[RFC PATCH] Arm64: introduce __hyp_func_call

[Marc Zyngier]

On 30/08/14 11:32, Arun Chandran wrote:
> Hi Marc,
> 
> On Wed, Aug 27, 2014 at 4:22 PM, Marc Zyngier <marc.zyngier at arm.com> wrote:
>> On 27/08/14 11:28, Arun Chandran wrote:
>>> This adds a mechanism to __hyp_stub_vectors to allow a hypercall to
>>> call a function at EL2. It is needed for users who want to
>>> run a part of code with EL2 permissions. The current usecase is for
>>> KVM and kexec.
>>>
>>> For kexec we need to move the final CPU up to the mode it started
>>> in before we branch to the new kernel. If we don't do that
>>>
>>> * We loose EL2 in the next boot
>>> * Arm64 bootwrapper may not be able to put CPUs at the spin-table
>>>   code. It expects the final jump from kernel to cpu-return-addr to be
>>>   done in EL2.
>>>
>>> KVM can use this to set/get VBAR_EL2
>>
>> Ah, looking at this a bit more, I see what you've done (missed the
>> #define trickery below).
>>
>>> Signed-off-by: Arun Chandran <achandran at mvista.com>
>>> ---
>>> Idea is from "Mark Rutland <mark.rutland at arm.com>"
>>> http://lists.infradead.org/pipermail/linux-arm-kernel/2014-August/280026.html
>>> ---
>>>  arch/arm64/include/asm/virt.h |   15 +++++++++++++++
>>>  arch/arm64/kernel/hyp-stub.S  |   33 +++++++++++++++++++++------------
>>>  2 files changed, 36 insertions(+), 12 deletions(-)
>>>
>>> diff --git a/arch/arm64/include/asm/virt.h b/arch/arm64/include/asm/virt.h
>>> index 7a5df52..910a163 100644
>>> --- a/arch/arm64/include/asm/virt.h
>>> +++ b/arch/arm64/include/asm/virt.h
>>> @@ -34,9 +34,24 @@
>>>   */
>>>  extern u32 __boot_cpu_mode[2];
>>>
>>> +void *__hyp_func_call(u64 __tmp, phys_addr_t func, ...);
>>>  void __hyp_set_vectors(phys_addr_t phys_vector_base);
>>>  phys_addr_t __hyp_get_vectors(void);
>>>
>>> +#define __hyp_set_vectors(__vbase)                                       \
>>> +({                                                                       \
>>> +     u64 __tmp = 0;                                                      \
>>> +     __hyp_func_call(__tmp, virt_to_phys(__hyp_set_vectors), __vbase);   \
>>> +})
>>> +
>>> +#define __hyp_get_vectors()                                              \
>>> +({                                                                       \
>>> +     u64 __tmp = 0;                                                      \
>>> +     phys_addr_t ret = (phys_addr_t) __hyp_func_call(__tmp,              \
>>> +                                     virt_to_phys(__hyp_get_vectors));   \
>>> +     ret;                                                                \
>>> +})
>>> +
>>
>> This is what has thrown me off the wrong path. Don't do that, this is
>> horrid. Just rename the assembly entry points, it will make things a lot
>> clearer.
>>
> 
> Ok. I will post another one.

Hold on, there are more problems.

You've changed the case that works when KVM is *not* running. But if you
look at the KVM code, you'll see that it has its own implementation of
__hyp_get_vectors. What will happen when KVM calls this version of
__hyp_get_vectors (on a CPU hotplug event, for example)? The code isn't
mapped in KVM's HYP code, so it will simply burst into flames.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...