@samitouri / QOSamiQemu / commits / 75d782a1e9

migration/multifd: cache migrate_multifd_channels() in send/recv hot paths

multifd_send() and multifd_recv() are on the per-page-batch hot path of live migration. Both functions call migrate_multifd_channels() multiple times (3-4 calls each) for modulo arithmetic in the round-robin channel selection loop. Each call goes through migrate_get_current() -> dereference MigrationState -> read parameters.multifd_channels. While each individual call is cheap, these functions execute for every page batch during the entire migration, easily millions of times. Cache the return value in a local variable at function entry. The channel count is fixed for the duration of a migration and cannot change mid-flight. For multifd_send(): 3 calls reduced to 1. For multifd_recv(): 4 calls reduced to 1. Signed-off-by: Bin Guo <guobin@linux.alibaba.com> Reviewed-by: Fabiano Rosas <farosas@suse.de> Link: https://lore.kernel.org/r/20260518110112.21395-8-guobin@linux.alibaba.com Signed-off-by: Peter Xu <peterx@redhat.com>

Bin Guo committed May 18, 2026 at 19:01 UTC 75d782a1e9f376a4dc1812f67101e9474cc133df

1 file changed +9 -6

migration/multifd.c

+9 -6

@@ -362,13 +362,15 @@ bool multifd_send(MultiFDSendData **send_data)

362 /* We wait here, until at least one channel is ready */

363 qemu_sem_wait(&multifd_send_state->channels_ready);

364

365 + int thread_count = migrate_multifd_channels();

366 +

367 /*

368 * next_channel can remain from a previous migration that was

369 * using more channels, so ensure it doesn't overflow if the

370 * limit is lower now.

371 */

370 - next_channel %= migrate_multifd_channels();

371 - for (i = next_channel;; i = (i + 1) % migrate_multifd_channels()) {

372 + next_channel %= thread_count;

373 + for (i = next_channel;; i = (i + 1) % thread_count) {

374 if (multifd_send_should_exit()) {

375 return false;

376 }

@@ -378,7 +380,7 @@ bool multifd_send(MultiFDSendData **send_data)

380 * sender thread can clear it.

381 */

382 if (qatomic_read(&p->pending_job) == false) {

381 - next_channel = (i + 1) % migrate_multifd_channels();

383 + next_channel = (i + 1) % thread_count;

384 break;

385 }

386 }

@@ -998,6 +1000,7 @@ bool multifd_recv(void)

1000 int i;

1001 static int next_recv_channel;

1002 MultiFDRecvParams *p = NULL;

1003 + int thread_count = migrate_multifd_channels();

1004 MultiFDRecvData *data = multifd_recv_state->data;

1005

1006 /*

@@ -1005,8 +1008,8 @@ bool multifd_recv(void)

1008 * using more channels, so ensure it doesn't overflow if the

1009 * limit is lower now.

1010 */

1008 - next_recv_channel %= migrate_multifd_channels();

1009 - for (i = next_recv_channel;; i = (i + 1) % migrate_multifd_channels()) {

1011 + next_recv_channel %= thread_count;

1012 + for (i = next_recv_channel;; i = (i + 1) % thread_count) {

1013 if (multifd_recv_should_exit()) {

1014 return false;

1015 }

@@ -1014,7 +1017,7 @@ bool multifd_recv(void)

1017 p = &multifd_recv_state->params[i];

1018

1019 if (qatomic_read(&p->pending_job) == false) {

1017 - next_recv_channel = (i + 1) % migrate_multifd_channels();

1020 + next_recv_channel = (i + 1) % thread_count;

1021 break;

1022 }

1023 }