Re: [PATCH v2 24/29] nios2: Module support

[Arnd Bergmann <arnd@arndb.de>]

On Tuesday 15 July 2014 13:12:37 Tobias Klauser wrote:
> On 2014-07-15 at 12:24:46 +0200, Arnd Bergmann <arnd@arndb.de> wrote:
> > On Tuesday 15 July 2014 16:45:51 Ley Foon Tan wrote:
> > > +void *module_alloc(unsigned long size)
> > > +{
> > > +   if (size == 0)
> > > +           return NULL;
> > > +   return kmalloc(size, GFP_KERNEL);
> > > +}
> > > +
> > > +/* Free memory returned from module_alloc */
> > > +void module_free(struct module *mod, void *module_region)
> > > +{
> > > +   kfree(module_region);
> > > +}
> > 
> > Any particular reason for defining these to use kmalloc rather than
> > the default vmalloc based functions?
> 
> AFAIR this is due to relocation issues, as the FIXME comment in the
> "original" code [1] states. I don't know whether this still applies,
> though (or what would be the proper fix for this to overcome the
> relocation issues).
> 
> [1] https://github.com/tklauser/linux-nios2/blob/nios2/arch/nios2/kernel/module.c#L45
> 

Ah, I see. Please at least copy over that comment then.

A long-term solution would be to copy what ARM does, since that
has the same problem. The modules are allocated with
__vmalloc_node_range, passing a virtual address range that
is just before the kernel virtual address, taken out of the
top 16MB of the user space addresses.

	Arnd