a question about slub in function __slab_free()

Xishi Qiu

author

2016-03-02 12:31 UTC #0 lore raw

___slab_alloc()

	deactivate_slab()
		add_full(s, n, page);

The page will be added to full list and the frozen is 0, right?

__slab_free()

	prior = page->freelist;  // prior is NULL
	was_frozen = new.frozen;  // was_frozen is 0
	...
		/*
		 * Slab was on no list before and will be
		 * partially empty
		 * We can defer the list move and instead
		 * freeze it.
		 */
		new.frozen = 1;
	...

I don't understand why "Slab was on no list before"?

Thanks,
Xishi Qiu

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Joonsoo Kim

2016-03-02 14:38 UTC #1 lore raw

2016-03-02 21:26 GMT+09:00 Xishi Qiu <[email protected]>:

Quoted 18 lines

___slab_alloc()

        deactivate_slab()
                add_full(s, n, page);

The page will be added to full list and the frozen is 0, right?

__slab_free()

        prior = page->freelist;  // prior is NULL
        was_frozen = new.frozen;  // was_frozen is 0
        ...
                /*
                 * Slab was on no list before and will be
                 * partially empty
                 * We can defer the list move and instead
                 * freeze it.
                 */
                new.frozen = 1;
        ...

Quoted 1 line

I don't understand why "Slab was on no list before"?

add_full() is defined only for CONFIG_SLUB_DEBUG. And, actual add happens if slub_debug=u is enabled. In other cases, fully used slab isn't attached on any list.

Thanks.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Xishi Qiu

author

2016-03-03 01:33 UTC #2 lore raw

Quoted 19 lines · Joonsoo Kim

2016-03-02 21:26 GMT+09:00 Xishi Qiu <[email protected]>:

Quoted 18 lines

___slab_alloc()

        deactivate_slab()
                add_full(s, n, page);

The page will be added to full list and the frozen is 0, right?

__slab_free()

        prior = page->freelist;  // prior is NULL
        was_frozen = new.frozen;  // was_frozen is 0
        ...
                /*
                 * Slab was on no list before and will be
                 * partially empty
                 * We can defer the list move and instead
                 * freeze it.
                 */
                new.frozen = 1;
        ...

Quoted 8 lines

Quoted 1 line
I don't understand why "Slab was on no list before"?
add_full() is defined only for CONFIG_SLUB_DEBUG. And, actual add happens if slub_debug=u is enabled. In other cases, fully used slab isn't attached on any list.
Thanks.

Hi Joonsoo,

You are right, thank you very much!

Quoted 1 line

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Jianyu Zhan

2016-03-02 15:13 UTC #3 lore raw

In reply to Xishi Qiu

Quoted 13 lines · Xishi Qiu

__slab_free()

        prior = page->freelist;  // prior is NULL
        was_frozen = new.frozen;  // was_frozen is 0
        ...
                /*
                 * Slab was on no list before and will be
                 * partially empty
                 * We can defer the list move and instead
                 * freeze it.
                 */
                new.frozen = 1;
        ...

Quoted 1 line

I don't understand why "Slab was on no list before"?

in this __slab_free() code path, we are freeing an object to a remote CPU.

Consider the condition that leads to this branch, that is:

new.inuse && !prior && !was_frozen

new.inuse means that slab page has free objects after we do this free operiton.

!prior && !was_frozen together means that slab page has previously depleted all objects
and forgotten(SLUB don't remember a slab page that has got all its object allocated).

All these 3 conditions mean that,

A slab page on a remote CPU has got all tis objects allocated and it was forgotten by SLUB,
so "Slab was on no list before",

and then at the present, we (on local CPU) are freeing object back to that CPU, that "will make
the slab page partially empty".

But we don't bother to immediately add it back to the node partial list( to avoid the list->lock contention),
so "we can defer the list move".

But how do we handle this? Easy, just mark it frozen, and latter that CPU's per-cpu freelist queue will
use it.

Regards,
Jianyu Zhan

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>