Re: [PATCH v4 2/2] arm64: kexec_file: add crash dump support

[AKASHI Takahiro]

Will, Pavel,

On Mon, Jan 13, 2020 at 11:21:06AM +0000, Will Deacon wrote:
> On Fri, Jan 10, 2020 at 11:19:16AM -0500, Pavel Tatashin wrote:
> > On Fri, Jan 10, 2020 at 11:05 AM Will Deacon <will@xxxxxxxxxx> wrote:
> > >
> > > On Thu, Jan 09, 2020 at 08:32:54AM +0000, Will Deacon wrote:
> > > > On Thu, Jan 09, 2020 at 09:46:55AM +0900, AKASHI Takahiro wrote:
> > > > > On Wed, Jan 08, 2020 at 05:48:39PM +0000, Will Deacon wrote:
> > > > > > On Mon, Dec 16, 2019 at 11:12:47AM +0900, AKASHI Takahiro wrote:
> > > > > > > diff --git a/arch/arm64/include/asm/kexec.h b/arch/arm64/include/asm/kexec.h
> > > > > > > index 12a561a54128..d24b527e8c00 100644
> > > > > > > --- a/arch/arm64/include/asm/kexec.h
> > > > > > > +++ b/arch/arm64/include/asm/kexec.h
> > > > > > > @@ -96,6 +96,10 @@ static inline void crash_post_resume(void) {}
> > > > > > >  struct kimage_arch {
> > > > > > >         void *dtb;
> > > > > > >         unsigned long dtb_mem;
> > > > > > > +       /* Core ELF header buffer */
> > > > > > > +       void *elf_headers;
> > > > > > > +       unsigned long elf_headers_mem;
> > > > > > > +       unsigned long elf_headers_sz;
> > > > > > >  };
> > > > > >
> > > > > > This conflicts with the cleanup work from Pavel. Please can you check my
> > > > > > resolution? [1]
> > > > >
> > > > > I don't know why we need to change a type of dtb_mem,
> > > > > otherwise it looks good.
> > > > >
> > > > > (I also assume that you notice that kimage_arch is of no use for kexec.)
> > > >
> > > > Yes, that's why I'd like the resolution checked. If you reckon it's cleaner
> > > > to drop Pavel's patch altogether in light of your changes, we can do that
> > > > instead.
> > > >
> > > > Thoughts?
> > >
> > > Well, I've reverted the cleanup patch so please shout if you'd prefer
> > > something else.
> > 
> > As I understand, the only concern was the type change for dtb_mem.
> > This was one of the review comments for my patch
> > https://lore.kernel.org/lkml/20191204155938.2279686-21-pasha.tatashin@xxxxxxxxxx/
> > 
> > (I believe it was from Marc Zyngier), I add a number of new fields,
> > and they all should be phys_addr_t, this is why I change dtb_mem to
> > phys_addr_t to be consistent.
> 
> Sure, but I've only queued the first part of your series and that cleanup
> patch doesn't make a lot of sense when applied against Akashi's work. I'm
> happy to take stuff on top if you both agree to it, but having half of the
> struct use unsigned long and the other half use phys_addr_t is messy.

Logically, whether dtb_mem is a "unsigned long" or phys_addr_t doesn't
matter unless the kernel is compiled under LLP64.
As far as the existing kexec code, either generic or arm64-specific,
is concerned, however, "unsigned long is widely used as a physical address
(For example, see kexec_buf definition) over the code.

(Oops, reboot_code_buffer_phys is a phys_addr_t :)

So as long as my kexec_file (and associated kdump) patch comes first
before Pavel's, I'd like to keep using "unsigned long".
Then, you can change "unsigned long" to phys_addr_t in your patch
for whatever reason it is.

Please note that, if you want to do that, it would be better to modify
not only kimage_arch but also all the occurrences of "unsigned long"
to phys_addr_t for maintaining the integrity.

In addition, in my kexec_file kdump code, I still believe that
"#ifdef CONFIG_KEXEC_FILE" should stay before the definition of
kimage_arch as kimage_arch is of no use for normal kexec code.

Again,
"#ifdef" statement may be moved forward once additional fields be
added later by Pavel's patch, say, "[PATCH v8 15/25] arm64: kexec:
move relocation function setup" for any reason.

I believe that this way gives us a logical and consistent view of
history of changes.
Make sense?

Thanks,
-Takahiro Akashi


> Will