[PATCH] mm/zsmalloc: don't need to save tag bit in handle

[PATCH] mm/zsmalloc: don't need to save tag bit in handle

[chengming.zhou]

From: Chengming Zhou <zhouchengming@bytedance.com>

We only need to save the position (pfn + obj_idx) in the handle, don't
need to save tag bit in handle. So one more bit can be used as obj_idx.

Actually, the tag bit is only useful in zspage's memory space, to tell
if an object is allocated or not.

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
---
 mm/zsmalloc.c | 5 +----
 1 file changed, 1 insertion(+), 4 deletions(-)

diff --git a/mm/zsmalloc.c b/mm/zsmalloc.c
index 63ec385cd670..7d7cb3eaabe0 100644
--- a/mm/zsmalloc.c
+++ b/mm/zsmalloc.c
@@ -110,7 +110,7 @@
 #define OBJ_TAG_BITS	1
 #define OBJ_TAG_MASK	OBJ_ALLOCATED_TAG
 
-#define OBJ_INDEX_BITS	(BITS_PER_LONG - _PFN_BITS - OBJ_TAG_BITS)
+#define OBJ_INDEX_BITS	(BITS_PER_LONG - _PFN_BITS)
 #define OBJ_INDEX_MASK	((_AC(1, UL) << OBJ_INDEX_BITS) - 1)
 
 #define HUGE_BITS	1
@@ -737,14 +737,12 @@ static struct page *get_next_page(struct page *page)
 static void obj_to_location(unsigned long obj, struct page **page,
 				unsigned int *obj_idx)
 {
-	obj >>= OBJ_TAG_BITS;
 	*page = pfn_to_page(obj >> OBJ_INDEX_BITS);
 	*obj_idx = (obj & OBJ_INDEX_MASK);
 }
 
 static void obj_to_page(unsigned long obj, struct page **page)
 {
-	obj >>= OBJ_TAG_BITS;
 	*page = pfn_to_page(obj >> OBJ_INDEX_BITS);
 }
 
@@ -759,7 +757,6 @@ static unsigned long location_to_obj(struct page *page, unsigned int obj_idx)
 
 	obj = page_to_pfn(page) << OBJ_INDEX_BITS;
 	obj |= obj_idx & OBJ_INDEX_MASK;
-	obj <<= OBJ_TAG_BITS;
 
 	return obj;
 }
-- 
2.40.1

Re: [PATCH] mm/zsmalloc: don't need to save tag bit in handle

[Sergey Senozhatsky]

On (24/02/27 03:00), chengming.zhou@linux.dev wrote:
> 
> We only need to save the position (pfn + obj_idx) in the handle, don't
> need to save tag bit in handle. So one more bit can be used as obj_idx.

[..]

> mm/zsmalloc: don't need to save tag bit in handle

Does this mean "don't need to reserve LSB for tag"?
We still save allocated tag in the handle, that's what

	handle |= OBJ_ALLOCATED_TAG;

does.

> Actually, the tag bit is only useful in zspage's memory space, to tell
> if an object is allocated or not.

I'm not completely sure if I follow this sentence.

Re: [PATCH] mm/zsmalloc: don't need to save tag bit in handle

[Chengming Zhou]

On 2024/2/27 15:52, Sergey Senozhatsky wrote:
> On (24/02/27 03:00), chengming.zhou@linux.dev wrote:
>>
>> We only need to save the position (pfn + obj_idx) in the handle, don't
>> need to save tag bit in handle. So one more bit can be used as obj_idx.
> 
> [..]
> 
>> mm/zsmalloc: don't need to save tag bit in handle
> 
> Does this mean "don't need to reserve LSB for tag"?
The head of object still need to reverve LSB, to save (handle | OBJ_ALLOCATED_TAG),
only the handle doesn't need to reserve LSB, which save (pfn | obj_idx).

> We still save allocated tag in the handle, that's what
> 
> 	handle |= OBJ_ALLOCATED_TAG;

Yes, this result will be saved in the head of each allocated object.

> 
> does.
> 
>> Actually, the tag bit is only useful in zspage's memory space, to tell
>> if an object is allocated or not.
> 
> I'm not completely sure if I follow this sentence.

What I mean is that only the head of each allocated object need to reverve LSB,
which is used to check if allocated or not.

handle address -> handle (pfn + obj_idx) -> object: (handle | tag), real_object start

I'm not sure if this makes it clearer?

Thanks.

Re: [PATCH] mm/zsmalloc: don't need to save tag bit in handle

[Sergey Senozhatsky]

On (24/02/27 16:16), Chengming Zhou wrote:
> On 2024/2/27 15:52, Sergey Senozhatsky wrote:
> > On (24/02/27 03:00), chengming.zhou@linux.dev wrote:
> >>
> >> We only need to save the position (pfn + obj_idx) in the handle, don't
> >> need to save tag bit in handle. So one more bit can be used as obj_idx.
> > 
> > [..]
> > 
> >> mm/zsmalloc: don't need to save tag bit in handle
> > 
> > Does this mean "don't need to reserve LSB for tag"?
> The head of object still need to reverve LSB, to save (handle | OBJ_ALLOCATED_TAG),
> only the handle doesn't need to reserve LSB, which save (pfn | obj_idx).

Correct.

> > We still save allocated tag in the handle, that's what
> > 
> > 	handle |= OBJ_ALLOCATED_TAG;
> 
> Yes, this result will be saved in the head of each allocated object.

Right, that's what I was talking about.

> >> Actually, the tag bit is only useful in zspage's memory space, to tell
> >> if an object is allocated or not.
> > 
> > I'm not completely sure if I follow this sentence.
> 
> What I mean is that only the head of each allocated object need to reverve LSB,
> which is used to check if allocated or not.
> 
> handle address -> handle (pfn + obj_idx) -> object: (handle | tag), real_object start
> 
> I'm not sure if this makes it clearer?

Yes, thanks. I think separating handle and object header in the commit
message will be helpful.

Re: [PATCH] mm/zsmalloc: don't need to save tag bit in handle

[Chengming Zhou]

On 2024/2/28 09:54, Sergey Senozhatsky wrote:
> On (24/02/27 16:16), Chengming Zhou wrote:
>> On 2024/2/27 15:52, Sergey Senozhatsky wrote:
>>> On (24/02/27 03:00), chengming.zhou@linux.dev wrote:
>>>>
>>>> We only need to save the position (pfn + obj_idx) in the handle, don't
>>>> need to save tag bit in handle. So one more bit can be used as obj_idx.
>>>
>>> [..]
>>>
>>>> mm/zsmalloc: don't need to save tag bit in handle
>>>
>>> Does this mean "don't need to reserve LSB for tag"?
>> The head of object still need to reverve LSB, to save (handle | OBJ_ALLOCATED_TAG),
>> only the handle doesn't need to reserve LSB, which save (pfn | obj_idx).
> 
> Correct.
> 
>>> We still save allocated tag in the handle, that's what
>>>
>>> 	handle |= OBJ_ALLOCATED_TAG;
>>
>> Yes, this result will be saved in the head of each allocated object.
> 
> Right, that's what I was talking about.
> 
>>>> Actually, the tag bit is only useful in zspage's memory space, to tell
>>>> if an object is allocated or not.
>>>
>>> I'm not completely sure if I follow this sentence.
>>
>> What I mean is that only the head of each allocated object need to reverve LSB,
>> which is used to check if allocated or not.
>>
>> handle address -> handle (pfn + obj_idx) -> object: (handle | tag), real_object start
>>
>> I'm not sure if this makes it clearer?
> 
> Yes, thanks. I think separating handle and object header in the commit
> message will be helpful.

Right, I will improve the commit message and send v2.

Thanks!