[PATCH v2 1/3] mm: vmalloc: Let user to control huge vmalloc default behavior

[Matthew Wilcox]

On Wed, Jan 19, 2022 at 09:44:20PM +0800, Kefeng Wang wrote:
> 
> On 2022/1/19 21:22, Matthew Wilcox wrote:
> > On Wed, Jan 19, 2022 at 08:57:58PM +0800, Kefeng Wang wrote:
> > > Only parts of our products wants this feature,  we add some interfaces which
> > > only
> > > 
> > > alloc hugevmalloc for them, eg,
> > > vmap_hugepage/vmalloc_hugepage/remap_vmalloc_hugepage_range..
> > > 
> > > for our products, but it's not the choice of most products, also add
> > > nohugevmalloc
> > > 
> > > for most products is expensive, so this is the reason for adding the patch.
> > > 
> > > more config/cmdline are more flexible for test/products，
> > But why do only some products want it?  What goes wrong if all products
> > enable it?  Features should be auto-tuning, not relying on admins to
> > understand them.
> 
> Because this feature will use more memory for vmalloc/vmap, that's why we
> add some explicit
> interfaces as said above in our kernel to control the user.

Have you validated that?  What sort of performance penalty do you see?