Re: [PATCH] arm/xen: Enable user access to the kernel before issuing a privcmd call

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

On 11/09/15 15:29, Ian Campbell wrote:
> On Fri, 2015-09-11 at 15:16 +0100, Julien Grall wrote:
>> When Xen is copyin data to/from the guest it will check if the kernel
> 
> "copying"
> 
>> has the right to do the access. If not, the hypercall will return an
>> error.
>>
>> After the commit a5e090acbf545c0a3b04080f8a488b17ec41fe02 "ARM:
>> software-based priviledged-no-access support", the kernel can't access
> 
> "privileged"
> 
>> anymore the user space by default. This will result to fail on every
> 
> "any more" (or "any longer")
> 
>> hypercall made by the userspace (i.e via privcmd).
>>
>> We have to enable the userspace access and then restore the correct
>> permission everytime the privmcd is used to made an hypercall.
> 
> "every time" and "privcmd"
> 
>>  HYPERCALL1(tmem_op);
>>  HYPERCALL2(multicall);
>>  
>> -ENTRY(privcmd_call)
>> +ENTRY(__privcmd_call)
> 
> arch/arm/include/asm/assembler.h seems to contain uaccess_* macros which
> could be used right here directly I think? That would be preferable to
> wrapping I think.

Looking to the uaccess_save macro:

        .macro  uaccess_save, tmp
#ifdef CONFIG_CPU_SW_DOMAIN_PAN
        mrc     p15, 0, \tmp, c3, c0, 0
        str     \tmp, [sp, #S_FRAME_SIZE]
#endif
        .endm


It's saving the register on the Stack with an offset S_FRAME_SIZE.
AFAICT, S_FRAME_SIZE is the size of the pt_regs structure.

So it looks like to me that they are unusable for any assembly functions
but entry point.

I though about using a static inline for privcmd_call but it was
introducing changes on the arm64 in order to decouple hypercall.h (it's
common right now).

Regards,

-- 
Julien Grall