Re: [PATCH v4 16/17] maple_tree: remove unneeded mas_wr_walk() in mas_store_prealloc()

[Sid Kumar <sidhartha.kumar@oracle.com>]

On 10/25/24 7:58 PM, Wei Yang wrote:
> On Fri, Oct 25, 2024 at 03:54:04PM -0400, Sid Kumar wrote:
>> On 10/23/24 9:20 PM, Wei Yang wrote:
>>> On Wed, Aug 14, 2024 at 12:19:43PM -0400, Sidhartha Kumar wrote:
>>>> Users of mas_store_prealloc() enter this function with nodes already
>>>> preallocated. This means the store type must be already set. We can then
>>>> remove the call to mas_wr_store_type() and initialize the write state to
>>>> continue the partial walk that was done when determining the store type.
>>>>
>>> May I ask what is the partial walk here means?
>>>
>>> It is the mas_wr_walk() in mas_wr_store_type() which is stopped because of it
>>> is spanning write?
>> Yes, this is what I meant by the partial walk that's already been started.
>> It's the walk done by mas_wr_store_type().
>>
>>> I may lost some background, so the assumption here is mas_wr_store_type() has
>>> already been invoked and the store type has been decided, right?
>> Ya users of mas_store_prealloc() should have already called mas_preallocate()
>> which does:
>>
>>      mas->store_type = mas_wr_store_type(&wr_mas);
>>      request = mas_prealloc_calc(&wr_mas, entry);
>>
>> to set the store type and allocate the nodes.
>>
>>
>>>> Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
>>>> Signed-off-by: Sidhartha Kumar <sidhartha.kumar@oracle.com>
>>>> ---
>>>> lib/maple_tree.c | 18 +++++++++++++-----
>>>> 1 file changed, 13 insertions(+), 5 deletions(-)
>>>>
>>>> diff --git a/lib/maple_tree.c b/lib/maple_tree.c
>>>> index 8c1a1a483395..73ce63d9c3a0 100644
>>>> --- a/lib/maple_tree.c
>>>> +++ b/lib/maple_tree.c
>>>> @@ -3979,9 +3979,6 @@ static inline void mas_wr_end_piv(struct ma_wr_state *wr_mas)
>>>> 		wr_mas->end_piv = wr_mas->pivots[wr_mas->offset_end];
>>>> 	else
>>>> 		wr_mas->end_piv = wr_mas->mas->max;
>>>> -
>>>> -	if (!wr_mas->entry)
>>>> -		mas_wr_extend_null(wr_mas);
>>>> }
>>>>
>>>> static inline unsigned char mas_wr_new_end(struct ma_wr_state *wr_mas)
>>>> @@ -5532,8 +5529,19 @@ void mas_store_prealloc(struct ma_state *mas, void *entry)
>>>> {
>>>> 	MA_WR_STATE(wr_mas, mas, entry);
>>>>
>>>> -	mas_wr_prealloc_setup(&wr_mas);
>>>> -	mas_wr_store_type(&wr_mas);
>>>> +	if (mas->store_type == wr_store_root) {
>>>> +		mas_wr_prealloc_setup(&wr_mas);
>>>> +		goto store;
>>>> +	}
>>>> +
>>>> +	mas_wr_walk_descend(&wr_mas);
>>> This one does not descend the tree, just locate the offset in a node and
>>> adjust min/max. So not look like to continue the partial walk to me.
>>>
>>>> +	if (mas->store_type != wr_spanning_store) {
>>>> +		/* set wr_mas->content to current slot */
>>>> +		wr_mas.content = mas_slot_locked(mas, wr_mas.slots, mas->offset);
>>>> +		mas_wr_end_piv(&wr_mas);
>>> If not a spanning write, the previous walk should reach a leaf node, right?
>> Ya that's true.
>>
>>> I am not sure why we don't need to check extend null here. Because we have
>>> already done it?
>>
>> Ya we extend null in mas_wr_store_type() which has already been called at
>> this point.
>>
>>
>>      /* At this point, we are at the leaf node that needs to be altered. */
>>      mas_wr_end_piv(wr_mas);
>>      if (!wr_mas->entry)
>>          mas_wr_extend_null(wr_mas);
>>
>> Thanks,
> Hmm... if we have already done this, why we need to do mas_wr_end_piv() again?

The maple write state here is local to this function:

void mas_store_prealloc(struct ma_state *mas, void *entry)
{
     MA_WR_STATE(wr_mas, mas, entry);

so we don't retain the wr_end information from the previous call to 
mas_preallocate() and have to repeat it here. The write state is not 
currently exposed so have to call mas_wr_end_piv() again.

Thanks,

Sid


>
>> Sid
>>
>>>> +	}
>>>> +
>>>> +store:
>>>> 	trace_ma_write(__func__, mas, 0, entry);
>>>> 	mas_wr_store_entry(&wr_mas);
>>>> 	MAS_WR_BUG_ON(&wr_mas, mas_is_err(mas));
>>>> -- 
>>>> 2.46.0
>>>>