Re: [PATCH v3 4/9] arm/vm_event: get/set registers

[Julien Grall <julien.grall@arm.com>]

On 16/05/16 17:26, Tamas K Lengyel wrote:
>
> On May 16, 2016 09:59, "Julien Grall" <julien.grall@arm.com
> <mailto:julien.grall@arm.com>> wrote:
>  >>  > By defining 2 distinct structures, it will be more difficult for the
>  >> introspection tools to inspect registers of an Aarch64 domain running in
>  >> AArch32 mode. They wouldn't be able to re-use code for AArch32 domain
>  >> because the structure fields are different.
>  >>  >
>  >>  > The ARM ARM (see D1.20.1 in ARM DDI 0487A.i) provides a mapping
>  >> between AArch32 state and AArch64 state. If you use it to define the
>  >> layout of a common structure, the support of AArch32 state for AArch64
>  >> domain will come free.
>  >>  >
>  >>
>  >> If the guest is running in 32-bit mode Xen will fill the 32-bit struct,
>  >> so doing a common struct is not strictly necessary. It also requires a
>  >> bunch if union declarations to match the names between that I would
>  >> prefer to avoid. IMHO it's cleaner to do the struct definitions
>  >> separately and then do a union on top.
>  >
>  >
>  > That is not true. is_domain_32bit will check if the domain is
> configured to run 32-bit or 64-bit in EL1 (i.e the kernel level).
>  >
>  > So if you have a guest with 64-bit kernel and 32-bit userspace, Xen
> will always fill the 64-bit structure, even when the userspace is running.
>  >
>
> Hm fair point. It complicates things a bit though either way as the
> event subscriber wouldn't know which mode it received the event from
> without doing some extra checks. So I guess Xen should transmit that
> information too, at which point it could also just pick the right struct
> to fill with the current setup.

Not really, CPSR.M[4] tells you if the execution state were AArch32 or 
AArch64. vm_event_regs_* contains the cpsr fields, so the event
subscriber can determine the mode.

Actually, not all the registers in vm_event_regs_arm32 makes sense when 
the guest is a 64-bit domain. For instance, spsr_svc is for the EL1 (i.e 
kernel).

Hence the suggestion to use a structure very similar to 
vcpu_guest_core_regs.

Regards,

-- 
Julien Grall

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel