MSG_ZEROCOPY + multifd
v7 cover MSG_ZEROCOPY + multifd
This patch series intends to enable MSG_ZEROCOPY in QIOChannel, and make use of it for multifd migration performance improvement, by reducing cpu usage.
Patch #1 creates new callbacks for QIOChannel, allowing the implementation of zero copy writing.
Patch #2 implements io_writev flags and io_flush() on QIOChannelSocket, making use of MSG_ZEROCOPY on Linux.
Patch #3 adds a "zero_copy" migration property, only available with CONFIG_LINUX, and compiled-out in any other architectures. This migration property has to be enabled before multifd migration starts.
Patch #4 adds a helper function that allows to see if TLS is going to be used. This helper will be later used in patch #5.
Patch #5 Makes use of QIOChannelSocket zero_copy implementation on nocomp multifd migration.
Results:
In preliminary tests, the resource usage of __sys_sendmsg() reduced 15 times, and the overall migration took 13-22% less time, based in synthetic cpu workload.
In further tests, it was noted that, on multifd migration with 8 channels:
- On idle hosts, migration time reduced in 10% to 21%.
- On hosts busy with heavy cpu stress (1 stress thread per cpu, but not cpu-pinned) migration time reduced in ~25% by enabling zero-copy.
- On hosts with heavy cpu-pinned workloads (1 stress thread per cpu, cpu-pinned), migration time reducted in ~66% by enabling zero-copy.
Above tests setup:
- Sending and Receiving hosts:
- CPU : Intel(R) Xeon(R) Platinum 8276L CPU @ 2.20GHz (448 CPUS)
- Network card: E810-C (100Gbps)
- >1TB RAM
- QEMU: Upstream master branch + This patchset
- Linux: Upstream v5.15
- VM configuration:
- 28 VCPUs
- 512GB RAM
---
Changes since v6:
- Remove io_writev_zero_copy(), and makes use of io_writev() new flags to achieve the same results.
- Rename io_flush_zero_copy() to io_flush()
- Previous patch #2 became too small, so it was squashed in previous patch #3 (now patch #2)
Changes since v5:
- flush_zero_copy now returns -1 on fail, 0 on success, and 1 when all processed writes were not able to use zerocopy in kernel.
- qio_channel_socket_poll() removed, using qio_channel_wait() instead
- ENOBUFS is now processed inside qio_channel_socket_writev_flags()
- Most zerocopy parameter validation moved to migrate_params_check(), leaving only feature test to socket_outgoing_migration() callback
- Naming went from *zerocopy to *zero_copy or *zero-copy, due to QAPI/QMP preferences
- Improved docs
Changes since v4:
- 3 patches got splitted in 6
- Flush is used for syncing after each iteration, instead of only at the end
- If zerocopy is not available, fail in connect instead of failing on write
- 'multifd-zerocopy' property renamed to 'zerocopy'
- Fail migrations that don't support zerocopy, if it's enabled.
- Instead of checking for zerocopy at each write, save the flags in MultiFDSendParams->write_flags and use them on write
- Reorganized flag usage in QIOChannelSocket
- A lot of typos fixed
- More doc on buffer restrictions
Changes since v3:
- QIOChannel interface names changed from io_async_{writev,flush} to io_{writev,flush}_zerocopy
- Instead of falling back in case zerocopy is not implemented, return error and abort operation.
- Flush now waits as long as needed, or return error in case anything goes wrong, aborting the operation.
- Zerocopy is now conditional in multifd, being set by parameter multifd-zerocopy
- Moves zerocopy_flush to multifd_send_sync_main() from multifd_save_cleanup so migration can abort if flush goes wrong.
- Several other small improvements
Changes since v2:
- Patch #1: One more fallback
- Patch #2: Fall back to sync if fails to lock buffer memory in MSG_ZEROCOPY send.
Changes since v1:
- Reimplemented the patchset using async_write + async_flush approach.
- Implemented a flush to be able to tell whenever all data was written.
Leonardo Bras (5):
QIOChannel: Add flags on io_writev and introduce io_flush callback
QIOChannelSocket: Implement io_writev zero copy flag & io_flush for
CONFIG_LINUX
migration: Add zero-copy parameter for QMP/HMP for Linux
migration: Add migrate_use_tls() helper
multifd: Implement zero copy write in multifd migration
(multifd-zero-copy)
qapi/migration.json | 24 ++++++++ include/io/channel-socket.h | 2 + include/io/channel.h | 67 +++++++++++++++++----- migration/migration.h | 6 ++ migration/multifd.h | 4 +- io/channel-buffer.c | 1 + io/channel-command.c | 1 + io/channel-file.c | 1 + io/channel-socket.c | 109 ++++++++++++++++++++++++++++++++++-- io/channel-tls.c | 1 + io/channel-websock.c | 1 + io/channel.c | 51 ++++++++++++----- migration/channel.c | 6 +- migration/migration.c | 52 ++++++++++++++++- migration/multifd.c | 45 ++++++++++++--- migration/ram.c | 29 +++++++--- migration/rdma.c | 1 + migration/socket.c | 6 ++ monitor/hmp-cmds.c | 6 ++ 19 files changed, 360 insertions(+), 53 deletions(-)
Signature · 1 line
2.34.1
PATCH v7 2/5 QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
For CONFIG_LINUX, implement the new zero copy flag and the optional callback io_flush on QIOChannelSocket, but enables it only when MSG_ZEROCOPY feature is available in the host kernel, which is checked on qio_channel_socket_connect_sync()
qio_channel_socket_flush() was implemented by counting how many times sendmsg(...,MSG_ZEROCOPY) was successfully called, and then reading the socket's error queue, in order to find how many of them finished sending. Flush will loop until those counters are the same, or until some error occurs.
Notes on using writev() with QIO_CHANNEL_WRITE_FLAG_ZERO_COPY:
1: Buffer
- As MSG_ZEROCOPY tells the kernel to use the same user buffer to avoid copying, some caution is necessary to avoid overwriting any buffer before it's sent. If something like this happen, a newer version of the buffer may be sent instead.
- If this is a problem, it's recommended to call qio_channel_flush() before freeing or re-using the buffer.
2: Locked memory
- When using MSG_ZERCOCOPY, the buffer memory will be locked after queued, and unlocked after it's sent.
- Depending on the size of each buffer, and how often it's sent, it may require a larger amount of locked memory than usually available to non-root user.
- If the required amount of locked memory is not available, writev_zero_copy will return an error, which can abort an operation like migration,
- Because of this, when an user code wants to add zero copy as a feature, it requires a mechanism to disable it, so it can still be accessible to less privileged users.
--- include/io/channel-socket.h | 2 + io/channel-socket.c | 107 ++++++++++++++++++++++++++++++++++-- 2 files changed, 105 insertions(+), 4 deletions(-)
include/io/channel-socket.h +2 -0
@@ -47,6 +47,8 @@ struct QIOChannelSocket {socklen_t localAddrLen;struct sockaddr_storage remoteAddr;socklen_t remoteAddrLen; +ssize_t zero_copy_queued; +ssize_t zero_copy_sent; };
io/channel-socket.c +103 -4
154 lines not on this page · raw message
Signature · 1 line
2.34.1
PATCH v7 1/5 QIOChannel: Add flags on io_writev and introduce io_flush callback
In reply to Leonardo BrasAdd flags to io_writev and introduce io_flush as optional callback to QIOChannelClass, allowing the implementation of zero copy writes by subclasses.
How to use them:
- Write data using qio_channel_writev(...,QIO_CHANNEL_WRITE_FLAG_ZERO_COPY),
- Wait write completion with qio_channel_flush().
Notes:
As some zero copy implementations work asynchronously, it's recommended to keep the write buffer untouched until the return of qio_channel_flush(), to avoid the risk of sending an updated buffer instead of the buffer state during write.
As io_flush callback is optional, if a subclass does not implement it, then:
- io_flush will return 0 without changing anything.
Also, some functions like qio_channel_writev_full_all() were adapted to receive a flag parameter. That allows shared code between zero copy and non-zero copy writev, and also an easier implementation on new flags.
--- include/io/channel.h | 67 +++++++++++++++++++++++++++++++++++--------- io/channel-buffer.c | 1 + io/channel-command.c | 1 + io/channel-file.c | 1 + io/channel-socket.c | 2 ++ io/channel-tls.c | 1 + io/channel-websock.c | 1 + io/channel.c | 51 +++++++++++++++++++++++---------- migration/rdma.c | 1 + 9 files changed, 98 insertions(+), 28 deletions(-)
include/io/channel.h +54 -13
@@ -32,12 +32,15 @@ OBJECT_DECLARE_TYPE(QIOChannel, QIOChannelClass,#define QIO_CHANNEL_ERR_BLOCK -2 +#define QIO_CHANNEL_WRITE_FLAG_ZERO_COPY 0x1 +typedef enum QIOChannelFeature QIOChannelFeature ;enum QIOChannelFeature {QIO_CHANNEL_FEATURE_FD_PASS ,QIO_CHANNEL_FEATURE_SHUTDOWN ,QIO_CHANNEL_FEATURE_LISTEN , +QIO_CHANNEL_FEATURE_WRITE_ZERO_COPY , };
@@ -104,6 +107,7 @@ struct QIOChannelClass {size_t niov, int *fds, size_t nfds, + int flags, Error **errp); ssize_t (*io_readv)(QIOChannel *ioc, const struct iovec *iov,
@@ -136,6 +140,8 @@ struct QIOChannelClass { IOHandler *io_read, IOHandler *io_write, void *opaque); +int (*io_flush)(QIOChannel *ioc, +Error **errp); };/* General I/O handling functions */
@@ -222,12 +228,13 @@ ssize_t qio_channel_readv_full(QIOChannel *ioc,/** -* qio_channel_writev_full: +* qio_channel_writev_full_flags: * @ioc: the channel object * @iov: the array of memory regions to write data from * @niov: the length of the @iov array * @fds: an array of file handles to send * @nfds: number of file handles in @fds +* @flags: write flags (QIO_CHANNEL_WRITE_FLAG_*) * @errp: pointer to a NULL-initialized error object * * Write data to the IO channel, reading it from the
@@ -255,12 +262,16 @@ ssize_t qio_channel_readv_full(QIOChannel *ioc,* or QIO_CHANNEL_ERR_BLOCK if no data is can be sent * and the channel is non-blocking */ -ssize_t qio_channel_writev_full (QIOChannel *ioc, -const struct iovec *iov, -size_t niov, -int *fds, -size_t nfds, -Error **errp); +ssize_t qio_channel_writev_full_flags (QIOChannel *ioc, +const struct iovec *iov, +size_t niov, +int *fds, +size_t nfds, +int flags, +Error **errp); + +#define qio_channel_writev_full (ioc, iov, niov, fds, nfds, errp) \ + qio_channel_writev_full_flags(ioc, iov, niov, fds, nfds, 0, errp)/** * qio_channel_readv_all_eof:
@@ -831,12 +842,13 @@ int qio_channel_readv_full_all(QIOChannel *ioc, Error **errp);/** -* qio_channel_writev_full_all: +* qio_channel_writev_full_all_flags: * @ioc: the channel object * @iov: the array of memory regions to write data from * @niov: the length of the @iov array * @fds: an array of file handles to send * @nfds: number of file handles in @fds +* @flags: write flags (QIO_CHANNEL_WRITE_FLAG_*) * @errp: pointer to a NULL-initialized error object * *
@@ -846,13 +858,42 @@ int qio_channel_readv_full_all(QIOChannel *ioc,* to be written, yielding from the current coroutine * if required. * +* If QIO_CHANNEL_WRITE_FLAG_ZERO_COPY is passed in flags, +* instead of waiting for all requested data to be written, +* this function will wait until it's all queued for writing. +* In this case, if the buffer gets changed between queueing and +* sending, the updated buffer will be sent. If this is not a +* desired behavior, it's suggested to call qio_channel_flush() +* before reusing the buffer. +* * Returns: 0 if all bytes were written, or -1 on error */ -int qio_channel_writev_full_all (QIOChannel *ioc, -const struct iovec *iov, -size_t niov, -int *fds,size_t nfds, -Error **errp); +int qio_channel_writev_full_all_flags (QIOChannel *ioc, +const struct iovec *iov, +size_t niov, +int *fds,size_t nfds, +int flags,Error **errp); +#define qio_channel_writev_full_all (ioc, iov, niov, fds, nfds, errp) \ + qio_channel_writev_full_all_flags(ioc, iov, niov, fds, nfds, 0, errp) + +/** +* qio_channel_flush: +* @ioc: the channel object +* @errp: pointer to a NULL-initialized error object +* +* Will block until every packet queued with +* qio_channel_writev_full_flags() + QIO_CHANNEL_WRITE_FLAG_ZERO_COPY +* is sent, or return in case of any error. +* +* If not implemented, acts as a no-op, and returns 0. +* +* Returns -1 if any error is found, +* 1 if every send failed to use zero copy. +* 0 otherwise. +*/ + +int qio_channel_flush (QIOChannel *ioc, +Error **errp);#endif /* QIO_CHANNEL_H */
io/channel-buffer.c +1 -0
@@ -81,6 +81,7 @@ static ssize_t qio_channel_buffer_writev(QIOChannel *ioc,size_t niov, int *fds, size_t nfds, + int flags, Error **errp) { QIOChannelBuffer *bioc = QIO_CHANNEL_BUFFER(ioc);
io/channel-command.c +1 -0
@@ -258,6 +258,7 @@ static ssize_t qio_channel_command_writev(QIOChannel *ioc,size_t niov, int *fds, size_t nfds, + int flags, Error **errp) { QIOChannelCommand *cioc = QIO_CHANNEL_COMMAND(ioc);
io/channel-file.c +1 -0
@@ -114,6 +114,7 @@ static ssize_t qio_channel_file_writev(QIOChannel *ioc,size_t niov, int *fds, size_t nfds, + int flags, Error **errp) { QIOChannelFile *fioc = QIO_CHANNEL_FILE(ioc);
io/channel-socket.c +2 -0
@@ -525,6 +525,7 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,size_t niov, int *fds, size_t nfds, + int flags, Error **errp) { QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc);
@@ -620,6 +621,7 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,size_t niov, int *fds, size_t nfds, + int flags, Error **errp) { QIOChannelSocket *sioc = QIO_CHANNEL_SOCKET(ioc);
io/channel-tls.c +1 -0
@@ -301,6 +301,7 @@ static ssize_t qio_channel_tls_writev(QIOChannel *ioc,size_t niov, int *fds, size_t nfds, + int flags, Error **errp) { QIOChannelTLS *tioc = QIO_CHANNEL_TLS(ioc);
io/channel-websock.c +1 -0
@@ -1127,6 +1127,7 @@ static ssize_t qio_channel_websock_writev(QIOChannel *ioc,size_t niov, int *fds, size_t nfds, + int flags, Error **errp) { QIOChannelWebsock *wioc = QIO_CHANNEL_WEBSOCK(ioc);
io/channel.c +36 -15
85 lines not on this page · raw message
migration/rdma.c +1 -0
@@ -2833,6 +2833,7 @@ static ssize_t qio_channel_rdma_writev(QIOChannel *ioc,size_t niov, int *fds, size_t nfds, + int flags, Error **errp) { QIOChannelRDMA *rioc = QIO_CHANNEL_RDMA(ioc);
Signature · 1 line
2.34.1
PATCH v7 5/5 multifd: Implement zero copy write in multifd migration (multifd-zero-copy)
In reply to Leonardo BrasImplement zero copy on nocomp_send_write(), by making use of QIOChannel writev + flags & flush interface.
Change multifd_send_sync_main() so it can distinguish each iteration sync from the setup and the completion, so a flush_zero_copy() can be called after each iteration in order to make sure all dirty pages are sent before a new iteration is started.
Also make it return -1 if flush_zero_copy() fails, in order to cancel the migration process, and avoid resuming the guest in the target host without receiving all current RAM.
This will work fine on RAM migration because the RAM pages are not usually freed, and there is no problem on changing the pages content between writev_zero_copy() and the actual sending of the buffer, because this change will dirty the page and cause it to be re-sent on a next iteration anyway.
A lot of locked memory may be needed in order to use multid migration with zero-copy enabled, so disabling the feature should be necessary for low-privileged users trying to perform multifd migrations.
--- migration/multifd.h | 4 +++- migration/migration.c | 11 ++++++++++- migration/multifd.c | 40 +++++++++++++++++++++++++++++++++++----- migration/ram.c | 29 ++++++++++++++++++++++------- migration/socket.c | 5 +++-- 5 files changed, 73 insertions(+), 16 deletions(-)
migration/multifd.h +3 -1
@@ -22,7 +22,7 @@ int multifd_load_cleanup(Error **errp);bool multifd_recv_all_channels_created (void );bool multifd_recv_new_channel (QIOChannel *ioc,Error **errp);void multifd_recv_sync_main (void ); -void multifd_send_sync_main (QEMUFile *f); +int multifd_send_sync_main (QEMUFile *f,bool sync);int multifd_queue_page (QEMUFile *f,RAMBlock *block,ram_addr_t offset);/* Multifd Compression flags */
@@ -97,6 +97,8 @@ typedef struct {uint32_t packet_len;/* pointer to the packet */ MultiFDPacket_t *packet; +/* multifd flags for sending ram */ +int write_flags;/* multifd flags for each packet */ uint32_t flags;/* size of the next packet that contains pages */
migration/migration.c +10 -1
@@ -1476,7 +1476,16 @@ static bool migrate_params_check(MigrationParameters *params, Error **errp)error_prepend (errp,"Invalid mapping given for block-bitmap-mapping: " );return false; } - +#ifdef CONFIG_LINUX +if (params->zero_copy && + (!migrate_use_multifd () || + params->multifd_compression !=MULTIFD_COMPRESSION_NONE || + (params->tls_creds && *params->tls_creds))) { +error_setg (errp, +"Zero copy only available for non-compressed non-TLS multifd migration" ); +return false; + } +#endif return true; }
migration/multifd.c +35 -5
@@ -104,7 +104,8 @@ static int nocomp_send_prepare(MultiFDSendParams *p, Error **errp)*/ static int nocomp_send_write (MultiFDSendParams *p,uint32_t used,Error **errp) { -return qio_channel_writev_all (p->c, p->pages->iov, used, errp); +return qio_channel_writev_full_all_flags (p->c, p->pages->iov, used, +NULL ,0 , p->write_flags, errp); }/**
@@ -582,19 +583,28 @@ void multifd_save_cleanup(void) multifd_send_state =NULL ; } -void multifd_send_sync_main (QEMUFile *f) +int multifd_send_sync_main (QEMUFile *f,bool sync) {int i; +bool flush_zero_copy;if (!migrate_use_multifd ()) { -return ; +return 0 ; }if (multifd_send_state->pages->num) {if (multifd_send_pages (f) <0 ) {error_report ("%s: multifd_send_pages fail" , __func__); -return ; +return 0 ; } } + +/* +* When using zero-copy, it's necessary to flush after each iteration to +* make sure pages from earlier iterations don't end up replacing newer +* pages. +*/ + flush_zero_copy = sync &&migrate_use_zero_copy (); +for (i =0 ; i <migrate_multifd_channels (); i++) {MultiFDSendParams *p = &multifd_send_state->params[i];
@@ -605,7 +615,7 @@ void multifd_send_sync_main(QEMUFile *f)if (p->quit) {error_report ("%s: channel %d has already quit" , __func__, i);qemu_mutex_unlock (&p->mutex); -return ; +return 0 ; } p->packet_num = multifd_send_state->packet_num++;
@@ -616,6 +626,17 @@ void multifd_send_sync_main(QEMUFile *f) ram_counters.transferred += p->packet_len;qemu_mutex_unlock (&p->mutex);qemu_sem_post (&p->sem); + +if (flush_zero_copy) { +int ret; +Error *err =NULL ; + + ret =qio_channel_flush (p->c, &err); +if (ret <0 ) { +error_report_err (err); +return -1 ; + } + } }for (i =0 ; i <migrate_multifd_channels (); i++) {MultiFDSendParams *p = &multifd_send_state->params[i];
@@ -624,6 +645,8 @@ void multifd_send_sync_main(QEMUFile *f)qemu_sem_wait (&p->sem_sync); }trace_multifd_send_sync_main (multifd_send_state->packet_num); + +return 0 ; }static void *multifd_send_thread (void *opaque)
@@ -919,6 +942,13 @@ int multifd_save_setup(Error **errp) p->packet->version =cpu_to_be32 (MULTIFD_VERSION ); p->name =g_strdup_printf ("multifdsend_%d" , i); p->tls_hostname =g_strdup (s->hostname); + +if (migrate_use_zero_copy ()) { + p->write_flags =QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ; + }else { + p->write_flags =0 ; + } +socket_send_channel_create (multifd_new_send_channel_async, p); }
migration/ram.c +22 -7
@@ -2987,6 +2987,7 @@ static int ram_save_setup(QEMUFile *f, void *opaque) {RAMState **rsp = opaque;RAMBlock *block; +int ret;if (compress_threads_save_setup ()) {return -1 ;
@@ -3021,7 +3022,11 @@ static int ram_save_setup(QEMUFile *f, void *opaque)ram_control_before_iterate (f,RAM_CONTROL_SETUP );ram_control_after_iterate (f,RAM_CONTROL_SETUP ); -multifd_send_sync_main (f); + ret =multifd_send_sync_main (f, false); +if (ret <0 ) { +return ret; + } +qemu_put_be64 (f,RAM_SAVE_FLAG_EOS );qemu_fflush (f);
@@ -3130,7 +3135,11 @@ static int ram_save_iterate(QEMUFile *f, void *opaque) out:if (ret >=0 &&migration_is_setup_or_active (migrate_get_current ()->state)) { -multifd_send_sync_main (rs->f); + ret =multifd_send_sync_main (rs->f, true); +if (ret <0 ) { +return ret; + } +qemu_put_be64 (f,RAM_SAVE_FLAG_EOS );qemu_fflush (f); ram_counters.transferred +=8 ;
@@ -3188,13 +3197,19 @@ static int ram_save_complete(QEMUFile *f, void *opaque)ram_control_after_iterate (f,RAM_CONTROL_FINISH ); } -if (ret >=0 ) { -multifd_send_sync_main (rs->f); -qemu_put_be64 (f,RAM_SAVE_FLAG_EOS ); -qemu_fflush (f); +if (ret <0 ) { +return ret; } -return ret; + ret =multifd_send_sync_main (rs->f, false); +if (ret <0 ) { +return ret; + } + +qemu_put_be64 (f,RAM_SAVE_FLAG_EOS ); +qemu_fflush (f); + +return 0 ; } static void ram_save_pending(QEMUFile *f, void *opaque, uint64_t max_size,
migration/socket.c +3 -2
@@ -78,8 +78,9 @@ static void socket_outgoing_migration(QIOTask *task,trace_migration_socket_outgoing_connected (data->hostname); } -if (migrate_use_zero_copy ()) { -error_setg (&err,"Zero copy not available in migration" ); +if (migrate_use_zero_copy () && + !qio_channel_has_feature (sioc,QIO_CHANNEL_FEATURE_WRITE_ZERO_COPY )) { +error_setg (&err,"Zero copy feature not detected in host kernel" ); }migration_channel_connect (data->s, sioc, data->hostname, err);
Signature · 1 line
2.34.1
PATCH v7 4/5 migration: Add migrate_use_tls() helper
In reply to Leonardo BrasA lot of places check parameters.tls_creds in order to evaluate if TLS is in use, and sometimes call migrate_get_current() just for that test.
Add new helper function migrate_use_tls() in order to simplify testing for TLS usage.
--- migration/migration.h | 1 + migration/channel.c | 6 +++--- migration/migration.c | 9 +++++++++ migration/multifd.c | 5 +---- 4 files changed, 14 insertions(+), 7 deletions(-)
migration/migration.h +1 -0
@@ -344,6 +344,7 @@ bool migrate_use_zero_copy(void);#else #define migrate_use_zero_copy () (false)#endif +int migrate_use_tls (void );int migrate_use_xbzrle (void );uint64_t migrate_xbzrle_cache_size (void );bool migrate_colo_enabled (void );
migration/channel.c +3 -3
@@ -32,16 +32,16 @@*/ void migration_channel_process_incoming (QIOChannel *ioc) { -MigrationState *s =migrate_get_current ();Error *local_err =NULL ;trace_migration_set_incoming_channel ( ioc,object_get_typename (OBJECT (ioc))); -if (s->parameters.tls_creds && - *s->parameters.tls_creds && +if (migrate_use_tls () && !object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_TLS )) { +MigrationState *s =migrate_get_current (); +migration_tls_channel_process_incoming (s, ioc, &local_err); }else {migration_ioc_register_yank (ioc);
migration/migration.c +9 -0
@@ -2573,6 +2573,15 @@ bool migrate_use_zero_copy(void) }#endif +int migrate_use_tls (void ) +{ +MigrationState *s; + + s =migrate_get_current (); + +return s->parameters.tls_creds && *s->parameters.tls_creds; +} +int migrate_use_xbzrle (void ) {MigrationState *s;
migration/multifd.c +1 -4
@@ -796,14 +796,11 @@ static bool multifd_channel_connect(MultiFDSendParams *p, QIOChannel *ioc, Error *error) { -MigrationState *s =migrate_get_current (); -trace_multifd_set_outgoing_channel ( ioc,object_get_typename (OBJECT (ioc)), p->tls_hostname, error);if (!error) { -if (s->parameters.tls_creds && - *s->parameters.tls_creds && +if (migrate_use_tls () && !object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_TLS )) {multifd_tls_channel_connect (p, ioc, &error);
Signature · 1 line
2.34.1
PATCH v7 3/5 migration: Add zero-copy parameter for QMP/HMP for Linux
In reply to Leonardo BrasAdd property that allows zero-copy migration of memory pages, and also includes a helper function migrate_use_zero_copy() to check if it's enabled.
No code is introduced to actually do the migration, but it allow future implementations to enable/disable this feature.
On non-Linux builds this parameter is compiled-out.
--- qapi/migration.json | 24 ++++++++++++++++++++++++ migration/migration.h | 5 +++++ migration/migration.c | 32 ++++++++++++++++++++++++++++++++ migration/socket.c | 5 +++++ monitor/hmp-cmds.c | 6 ++++++ 5 files changed, 72 insertions(+)
qapi/migration.json +24 -0
@@ -730,6 +730,13 @@ # will consume more CPU. # Defaults to 1. (Since 5.0) # +# @zero-copy: Controls behavior on sending memory pages on migration. +# When true, enables a zero-copy mechanism for sending memory +# pages, if host supports it. +# Requires that QEMU be permitted to use locked memory for guest +# RAM pages. +# Defaults to false. (Since 7.0) +# # @block-bitmap-mapping: Maps block nodes and bitmaps on them to # aliases for the purpose of dirty bitmap migration. Such # aliases may for example be the corresponding names on the
@@ -769,6 +776,7 @@ 'xbzrle-cache-size', 'max-postcopy-bandwidth', 'max-cpu-throttle', 'multifd-compression', 'multifd-zlib-level' ,'multifd-zstd-level', + { 'name': 'zero-copy', 'if' : 'CONFIG_LINUX'}, 'block-bitmap-mapping' ] } ##
@@ -895,6 +903,13 @@ # will consume more CPU. # Defaults to 1. (Since 5.0) # +# @zero-copy: Controls behavior on sending memory pages on migration. +# When true, enables a zero-copy mechanism for sending memory +# pages, if host supports it. +# Requires that QEMU be permitted to use locked memory for guest +# RAM pages. +# Defaults to false. (Since 7.0) +# # @block-bitmap-mapping: Maps block nodes and bitmaps on them to # aliases for the purpose of dirty bitmap migration. Such # aliases may for example be the corresponding names on the
@@ -949,6 +964,7 @@ '*multifd-compression': 'MultiFDCompression', '*multifd-zlib-level': 'uint8', '*multifd-zstd-level': 'uint8', + '*zero-copy': { 'type': 'bool', 'if': 'CONFIG_LINUX' }, '*block-bitmap-mapping': [ 'BitmapMigrationNodeAlias' ] } } ##
@@ -1095,6 +1111,13 @@ # will consume more CPU. # Defaults to 1. (Since 5.0) # +# @zero-copy: Controls behavior on sending memory pages on migration. +# When true, enables a zero-copy mechanism for sending memory +# pages, if host supports it. +# Requires that QEMU be permitted to use locked memory for guest +# RAM pages. +# Defaults to false. (Since 7.0) +# # @block-bitmap-mapping: Maps block nodes and bitmaps on them to # aliases for the purpose of dirty bitmap migration. Such # aliases may for example be the corresponding names on the
@@ -1147,6 +1170,7 @@ '*multifd-compression': 'MultiFDCompression', '*multifd-zlib-level': 'uint8', '*multifd-zstd-level': 'uint8', + '*zero-copy': { 'type': 'bool', 'if': 'CONFIG_LINUX' }, '*block-bitmap-mapping': [ 'BitmapMigrationNodeAlias' ] } } ##
migration/migration.h +5 -0
@@ -339,6 +339,11 @@ MultiFDCompression migrate_multifd_compression(void);int migrate_multifd_zlib_level (void );int migrate_multifd_zstd_level (void ); +#ifdef CONFIG_LINUX +bool migrate_use_zero_copy (void ); +#else +#define migrate_use_zero_copy () (false) +#endif int migrate_use_xbzrle (void );uint64_t migrate_xbzrle_cache_size (void );bool migrate_colo_enabled (void );
migration/migration.c +32 -0
@@ -893,6 +893,10 @@ MigrationParameters *qmp_query_migrate_parameters(Error **errp) params->multifd_zlib_level = s->parameters.multifd_zlib_level; params->has_multifd_zstd_level = true; params->multifd_zstd_level = s->parameters.multifd_zstd_level; +#ifdef CONFIG_LINUX + params->has_zero_copy = true; + params->zero_copy = s->parameters.zero_copy; +#endif params->has_xbzrle_cache_size = true; params->xbzrle_cache_size = s->parameters.xbzrle_cache_size; params->has_max_postcopy_bandwidth = true;
@@ -1546,6 +1550,11 @@ static void migrate_params_test_apply(MigrateSetParameters *params,if (params->has_multifd_compression) { dest->multifd_compression = params->multifd_compression; } +#ifdef CONFIG_LINUX +if (params->has_zero_copy) { + dest->zero_copy = params->zero_copy; + } +#endif if (params->has_xbzrle_cache_size) { dest->xbzrle_cache_size = params->xbzrle_cache_size; }
@@ -1658,6 +1667,11 @@ static void migrate_params_apply(MigrateSetParameters *params, Error **errp)if (params->has_multifd_compression) { s->parameters.multifd_compression = params->multifd_compression; } +#ifdef CONFIG_LINUX +if (params->has_zero_copy) { + s->parameters.zero_copy = params->zero_copy; + } +#endif if (params->has_xbzrle_cache_size) { s->parameters.xbzrle_cache_size = params->xbzrle_cache_size;xbzrle_cache_resize (params->xbzrle_cache_size, errp);
@@ -2548,6 +2562,17 @@ int migrate_multifd_zstd_level(void)return s->parameters.multifd_zstd_level; } +#ifdef CONFIG_LINUX +bool migrate_use_zero_copy (void ) +{ +MigrationState *s; + + s =migrate_get_current (); + +return s->parameters.zero_copy; +} +#endif +int migrate_use_xbzrle (void ) {MigrationState *s;
@@ -4200,6 +4225,10 @@ static Property migration_properties[] = { DEFINE_PROP_UINT8("multifd-zstd-level", MigrationState, parameters.multifd_zstd_level, DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL), +#ifdef CONFIG_LINUX +DEFINE_PROP_BOOL ("zero_copy" , MigrationState, + parameters.zero_copy, false), +#endifDEFINE_PROP_SIZE ("xbzrle-cache-size" , MigrationState, parameters.xbzrle_cache_size, DEFAULT_MIGRATE_XBZRLE_CACHE_SIZE),
@@ -4297,6 +4326,9 @@ static void migration_instance_init(Object *obj) params->has_multifd_compression = true; params->has_multifd_zlib_level = true; params->has_multifd_zstd_level = true; +#ifdef CONFIG_LINUX + params->has_zero_copy = true; +#endif params->has_xbzrle_cache_size = true; params->has_max_postcopy_bandwidth = true; params->has_max_cpu_throttle = true;
migration/socket.c +5 -0
@@ -77,6 +77,11 @@ static void socket_outgoing_migration(QIOTask *task, }else {trace_migration_socket_outgoing_connected (data->hostname); } + +if (migrate_use_zero_copy ()) { +error_setg (&err,"Zero copy not available in migration" ); + } +migration_channel_connect (data->s, sioc, data->hostname, err);object_unref (OBJECT (sioc)); }
monitor/hmp-cmds.c +6 -0
@@ -1297,6 +1297,12 @@ void hmp_migrate_set_parameter(Monitor *mon, const QDict *qdict) p->has_multifd_zstd_level = true;visit_type_uint8 (v, param, &p->multifd_zstd_level, &err);break ; +#ifdef CONFIG_LINUX +case MIGRATION_PARAMETER_ZERO_COPY : + p->has_zero_copy = true; +visit_type_bool (v, param, &p->zero_copy, &err); +break ; +#endif case MIGRATION_PARAMETER_XBZRLE_CACHE_SIZE : p->has_xbzrle_cache_size = true;if (!visit_type_size (v, param, &cache_size, &err)) {
Signature · 1 line
2.34.1
Re: [PATCH v7 1/5] QIOChannel: Add flags on io_writev and introduce io_flush callback
In reply to Leonardo BrasQuoted 25 lines · Leonardo Bras
io/channel.c +7
-6@@ -67,12 +67,13 @@ ssize_t qio_channel_readv_full(QIOChannel *ioc, } -ssize_t qio_channel_writev_full (QIOChannel *ioc, -const struct iovec *iov, -size_t niov, -int *fds, -size_t nfds, -Error **errp) +ssize_t qio_channel_writev_full_flags (QIOChannel *ioc, +const struct iovec *iov, +size_t niov, +int *fds, +size_t nfds, +int flags, +Error **errp) {QIOChannelClass *klass =QIO_CHANNEL_GET_CLASS (ioc);
io/channel.c ssize_t qio_channel_writev_full(QIOChannel *ioc, · line 85
Quoted 2 lines
io/channel.c +0
-0@@ -83,7 +84,7 @@ ssize_t qio_channel_writev_full(QIOChannel *ioc,return -1 ; }
Should we better also check QIO_CHANNEL_FEATURE_WRITE_ZERO_COPY here when QIO_CHANNEL_WRITE_FLAG_ZERO_COPY is set? Just like what we do with:
if ((fds || nfds) &&
!qio_channel_has_feature(ioc, QIO_CHANNEL_FEATURE_FD_PASS)) {
error_setg_errno(errp, EINVAL,
"Channel does not support file descriptor passing");
return -1;
}I still think it's better to have the caller be crystal clear when to use zero_copy feature because it has implication on buffer lifetime.
I might have commented similar things before, but I have missed a few versions so I could also have missed some previous discussions..
Quoted 4 lines
io/channel.c +1
-1-return klass->io_writev (ioc, iov, niov, fds, nfds, errp); +return klass->io_writev (ioc, iov, niov, fds, nfds, flags, errp); }
Signature · 1 line
Peter Xu
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
In reply to Leonardo BrasQuoted 16 lines · Leonardo Bras
io/channel-socket.c +7
-2@@ -558,15 +575,26 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,memcpy (CMSG_DATA (cmsg), fds, fdsize); } +if (flags &QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ) { + sflags =MSG_ZEROCOPY ; + } + retry: - ret =sendmsg (sioc->fd, &msg,0 ); + ret =sendmsg (sioc->fd, &msg, sflags);if (ret <=0 ) { -if (errno ==EAGAIN ) { +switch (errno) { +case EAGAIN :return QIO_CHANNEL_ERR_BLOCK ;
io/channel-socket.c static ssize_t qio_channel_socket_writev(QIOChannel *ioc, · line 595
Quoted 10 lines
io/channel-socket.c +7
-2- } -if (errno ==EINTR ) { +case EINTR : goto retry; +case ENOBUFS : +if (sflags &MSG_ZEROCOPY ) { +error_setg_errno (errp, errno, +"Process can't lock enough memory for using MSG_ZEROCOPY" ); +return -1 ; + }
I have no idea whether it'll make a real differnece, but - should we better add a "break" here? If you agree and with that fixed, feel free to add:
I also wonder whether you hit ENOBUFS in any of the environments. On Fedora here it's by default unlimited, but just curious when we should keep an eye.
Thanks,
Signature · 1 line
Peter Xu
Re: [PATCH v7 3/5] migration: Add zero-copy parameter for QMP/HMP for Linux
In reply to Leonardo BrasQuoted 8 lines · Leonardo Bras
Add property that allows zero-copy migration of memory pages, and also includes a helper function
migrate_use_zero_copy()to check if it's enabled.No code is introduced to actually do the migration, but it allow future implementations to enable/disable this feature.
On non-Linux builds this parameter is compiled-out.
I feel sad every time seeing a new parameter needs to be mostly duplicated 3 times in the code. :(
migration/socket.c static void socket_outgoing_migration(QIOTask *task, · line 83
Quoted 12 lines
migration/socket.c +4
-0@@ -77,6 +77,11 @@ static void socket_outgoing_migration(QIOTask *task, }else {trace_migration_socket_outgoing_connected (data->hostname); } + +if (migrate_use_zero_copy ()) { +error_setg (&err,"Zero copy not available in migration" ); + }
I got confused the 1st time looking at it.. I think this is not strongly needed, but that's okay:
Thanks,
Signature · 1 line
Peter Xu
Re: [PATCH v7 4/5] migration: Add migrate_use_tls() helper
In reply to Leonardo BrasQuoted 5 lines · Leonardo Bras
void
migration_channel_process_incoming(QIOChannel *ioc){migration/channel.c +0
-1-MigrationState *s =migrate_get_current ();Error *local_err =NULL ;
migration/channel.c line 44
Quoted 10 lines
migration/channel.c +3
-2trace_migration_set_incoming_channel ( ioc,object_get_typename (OBJECT (ioc))); -if (s->parameters.tls_creds && - *s->parameters.tls_creds && +if (migrate_use_tls () && !object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_TLS )) { +MigrationState *s =migrate_get_current (); +
Trivial nit: I'd rather keep the line there; as the movement offers nothing, imho..
Quoted 3 lines
migration_tls_channel_process_incoming(s, ioc, &local_err);
} else {
migration_ioc_register_yank(ioc);Signature · 1 line
Peter Xu
Re: [PATCH v7 5/5] multifd: Implement zero copy write in multifd migration (multifd-zero-copy)
In reply to Leonardo BrasQuoted 7 lines · Leonardo Bras
Implement zero copy on
nocomp_send_write(), by making use of QIOChannel writev + flags & flush interface.Change
multifd_send_sync_main()so it can distinguish each iteration sync from the setup and the completion, so aflush_zero_copy()can be called after each iteration in order to make sure all dirty pages are sent before a new iteration is started.
Leo - could you remind me (since I remembered we've discussed something similar) on why we can't simply do the sync() unconditionally for zero copy?
I remember why we need the sync(), but I can't remember what's the matter if we simply sync() too during setup and complete of migration.
Another trivial nit here:
migration/multifd.c void multifd_save_cleanup(void) · line 586
Quoted 2 lines
migration/multifd.c +1
-1-void multifd_send_sync_main (QEMUFile *f) +int multifd_send_sync_main (QEMUFile *f,bool sync)
I'd name it "bool full" or anything not called "sync", because the function already has a name that contains "sync", then it's werid to sync(sync==false).
The rest looks good to me. Thanks.
Signature · 1 line
Peter Xu
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
In reply to Peter XuQuoted 31 lines · Peter Xu
Quoted 16 lines · Leonardo Bras
io/channel-socket.c +7
-2@@ -558,15 +575,26 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,memcpy (CMSG_DATA (cmsg), fds, fdsize); } +if (flags &QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ) { + sflags =MSG_ZEROCOPY ; + } + retry: - ret =sendmsg (sioc->fd, &msg,0 ); + ret =sendmsg (sioc->fd, &msg, sflags);if (ret <=0 ) { -if (errno ==EAGAIN ) { +switch (errno) { +case EAGAIN :return QIO_CHANNEL_ERR_BLOCK ;io/channel-socket.c static ssize_t qio_channel_socket_writev(QIOChannel *ioc, · line 595
Quoted 10 lines
io/channel-socket.c +7
-2- } -if (errno ==EINTR ) { +case EINTR : goto retry; +case ENOBUFS : +if (sflags &MSG_ZEROCOPY ) { +error_setg_errno (errp, errno, +"Process can't lock enough memory for using MSG_ZEROCOPY" ); +return -1 ; + }I have no idea whether it'll make a real differnece, but - should we better add a "break" here? If you agree and with that fixed, feel free to add:
Quoted 4 lines
Reviewed-by Peter Xu <[email protected]>I also wonder whether you hit ENOBUFS in any of the environments. On Fedora here it's by default unlimited, but just curious when we should keep an eye.
Fedora doesn't allow unlimited locked memory by default
$ grep "locked memory" /proc/self/limits Max locked memory 65536 65536 bytes
And regardless of Fedora defaults, libvirt will set a limit for the guest. It will only be unlimited if requiring certain things like VFIO.
Regards,
Daniel
Signature · 3 lines
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
Quoted 37 lines · Daniel P. Berrangé
Quoted 31 lines · Peter Xu
Quoted 16 lines · Leonardo Bras
io/channel-socket.c +7
-2@@ -558,15 +575,26 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,memcpy (CMSG_DATA (cmsg), fds, fdsize); } +if (flags &QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ) { + sflags =MSG_ZEROCOPY ; + } + retry: - ret =sendmsg (sioc->fd, &msg,0 ); + ret =sendmsg (sioc->fd, &msg, sflags);if (ret <=0 ) { -if (errno ==EAGAIN ) { +switch (errno) { +case EAGAIN :return QIO_CHANNEL_ERR_BLOCK ;io/channel-socket.c static ssize_t qio_channel_socket_writev(QIOChannel *ioc, · line 595
Quoted 10 lines
io/channel-socket.c +7
-2- } -if (errno ==EINTR ) { +case EINTR : goto retry; +case ENOBUFS : +if (sflags &MSG_ZEROCOPY ) { +error_setg_errno (errp, errno, +"Process can't lock enough memory for using MSG_ZEROCOPY" ); +return -1 ; + }I have no idea whether it'll make a real differnece, but - should we better add a "break" here? If you agree and with that fixed, feel free to add:
Quoted 4 lines
Reviewed-by Peter Xu <[email protected]>I also wonder whether you hit ENOBUFS in any of the environments. On Fedora here it's by default unlimited, but just curious when we should keep an eye.
Quoted 8 lines
Fedora doesn't allow unlimited locked memory by default
$ grep "locked memory" /proc/self/limits Max locked memory 65536 65536 bytesAnd regardless of Fedora defaults, libvirt will set a limit for the guest. It will only be unlimited if requiring certain things like VFIO.
Thanks, I obviously checked up the wrong host..
Leo, do you know how much locked memory will be needed by zero copy? Will there be a limit? Is it linear to the number of sockets/channels?
It'll be better if we can fail at enabling the feature when we detected that the specified locked memory limit may not be suffice.
Signature · 1 line
Peter Xu
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
Quoted 43 lines · Peter Xu
Quoted 42 lines · Daniel P. Berrangé
Quoted 31 lines · Peter Xu
Quoted 16 lines · Leonardo Bras
io/channel-socket.c +7
-2@@ -558,15 +575,26 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,memcpy (CMSG_DATA (cmsg), fds, fdsize); } +if (flags &QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ) { + sflags =MSG_ZEROCOPY ; + } + retry: - ret =sendmsg (sioc->fd, &msg,0 ); + ret =sendmsg (sioc->fd, &msg, sflags);if (ret <=0 ) { -if (errno ==EAGAIN ) { +switch (errno) { +case EAGAIN :return QIO_CHANNEL_ERR_BLOCK ;io/channel-socket.c static ssize_t qio_channel_socket_writev(QIOChannel *ioc, · line 595
Quoted 10 lines
io/channel-socket.c +7
-2- } -if (errno ==EINTR ) { +case EINTR : goto retry; +case ENOBUFS : +if (sflags &MSG_ZEROCOPY ) { +error_setg_errno (errp, errno, +"Process can't lock enough memory for using MSG_ZEROCOPY" ); +return -1 ; + }I have no idea whether it'll make a real differnece, but - should we better add a "break" here? If you agree and with that fixed, feel free to add:
Quoted 4 lines
Reviewed-by Peter Xu <[email protected]>I also wonder whether you hit ENOBUFS in any of the environments. On Fedora here it's by default unlimited, but just curious when we should keep an eye.
Fedora doesn't allow unlimited locked memory by default
$ grep "locked memory" /proc/self/limits Max locked memory 65536 65536 bytes
Quoted 8 lines
Quoted 3 lines
And regardless of Fedora defaults, libvirt will set a limit for the guest. It will only be unlimited if requiring certain things like VFIO.
Thanks, I obviously checked up the wrong host..
Leo, do you know how much locked memory will be needed by zero copy? Will there be a limit? Is it linear to the number of sockets/channels?
IIRC we decided it would be limited by the socket send buffer size, rather than guest RAM, because writes will block once the send buffer is full.
This has a default global setting, with per-socket override. On one box I have it is 200 Kb. With multifd you'll need "num-sockets * send buffer".
Quoted 2 lines
It'll be better if we can fail at enabling the feature when we detected that the specified locked memory limit may not be suffice.
Checking this value against available locked memory though will always have an error margin because other things in QEMU can use locked memory too
Regards,
Daniel
Signature · 3 lines
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Re: [PATCH v7 1/5] QIOChannel: Add flags on io_writev and introduce io_flush callback
In reply to Leonardo BrasQuoted 96 lines · Leonardo Bras
Add flags to
io_writevand introduceio_flushas optional callback to QIOChannelClass, allowing the implementation of zero copy writes by subclasses.How to use them:
- Write data usingqio_channel_writev(...,QIO_CHANNEL_WRITE_FLAG_ZERO_COPY),
- Wait write completion withqio_channel_flush().Notes:
As some zero copy implementations work asynchronously, it's recommended to keep the write buffer untouched until the return ofqio_channel_flush(), to avoid the risk of sending an updated buffer instead of the buffer state during write.As
io_flushcallback is optional, if a subclass does not implement it, then:
-io_flushwill return 0 without changing anything.Also, some functions like
qio_channel_writev_full_all()were adapted to receive a flag parameter. That allows shared code between zero copy and non-zero copy writev, and also an easier implementation on new flags.Signed-off-by Leonardo Bras <[email protected]>--- include/io/channel.h | 67 +++++++++++++++++++++++++++++++++++--------- io/channel-buffer.c | 1 + io/channel-command.c | 1 + io/channel-file.c | 1 + io/channel-socket.c | 2 ++ io/channel-tls.c | 1 + io/channel-websock.c | 1 + io/channel.c | 51 +++++++++++++++++++++++---------- migration/rdma.c | 1 + 9 files changed, 98 insertions(+), 28 deletions(-)include/io/channel.h +8
-7@@ -32,12 +32,15 @@ OBJECT_DECLARE_TYPE(QIOChannel, QIOChannelClass,#define QIO_CHANNEL_ERR_BLOCK -2 +#define QIO_CHANNEL_WRITE_FLAG_ZERO_COPY 0x1 +typedef enum QIOChannelFeature QIOChannelFeature ;enum QIOChannelFeature {QIO_CHANNEL_FEATURE_FD_PASS ,QIO_CHANNEL_FEATURE_SHUTDOWN ,QIO_CHANNEL_FEATURE_LISTEN , +QIO_CHANNEL_FEATURE_WRITE_ZERO_COPY , };@@ -104,6 +107,7 @@ struct QIOChannelClass {size_t niov, int *fds, size_t nfds, + int flags, Error **errp); ssize_t (*io_readv)(QIOChannel *ioc, const struct iovec *iov,@@ -136,6 +140,8 @@ struct QIOChannelClass { IOHandler *io_read, IOHandler *io_write, void *opaque); +int (*io_flush)(QIOChannel *ioc, +Error **errp); };/* General I/O handling functions */ @@ -222,12 +228,13 @@ ssize_t qio_channel_readv_full(QIOChannel *ioc,/** -* qio_channel_writev_full: +* qio_channel_writev_full_flags: * @ioc: the channel object * @iov: the array of memory regions to write data from * @niov: the length of the @iov array * @fds: an array of file handles to send * @nfds: number of file handles in @fds +* @flags: write flags (QIO_CHANNEL_WRITE_FLAG_*) * @errp: pointer to a NULL-initialized error object * * Write data to the IO channel, reading it from the @@ -255,12 +262,16 @@ ssize_t qio_channel_readv_full(QIOChannel *ioc,* or QIO_CHANNEL_ERR_BLOCK if no data is can be sent * and the channel is non-blocking */ -ssize_t qio_channel_writev_full (QIOChannel *ioc, -const struct iovec *iov, -size_t niov, -int *fds, -size_t nfds, -Error **errp);
include/io/channel.h ssize_t qio_channel_readv_full(QIOChannel *ioc, · line 274
Quoted 10 lines
include/io/channel.h +10
-0+ssize_t qio_channel_writev_full_flags (QIOChannel *ioc, +const struct iovec *iov, +size_t niov, +int *fds, +size_t nfds, +int flags, +Error **errp); + +#define qio_channel_writev_full (ioc, iov, niov, fds, nfds, errp) \ + qio_channel_writev_full_flags(ioc, iov, niov, fds, nfds, 0, errp)
Don't introduce yet another API variant here. Just add flags to all the existing write APIs with "full" in their name. The word "full" in their name was intended to indicate that they are accepting all possible parameters, so it doesn't mean sense to add APIs which take even more possible parameters.
include/io/channel.h int qio_channel_readv_full_all(QIOChannel *ioc, · line 878
Quoted 7 lines
include/io/channel.h +7
-0+int qio_channel_writev_full_all_flags (QIOChannel *ioc, +const struct iovec *iov, +size_t niov, +int *fds,size_t nfds, +int flags,Error **errp); +#define qio_channel_writev_full_all (ioc, iov, niov, fds, nfds, errp) \ + qio_channel_writev_full_all_flags(ioc, iov, niov, fds, nfds, 0, errp)
Same note.
Quoted 8 lines
include/io/channel.h +8
-0+/** +* qio_channel_flush: +* @ioc: the channel object +* @errp: pointer to a NULL-initialized error object +* +* Will block until every packet queued with +* qio_channel_writev_full_flags() + QIO_CHANNEL_WRITE_FLAG_ZERO_COPY +* is sent, or return in case of any error.
include/io/channel.h int qio_channel_readv_full_all(QIOChannel *ioc, · line 897
Quoted 10 lines
include/io/channel.h +10
-0+* +* If not implemented, acts as a no-op, and returns 0. +* +* Returns -1 if any error is found, +* 1 if every send failed to use zero copy. +* 0 otherwise. +*/ + +int qio_channel_flush (QIOChannel *ioc, +Error **errp);
Regards,
Daniel
Signature · 3 lines
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
In reply to Daniel P. BerrangéQuoted 59 lines · Daniel P. Berrangé
Quoted 43 lines · Peter Xu
Quoted 42 lines · Daniel P. Berrangé
Quoted 31 lines · Peter Xu
Quoted 16 lines · Leonardo Bras
io/channel-socket.c +7
-2@@ -558,15 +575,26 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,memcpy (CMSG_DATA (cmsg), fds, fdsize); } +if (flags &QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ) { + sflags =MSG_ZEROCOPY ; + } + retry: - ret =sendmsg (sioc->fd, &msg,0 ); + ret =sendmsg (sioc->fd, &msg, sflags);if (ret <=0 ) { -if (errno ==EAGAIN ) { +switch (errno) { +case EAGAIN :return QIO_CHANNEL_ERR_BLOCK ;io/channel-socket.c static ssize_t qio_channel_socket_writev(QIOChannel *ioc, · line 595
Quoted 10 lines
io/channel-socket.c +7
-2- } -if (errno ==EINTR ) { +case EINTR : goto retry; +case ENOBUFS : +if (sflags &MSG_ZEROCOPY ) { +error_setg_errno (errp, errno, +"Process can't lock enough memory for using MSG_ZEROCOPY" ); +return -1 ; + }I have no idea whether it'll make a real differnece, but - should we better add a "break" here? If you agree and with that fixed, feel free to add:
Quoted 4 lines
Reviewed-by Peter Xu <[email protected]>I also wonder whether you hit ENOBUFS in any of the environments. On Fedora here it's by default unlimited, but just curious when we should keep an eye.
Fedora doesn't allow unlimited locked memory by default
$ grep "locked memory" /proc/self/limits Max locked memory 65536 65536 bytesQuoted 8 lines
Quoted 3 lines
And regardless of Fedora defaults, libvirt will set a limit for the guest. It will only be unlimited if requiring certain things like VFIO.
Thanks, I obviously checked up the wrong host..
Leo, do you know how much locked memory will be needed by zero copy? Will there be a limit? Is it linear to the number of sockets/channels?
IIRC we decided it would be limited by the socket send buffer size, rather than guest RAM, because writes will block once the send buffer is full.
This has a default global setting, with per-socket override. On one box I have it is 200 Kb. With multifd you'll need "num-sockets * send buffer".
Quoted 6 lines
Quoted 2 lines
It'll be better if we can fail at enabling the feature when we detected that the specified locked memory limit may not be suffice.
Checking this value against available locked memory though will always have an error margin because other things in QEMU can use locked memory too
We could always still allow false positive in this check, so we can fail if we have a solid clue to know we'll fail later (e.g. minimum locked_vm needed is already less than total). But no strong opinion; we could have this merged and see whether that's needed in real life. Thanks,
Signature · 1 line
Peter Xu
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
In reply to Leonardo BrasQuoted 29 lines · Leonardo Bras
For
CONFIG_LINUX, implement the new zero copy flag and the optional callbackio_flushon QIOChannelSocket, but enables it only whenMSG_ZEROCOPYfeature is available in the host kernel, which is checked onqio_channel_socket_connect_sync()
qio_channel_socket_flush()was implemented by counting how many timessendmsg(...,MSG_ZEROCOPY)was successfully called, and then reading the socket's error queue, in order to find how many of them finished sending. Flush will loop until those counters are the same, or until some error occurs.Notes on using
writev()withQIO_CHANNEL_WRITE_FLAG_ZERO_COPY:
1: Buffer
- AsMSG_ZEROCOPYtells the kernel to use the same user buffer to avoid copying, some caution is necessary to avoid overwriting any buffer before it's sent. If something like this happen, a newer version of the buffer may be sent instead.
- If this is a problem, it's recommended to callqio_channel_flush()before freeing or re-using the buffer.2: Locked memory
- When usingMSG_ZERCOCOPY, the buffer memory will be locked after queued, and unlocked after it's sent.
- Depending on the size of each buffer, and how often it's sent, it may require a larger amount of locked memory than usually available to non-root user.
- If the required amount of locked memory is not available,writev_zero_copywill return an error, which can abort an operation like migration,
- Because of this, when an user code wants to add zero copy as a feature, it requires a mechanism to disable it, so it can still be accessible to less privileged users.
Quoted 5 lines
Signed-off-by Leonardo Bras <[email protected]>--- include/io/channel-socket.h | 2 + io/channel-socket.c | 107 ++++++++++++++++++++++++++++++++++-- 2 files changed, 105 insertions(+), 4 deletions(-)
Regards,
Daniel
Signature · 3 lines
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Re: [PATCH v7 3/5] migration: Add zero-copy parameter for QMP/HMP for Linux
In reply to Leonardo BrasQuoted 9 lines · Leonardo Bras
Add property that allows zero-copy migration of memory pages, and also includes a helper function
migrate_use_zero_copy()to check if it's enabled.No code is introduced to actually do the migration, but it allow future implementations to enable/disable this feature.
On non-Linux builds this parameter is compiled-out.
Quoted 8 lines
Signed-off-by Leonardo Bras <[email protected]>--- qapi/migration.json | 24 ++++++++++++++++++++++++ migration/migration.h | 5 +++++ migration/migration.c | 32 ++++++++++++++++++++++++++++++++ migration/socket.c | 5 +++++ monitor/hmp-cmds.c | 6 ++++++ 5 files changed, 72 insertions(+)
Quoted 42 lines
qapi/migration.json +15
-0@@ -730,6 +730,13 @@ # will consume more CPU. # Defaults to 1. (Since 5.0) # +# @zero-copy: Controls behavior on sending memory pages on migration. +# When true, enables a zero-copy mechanism for sending memory +# pages, if host supports it. +# Requires that QEMU be permitted to use locked memory for guest +# RAM pages. +# Defaults to false. (Since 7.0) +# # @block-bitmap-mapping: Maps block nodes and bitmaps on them to # aliases for the purpose of dirty bitmap migration. Such # aliases may for example be the corresponding names on the@@ -769,6 +776,7 @@ 'xbzrle-cache-size', 'max-postcopy-bandwidth', 'max-cpu-throttle', 'multifd-compression', 'multifd-zlib-level' ,'multifd-zstd-level', + { 'name': 'zero-copy', 'if' : 'CONFIG_LINUX'}, 'block-bitmap-mapping' ] } ##@@ -895,6 +903,13 @@ # will consume more CPU. # Defaults to 1. (Since 5.0) # +# @zero-copy: Controls behavior on sending memory pages on migration. +# When true, enables a zero-copy mechanism for sending memory +# pages, if host supports it. +# Requires that QEMU be permitted to use locked memory for guest +# RAM pages. +# Defaults to false. (Since 7.0) +# # @block-bitmap-mapping: Maps block nodes and bitmaps on them to # aliases for the purpose of dirty bitmap migration. Such # aliases may for example be the corresponding names on the
qapi/migration.json line 968
Quoted 5 lines
qapi/migration.json +1
-0@@ -949,6 +964,7 @@ '*multifd-compression': 'MultiFDCompression', '*multifd-zlib-level': 'uint8', '*multifd-zstd-level': 'uint8', + '*zero-copy': { 'type': 'bool', 'if': 'CONFIG_LINUX' }, '*block-bitmap-mapping': [ 'BitmapMigrationNodeAlias' ] } }
The current zerocopy impl is for the send path.
Do you expect we might get zerocopy in the receive path later ?
If so then either call this 'send-zero-copy', or change it from a bool to an enum taking '["send", "recv", "both"]'.
I'd probably take the former and just rename it.
Regards,
Daniel
Signature · 3 lines
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Re: [PATCH v7 4/5] migration: Add migrate_use_tls() helper
In reply to Leonardo BrasQuoted 6 lines · Leonardo Bras
A lot of places check parameters.tls_creds in order to evaluate if TLS is in use, and sometimes call
migrate_get_current()just for that test.Add new helper function
migrate_use_tls()in order to simplify testing for TLS usage.
Quoted 8 lines
Signed-off-by Leonardo Bras <[email protected]>Reviewed-by Juan Quintela <[email protected]>--- migration/migration.h | 1 + migration/channel.c | 6 +++--- migration/migration.c | 9 +++++++++ migration/multifd.c | 5 +---- 4 files changed, 14 insertions(+), 7 deletions(-)
Regards,
Daniel
Signature · 3 lines
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Re: [PATCH v7 1/5] QIOChannel: Add flags on io_writev and introduce io_flush callback
In reply to Peter XuHello Peter,
Quoted 26 lines · Peter Xu
Quoted 24 lines · Leonardo Bras
io/channel.c +7
-6@@ -67,12 +67,13 @@ ssize_t qio_channel_readv_full(QIOChannel *ioc, } -ssize_t qio_channel_writev_full (QIOChannel *ioc, -const struct iovec *iov, -size_t niov, -int *fds, -size_t nfds, -Error **errp) +ssize_t qio_channel_writev_full_flags (QIOChannel *ioc, +const struct iovec *iov, +size_t niov, +int *fds, +size_t nfds, +int flags, +Error **errp) {QIOChannelClass *klass =QIO_CHANNEL_GET_CLASS (ioc);
Quoted 6 lines
io/channel.c ssize_t qio_channel_writev_full(QIOChannel *ioc, · line 85
Quoted 3 lines
io/channel.c +0
-0@@ -83,7 +84,7 @@ ssize_t qio_channel_writev_full(QIOChannel *ioc,return -1 ; }Should we better also check
QIO_CHANNEL_FEATURE_WRITE_ZERO_COPYhere whenQIO_CHANNEL_WRITE_FLAG_ZERO_COPYis set? Just like what we do with:
Yes, that's correct.
I will also test for fds + zerocopy_flag , which should also fail here.
Quoted 10 lines
if ((fds || nfds) && !qio_channel_has_feature(ioc, QIO_CHANNEL_FEATURE_FD_PASS)) { error_setg_errno(errp, EINVAL, "Channel does not support file descriptor passing"); return -1; }I still think it's better to have the caller be crystal clear when to use
zero_copyfeature because it has implication on buffer lifetime.
I don't disagree with that suggestion.
But the buffer lifetime limitation is something on the socket implementation, right?
There could be some synchronous zerocopy implementation that does not require flush, and thus
don't require the buffer to be treated any special. Or am I missing something?
Quoted 4 lines
I might have commented similar things before, but I have missed a few versions so I could also have missed some previous discussions..
That's all great suggestions Peter! Thanks for that!
Some of the previous suggestions may have been missed because a lot of code moved.
Sorry about that.
Best regards,
Leo
Re: [PATCH v7 1/5] QIOChannel: Add flags on io_writev and introduce io_flush callback
Quoted 50 lines · Leonardo Bras Soares Passos
Hello Peter,
Quoted 26 lines · Peter Xu
Quoted 24 lines · Leonardo Bras
io/channel.c +7
-6@@ -67,12 +67,13 @@ ssize_t qio_channel_readv_full(QIOChannel *ioc, } -ssize_t qio_channel_writev_full (QIOChannel *ioc, -const struct iovec *iov, -size_t niov, -int *fds, -size_t nfds, -Error **errp) +ssize_t qio_channel_writev_full_flags (QIOChannel *ioc, +const struct iovec *iov, +size_t niov, +int *fds, +size_t nfds, +int flags, +Error **errp) {QIOChannelClass *klass =QIO_CHANNEL_GET_CLASS (ioc);Quoted 6 lines
io/channel.c ssize_t qio_channel_writev_full(QIOChannel *ioc, · line 85
Quoted 3 lines
io/channel.c +0
-0@@ -83,7 +84,7 @@ ssize_t qio_channel_writev_full(QIOChannel *ioc,return -1 ; }Should we better also check
QIO_CHANNEL_FEATURE_WRITE_ZERO_COPYhere whenQIO_CHANNEL_WRITE_FLAG_ZERO_COPYis set? Just like what we do with:Yes, that's correct.
I will also test for fds +zerocopy_flag, which should also fail here.Quoted 10 lines
if ((fds || nfds) && !qio_channel_has_feature(ioc, QIO_CHANNEL_FEATURE_FD_PASS)) { error_setg_errno(errp, EINVAL, "Channel does not support file descriptor passing"); return -1; }I still think it's better to have the caller be crystal clear when to use
zero_copyfeature because it has implication on buffer lifetime.
Quoted 7 lines
I don't disagree with that suggestion.
But the buffer lifetime limitation is something on the socket implementation, right?
There could be some synchronous zerocopy implementation that does not require flush, and thus
don't require the buffer to be treated any special. Or am I missing something?
Currently the flush() is required for zerocopy and not required for all the existing non-zerocopy use cases, that's already an API difference so the caller needs to identify it anyway. Then I think it's simpler we expose all of it to the user.
Not to mention IIUC if we don't fail here, it will just fail later when the code will unconditionally convert the flags=ZEROCOPY into MSG_ZEROCOPY in your next patch:
if (flags & QIO_CHANNEL_WRITE_FLAG_ZERO_COPY) {
sflags = MSG_ZEROCOPY;
}So AFAIU it'll fail anyway, either here with the cap check I mentioned, or later in sendmsg().
IOW, I think it fails cleaner here, rather than reaching sendmsg().
Quoted 11 lines
Quoted 5 lines
I might have commented similar things before, but I have missed a few versions so I could also have missed some previous discussions..
That's all great suggestions Peter! Thanks for that!
Some of the previous suggestions may have been missed because a lot of code moved.
Sorry about that.
Not a problem at all, I just want to make sure my question still makes sense. :)
Signature · 1 line
Peter Xu
Re: [PATCH v7 1/5] QIOChannel: Add flags on io_writev and introduce io_flush callback
In reply to Daniel P. BerrangéHello Daniel,
Quoted 106 lines · Daniel P. Berrangé
Quoted 104 lines · Leonardo Bras
Add flags to
io_writevand introduceio_flushas optional callback to QIOChannelClass, allowing the implementation of zero copy writes by subclasses.How to use them:
- Write data usingqio_channel_writev(...,QIO_CHANNEL_WRITE_FLAG_ZERO_COPY),
- Wait write completion withqio_channel_flush().Notes:
As some zero copy implementations work asynchronously, it's recommended to keep the write buffer untouched until the return ofqio_channel_flush(), to avoid the risk of sending an updated buffer instead of the buffer state during write.As
io_flushcallback is optional, if a subclass does not implement it, then:
-io_flushwill return 0 without changing anything.Also, some functions like
qio_channel_writev_full_all()were adapted to receive a flag parameter. That allows shared code between zero copy and non-zero copy writev, and also an easier implementation on new flags.Signed-off-by Leonardo Bras <[email protected]>--- include/io/channel.h | 67 +++++++++++++++++++++++++++++++++++--------- io/channel-buffer.c | 1 + io/channel-command.c | 1 + io/channel-file.c | 1 + io/channel-socket.c | 2 ++ io/channel-tls.c | 1 + io/channel-websock.c | 1 + io/channel.c | 51 +++++++++++++++++++++++---------- migration/rdma.c | 1 + 9 files changed, 98 insertions(+), 28 deletions(-)include/io/channel.h +16
-7@@ -32,12 +32,15 @@ OBJECT_DECLARE_TYPE(QIOChannel, QIOChannelClass,#define QIO_CHANNEL_ERR_BLOCK -2 +#define QIO_CHANNEL_WRITE_FLAG_ZERO_COPY 0x1 +typedef enum QIOChannelFeature QIOChannelFeature ;enum QIOChannelFeature {QIO_CHANNEL_FEATURE_FD_PASS ,QIO_CHANNEL_FEATURE_SHUTDOWN ,QIO_CHANNEL_FEATURE_LISTEN , +QIO_CHANNEL_FEATURE_WRITE_ZERO_COPY , };@@ -104,6 +107,7 @@ struct QIOChannelClass {size_t niov, int *fds, size_t nfds, + int flags, Error **errp); ssize_t (*io_readv)(QIOChannel *ioc, const struct iovec *iov,@@ -136,6 +140,8 @@ struct QIOChannelClass { IOHandler *io_read, IOHandler *io_write, void *opaque); +int (*io_flush)(QIOChannel *ioc, +Error **errp); };/* General I/O handling functions */ @@ -222,12 +228,13 @@ ssize_t qio_channel_readv_full(QIOChannel *ioc,/** -* qio_channel_writev_full: +* qio_channel_writev_full_flags: * @ioc: the channel object * @iov: the array of memory regions to write data from * @niov: the length of the @iov array * @fds: an array of file handles to send * @nfds: number of file handles in @fds +* @flags: write flags (QIO_CHANNEL_WRITE_FLAG_*) * @errp: pointer to a NULL-initialized error object * * Write data to the IO channel, reading it from the @@ -255,12 +262,16 @@ ssize_t qio_channel_readv_full(QIOChannel *ioc,* or QIO_CHANNEL_ERR_BLOCK if no data is can be sent * and the channel is non-blocking */ -ssize_t qio_channel_writev_full (QIOChannel *ioc, -const struct iovec *iov, -size_t niov, -int *fds, -size_t nfds, -Error **errp); +ssize_t qio_channel_writev_full_flags (QIOChannel *ioc, +const struct iovec *iov, +size_t niov, +int *fds, +size_t nfds, +int flags, +Error **errp); +
Quoted 8 lines
include/io/channel.h ssize_t qio_channel_readv_full(QIOChannel *ioc, · line 274
Quoted 2 lines
include/io/channel.h +2
-0+#define qio_channel_writev_full (ioc, iov, niov, fds, nfds, errp) \ + qio_channel_writev_full_flags(ioc, iov, niov, fds, nfds, 0, errp)Don't introduce yet another API variant here. Just add flags to all the existing write APIs with "full" in their name. The word "full" in their name was intended to indicate that they are accepting all possible parameters, so it doesn't mean sense to add APIs which take even more possible parameters.
Oh, I was not aware of that. Thanks for letting me know!
Sure, I will do this change for v8.
Quoted 21 lines
include/io/channel.h int qio_channel_readv_full_all(QIOChannel *ioc, · line 878
Quoted 8 lines
include/io/channel.h +7
-0+int qio_channel_writev_full_all_flags (QIOChannel *ioc, +const struct iovec *iov, +size_t niov, +int *fds,size_t nfds, +int flags,Error **errp); +#define qio_channel_writev_full_all (ioc, iov, niov, fds, nfds, errp) \ + qio_channel_writev_full_all_flags(ioc, iov, niov, fds, nfds, 0, errp)Same note.
Quoted 10 lines
+/**
+ *qio_channel_flush:
+ * @ioc: the channel object
+ * @errp: pointer to a NULL-initialized error object
+ *
+ * Will block until every packet queued with
+ *qio_channel_writev_full_flags()+QIO_CHANNEL_WRITE_FLAG_ZERO_COPY
+ * is sent, or return in case of any error.
+ *
+ * If not implemented, acts as a no-op, and returns 0.
Quoted 10 lines
include/io/channel.h int qio_channel_readv_full_all(QIOChannel *ioc, · line 897
Quoted 8 lines
+ *
+ * Returns -1 if any error is found,
+ * 1 if every send failed to use zero copy.
+ * 0 otherwise.
+ */include/io/channel.h +3
-0+ +int qio_channel_flush (QIOChannel *ioc, +Error **errp);
Best regards,
Leo
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
In reply to Peter XuQuoted 44 lines · Peter Xu
Quoted 42 lines · Daniel P. Berrangé
Quoted 31 lines · Peter Xu
Quoted 16 lines · Leonardo Bras
io/channel-socket.c +7
-2@@ -558,15 +575,26 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,memcpy (CMSG_DATA (cmsg), fds, fdsize); } +if (flags &QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ) { + sflags =MSG_ZEROCOPY ; + } + retry: - ret =sendmsg (sioc->fd, &msg,0 ); + ret =sendmsg (sioc->fd, &msg, sflags);if (ret <=0 ) { -if (errno ==EAGAIN ) { +switch (errno) { +case EAGAIN :return QIO_CHANNEL_ERR_BLOCK ;io/channel-socket.c static ssize_t qio_channel_socket_writev(QIOChannel *ioc, · line 595
Quoted 10 lines
io/channel-socket.c +7
-2- } -if (errno ==EINTR ) { +case EINTR : goto retry; +case ENOBUFS : +if (sflags &MSG_ZEROCOPY ) { +error_setg_errno (errp, errno, +"Process can't lock enough memory for using MSG_ZEROCOPY" ); +return -1 ; + }I have no idea whether it'll make a real differnece, but - should we better add a "break" here? If you agree and with that fixed, feel free to add:
Quoted 4 lines
Reviewed-by Peter Xu <[email protected]>I also wonder whether you hit ENOBUFS in any of the environments. On Fedora here it's by default unlimited, but just curious when we should keep an eye.
Fedora doesn't allow unlimited locked memory by default
$ grep "locked memory" /proc/self/limits Max locked memory 65536 65536 bytes
Quoted 8 lines
Quoted 3 lines
And regardless of Fedora defaults, libvirt will set a limit for the guest. It will only be unlimited if requiring certain things like VFIO.
Thanks, I obviously checked up the wrong host..
Leo, do you know how much locked memory will be needed by zero copy? Will there be a limit? Is it linear to the number of sockets/channels?
It depends on the number of channels, of course, but there are influencing factors, like network bandwidth & usage, and cpu speed & usage, network queue size, VM pagesize and so on.
A simple exemple:
If the cpu is free/fast, but there are other applications using the network, we may enqueue a lot of stuff for sending, and end up needing a lot of locked memory.
I don't think it's easy to calculate a good reference value for locked memory here.
Quoted 3 lines
It'll be better if we can fail at enabling the feature when we detected that the specified locked memory limit may not be suffice.
I agree it's a good idea. But having this reference value calculated is not much simple, IIUC.
Quoted 4 lines
--
Peter Xu
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
In reply to Peter XuHello Peter,
Quoted 23 lines · Peter Xu
Quoted 21 lines · Leonardo Bras
io/channel-socket.c +9
-4@@ -558,15 +575,26 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,memcpy (CMSG_DATA (cmsg), fds, fdsize); } +if (flags &QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ) { + sflags =MSG_ZEROCOPY ; + } + retry: - ret =sendmsg (sioc->fd, &msg,0 ); + ret =sendmsg (sioc->fd, &msg, sflags);if (ret <=0 ) { -if (errno ==EAGAIN ) { +switch (errno) { +case EAGAIN :return QIO_CHANNEL_ERR_BLOCK ; - } -if (errno ==EINTR ) { +case EINTR : goto retry; +case ENOBUFS :
Quoted 8 lines
io/channel-socket.c static ssize_t qio_channel_socket_writev(QIOChannel *ioc, · line 595
Quoted 5 lines
io/channel-socket.c +5
-0+if (sflags &MSG_ZEROCOPY ) { +error_setg_errno (errp, errno, +"Process can't lock enough memory for using MSG_ZEROCOPY" ); +return -1 ; + }I have no idea whether it'll make a real differnece, but - should we better add a "break" here?
Here I followed the standard of the EAGAIN error, that's why I just returned -1.
IIUC A break here would cause the errp to be re-set to the default message, after the switch.
Another option would be to add a 'default' clause, and move the default error msg there, and return the -1
after the switch.
In the end I thought the current way was simpler, but it's no issue to change if you think the 'default' idea would be better.
Quoted 4 lines
If you agree and with that fixed, feel free to add:
Reviewed-by Peter Xu <[email protected]>
Thanks!
Quoted 2 lines
I also wonder whether you hit ENOBUFS in any of the environments. On Fedora here it's by default unlimited, but just curious when we should keep an eye.
It's unlimited if you run as root IIRC.
Quoted 6 lines
Thanks,
--
Peter Xu
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
In reply to Peter XuQuoted 64 lines · Peter Xu
Quoted 62 lines · Daniel P. Berrangé
Quoted 43 lines · Peter Xu
Quoted 42 lines · Daniel P. Berrangé
Quoted 31 lines · Peter Xu
Quoted 16 lines · Leonardo Bras
io/channel-socket.c +7
-2@@ -558,15 +575,26 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,memcpy (CMSG_DATA (cmsg), fds, fdsize); } +if (flags &QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ) { + sflags =MSG_ZEROCOPY ; + } + retry: - ret =sendmsg (sioc->fd, &msg,0 ); + ret =sendmsg (sioc->fd, &msg, sflags);if (ret <=0 ) { -if (errno ==EAGAIN ) { +switch (errno) { +case EAGAIN :return QIO_CHANNEL_ERR_BLOCK ;io/channel-socket.c static ssize_t qio_channel_socket_writev(QIOChannel *ioc, · line 595
Quoted 10 lines
io/channel-socket.c +7
-2- } -if (errno ==EINTR ) { +case EINTR : goto retry; +case ENOBUFS : +if (sflags &MSG_ZEROCOPY ) { +error_setg_errno (errp, errno, +"Process can't lock enough memory for using MSG_ZEROCOPY" ); +return -1 ; + }I have no idea whether it'll make a real differnece, but - should we better add a "break" here? If you agree and with that fixed, feel free to add:
Quoted 4 lines
Reviewed-by Peter Xu <[email protected]>I also wonder whether you hit ENOBUFS in any of the environments. On Fedora here it's by default unlimited, but just curious when we should keep an eye.
Fedora doesn't allow unlimited locked memory by default
$ grep "locked memory" /proc/self/limits Max locked memory 65536 65536 bytesQuoted 8 lines
Quoted 3 lines
And regardless of Fedora defaults, libvirt will set a limit for the guest. It will only be unlimited if requiring certain things like VFIO.
Thanks, I obviously checked up the wrong host..
Leo, do you know how much locked memory will be needed by zero copy? Will there be a limit? Is it linear to the number of sockets/channels?
IIRC we decided it would be limited by the socket send buffer size, rather than guest RAM, because writes will block once the send buffer is full.
This has a default global setting, with per-socket override. On one box I have it is 200 Kb. With multifd you'll need "num-sockets * send buffer".
Quoted 2 lines
It'll be better if we can fail at enabling the feature when we detected that the specified locked memory limit may not be suffice.
Quoted 8 lines
Quoted 3 lines
Checking this value against available locked memory though will always have an error margin because other things in QEMU can use locked memory too
We could always still allow false positive in this check, so we can fail if we have a solid clue to know we'll fail later (e.g. minimum
locked_vmneeded is already less than total). But no strong opinion; we could have this merged and see whether that's needed in real life. Thanks,
I agree, this is a good approach.
Leo
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
In reply to Daniel P. BerrangéHello Daniel,
Quoted 51 lines · Daniel P. Berrangé
Quoted 49 lines · Peter Xu
Quoted 37 lines · Daniel P. Berrangé
Quoted 31 lines · Peter Xu
Quoted 16 lines · Leonardo Bras
io/channel-socket.c +7
-2@@ -558,15 +575,26 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,memcpy (CMSG_DATA (cmsg), fds, fdsize); } +if (flags &QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ) { + sflags =MSG_ZEROCOPY ; + } + retry: - ret =sendmsg (sioc->fd, &msg,0 ); + ret =sendmsg (sioc->fd, &msg, sflags);if (ret <=0 ) { -if (errno ==EAGAIN ) { +switch (errno) { +case EAGAIN :return QIO_CHANNEL_ERR_BLOCK ;io/channel-socket.c static ssize_t qio_channel_socket_writev(QIOChannel *ioc, · line 595
Quoted 10 lines
io/channel-socket.c +7
-2- } -if (errno ==EINTR ) { +case EINTR : goto retry; +case ENOBUFS : +if (sflags &MSG_ZEROCOPY ) { +error_setg_errno (errp, errno, +"Process can't lock enough memory for using MSG_ZEROCOPY" ); +return -1 ; + }I have no idea whether it'll make a real differnece, but - should we better add a "break" here? If you agree and with that fixed, feel free to add:
Quoted 4 lines
Reviewed-by Peter Xu <[email protected]>I also wonder whether you hit ENOBUFS in any of the environments. On Fedora here it's by default unlimited, but just curious when we should keep an eye.
Quoted 8 lines
Fedora doesn't allow unlimited locked memory by default
$ grep "locked memory" /proc/self/limits Max locked memory 65536 65536 bytesAnd regardless of Fedora defaults, libvirt will set a limit for the guest. It will only be unlimited if requiring certain things like VFIO.
Thanks, I obviously checked up the wrong host..
Quoted 8 lines
Quoted 2 lines
Leo, do you know how much locked memory will be needed by zero copy? Will there be a limit? Is it linear to the number of sockets/channels?
IIRC we decided it would be limited by the socket send buffer size, rather than guest RAM, because writes will block once the send buffer is full.
This has a default global setting, with per-socket override. On one box I have it is 200 Kb. With multifd you'll need "num-sockets * send buffer".
Oh, I was not aware there is a send buffer size (or maybe I am unable to recall).
That sure makes things much easier.
Quoted 3 lines
Quoted 3 lines
It'll be better if we can fail at enabling the feature when we detected that the specified locked memory limit may not be suffice.
sure
Quoted 4 lines
Checking this value against available locked memory though will always have an error margin because other things in QEMU can use locked memory too
We can get the current limit (before zerocopy) as an error margin:req_lock_mem = num-sockets * send buffer + BASE_LOCKED
Where BASE_LOCKED is the current libvirt value, or so on.
What do you think?
Best regards,
Leo
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
In reply to Daniel P. BerrangéQuoted 31 lines · Daniel P. Berrangé
Quoted 29 lines · Leonardo Bras
For
CONFIG_LINUX, implement the new zero copy flag and the optional callbackio_flushon QIOChannelSocket, but enables it only whenMSG_ZEROCOPYfeature is available in the host kernel, which is checked onqio_channel_socket_connect_sync()
qio_channel_socket_flush()was implemented by counting how many timessendmsg(...,MSG_ZEROCOPY)was successfully called, and then reading the socket's error queue, in order to find how many of them finished sending. Flush will loop until those counters are the same, or until some error occurs.Notes on using
writev()withQIO_CHANNEL_WRITE_FLAG_ZERO_COPY:
1: Buffer
- AsMSG_ZEROCOPYtells the kernel to use the same user buffer to avoid copying, some caution is necessary to avoid overwriting any buffer before it's sent. If something like this happen, a newer version of the buffer may be sent instead.
- If this is a problem, it's recommended to callqio_channel_flush()before freeing or re-using the buffer.2: Locked memory
- When usingMSG_ZERCOCOPY, the buffer memory will be locked after queued, and unlocked after it's sent.
- Depending on the size of each buffer, and how often it's sent, it may require a larger amount of locked memory than usually available to non-root user.
- If the required amount of locked memory is not available,writev_zero_copywill return an error, which can abort an operation like migration,
- Because of this, when an user code wants to add zero copy as a feature, it requires a mechanism to disable it, so it can still be accessible to less privileged users.
Quoted 8 lines
Quoted 5 lines
Signed-off-by Leonardo Bras <[email protected]>--- include/io/channel-socket.h | 2 + io/channel-socket.c | 107 ++++++++++++++++++++++++++++++++++-- 2 files changed, 105 insertions(+), 4 deletions(-)Reviewed-by Daniel P. Berrangé <[email protected]>
Thanks!
Quoted 8 lines
Regards,
Daniel
--|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Re: [PATCH v7 3/5] migration: Add zero-copy parameter for QMP/HMP for Linux
In reply to Peter XuHello Peter,
Quoted 14 lines · Peter Xu
Quoted 8 lines · Leonardo Bras
Add property that allows zero-copy migration of memory pages, and also includes a helper function
migrate_use_zero_copy()to check if it's enabled.No code is introduced to actually do the migration, but it allow future implementations to enable/disable this feature.
On non-Linux builds this parameter is compiled-out.
I feel sad every time seeing a new parameter needs to be mostly duplicated 3 times in the code. :(
Quoted 15 lines
migration/socket.c static void socket_outgoing_migration(QIOTask *task, · line 83
Quoted 12 lines
migration/socket.c +4
-0@@ -77,6 +77,11 @@ static void socket_outgoing_migration(QIOTask *task, }else {trace_migration_socket_outgoing_connected (data->hostname); } + +if (migrate_use_zero_copy ()) { +error_setg (&err,"Zero copy not available in migration" ); + }I got confused the 1st time looking at it.. I think this is not strongly needed, but that's okay:
The idea is to avoid some future issues on testing migration while bisecting.
Quoted 2 lines
Reviewed-by Peter Xu <[email protected]>
Thanks Peter!
Quoted 6 lines
Thanks,
--
Peter Xu
Re: [PATCH v7 3/5] migration: Add zero-copy parameter for QMP/HMP for Linux
In reply to Daniel P. BerrangéHello Daniel,
Quoted 20 lines · Daniel P. Berrangé
Quoted 9 lines · Leonardo Bras
Add property that allows zero-copy migration of memory pages, and also includes a helper function
migrate_use_zero_copy()to check if it's enabled.No code is introduced to actually do the migration, but it allow future implementations to enable/disable this feature.
On non-Linux builds this parameter is compiled-out.
Quoted 8 lines
Signed-off-by Leonardo Bras <[email protected]>--- qapi/migration.json | 24 ++++++++++++++++++++++++ migration/migration.h | 5 +++++ migration/migration.c | 32 ++++++++++++++++++++++++++++++++ migration/socket.c | 5 +++++ monitor/hmp-cmds.c | 6 ++++++ 5 files changed, 72 insertions(+)
Quoted 1 line
Reviewed-by Daniel P. Berrangé <[email protected]>
Thanks!
Quoted 51 lines
Quoted 43 lines
qapi/migration.json +15
-0@@ -730,6 +730,13 @@ # will consume more CPU. # Defaults to 1. (Since 5.0) # +# @zero-copy: Controls behavior on sending memory pages on migration. +# When true, enables a zero-copy mechanism for sending memory +# pages, if host supports it. +# Requires that QEMU be permitted to use locked memory for guest +# RAM pages. +# Defaults to false. (Since 7.0) +# # @block-bitmap-mapping: Maps block nodes and bitmaps on them to # aliases for the purpose of dirty bitmap migration. Such # aliases may for example be the corresponding names on the@@ -769,6 +776,7 @@ 'xbzrle-cache-size', 'max-postcopy-bandwidth', 'max-cpu-throttle', 'multifd-compression', 'multifd-zlib-level' ,'multifd-zstd-level', + { 'name': 'zero-copy', 'if' : 'CONFIG_LINUX'}, 'block-bitmap-mapping' ] } ##@@ -895,6 +903,13 @@ # will consume more CPU. # Defaults to 1. (Since 5.0) # +# @zero-copy: Controls behavior on sending memory pages on migration. +# When true, enables a zero-copy mechanism for sending memory +# pages, if host supports it. +# Requires that QEMU be permitted to use locked memory for guest +# RAM pages. +# Defaults to false. (Since 7.0) +# # @block-bitmap-mapping: Maps block nodes and bitmaps on them to # aliases for the purpose of dirty bitmap migration. Such # aliases may for example be the corresponding names on theqapi/migration.json line 968
Quoted 5 lines
qapi/migration.json +1
-0@@ -949,6 +964,7 @@ '*multifd-compression': 'MultiFDCompression', '*multifd-zlib-level': 'uint8', '*multifd-zstd-level': 'uint8', + '*zero-copy': { 'type': 'bool', 'if': 'CONFIG_LINUX' }, '*block-bitmap-mapping': [ 'BitmapMigrationNodeAlias' ] } }The current zerocopy impl is for the send path.
Quoted 2 lines
Do you expect we might get zerocopy in the receive path later ?
It's possible, but I haven't started the implementation yet.
Quoted 6 lines
If so then either call this 'send-zero-copy', or change it from a bool to an enum taking '["send", "recv", "both"]'.
I'd probably take the former and just rename it.
Well, my rationale:
- I want to set zero copy sending:
zero-copy is set in the sending host, start migration.
- I want to set zero copy receiving:
zero-copy is set in the receiving host, wait for migration. (Of course host support is checked when setting the parameter).
The problem with the current approach is trying to enable zero-copy on receive before it's implemented, which will 'fail' silently . A possible solution would be to add a patch to check in the receiving path if zero-copy is enabled, and fail for now.
What do you think?
Best regards,
Leo
Quoted 8 lines
Regards,
Daniel
--|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Re: [PATCH v7 4/5] migration: Add migrate_use_tls() helper
In reply to Peter XuHello Peter,
Quoted 18 lines · Peter Xu
Quoted 5 lines · Leonardo Bras
void
migration_channel_process_incoming(QIOChannel *ioc){migration/channel.c +0
-1-MigrationState *s =migrate_get_current ();Error *local_err =NULL ;migration/channel.c line 44
Quoted 10 lines
migration/channel.c +3
-2trace_migration_set_incoming_channel ( ioc,object_get_typename (OBJECT (ioc))); -if (s->parameters.tls_creds && - *s->parameters.tls_creds && +if (migrate_use_tls () && !object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_TLS )) { +MigrationState *s =migrate_get_current (); +
Quoted 2 lines
Trivial nit: I'd rather keep the line there; as the movement offers nothing, imho..
The idea to move the 's' to inside the if block is to make it clear it's only used in this case.
But if you think it's better to keep it at the beginning of the function, sure, I can change that.
Just let me know.
Quoted 7 lines
Quoted 4 lines
migration_tls_channel_process_incoming(s, ioc, &local_err); } else { migration_ioc_register_yank(ioc);Reviewed-by Peter Xu <[email protected]>
Thanks!
Quoted 3 lines
--
Peter Xu
Best regards,
Leo
Re: [PATCH v7 4/5] migration: Add migrate_use_tls() helper
In reply to Daniel P. BerrangéHello Daniel,
Quoted 17 lines · Daniel P. Berrangé
Quoted 6 lines · Leonardo Bras
A lot of places check parameters.tls_creds in order to evaluate if TLS is in use, and sometimes call
migrate_get_current()just for that test.Add new helper function
migrate_use_tls()in order to simplify testing for TLS usage.Quoted 8 lines
Signed-off-by Leonardo Bras <[email protected]>Reviewed-by Juan Quintela <[email protected]>--- migration/migration.h | 1 + migration/channel.c | 6 +++--- migration/migration.c | 9 +++++++++ migration/multifd.c | 5 +---- 4 files changed, 14 insertions(+), 7 deletions(-)
Quoted 2 lines
Reviewed-by Daniel P. Berrangé <[email protected]>
Thanks!
Quoted 7 lines
Regards,
Daniel
--|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Best regards,
Leo
Re: [PATCH v7 3/5] migration: Add zero-copy parameter for QMP/HMP for Linux
In reply to Leonardo Bras Soares PassosQuoted 94 lines · Leonardo Bras Soares Passos
Hello Daniel,
Quoted 20 lines · Daniel P. Berrangé
Quoted 9 lines · Leonardo Bras
Add property that allows zero-copy migration of memory pages, and also includes a helper function
migrate_use_zero_copy()to check if it's enabled.No code is introduced to actually do the migration, but it allow future implementations to enable/disable this feature.
On non-Linux builds this parameter is compiled-out.
Quoted 8 lines
Signed-off-by Leonardo Bras <[email protected]>--- qapi/migration.json | 24 ++++++++++++++++++++++++ migration/migration.h | 5 +++++ migration/migration.c | 32 ++++++++++++++++++++++++++++++++ migration/socket.c | 5 +++++ monitor/hmp-cmds.c | 6 ++++++ 5 files changed, 72 insertions(+)Quoted 1 line
Reviewed-by Daniel P. Berrangé <[email protected]>Thanks!
Quoted 51 lines
Quoted 43 lines
qapi/migration.json +15
-0@@ -730,6 +730,13 @@ # will consume more CPU. # Defaults to 1. (Since 5.0) # +# @zero-copy: Controls behavior on sending memory pages on migration. +# When true, enables a zero-copy mechanism for sending memory +# pages, if host supports it. +# Requires that QEMU be permitted to use locked memory for guest +# RAM pages. +# Defaults to false. (Since 7.0) +# # @block-bitmap-mapping: Maps block nodes and bitmaps on them to # aliases for the purpose of dirty bitmap migration. Such # aliases may for example be the corresponding names on the@@ -769,6 +776,7 @@ 'xbzrle-cache-size', 'max-postcopy-bandwidth', 'max-cpu-throttle', 'multifd-compression', 'multifd-zlib-level' ,'multifd-zstd-level', + { 'name': 'zero-copy', 'if' : 'CONFIG_LINUX'}, 'block-bitmap-mapping' ] } ##@@ -895,6 +903,13 @@ # will consume more CPU. # Defaults to 1. (Since 5.0) # +# @zero-copy: Controls behavior on sending memory pages on migration. +# When true, enables a zero-copy mechanism for sending memory +# pages, if host supports it. +# Requires that QEMU be permitted to use locked memory for guest +# RAM pages. +# Defaults to false. (Since 7.0) +# # @block-bitmap-mapping: Maps block nodes and bitmaps on them to # aliases for the purpose of dirty bitmap migration. Such # aliases may for example be the corresponding names on theqapi/migration.json line 968
Quoted 5 lines
qapi/migration.json +1
-0@@ -949,6 +964,7 @@ '*multifd-compression': 'MultiFDCompression', '*multifd-zlib-level': 'uint8', '*multifd-zstd-level': 'uint8', + '*zero-copy': { 'type': 'bool', 'if': 'CONFIG_LINUX' }, '*block-bitmap-mapping': [ 'BitmapMigrationNodeAlias' ] } }The current zerocopy impl is for the send path.
Quoted 2 lines
Do you expect we might get zerocopy in the receive path later ?
It's possible, but I haven't started the implementation yet.
Quoted 6 lines
If so then either call this 'send-zero-copy', or change it from a bool to an enum taking '["send", "recv", "both"]'.
I'd probably take the former and just rename it.
Well, my rationale:
- I want to set zero copy sending:
zero-copy is set in the sending host, start migration.
Quoted 8 lines
- I want to set zero copy receiving:
zero-copy is set in the receiving host, wait for migration. (Of course host support is checked when setting the parameter).The problem with the current approach is trying to enable zero-copy on receive before it's implemented, which will 'fail' silently . A possible solution would be to add a patch to check in the receiving path if zero-copy is enabled, and fail for now.
That's not good because mgmt apps cannot query the QAPI schema to find out if this feature is supported or not.
If we wantt o support zero copy recv, then we need an explicit flag for it that is distinct from zero copy send, so that apps can introspect whether the feature is implemneted in QEMU or not.
Distinct named bool flags feels better, and also makes it clear to anyone not familiar with the impl that the current code is strictly send only.
Regards,
Daniel
Signature · 3 lines
|: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|
Re: [PATCH v7 5/5] multifd: Implement zero copy write in multifd migration (multifd-zero-copy)
In reply to Peter XuHello Peter,
Quoted 5 lines · Peter Xu
Quoted 3 lines · Leonardo Bras
Implement zero copy on
nocomp_send_write(), by making use of QIOChannel writev + flags & flush interface.
Quoted 7 lines
Quoted 4 lines
Change
multifd_send_sync_main()so it can distinguish each iteration sync from the setup and the completion, so aflush_zero_copy()can be called after each iteration in order to make sure all dirty pages are sent before a new iteration is started.Leo - could you remind me (since I remembered we've discussed something similar) on why we can't simply do the
sync()unconditionally for zero copy?
On previous implementations, it would get stuck on the setup, since it was waiting for any movement on the error queue before even starting the sending process.
At the time we would sync only at 'complete', so it would only need to run once. Running every iteration seemed a waste at the time.
Then, after some talk with Juan, it was decided to sync after each migration, so on 'complete' it was unnecessary.
But sure, now it would add just 2 syncs in the whole migration, and those should not even get to the syscall due to queued/sent counters.
Quoted 7 lines
I remember why we need the
sync(), but I can't remember what's the matter if we simplysync()too during setup and complete of migration.
Quoted 8 lines
Another trivial nit here:
migration/multifd.c void multifd_save_cleanup(void) · line 586
Quoted 2 lines
migration/multifd.c +1
-1-void multifd_send_sync_main (QEMUFile *f) +int multifd_send_sync_main (QEMUFile *f,bool sync)I'd name it "bool full" or anything not called "sync", because the function already has a name that contains "sync", then it's werid to sync(sync==false).
Yeah, I agree.
But if we will flush every time, then there is no need for such parameter :).
Quoted 2 lines
The rest looks good to me. Thanks.
Thanks!
Quoted 3 lines
--
Peter Xu
Best regards,
Leo
Re: [PATCH v7 1/5] QIOChannel: Add flags on io_writev and introduce io_flush callback
In reply to Peter XuQuoted 60 lines · Peter Xu
Quoted 50 lines · Leonardo Bras Soares Passos
Hello Peter,
Quoted 26 lines · Peter Xu
Quoted 24 lines · Leonardo Bras
io/channel.c +7
-6@@ -67,12 +67,13 @@ ssize_t qio_channel_readv_full(QIOChannel *ioc, } -ssize_t qio_channel_writev_full (QIOChannel *ioc, -const struct iovec *iov, -size_t niov, -int *fds, -size_t nfds, -Error **errp) +ssize_t qio_channel_writev_full_flags (QIOChannel *ioc, +const struct iovec *iov, +size_t niov, +int *fds, +size_t nfds, +int flags, +Error **errp) {QIOChannelClass *klass =QIO_CHANNEL_GET_CLASS (ioc);Quoted 6 lines
io/channel.c ssize_t qio_channel_writev_full(QIOChannel *ioc, · line 85
Quoted 3 lines
io/channel.c +0
-0@@ -83,7 +84,7 @@ ssize_t qio_channel_writev_full(QIOChannel *ioc,return -1 ; }Should we better also check
QIO_CHANNEL_FEATURE_WRITE_ZERO_COPYhere whenQIO_CHANNEL_WRITE_FLAG_ZERO_COPYis set? Just like what we do with:Yes, that's correct.
I will also test for fds +zerocopy_flag, which should also fail here.Quoted 10 lines
if ((fds || nfds) && !qio_channel_has_feature(ioc, QIO_CHANNEL_FEATURE_FD_PASS)) { error_setg_errno(errp, EINVAL, "Channel does not support file descriptor passing"); return -1; }I still think it's better to have the caller be crystal clear when to use
zero_copyfeature because it has implication on buffer lifetime.Quoted 7 lines
I don't disagree with that suggestion.
But the buffer lifetime limitation is something on the socket implementation, right?
There could be some synchronous zerocopy implementation that does not require flush, and thus
don't require the buffer to be treated any special. Or am I missing something?
Quoted 4 lines
Currently the
flush()is required for zerocopy and not required for all the existing non-zerocopy use cases, that's already an API difference so the caller needs to identify it anyway. Then I think it's simpler we expose all of it to the user.
Yeah, I agree.
Since one ZC implementation uses flush, all should use them. Even if it's a no-op.
It was just an observation that not all ZC implementations have buffer limitations, but I agree the user should expect them anyway, since they will exist in some implementations.
Quoted 9 lines
Not to mention IIUC if we don't fail here, it will just fail later when the code will unconditionally convert the flags=ZEROCOPY into
MSG_ZEROCOPYin your next patch:if (flags & QIO_CHANNEL_WRITE_FLAG_ZERO_COPY) { sflags = MSG_ZEROCOPY; }
Correct.
Quoted 4 lines
So AFAIU it'll fail anyway, either here with the cap check I mentioned, or later in
sendmsg().IOW, I think it fails cleaner here, rather than reaching
sendmsg().
I Agree.
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
I might have commented similar things before, but I have missed a few versions so I could also have missed some previous discussions..
Quoted 8 lines
Quoted 5 lines
That's all great suggestions Peter! Thanks for that!
Some of the previous suggestions may have been missed because a lot of code moved.
Sorry about that.Not a problem at all, I just want to make sure my question still makes sense. :)
Thanks for asking them!
Quoted 4 lines
--
Peter Xu
Best regards,
Leo
Re: [PATCH v7 3/5] migration: Add zero-copy parameter for QMP/HMP for Linux
In reply to Daniel P. BerrangéQuoted 108 lines · Daniel P. Berrangé
Quoted 94 lines · Leonardo Bras Soares Passos
Hello Daniel,
Quoted 20 lines · Daniel P. Berrangé
Quoted 9 lines · Leonardo Bras
Add property that allows zero-copy migration of memory pages, and also includes a helper function
migrate_use_zero_copy()to check if it's enabled.No code is introduced to actually do the migration, but it allow future implementations to enable/disable this feature.
On non-Linux builds this parameter is compiled-out.
Quoted 8 lines
Signed-off-by Leonardo Bras <[email protected]>--- qapi/migration.json | 24 ++++++++++++++++++++++++ migration/migration.h | 5 +++++ migration/migration.c | 32 ++++++++++++++++++++++++++++++++ migration/socket.c | 5 +++++ monitor/hmp-cmds.c | 6 ++++++ 5 files changed, 72 insertions(+)Quoted 1 line
Reviewed-by Daniel P. Berrangé <[email protected]>Thanks!
Quoted 51 lines
Quoted 43 lines
qapi/migration.json +15
-0@@ -730,6 +730,13 @@ # will consume more CPU. # Defaults to 1. (Since 5.0) # +# @zero-copy: Controls behavior on sending memory pages on migration. +# When true, enables a zero-copy mechanism for sending memory +# pages, if host supports it. +# Requires that QEMU be permitted to use locked memory for guest +# RAM pages. +# Defaults to false. (Since 7.0) +# # @block-bitmap-mapping: Maps block nodes and bitmaps on them to # aliases for the purpose of dirty bitmap migration. Such # aliases may for example be the corresponding names on the@@ -769,6 +776,7 @@ 'xbzrle-cache-size', 'max-postcopy-bandwidth', 'max-cpu-throttle', 'multifd-compression', 'multifd-zlib-level' ,'multifd-zstd-level', + { 'name': 'zero-copy', 'if' : 'CONFIG_LINUX'}, 'block-bitmap-mapping' ] } ##@@ -895,6 +903,13 @@ # will consume more CPU. # Defaults to 1. (Since 5.0) # +# @zero-copy: Controls behavior on sending memory pages on migration. +# When true, enables a zero-copy mechanism for sending memory +# pages, if host supports it. +# Requires that QEMU be permitted to use locked memory for guest +# RAM pages. +# Defaults to false. (Since 7.0) +# # @block-bitmap-mapping: Maps block nodes and bitmaps on them to # aliases for the purpose of dirty bitmap migration. Such # aliases may for example be the corresponding names on theqapi/migration.json line 968
Quoted 5 lines
qapi/migration.json +1
-0@@ -949,6 +964,7 @@ '*multifd-compression': 'MultiFDCompression', '*multifd-zlib-level': 'uint8', '*multifd-zstd-level': 'uint8', + '*zero-copy': { 'type': 'bool', 'if': 'CONFIG_LINUX' }, '*block-bitmap-mapping': [ 'BitmapMigrationNodeAlias' ] } }The current zerocopy impl is for the send path.
Quoted 2 lines
Do you expect we might get zerocopy in the receive path later ?
It's possible, but I haven't started the implementation yet.
Quoted 6 lines
If so then either call this 'send-zero-copy', or change it from a bool to an enum taking '["send", "recv", "both"]'.
I'd probably take the former and just rename it.
Well, my rationale:
- I want to set zero copy sending:
zero-copy is set in the sending host, start migration.Quoted 8 lines
- I want to set zero copy receiving:
zero-copy is set in the receiving host, wait for migration. (Of course host support is checked when setting the parameter).The problem with the current approach is trying to enable zero-copy on receive before it's implemented, which will 'fail' silently . A possible solution would be to add a patch to check in the receiving path if zero-copy is enabled, and fail for now.
That's not good because mgmt apps cannot query the QAPI schema to find out if this feature is supported or not.
Quoted 9 lines
If we wantt o support zero copy recv, then we need an explicit flag for it that is distinct from zero copy send, so that apps can introspect whether the feature is implemneted in QEMU or not.
Distinct named bool flags feels better, and also makes it clear to anyone not familiar with the impl that the current code is strictly send only.
Ok, I see the point.
I will try to refactor the code changing zero-copy to zero-copy-send or something like that.
Best regards,
Leo
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
In reply to Leonardo Bras Soares PassosQuoted 74 lines · Leonardo Bras Soares Passos
Hello Daniel,
Quoted 51 lines · Daniel P. Berrangé
Quoted 49 lines · Peter Xu
Quoted 37 lines · Daniel P. Berrangé
Quoted 31 lines · Peter Xu
Quoted 16 lines · Leonardo Bras
io/channel-socket.c +7
-2@@ -558,15 +575,26 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,memcpy (CMSG_DATA (cmsg), fds, fdsize); } +if (flags &QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ) { + sflags =MSG_ZEROCOPY ; + } + retry: - ret =sendmsg (sioc->fd, &msg,0 ); + ret =sendmsg (sioc->fd, &msg, sflags);if (ret <=0 ) { -if (errno ==EAGAIN ) { +switch (errno) { +case EAGAIN :return QIO_CHANNEL_ERR_BLOCK ;io/channel-socket.c static ssize_t qio_channel_socket_writev(QIOChannel *ioc, · line 595
Quoted 10 lines
io/channel-socket.c +7
-2- } -if (errno ==EINTR ) { +case EINTR : goto retry; +case ENOBUFS : +if (sflags &MSG_ZEROCOPY ) { +error_setg_errno (errp, errno, +"Process can't lock enough memory for using MSG_ZEROCOPY" ); +return -1 ; + }I have no idea whether it'll make a real differnece, but - should we better add a "break" here? If you agree and with that fixed, feel free to add:
Quoted 4 lines
Reviewed-by Peter Xu <[email protected]>I also wonder whether you hit ENOBUFS in any of the environments. On Fedora here it's by default unlimited, but just curious when we should keep an eye.
Quoted 8 lines
Fedora doesn't allow unlimited locked memory by default
$ grep "locked memory" /proc/self/limits Max locked memory 65536 65536 bytesAnd regardless of Fedora defaults, libvirt will set a limit for the guest. It will only be unlimited if requiring certain things like VFIO.
Thanks, I obviously checked up the wrong host..
Quoted 8 lines
Quoted 2 lines
Leo, do you know how much locked memory will be needed by zero copy? Will there be a limit? Is it linear to the number of sockets/channels?
IIRC we decided it would be limited by the socket send buffer size, rather than guest RAM, because writes will block once the send buffer is full.
This has a default global setting, with per-socket override. On one box I have it is 200 Kb. With multifd you'll need "num-sockets * send buffer".
Oh, I was not aware there is a send buffer size (or maybe I am unable to recall).
That sure makes things much easier.Quoted 3 lines
Quoted 3 lines
It'll be better if we can fail at enabling the feature when we detected that the specified locked memory limit may not be suffice.
sure
Quoted 1 line
Quoted 8 lines
Quoted 3 lines
Checking this value against available locked memory though will always have an error margin because other things in QEMU can use locked memory too
We can get the current limit (before zerocopy) as an error margin:
req_lock_mem= num-sockets * send buffer +BASE_LOCKEDWhere
BASE_LOCKEDis the current libvirt value, or so on.
Hmm.. not familiar with libvirt, so I'm curious whether libvirt is actually enlarging the allowed locked mem on Fedora since the default is 64KB?
I think it'll be great to capture the very major going-to-fail scenarios. For example, I'm wondering whether a qemu (without libvirt) will simply fail directly on Fedora using non-root even with 1 channel due to the 64K limit, or the other extreme case is when the user does not allow locking mem at all in some container environment (when we see max locked mem is zero).
It's not only about failing early, it's also about failing with a meaningful error so the user knows what to tune, while I'm not very sure that'll be easily understandable when we wait until the failure of io_writev().
Thanks,
Signature · 1 line
Peter Xu
Re: [PATCH v7 4/5] migration: Add migrate_use_tls() helper
In reply to Leonardo Bras Soares PassosQuoted 21 lines · Leonardo Bras Soares Passos
Hello Peter,
Quoted 18 lines · Peter Xu
Quoted 15 lines · Leonardo Bras
void
migration_channel_process_incoming(QIOChannel *ioc){migration/channel.c +3
-3-MigrationState *s =migrate_get_current ();Error *local_err =NULL ;trace_migration_set_incoming_channel ( ioc,object_get_typename (OBJECT (ioc))); -if (s->parameters.tls_creds && - *s->parameters.tls_creds && +if (migrate_use_tls () && !object_dynamic_cast (OBJECT (ioc),TYPE_QIO_CHANNEL_TLS )) { +MigrationState *s =migrate_get_current (); +
Quoted 5 lines
Quoted 2 lines
Trivial nit: I'd rather keep the line there; as the movement offers nothing, imho..
The idea to move the 's' to inside the if block is to make it clear it's only used in this case.
IMHO not necessary; I hardly read declarations for this, unless there's a bug, e.g. on variable shadowing. Moving it downwards makes it easier to happen. :)
Quoted 4 lines
But if you think it's better to keep it at the beginning of the function, sure, I can change that.
Just let me know.
Since there'll be a new version, that definitely looks nicer.
Thanks,
Signature · 1 line
Peter Xu
Re: [PATCH v7 2/5] QIOChannelSocket: Implement io_writev zero copy flag & io_flush for CONFIG_LINUX
In reply to Peter XuHello Peter,
Re-reading everything before submitting the next version. I think I finally got that you are suggesting to just add a break at the end of the case, after the if :)
Sorry I misunderstand that before,
Best regards,
Leo
Quoted 42 lines · Peter Xu
Quoted 16 lines · Leonardo Bras
io/channel-socket.c +7
-2@@ -558,15 +575,26 @@ static ssize_t qio_channel_socket_writev(QIOChannel *ioc,memcpy (CMSG_DATA (cmsg), fds, fdsize); } +if (flags &QIO_CHANNEL_WRITE_FLAG_ZERO_COPY ) { + sflags =MSG_ZEROCOPY ; + } + retry: - ret =sendmsg (sioc->fd, &msg,0 ); + ret =sendmsg (sioc->fd, &msg, sflags);if (ret <=0 ) { -if (errno ==EAGAIN ) { +switch (errno) { +case EAGAIN :return QIO_CHANNEL_ERR_BLOCK ;io/channel-socket.c static ssize_t qio_channel_socket_writev(QIOChannel *ioc, · line 595
Quoted 10 lines
io/channel-socket.c +7
-2- } -if (errno ==EINTR ) { +case EINTR : goto retry; +case ENOBUFS : +if (sflags &MSG_ZEROCOPY ) { +error_setg_errno (errp, errno, +"Process can't lock enough memory for using MSG_ZEROCOPY" ); +return -1 ; + }I have no idea whether it'll make a real differnece, but - should we better add a "break" here? If you agree and with that fixed, feel free to add:
Reviewed-by Peter Xu <[email protected]>I also wonder whether you hit ENOBUFS in any of the environments. On Fedora here it's by default unlimited, but just curious when we should keep an eye.
Thanks,
--
Peter Xu
Re: [PATCH v7 3/5] migration: Add zero-copy parameter for QMP/HMP for Linux
In reply to Leonardo Bras Soares PassosQuoted 26 lines · Leonardo Bras Soares Passos
Quoted 25 lines · Daniel P. Berrangé
Quoted 23 lines · Leonardo Bras Soares Passos
Hello Daniel,
Quoted 20 lines · Daniel P. Berrangé
Quoted 17 lines · Leonardo Bras
Add property that allows zero-copy migration of memory pages, and also includes a helper function
migrate_use_zero_copy()to check if it's enabled.No code is introduced to actually do the migration, but it allow future implementations to enable/disable this feature.
On non-Linux builds this parameter is compiled-out.
Signed-off-by Leonardo Bras <[email protected]>--- qapi/migration.json | 24 ++++++++++++++++++++++++ migration/migration.h | 5 +++++ migration/migration.c | 32 ++++++++++++++++++++++++++++++++ migration/socket.c | 5 +++++ monitor/hmp-cmds.c | 6 ++++++ 5 files changed, 72 insertions(+)
Quoted 8 lines
Quoted 4 lines
Quoted 3 lines
Quoted 1 line
Reviewed-by Daniel P. Berrangé <[email protected]>Thanks!
Ok, I see the point.
I will try to refactor the code changing zero-copy to zero-copy-send or something like that.
Hi
I am late to the party, but I agree with Dan that we need two flags.
Thre reason is that you can be the target of one migration, and later be the source of a next one. If we only have one flag that means different things on the source and destination side, things become really complicated.
Later, Juan.
Re: [PATCH v7 3/5] migration: Add zero-copy parameter for QMP/HMP for Linux
Hello Juan,
Thanks for thew feedback!
Quoted 39 lines · Juan Quintela
Quoted 26 lines · Leonardo Bras Soares Passos
Quoted 25 lines · Daniel P. Berrangé
Quoted 23 lines · Leonardo Bras Soares Passos
Hello Daniel,
Quoted 20 lines · Daniel P. Berrangé
Quoted 17 lines · Leonardo Bras
Add property that allows zero-copy migration of memory pages, and also includes a helper function
migrate_use_zero_copy()to check if it's enabled.No code is introduced to actually do the migration, but it allow future implementations to enable/disable this feature.
On non-Linux builds this parameter is compiled-out.
Signed-off-by Leonardo Bras <[email protected]>--- qapi/migration.json | 24 ++++++++++++++++++++++++ migration/migration.h | 5 +++++ migration/migration.c | 32 ++++++++++++++++++++++++++++++++ migration/socket.c | 5 +++++ monitor/hmp-cmds.c | 6 ++++++ 5 files changed, 72 insertions(+)Quoted 8 lines
Quoted 4 lines
Quoted 3 lines
Quoted 1 line
Reviewed-by Daniel P. Berrangé <[email protected]>Thanks!
Ok, I see the point.
I will try to refactor the code changing zero-copy to zero-copy-send or something like that.Hi
Quoted 9 lines
I am late to the party, but I agree with Dan that we need two flags.
Thre reason is that you can be the target of one migration, and later be the source of a next one. If we only have one flag that means different things on the source and destination side, things become really complicated.
Later, Juan.
Yeah, that makes sense. :)
Best regards,
Leo