On Thu, May 25, 2023 at 02:57:09PM +0200, Thomas Gleixner wrote:
> Purging fragmented blocks is done unconditionally in several contexts:
>
> 1) From drain_vmap_area_work(), when the number of lazy to be freed
> vmap_areas reached the threshold
>
> 2) Reclaiming vmalloc address space from pcpu_get_vm_areas()
>
> 3) _vm_unmap_aliases()
>
> #1 There is no reason to zap fragmented vmap blocks unconditionally, simply
> because reclaiming all lazy areas drains at least
>
> 32MB * fls(num_online_cpus())
>
> per invocation which is plenty.
>
> #2 Reclaiming when running out of space or due to memory pressure makes a
> lot of sense
>
> #3 _unmap_aliases() requires to touch everything because the caller has no
> clue which vmap_area used a particular page last and the vmap_area lost
> that information too.
>
> Except for the vfree + VM_FLUSH_RESET_PERMS case, which removes the
> vmap area first and then cares about the flush. That in turn requires
> a full walk of _all_ vmap areas including the one which was just
> added to the purge list.
>
> But as this has to be flushed anyway this is an opportunity to combine
> outstanding TLB flushes and do the housekeeping of purging freed areas,
> but like #1 there is no real good reason to zap usable vmap blocks
> unconditionally.
>
> Add a @force_purge argument to the newly split out block purge function and
> if not true only purge fragmented blocks which have less than 1/4 of their
> capacity left.
>
> Rename purge_vmap_area_lazy() to reclaim_and_purge_vmap_areas() to make it
> clear what the function does.
>
> Signed-off-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> ---
> V2: Add the missing force_purge argument in _vm_unmap_aliases()
> Remove force_purge argument from the reclaim path - Baoquan
> ---
> mm/vmalloc.c | 28 ++++++++++++++++++++--------
> 1 file changed, 20 insertions(+), 8 deletions(-)
>
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -791,7 +791,7 @@ get_subtree_max_size(struct rb_node *nod
> RB_DECLARE_CALLBACKS_MAX(static, free_vmap_area_rb_augment_cb,
> struct vmap_area, rb_node, unsigned long, subtree_max_size, va_size)
>
> -static void purge_vmap_area_lazy(void);
> +static void reclaim_and_purge_vmap_areas(void);
> static BLOCKING_NOTIFIER_HEAD(vmap_notify_list);
> static void drain_vmap_area_work(struct work_struct *work);
> static DECLARE_WORK(drain_vmap_work, drain_vmap_area_work);
> @@ -1649,7 +1649,7 @@ static struct vmap_area *alloc_vmap_area
>
> overflow:
> if (!purged) {
> - purge_vmap_area_lazy();
> + reclaim_and_purge_vmap_areas();
> purged = 1;
> goto retry;
> }
> @@ -1785,9 +1785,10 @@ static bool __purge_vmap_area_lazy(unsig
> }
>
> /*
> - * Kick off a purge of the outstanding lazy areas.
> + * Reclaim vmap areas by purging fragmented blocks and purge_vmap_area_list.
> */
This comment feels a little redundant as it simply describes the code
below. Perhaps worth highlighting that the purge is unconditional and why it
would be invoked?
Nitty, but I'm also not entirely sure about this use of 'reclaim', as it is
rather an overloaded term and you'd expect it to refer to the activities of
vmscan.c :) yes we're trying to 'reclaim' vmap areas in a sense but perhaps
better to either stick with original or something less overloaded (I would
suggest something but naming is hard... :)
> -static void purge_vmap_area_lazy(void)
> +static void reclaim_and_purge_vmap_areas(void)
> +
> {
> mutex_lock(&vmap_purge_lock);
> purge_fragmented_blocks_allcpus();
> @@ -1908,6 +1909,12 @@ static struct vmap_area *find_unlink_vma
>
> #define VMAP_BLOCK_SIZE (VMAP_BBMAP_BITS * PAGE_SIZE)
>
> +/*
> + * Purge threshold to prevent overeager purging of fragmented blocks for
> + * regular operations: Purge if vb->free is less than 1/4 of the capacity.
> + */
'Purge if vb->free is less than 1/4 of the capacity' appears to be contradicted
by the actual conditional in purge_fragmented_block() below (see comment below)
unless I'm missing something/simply confused :)
> +#define VMAP_PURGE_THRESHOLD (VMAP_BBMAP_BITS / 4)
> +
> #define VMAP_RAM 0x1 /* indicates vm_map_ram area*/
> #define VMAP_BLOCK 0x2 /* mark out the vmap_block sub-type*/
> #define VMAP_FLAGS_MASK 0x3
> @@ -2087,12 +2094,17 @@ static void free_vmap_block(struct vmap_
> }
>
> static bool purge_fragmented_block(struct vmap_block *vb,
> - struct vmap_block_queue *vbq, struct list_head *purge_list)
> + struct vmap_block_queue *vbq, struct list_head *purge_list,
> + bool force_purge)
> {
> if (vb->free + vb->dirty != VMAP_BBMAP_BITS ||
> vb->dirty == VMAP_BBMAP_BITS)
> return false;
>
> + /* Don't overeagerly purge usable blocks unless requested */
> + if (!force_purge && vb->free < VMAP_PURGE_THRESHOLD)
Little confused by this - if free capacity is less than the threshold, we
_don't_ purge? I thought we should purge in this instance? Should this be an >=?
> + return false;
> +
> /* prevent further allocs after releasing lock */
> WRITE_ONCE(vb->free, 0);
> /* prevent purging it again */
> @@ -2132,7 +2144,7 @@ static void purge_fragmented_blocks(int
> continue;
>
> spin_lock(&vb->lock);
> - purge_fragmented_block(vb, vbq, &purge);
> + purge_fragmented_block(vb, vbq, &purge, true);
> spin_unlock(&vb->lock);
> }
> rcu_read_unlock();
> @@ -2269,7 +2281,7 @@ static void _vm_unmap_aliases(unsigned l
> * not purgeable, check whether there is dirty
> * space to be flushed.
> */
> - if (!purge_fragmented_block(vb, vbq, &purge_list) &&
> + if (!purge_fragmented_block(vb, vbq, &purge_list, false) &&
> vb->dirty_max && vb->dirty != VMAP_BBMAP_BITS) {
> unsigned long va_start = vb->va->va_start;
> unsigned long s, e;
> @@ -4175,7 +4187,7 @@ struct vm_struct **pcpu_get_vm_areas(con
> overflow:
> spin_unlock(&free_vmap_area_lock);
> if (!purged) {
> - purge_vmap_area_lazy();
> + reclaim_and_purge_vmap_areas();
> purged = true;
>
> /* Before "retry", check if we recover. */
>
>
