Re: [PATCH] target/arm: do not use cc->do_interrupt for KVM directly

[Claudio Fontana]

On 12/7/20 7:14 PM, Peter Maydell wrote:
> On Mon, 7 Dec 2020 at 18:08, Claudio Fontana <cfontana@suse.de> wrote:
>> what about also the existing code with qemu-arm (user mode)?
>>
>> In that case do_interrupt is not set at all in target/arm/cpu.c, since it's 
>> protected by #ifndef CONFIG_USER_ONLY
>>
>> Did we have a potential NULL pointer trying to be dereferenced there?
> 
> No, because in user-mode there are never any interrupts or
> exceptions invoked this way. The code in these methods is
> strictly system-emulation only.
> 
>> Commit 0adf7d3cc3f724e1e9ce5aaa008bd9daeb90f19 says:
>>
>>  target-arm: do not set do_interrupt handlers for ARM and AArch64 user modes
>>
>>  User mode emulation should never get interrupts and thus should not
>>  use the system emulation exception handler function.
>>
>> --
>>
>> But this was 2014. Is the comment above true today?
> 
> Yes.
> 
>> Looking at this commit in 2017, it does not seem to me to be the case:
>>
>> commit 17b50b0c299f1266578b01f7134810362418ac2e
>> Author: Pavel Dovgaluk <Pavel.Dovgaluk@ispras.ru>
>> Date:   Tue Nov 14 11:18:18 2017 +0300
>>
>>     cpu-exec: avoid cpu_exec_nocache infinite loop with record/replay
>>
>>     This patch
>>     [...]
>>     Second, try to cause the exception at the beginning of
>>     cpu_handle_exception, and exit immediately if the TB cannot
>>     execute.  With this change, interrupts are processed and
>>     cpu_exec_nocache can make process.
> 
> This code only invokes cc->do_interrupt() in CONFIG_USER_ONLY
> if TARGET_I386 is true. i386 does this stuff in a weird way
> that's different to all the other target architectures.
> (One day we should fix this inconsistency I suppose.)
> 
>> Is cc->do_interrupt supposed to be !CONFIG_USER_ONLY or not?
> 
> It's !CONFIG_USER_ONLY.
> 
> thanks
> -- PMM
> 

Ah right, I somehow missed the #if defined(TARGET_I386),

thanks!