Question to hv_vtl_real_mode_header in hv_vtl.c

Question to hv_vtl_real_mode_header in hv_vtl.c

[Yunhong Jiang]

Hi, Srinivasan and Dexuan,
	I have a question to the hv_vtl_real_mode_header in
arch/x86//hyperv/hv_vtl.c when addressing one patch review comment.
	In hv_vtl_early_init(), the real_mode_header is set to
hv_vtl_real_mode_header, but there is no setup to the real_mode_header, since
the realmode_init() is marked x86_init_noop in hv_vtl_init_platform.
	How is the real_mode_header(in another word, hv_vtl_real_mode_header)
used? Is it to meet the access requirement from do_boot_cpu(), so that
real_mode_header->trampoline_start64 will work, although the start_ip is not
used?
	If it's really to support the do_boot_cpu() requirement, how does the
non-VTL guest meet the access requirement? The hv_vtl_init_platform() is
unconditionally called from ms_hyperv_init_platform(), so I assume all hyperv
guest will have the realmode_init() set as x86_init_noop.

Thank you
--jyh

Re: Question to hv_vtl_real_mode_header in hv_vtl.c

[Yunhong Jiang]

On Tue, Aug 13, 2024 at 10:28:19AM -0700, Yunhong Jiang wrote:
> Hi, Srinivasan and Dexuan,
> 	I have a question to the hv_vtl_real_mode_header in
> arch/x86//hyperv/hv_vtl.c when addressing one patch review comment.
> 	In hv_vtl_early_init(), the real_mode_header is set to
> hv_vtl_real_mode_header, but there is no setup to the real_mode_header, since
> the realmode_init() is marked x86_init_noop in hv_vtl_init_platform.
> 	How is the real_mode_header(in another word, hv_vtl_real_mode_header)
> used? Is it to meet the access requirement from do_boot_cpu(), so that
> real_mode_header->trampoline_start64 will work, although the start_ip is not
> used?
> 	If it's really to support the do_boot_cpu() requirement, how does the
> non-VTL guest meet the access requirement? The hv_vtl_init_platform() is
> unconditionally called from ms_hyperv_init_platform(), so I assume all hyperv
> guest will have the realmode_init() set as x86_init_noop.

The patch review mentioned above is
https://lore.kernel.org/lkml/87a5ho2q6x.ffs@tglx/ . Can we set real_mode_header
to hv_vtl_real_mode_header in hv_vtl_init_platform(), instead of
hv_vtl_early_init()? I'm not sure if such move is safe, because
hv_vtl_early_init() is invoked only for vtl!=0 guest while
hv_vtl_init_platform() is invoked for all the hyperv guest.

> 
> Thank you
> --jyh
> 

Re: Question to hv_vtl_real_mode_header in hv_vtl.c

[Saurabh Singh Sengar]

On Fri, Aug 23, 2024 at 10:41:22PM +0000, Wei Liu wrote:
> On Tue, Aug 13, 2024 at 02:36:28PM -0700, Yunhong Jiang wrote:
> > On Tue, Aug 13, 2024 at 10:28:19AM -0700, Yunhong Jiang wrote:
> > > Hi, Srinivasan and Dexuan,
> > > 	I have a question to the hv_vtl_real_mode_header in
> > > arch/x86//hyperv/hv_vtl.c when addressing one patch review comment.
> > > 	In hv_vtl_early_init(), the real_mode_header is set to
> > > hv_vtl_real_mode_header, but there is no setup to the real_mode_header, since
> > > the realmode_init() is marked x86_init_noop in hv_vtl_init_platform.
> > > 	How is the real_mode_header(in another word, hv_vtl_real_mode_header)
> > > used? Is it to meet the access requirement from do_boot_cpu(), so that
> > > real_mode_header->trampoline_start64 will work, although the start_ip is not
> > > used?
> > > 	If it's really to support the do_boot_cpu() requirement, how does the
> > > non-VTL guest meet the access requirement? The hv_vtl_init_platform() is
> > > unconditionally called from ms_hyperv_init_platform(), so I assume all hyperv
> > > guest will have the realmode_init() set as x86_init_noop.
> > 
> > The patch review mentioned above is
> > https://lore.kernel.org/lkml/87a5ho2q6x.ffs@tglx/ . Can we set real_mode_header
> > to hv_vtl_real_mode_header in hv_vtl_init_platform(), instead of
> > hv_vtl_early_init()?

Yes, fine to move.

- Saurabh