Re: [PATCH v2 4/4] efi: Fix handling of multiple efi_fake_mem= entries

[Dave Young <dyoung@redhat.com>]

On 01/01/20 at 10:36am, Dan Williams wrote:
> On Tue, Dec 31, 2019 at 10:21 PM Dave Young <dyoung@redhat.com> wrote:
> >
> > > > Does kexec preserve iomem? I.e. as long as the initial translation of
> > > > efi entries to e820, and resulting resource tree, is preserved by
> > > > successive kexec cycles then I think we're ok.
> > >
> > > It will not preserve them automatically, but that can be fixed if
> > > needed.
> > >
> > > There are two places:
> > > 1. the in kernel loader, we can do similar with below commit (for Soft
> > > Reseved instead):
> > > commit 980621daf368f2b9aa69c7ea01baa654edb7577b
> > > Author: Lianbo Jiang <lijiang@redhat.com>
> > > Date:   Tue Apr 23 09:30:07 2019 +0800
> > >
> > >     x86/crash: Add e820 reserved ranges to kdump kernel's e820 table
> >
> > Oops, that is for kdump only, for kexec, should update the kexec e820
> > table.  But before doing that we need first to see if this is necessary.
> 
> We can cross that bridge later, but I expect it will eventually be
> necessary. The soft-reservation facility will become more prevalent as
> more platforms ship with DRAM differentiated memory ranges, like
> high-bandwidth-memory, and the system needs to reserve it from general
> kernel allocations. See commit 262b45ae3ab4 "x86/efi: EFI soft
> reservation to E820 enumeration" and commit fe3e5e65c06e "efi:
> Enumerate EFI_MEMORY_SP" for more details.

Ok, agreed the EFI_MEMORY_SP should be preserved across kexec reboot,
I think those firmware provided EFI_MEMORY_SP should be persistent
because the e820 table is just a copy.   But I have no such hardware
to test,  could you do a test to confirm if possible?

The test steps should be:
# -s means to use kexec_file_load syscall
kexec -s -l bzImage --initrd initramfs-file --reuse-cmdline 
reboot

Maybe this should be fine for the time being.  And
leave the faked mem only works once during the physical boot?

Thanks
Dave