Re: [RFC PATCH v2] mm/vmalloc: fix incorrect __vmap_pages_range_noflush() if vm_area_alloc_pages() from high order fallback to order0

[Baoquan He <bhe@redhat.com>]

On 07/30/24 at 11:24am, Hailong Liu wrote:
> On Mon, 29. Jul 09:48, Baoquan He wrote:
> [...]
> > Hi Hailong,
> >
> > Please feel free to collect them to post formal patch, maybe two
> > patches, one is to allow non-nofail to fallback to order-0 in
> > vm_area_alloc_pages(), the other is passing out the fallbacked
> > page_order to vmap_pages_range() if it's OK.
> Sorry for late response. I personally prefer to revert part of
> the problematic patch. There are several reasons:
> - Save memory usage if high order allocation failed.
> - If nofail and fallback to order0 in vmalloc huge allocation,
> actually the allocation is alighed with PMD_SIZE or not PAGE_SHIFT.

It's OK, maybe you can post patch to show what it looks like,
we can review and discuss there.

> 
> You might be concerned about performance issues. But IMO,
> - If we fallback to order0, we can make use of bulk allocator.
> - Maybe we can remove VM_ALLOW_HUGE_VMAP in kvmalloc.
> 
> I am not sure if I have misunderstood anything. If I have, please
> let me know.
> >
> > Thanks
> > Baoquan
> >
> >
> 
> --
> help you, help me,
> Hailong.
>