warn("vmap allocation for size %lu failed\n", size);
> > +#else
> >            pr_warn("vmap allocation for size %lu failed: use vmalloc=<size>
> > to increase size\n",
> >                size);
> > +#endif
> I do not have a strong opinion here, but counting arches seems odd.
> Maybe modify the string with following message:
>
> <snip>
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index 1f46c3b86f9f..0aa26bc128d7 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -1202,7 +1202,7 @@ static struct vmap_area *alloc_vmap_area(unsigned long size,
>          }
>   
>          if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit())
> -               pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size\n",
> +               pr_warn("vmap allocation for size %lu failed: use vmalloc=<size> to increase size, if your ARCH supports it\n",
>                          size);
>   
>          kmem_cache_free(vmap_area_cachep, va);
> <snip>
Thanks, that's a good suggestion, I'll send v2 as you suggested!
> --
> Vlad Rezki
>
> .