on 3/15/2025 4:27 AM, Tim Chen wrote:
> On Fri, 2025-03-14 at 05:05 +0800, Kemeng Shi wrote:
>> Hi All,
>> This series contains some cleanups and improvements which are made
>> during learning swapfile. Here is a summary of the changes:
>
Hello,
> Nice work>
>> 1. Function nameing improvments.
>> -Use "put" instead of "free" to name functions which only do actual free
>> when count drops to zero.
>> -Use "entry" to name function only frees one swap slot. Use "entries" to
>> name function could may free multi swap slots within one cluster. Use
>> "_nr" suffix to name function which could free multi swap slots spanning
>> cross multi clusters.
>
> Will be nice to add some comments in the code about functions with _nr
> crossing the cluster boundaries and those without stay within a cluster.
Thanks for reviewing and all advises to this series. Will improve in next
version.
>
>> 2. Eliminate the need to set swap slot to intermediate SWAP_HAS_CACHE
>> value before do actual free by using swap_entry_range_free()
>> 3. Add helpers swap_entries_put_map() and swap_entries_put_cache() as a
>> general-purpose routine to free swap entries within a single cluster
>> which will try batch-remove first and fallback to put eatch entry
>> indvidually with cluster lock acquired/released only once. By using
>> these helpers, we could remove repeated code, levarage batch-remove in
>> more cases and aoivd to acquire/release cluster lock for each single
>> swap entry.
>
> Wonder if the batching shows up in any swap performance improvement
I have a simple test which is roughly as following:
#define SIZE 1024*1024*1024
#define HP_SIZE 2 * 1024 * 1024
p = mmap(...,SIZE,...)
madvise(p, SIZE, MADV_HUGEPAGE)
memset(p, 0x11, SIZE); /* alloc page */
madvise(p, SIZE, MADV_PAGEOUT); /* swap out */
gettimeofday(&tv_b, NULL);
for (j = 0; j < SIZE; j+= HP_SIZE) {
((char *)p)[j] = 0; /* swap in and free swap entry */
}
The time for swap-in and free swap entry shows no significant change.
Since this is more of a cleanup series, no further testing has been
conducted yet. I would appreciate it if you have any additional test
cases that could benefit from batching.
Thank,
Kemeng
.
>
> Tim
>
>>
>> Kemeng Shi (9):
>> mm: swap: rename __swap_[entry/entries]_free[_locked] to
>> swap_[entry/entries]_put[_locked]
>> mm: swap: factor out the actual swap entry freeing logic to new helper
>> mm: swap: use __swap_entry_free() to free swap entry in
>> swap_entry_put_locked()
>> mm: swap: remove unneeded VM_BUG_ON(*map != SWAP_HAS_CACHE) in
>> swap_entry_range_free()
>> mm: swap: use swap_entries_free() drop last 1 flag in
>> swap_entries_put_nr()
>> mm: swap: drop last SWAP_MAP_SHMEM flag in batch in
>> swap_entries_put_nr()
>> mm: swap: free each cluster individually in swap_entries_put_map_nr()
>> mm: swap: factor out helper to drop cache of entries within a single
>> cluster
>> mm: swap: replace cluster_swap_free_nr() with
>> swap_entries_put_[map/cache]()
>>
>> mm/swapfile.c | 173 +++++++++++++++++++++++++-------------------------
>> 1 file changed, 86 insertions(+), 87 deletions(-)
>>
>
>