Re: [PATCH 01/11] kexec_file: allow archs to handle special regions while locating memory hole

[Dave Young]

On 07/02/20 at 12:01am, Hari Bathini wrote:
> 
> 
> On 01/07/20 1:16 pm, Dave Young wrote:
> > On 06/29/20 at 05:26pm, Hari Bathini wrote:
> >> Hi Petr,
> >>
> >> On 29/06/20 5:09 pm, Petr Tesarik wrote:
> >>> Hi Hari,
> >>>
> >>> is there any good reason to add two more functions with a very similar
> >>> name to an existing function? AFAICS all you need is a way to call a
> >>> PPC64-specific function from within kexec_add_buffer (PATCH 4/11), so
> >>> you could add something like this:
> >>>
> >>> int __weak arch_kexec_locate_mem_hole(struct kexec_buf *kbuf)
> >>> {
> >>> 	return 0;
> >>> }
> >>>
> >>> Call this function from kexec_add_buffer where appropriate and then
> >>> override it for PPC64 (it roughly corresponds to your
> >>> kexec_locate_mem_hole_ppc64() from PATCH 4/11).
> >>>
> >>> FWIW it would make it easier for me to follow the resulting code.
> >>
> >> Right, Petr.
> >>
> >> I was trying out a few things before I ended up with what I sent here.
> >> Bu yeah.. I did realize arch_kexec_locate_mem_hole() would have been better
> >> after sending out v1. Will take care of that in v2.
> > 
> > Another way is use arch private function to locate mem hole, then set
> > kbuf->mem, and then call kexec_add_buf, it will skip the common locate
> > hole function.
> 
> Dave, I did think about it. But there are a couple of places this can get
> tricky. One is ima_add_kexec_buffer() and the other is kexec_elf_load().
> These call sites could be updated to set kbuf->mem before kexec_add_buffer().
> But the current approach seemed like the better option for it creates a
> single point of control in setting up segment buffers and also, makes adding
> any new segments simpler, arch-specific segments or otherwise.
> 

Ok, thanks for the explanation.