Re: [PATCH 1/4] mm: ioremap: Setup phys_addr of struct vm_struct

[Kefeng Wang]

On 2022/4/28 1:10, Andrew Morton wrote:
> On Wed, 27 Apr 2022 20:14:10 +0800 Kefeng Wang <wangkefeng.wang@xxxxxxxxxx> wrote:
>
> > Show physical address in /proc/vmallocinfo.
> >
> > Signed-off-by: Kefeng Wang <wangkefeng.wang@xxxxxxxxxx>
> > ---
> >   mm/ioremap.c | 1 +
> >   1 file changed, 1 insertion(+)
> >
> > diff --git a/mm/ioremap.c b/mm/ioremap.c
> > index 5fe598ecd9b7..522ef899c35f 100644
> > --- a/mm/ioremap.c
> > +++ b/mm/ioremap.c
> > @@ -32,6 +32,7 @@ void __iomem *ioremap_prot(phys_addr_t addr, size_t size, unsigned long prot)
> >   	if (!area)
> >   		return NULL;
> >   	vaddr = (unsigned long)area->addr;
> > +	area->phys_addr = addr;
> >   
> >   	if (ioremap_page_range(vaddr, vaddr + size, addr, __pgprot(prot))) {
> >   		free_vm_area(area);
> Acked-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>
> I checked a bunch of arch-specific implementations of ioremap_prot()
> and they're already doing this.  As far as I can tell, only csky and
> riscv actually use this file (CONFIG_GENERIC_IOREMAP=y).  But you're
> ARM(?) so I'm wondering how come you're patching it?

Hi Andrew,

I found this via reading code when debug some other issue, meanwhile,  there

are some code duplication of ioremap between arm64 and generic ioremap, so

1) bugfix: fix the above issue and test on riscv

2) cleanup: convert arm64 to use GENERIC_IOREMAP,

3) feature: after that, enable HAVE_IOREMAP_PROT on arm64

> Someone should do s/addr/phys_addr/ in this function, like the rest of
> the world (sensibly) does.
I will make it in next version.
> .