Re: [PATCH v7 6/6] memory: Introduce address_space_to_flatview_rcu()

[Chuang Xu <xuchuangxclwt@bytedance.com>]

Hi, Peter,

On 2023/3/10 下午11:08, Peter Xu wrote:
> On Fri, Mar 10, 2023 at 10:24:25AM +0800, Chuang Xu wrote:
>> In last patch, we wrap vm_load with begin/commit, here we introduce
>> address_space_to_flatview_rcu() to avoid unnecessary enforce commit
>> during vm_load.
>>
>> Signed-off-by: Chuang Xu <xuchuangxclwt@bytedance.com>
>> ---
>>   include/exec/memory-internal.h |  2 +-
>>   include/exec/memory.h          | 20 ++++++++++++++++++++
>>   softmmu/memory.c               |  2 +-
>>   3 files changed, 22 insertions(+), 2 deletions(-)
>>
>> diff --git a/include/exec/memory-internal.h b/include/exec/memory-internal.h
>> index 100c1237ac..1432240449 100644
>> --- a/include/exec/memory-internal.h
>> +++ b/include/exec/memory-internal.h
>> @@ -30,7 +30,7 @@ static inline AddressSpaceDispatch *flatview_to_dispatch(FlatView *fv)
>>   
>>   static inline AddressSpaceDispatch *address_space_to_dispatch(AddressSpace *as)
>>   {
>> -    return flatview_to_dispatch(address_space_to_flatview(as));
>> +    return flatview_to_dispatch(address_space_to_flatview_rcu(as));
>>   }
> I'm not sure whether this one is always safe.

Previously I considered that there was no address_space_translate_iommu()
traced during vm_load, so I took it as safe. But my trace may not be
able to obtain all cases of triggering do_commit() during vm_load..

>
> tcg_commit() seems to be safe, but maybe address_space_translate_iommu() is
> not?  Maybe easier to leave this untouched?

Yes, I'll fix it in v8 later.

>>   
>>   FlatView *address_space_get_flatview(AddressSpace *as);
>> diff --git a/include/exec/memory.h b/include/exec/memory.h
>> index d6fd89db64..235e3017bc 100644
>> --- a/include/exec/memory.h
>> +++ b/include/exec/memory.h
>> @@ -1100,6 +1100,9 @@ bool memory_region_transaction_in_progress(void);
>>   
>>   void memory_region_transaction_do_commit(void);
>>   
>> +/*
>> + * We recommend using this by default.
>> + */
> I think this comment doesn't really help.. drop it?
>
>>   static inline FlatView *address_space_to_flatview(AddressSpace *as)

Thanks!