Re: [Qemu-devel] [PATCH] cpu_physical_memory_write_rom() needs to do TB invalidates

[Alexander Graf]

On 22.08.2012, at 08:10, David Gibson wrote:

> On Wed, Aug 22, 2012 at 08:02:11AM +0200, Alexander Graf wrote:
>> 
>> On 22.08.2012, at 07:57, David Gibson wrote:
>> 
>>> On Wed, Aug 22, 2012 at 07:55:31AM +0200, Alexander Graf wrote:
>>>> 
>>>> On 22.08.2012, at 06:59, David Gibson wrote:
>>>> 
>>>>> cpu_physical_memory_write_rom(), despite the name, can also be used to
>>>>> write images into RAM - and will often be used that way if the machine
>>>>> uses load_image_targphys() into RAM addresses.
>>>>> 
>>>>> However, cpu_physical_memory_write_rom(), unlike cpu_physical_memory_rw()
>>>>> does invalidate any cached TBs which might be affected by the region
>>>>> written.
>>>>> 
>>>>> This was breaking reset (under full emu) on the pseries machine - we 
>>>>> loaded
>>>>> our firmware image into RAM, and while executing it rewrite the code at
>>>>> the entry point (correctly causing a TB invalidate/refresh).  When we
>>>>> reset the firmware image was reloaded, but the TB from the rewrite was
>>>>> still active and caused us to get an illegal instruction trap.
>>>>> 
>>>>> This patch fixes the bug by duplicating the tb invalidate code from
>>>>> cpu_physical_memory_rw() in cpu_physical_memory_write_rom().
>>>>> 
>>>>> Signed-off-by: David Gibson <address@hidden>
>>>>> ---
>>>>> exec.c |    7 +++++++
>>>>> 1 file changed, 7 insertions(+)
>>>>> 
>>>>> diff --git a/exec.c b/exec.c
>>>>> index 5834766..eff40d7 100644
>>>>> --- a/exec.c
>>>>> +++ b/exec.c
>>>>> @@ -3523,6 +3523,13 @@ void 
>>>>> cpu_physical_memory_write_rom(target_phys_addr_t addr,
>>>>>           /* ROM/RAM case */
>>>>>           ptr = qemu_get_ram_ptr(addr1);
>>>>>           memcpy(ptr, buf, l);
>>>>> +            if (!cpu_physical_memory_is_dirty(addr1)) {
>>>>> +                /* invalidate code */
>>>>> +                tb_invalidate_phys_page_range(addr1, addr1 + l, 0);
>>>>> +                /* set dirty bit */
>>>>> +                cpu_physical_memory_set_dirty_flags(
>>>>> +                    addr1, (0xff & ~CODE_DIRTY_FLAG));
>>>>> +            }
>>>> 
>>>> Can't we just call cpu_physical_memory_rw in the RAM case? The
>>>> function only tries to not do MMIO accesses on ROM pages, right?
>>> 
>>> Maybe.  It's not clear at all to me what cases
>>> cpu_physical_memory_write_rom() is supposed to be for, as opposed to
>>> just using cpu_physical_memory_rw().
>> 
>> I can only guess, but the code looks to me as if it wants to be a
>> nop on ROM pages, while basically doing cpu_physical_memory_rw for
>> RAM pages. Usually in QEMU, every non-RAM page gets treated as MMIO
>> which might eventually lead to machine checks.
> 
> Maybe.  Anthony, can you make a ruling on this, or tell me who can.  I
> don't really care how I fix it, but it's definitely broken right now.

Also, does tb_invalidate_phys_page_range() do an icache flush on KVM?


Alex