leftover from 2020
v20210616 cover leftover from 2020
Various unreviewed changes, rebase to 4bcf6433ee.
Olaf Hering (36):
hotplug/Linux: fix starting of xenstored with restarting systemd
tools: add API to work with sevaral bits at once
xl: fix description of migrate --debug
tools: create libxensaverestore
MAINTAINERS: add myself as saverestore maintainer
tools: add readv_exact to libxenctrl
tools: add sr_is_known_page_type to libsaverestore
tools: use sr_is_known_page_type
tools: unify type checking for data pfns in migration stream
tools: show migration transfer rate in send_dirty_pages
tools: prepare to allocate saverestore arrays once
tools: save: move mfns array
tools: save: move types array
tools: save: move errors array
tools: save: move iov array
tools: save: move rec_pfns array
tools: save: move guest_data array
tools: save: move local_pages array
tools: restore: move types array
tools: restore: move mfns array
tools: restore: move map_errs array
tools: restore: move mfns array in populate_pfns
tools: restore: move pfns array in populate_pfns
tools: restore: split record processing
tools: restore: split handle_page_data
tools: restore: write data directly into guest
tools: recognize LIBXL_API_VERSION for 4.16
tools: adjust libxl_domain_suspend to receive a struct props
tools: change struct precopy_stats to precopy_stats_t
tools: add callback to libxl for precopy_policy and precopy_stats_t
tools: add --max_iters to libxl_domain_suspend
tools: add --min_remaining to libxl_domain_suspend
tools: add --abort_if_busy to libxl_domain_suspend
tools: add API for expandable bitmaps
tools: use sr_bitmap for populated_pfns
tools: use superpages during restore of HVM guest
.gitignore | 2 +
MAINTAINERS | 6 +
docs/man/xl.1.pod.in | 22 +-
tools/hotplug/Linux/init.d/xencommons.in | 2 +-
tools/hotplug/Linux/launch-xenstore.in | 40 +-
.../Linux/systemd/xenstored.service.in | 2 +-
tools/include/libxl.h | 32 +-
tools/include/xenguest.h | 186 -----
tools/include/xensaverestore.h | 207 ++++++
tools/libs/Makefile | 1 +
tools/libs/ctrl/xc_bitops.h | 28 +
tools/libs/ctrl/xc_private.c | 57 +-
tools/libs/ctrl/xc_private.h | 1 +
tools/libs/guest/Makefile | 11 -
tools/libs/guest/xg_dom_x86.c | 5 -
tools/libs/guest/xg_offline_page.c | 1 -
tools/libs/guest/xg_private.h | 5 +
tools/libs/guest/xg_sr_restore_x86_hvm.c | 274 --------
tools/libs/light/Makefile | 4 +-
tools/libs/light/libxl_dom_save.c | 24 +
tools/libs/light/libxl_domain.c | 10 +-
tools/libs/light/libxl_internal.h | 7 +
tools/libs/light/libxl_save_helper.c | 1 +
tools/libs/light/libxl_save_msgs_gen.pl | 5 +-
tools/libs/light/libxl_stream_write.c | 9 +-
tools/libs/light/libxl_types.idl | 1 +
tools/libs/saverestore/Makefile | 38 ++
.../xg_sr_common.c => saverestore/common.c} | 76 ++-
.../xg_sr_common.h => saverestore/common.h} | 253 ++++++-
.../common_x86.c} | 2 +-
.../common_x86.h} | 2 +-
.../common_x86_pv.c} | 2 +-
.../common_x86_pv.h} | 2 +-
.../nomigrate.c} | 0
.../xg_sr_restore.c => saverestore/restore.c} | 617 +++++++++--------
tools/libs/saverestore/restore_x86_hvm.c | 645 ++++++++++++++++++
.../restore_x86_pv.c} | 70 +-
.../xg_sr_save.c => saverestore/save.c} | 165 ++---
.../save_restore.h} | 2 -
.../save_x86_hvm.c} | 7 +-
.../save_x86_pv.c} | 33 +-
.../stream_format.h} | 0
tools/libs/uselibs.mk | 4 +-
tools/ocaml/libs/xl/xenlight_stubs.c | 3 +-
tools/xl/xl_cmdtable.c | 26 +-
tools/xl/xl_migrate.c | 54 +-
tools/xl/xl_saverestore.c | 3 +-
47 files changed, 2006 insertions(+), 941 deletions(-)
create mode 100644 tools/include/xensaverestore.h
delete mode 100644 tools/libs/guest/xg_sr_restore_x86_hvm.c
create mode 100644 tools/libs/saverestore/Makefile
rename tools/libs/{guest/xg_sr_common.c => saverestore/common.c} (72%)
rename tools/libs/{guest/xg_sr_common.h => saverestore/common.h} (68%)
rename tools/libs/{guest/xg_sr_common_x86.c => saverestore/common_x86.c} (99%)
rename tools/libs/{guest/xg_sr_common_x86.h => saverestore/common_x86.h} (98%)
rename tools/libs/{guest/xg_sr_common_x86_pv.c => saverestore/common_x86_pv.c} (99%)
rename tools/libs/{guest/xg_sr_common_x86_pv.h => saverestore/common_x86_pv.h} (98%)
rename tools/libs/{guest/xg_nomigrate.c => saverestore/nomigrate.c} (100%)
rename tools/libs/{guest/xg_sr_restore.c => saverestore/restore.c} (66%)
create mode 100644 tools/libs/saverestore/restore_x86_hvm.c
rename tools/libs/{guest/xg_sr_restore_x86_pv.c => saverestore/restore_x86_pv.c} (94%)
rename tools/libs/{guest/xg_sr_save.c => saverestore/save.c} (88%)
rename tools/libs/{guest/xg_save_restore.h => saverestore/save_restore.h} (98%)
rename tools/libs/{guest/xg_sr_save_x86_hvm.c => saverestore/save_x86_hvm.c} (96%)
rename tools/libs/{guest/xg_sr_save_x86_pv.c => saverestore/save_x86_pv.c} (97%)
rename tools/libs/{guest/xg_sr_stream_format.h => saverestore/stream_format.h} (100%)PATCH v20210616 1/36 hotplug/Linux: fix starting of xenstored with restarting systemd
A hard to trigger race with another unrelated systemd service and xenstored.service unveiled a bug in the way how xenstored is launched with systemd.
launch-xenstore may start either a daemon or a domain. In case a domain is used, systemd-notify was called. If another service triggered a restart of systemd while xenstored.service was executed, systemd may temporary lose track of services with Type=notify. As a result, xenstored.service would be marked as failed and units that depend on it will not be started. This breaks the enire Xen toolstack.
The chain of events is basically: xenstored.service sends the notification to systemd, this is a one-way event. Then systemd may be restarted by the other unit. During this time, xenstored.service is done and exits. Once systemd is done with its restart, it collects the pending notifications and childs. If it does not find the unit which sent the notification it will declare it as failed.
A workaround for this scenario is to leave the child processes running for a short time after sending the "READY=1" notification. If systemd happens to restart it will still find the unit it launched.
Adjust the callers of launch-xenstore to specifiy the init system:
Do not fork xenstored with systemd, preserve pid. This wil also avoid the need for a sleep because the process which sent the "READY=1" (the previously forked child) is still alive.
Remove the --pid-file in the systemd case because the pid of the child is known, and the file had probably little effect anyway due to lack of PidFile= and Type=forking in the unit file.
Be verbose about xenstored startup only with sysv to avoid interleaved output in systemd journal. Do the same also for domain case, even if is not strictly needed because init-xenstore-domain has no output.
The fix for upstream systemd which is supposed to fix it:575b300b795b6 ("pid1: rework how we dispatch SIGCHLD and other signals")
--
v04:
- do mkdir unconditionally because init-xenstore-domain writes the domid to xenstored.pid
v03:
- remove run_xenstored function, follow style of shell built-in test function v02:
- preserve Type=notify
--- tools/hotplug/Linux/init.d/xencommons.in | 2 +- tools/hotplug/Linux/launch-xenstore.in | 40 ++++++++++++++----- .../Linux/systemd/xenstored.service.in | 2 +- 3 files changed, 31 insertions(+), 13 deletions(-)
tools/hotplug/Linux/init.d/xencommons.in +1 -1
@@ -60,7 +60,7 @@ do_start () { mkdir -m700 -p ${XEN_LOCK_DIR} mkdir -p ${XEN_LOG_DIR} - @XEN_SCRIPT_DIR@/launch-xenstore || exit 1 + @XEN_SCRIPT_DIR@/launch-xenstore 'sysv' || exit 1 echo Setting domain 0 name, domid and JSON config... ${LIBEXEC_BIN}/xen-init-dom0 ${XEN_DOM0_UUID}
tools/hotplug/Linux/launch-xenstore.in +29 -11
@@ -15,6 +15,17 @@ # License along with this library; If not, see <http://www.gnu.org/licenses/>. # +initd=$1 + +case "$initd" in + sysv) nonl='-n' ;; + systemd) nonl= ;; + *) + echo "first argument must be 'sysv' or 'systemd'" + exit 1 + ;; +esac + XENSTORED=@XENSTORED@ . @XEN_SCRIPT_DIR@/hotplugpath.sh
@@ -44,14 +55,16 @@ timeout_xenstore () { return 0 } -test_xenstore && exit 0 +mkdir -p @XEN_RUN_DIR@ + +if test "$initd" = 'sysv' ; then + test_xenstore && exit 0 +fi test -f @CONFIG_DIR@/@CONFIG_LEAF_DIR@/xencommons && . @CONFIG_DIR@/@CONFIG_LEAF_DIR@/xencommons [ "$XENSTORETYPE" = "" ] && XENSTORETYPE=daemon -/bin/mkdir -p @XEN_RUN_DIR@ - [ "$XENSTORETYPE" = "daemon" ] && { [ -z "$XENSTORED_TRACE" ] || XENSTORED_ARGS="$XENSTORED_ARGS -T @XEN_LOG_DIR@/xenstored-trace.log" [ -z "$XENSTORED" ] && XENSTORED=@XENSTORED@
@@ -59,13 +72,15 @@ test -f @CONFIG_DIR@/@CONFIG_LEAF_DIR@/xencommons && . @CONFIG_DIR@/@CONFIG_LEAF echo "No xenstored found" exit 1 } + [ "$initd" = 'sysv' ] && { + echo $nonl Starting $XENSTORED... + $XENSTORED --pid-file @XEN_RUN_DIR@/xenstored.pid $XENSTORED_ARGS + timeout_xenstore $XENSTORED || exit 1 + exit 0 + } - echo -n Starting $XENSTORED... - $XENSTORED --pid-file @XEN_RUN_DIR@/xenstored.pid $XENSTORED_ARGS - - systemd-notify --booted 2>/dev/null || timeout_xenstore $XENSTORED || exit 1 - - exit 0 + exec $XENSTORED -N $XENSTORED_ARGS + exit 1 } [ "$XENSTORETYPE" = "domain" ] && {
@@ -75,9 +90,12 @@ test -f @CONFIG_DIR@/@CONFIG_LEAF_DIR@/xencommons && . @CONFIG_DIR@/@CONFIG_LEAF XENSTORE_DOMAIN_ARGS="$XENSTORE_DOMAIN_ARGS --memory $XENSTORE_DOMAIN_SIZE" [ -z "$XENSTORE_MAX_DOMAIN_SIZE" ] || XENSTORE_DOMAIN_ARGS="$XENSTORE_DOMAIN_ARGS --maxmem $XENSTORE_MAX_DOMAIN_SIZE" - echo -n Starting $XENSTORE_DOMAIN_KERNEL... + echo $nonl Starting $XENSTORE_DOMAIN_KERNEL... ${LIBEXEC_BIN}/init-xenstore-domain $XENSTORE_DOMAIN_ARGS || exit 1 - systemd-notify --ready 2>/dev/null + [ "$initd" = 'systemd' ] && { + systemd-notify --ready + sleep 9 + } exit 0 }
tools/hotplug/Linux/systemd/xenstored.service.in +1 -1
@@ -11,7 +11,7 @@ Type=notify NotifyAccess=all RemainAfterExit=true ExecStartPre=/bin/grep -q control_d /proc/xen/capabilities -ExecStart=@XEN_SCRIPT_DIR@/launch-xenstore +ExecStart=@XEN_SCRIPT_DIR@/launch-xenstore 'systemd' [Install] WantedBy=multi-user.target
PATCH v20210616 2/36 tools: add API to work with sevaral bits at once
In reply to Olaf HeringIntroduce new API to test if a fixed number of bits is clear or set, and clear or set them all at once.
The caller has to make sure the input bitnumber is a multiple of BITS_PER_LONG.
This API avoids the loop over each bit in a known range just to see if all of them are either clear or set.
v02:
- change return type from int to bool (jgross)
--- tools/libs/ctrl/xc_bitops.h | 28 ++++++++++++++++++++++++++++ 1 file changed, 28 insertions(+)
tools/libs/ctrl/xc_bitops.h +28 -0
@@ -3,6 +3,7 @@/* bitmap operations for single threaded access */ +#include <stdbool.h> #include <stdlib.h> #include <string.h>
@@ -77,4 +78,31 @@ static inline void bitmap_or(void *_dst, const void *_other, dst[i] |= other[i]; } +static inline bool test_bit_long_set (unsigned long nr_base,const void *_addr) +{ +const unsigned long *addr = _addr; +unsigned long val = addr[nr_base /BITS_PER_LONG ]; + +return val == ~0 ; +} + +static inline bool test_bit_long_clear (unsigned long nr_base,const void *_addr) +{ +const unsigned long *addr = _addr; +unsigned long val = addr[nr_base /BITS_PER_LONG ]; + +return val ==0 ; +} + +static inline void clear_bit_long (unsigned long nr_base,void *_addr) +{ +unsigned long *addr = _addr; + addr[nr_base /BITS_PER_LONG ] =0 ; +} + +static inline void set_bit_long (unsigned long nr_base,void *_addr) +{ +unsigned long *addr = _addr; + addr[nr_base /BITS_PER_LONG ] = ~0 ; +}#endif /* XC_BITOPS_H */
PATCH v20210616 3/36 xl: fix description of migrate --debug
In reply to Olaf Heringxl migrate --debug used to track every pfn in every batch of pages. But these times are gone. The code in xc_domain_save is the consumer of this knob, but it considers it only for the remus and colo case.
Adjust the help text to tell what --debug does today: Nothing.
v02:
- the option has no effect anymore
--- docs/man/xl.1.pod.in | 2 +- tools/xl/xl_cmdtable.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
docs/man/xl.1.pod.in +1 -1
@@ -481,7 +481,7 @@ domain. =item B<--debug> -Display huge (!) amount of debug information during the migration process. +This option has no effect. It is preserved for compatibility reasons. =item B<-p>
tools/xl/xl_cmdtable.c +1 -1
@@ -172,7 +172,7 @@ const struct cmd_spec cmd_table[] = {" migrate-receive [-d -e]\n" "-e Do not wait in the background (on <host>) for the death\n" " of the domain.\n" -"--debug Print huge (!) amount of debug during the migration process.\n" +"--debug Ignored.\n" "-p Do not unpause domain after migrating it.\n" "-D Preserve the domain id" },
PATCH v20210616 4/36 tools: create libxensaverestore
In reply to Olaf HeringMove all save/restore related code from libxenguest.so into a separate library libxensaverestore.so. The only consumer is libxl-save-helper. There is no need to have the moved code mapped all the time in binaries where libxenguest.so is used.
According to size(1) the change is:
text data bss dec hex filename
187183 4304 48 191535 2ec2f guest/libxenguest.so.4.15.0
124106 3376 48 127530 1f22a guest/libxenguest.so.4.15.0 67841 1872 8 69721 11059 saverestore/libxensaverestore.so.4.15.0
While touching the files anyway, take the opportunity to drop the redundant xg_sr_ filename prefix.
v5:
- fix spelling in description
v4:
- drop xg_ prefix from filenames (jgross)
- drop sr_ prefix from filenames (jbeulich)
v3:
- repost in time for 4.16
v2:
- copy also license header
- move xg_nomigrate.c
- add size(1) output to commit msg
- remove change from libxl_create.c
---
.gitignore | 2 +
tools/include/xenguest.h | 186 ----------------
tools/include/xensaverestore.h | 208 ++++++++++++++++++
tools/libs/Makefile | 1 +
tools/libs/guest/Makefile | 11 -
tools/libs/guest/xg_offline_page.c | 1 -
tools/libs/light/Makefile | 4 +-
tools/libs/light/libxl_internal.h | 1 +
tools/libs/light/libxl_save_helper.c | 1 +
tools/libs/light/libxl_save_msgs_gen.pl | 2 +-
tools/libs/saverestore/Makefile | 38 ++++
.../xg_sr_common.c => saverestore/common.c} | 2 +-
.../xg_sr_common.h => saverestore/common.h} | 16 +-
.../common_x86.c} | 2 +-
.../common_x86.h} | 2 +-
.../common_x86_pv.c} | 2 +-
.../common_x86_pv.h} | 2 +-
.../nomigrate.c} | 0
.../xg_sr_restore.c => saverestore/restore.c} | 2 +-
.../restore_x86_hvm.c} | 2 +-
.../restore_x86_pv.c} | 2 +-
.../xg_sr_save.c => saverestore/save.c} | 2 +-
.../save_restore.h} | 2 -
.../save_x86_hvm.c} | 2 +-
.../save_x86_pv.c} | 2 +-
.../stream_format.h} | 0
tools/libs/uselibs.mk | 4 +-
27 files changed, 282 insertions(+), 217 deletions(-)
create mode 100644 tools/include/xensaverestore.h
create mode 100644 tools/libs/saverestore/Makefile
rename tools/libs/{guest/xg_sr_common.c => saverestore/common.c} (99%)
rename tools/libs/{guest/xg_sr_common.h => saverestore/common.h} (98%)
rename tools/libs/{guest/xg_sr_common_x86.c => saverestore/common_x86.c} (99%)
rename tools/libs/{guest/xg_sr_common_x86.h => saverestore/common_x86.h} (98%)
rename tools/libs/{guest/xg_sr_common_x86_pv.c => saverestore/common_x86_pv.c} (99%)
rename tools/libs/{guest/xg_sr_common_x86_pv.h => saverestore/common_x86_pv.h} (98%)
rename tools/libs/{guest/xg_nomigrate.c => saverestore/nomigrate.c} (100%)
rename tools/libs/{guest/xg_sr_restore.c => saverestore/restore.c} (99%)
rename tools/libs/{guest/xg_sr_restore_x86_hvm.c => saverestore/restore_x86_hvm.c} (99%)
rename tools/libs/{guest/xg_sr_restore_x86_pv.c => saverestore/restore_x86_pv.c} (99%)
rename tools/libs/{guest/xg_sr_save.c => saverestore/save.c} (99%)
rename tools/libs/{guest/xg_save_restore.h => saverestore/save_restore.h} (98%)
rename tools/libs/{guest/xg_sr_save_x86_hvm.c => saverestore/save_x86_hvm.c} (99%)
rename tools/libs/{guest/xg_sr_save_x86_pv.c => saverestore/save_x86_pv.c} (99%)
rename tools/libs/{guest/xg_sr_stream_format.h => saverestore/stream_format.h} (100%).gitignore +2 -0
@@ -147,6 +147,8 @@ tools/libs/light/test_timedereg tools/libs/light/test_fdderegrace tools/libs/light/tmp.* tools/libs/light/xenlight.pc +tools/libs/saverestore/libxensaverestore.map +tools/libs/saverestore/xensaverestore.pc tools/libs/stat/_paths.h tools/libs/stat/headers.chk tools/libs/stat/libxenstat.map
tools/include/xenguest.h +0 -186
198 lines not on this page · raw message
tools/include/xensaverestore.h new file +208 -0
208 lines not on this page · raw message
tools/libs/Makefile +1 -0
@@ -12,6 +12,7 @@ SUBDIRS-y += devicemodel SUBDIRS-y += ctrl SUBDIRS-y += guest SUBDIRS-y += hypfs +SUBDIRS-y += saverestore SUBDIRS-y += store SUBDIRS-y += stat SUBDIRS-$(CONFIG_Linux) += vchan
tools/libs/guest/Makefile +0 -11
@@ -11,18 +11,7 @@ SRCS-y += xg_domain.c SRCS-y += xg_suspend.c SRCS-y += xg_resume.c ifeq ($(CONFIG_MIGRATE),y) -SRCS-y += xg_sr_common.c -SRCS-$(CONFIG_X86) += xg_sr_common_x86.c -SRCS-$(CONFIG_X86) += xg_sr_common_x86_pv.c -SRCS-$(CONFIG_X86) += xg_sr_restore_x86_pv.c -SRCS-$(CONFIG_X86) += xg_sr_restore_x86_hvm.c -SRCS-$(CONFIG_X86) += xg_sr_save_x86_pv.c -SRCS-$(CONFIG_X86) += xg_sr_save_x86_hvm.c -SRCS-y += xg_sr_restore.c -SRCS-y += xg_sr_save.c SRCS-y += xg_offline_page.c -else -SRCS-y += xg_nomigrate.c endif SRCS-y += xg_core.c SRCS-$(CONFIG_X86) += xg_core_x86.c
tools/libs/guest/xg_offline_page.c +0 -1
@@ -29,7 +29,6 @@#include "xc_private.h" #include "xg_private.h" -#include "xg_save_restore.h" struct pte_backup_entry {
tools/libs/light/Makefile +2 -2
@@ -179,7 +179,7 @@ $(ACPI_OBJS) $(ACPI_PIC_OBJS): CFLAGS += -I. -DLIBACPI_STDUTILS=\"$(CURDIR)/libx $(TEST_PROG_OBJS) _libxl.api-for-check: CFLAGS += $(CFLAGS_libxentoollog) $(CFLAGS_libxentoolcore) libxl_dom.o libxl_dom.opic: CFLAGS += -I$(XEN_ROOT)/tools # include libacpi/x86.h libxl_x86_acpi.o libxl_x86_acpi.opic: CFLAGS += -I$(XEN_ROOT)/tools -$(SAVE_HELPER_OBJS): CFLAGS += $(CFLAGS_libxenctrl) $(CFLAGS_libxenevtchn) $(CFLAGS_libxenguest) +$(SAVE_HELPER_OBJS): CFLAGS += $(CFLAGS_libxenctrl) $(CFLAGS_libxenevtchn) $(CFLAGS_libxensaverestore) testidl.o: CFLAGS += $(CFLAGS_libxenctrl) $(CFLAGS_libxenlight) testidl.c: libxl_types.idl gentest.py $(XEN_INCLUDE)/libxl.h $(AUTOINCS)
@@ -241,7 +241,7 @@ test_%: test_%.o test_common.o libxenlight_test.so $(CC) $(LDFLAGS) -o $@ $^ $(filter-out %libxenlight.so, $(LDLIBS_libxenlight)) $(LDLIBS_libxentoollog) $(LDLIBS_libxentoolcore) -lyajl $(APPEND_LDFLAGS) libxl-save-helper: $(SAVE_HELPER_OBJS) libxenlight.so - $(CC) $(LDFLAGS) -o $@ $(SAVE_HELPER_OBJS) $(LDLIBS_libxentoollog) $(LDLIBS_libxenctrl) $(LDLIBS_libxenguest) $(LDLIBS_libxentoolcore) $(APPEND_LDFLAGS) + $(CC) $(LDFLAGS) -o $@ $(SAVE_HELPER_OBJS) $(LDLIBS_libxentoollog) $(LDLIBS_libxenctrl) $(LDLIBS_libxensaverestore) $(LDLIBS_libxentoolcore) $(APPEND_LDFLAGS) testidl: testidl.o libxenlight.so $(CC) $(LDFLAGS) -o $@ testidl.o $(LDLIBS_libxenlight) $(LDLIBS_libxentoollog) $(LDLIBS_libxentoolcore) $(APPEND_LDFLAGS)
tools/libs/light/libxl_internal.h +1 -0
@@ -56,6 +56,7 @@#define XC_WANT_COMPAT_MAP_FOREIGN_API #include <xenctrl.h> #include <xenguest.h> +#include <xensaverestore.h> #include <xenhypfs.h> #include <xen-tools/libs.h>
tools/libs/light/libxl_save_helper.c +1 -0
@@ -48,6 +48,7 @@#include "xenctrl.h" #include "xenguest.h" +#include "xensaverestore.h" #include "_libxl_save_msgs_helper.h" /*----- logger -----*/
tools/libs/light/libxl_save_msgs_gen.pl +1 -1
@@ -72,7 +72,7 @@ END_BOTH END_CALLOUT #include <xenctrl.h> -#include <xenguest.h> +#include <xensaverestore.h> #include "_libxl_save_msgs_${ah}.h" END_HELPER
tools/libs/saverestore/Makefile new file +38 -0
@@ -0,0 +1,38 @@ +XEN_ROOT = $(CURDIR)/../../.. +include $(XEN_ROOT)/tools/Rules.mk + +ifeq ($(CONFIG_MIGRATE),y) +SRCS-y += common.c +SRCS-$(CONFIG_X86) += common_x86.c +SRCS-$(CONFIG_X86) += common_x86_pv.c +SRCS-$(CONFIG_X86) += restore_x86_pv.c +SRCS-$(CONFIG_X86) += restore_x86_hvm.c +SRCS-$(CONFIG_X86) += save_x86_pv.c +SRCS-$(CONFIG_X86) += save_x86_hvm.c +SRCS-y += restore.c +SRCS-y += save.c +else +SRCS-y += nomigrate.c +endif + +CFLAGS += -I$(XEN_libxenctrl) +CFLAGS += -I$(XEN_libxenguest) + +-include $(XEN_TARGET_ARCH)/Makefile + +CFLAGS += -Werror -Wmissing-prototypes +CFLAGS += -I. -I./include $(CFLAGS_xeninclude) +CFLAGS += -D__XEN_TOOLS__ +CFLAGS += -include $(XEN_ROOT)/tools/config.h +# Needed for asprintf() +CFLAGS-$(CONFIG_Linux) += -D_GNU_SOURCE + +LIBHEADER := xensaverestore.h + +NO_HEADERS_CHK := y + +include $(XEN_ROOT)/tools/libs/libs.mk + +.PHONY: cleanlocal +cleanlocal: + rm -f libxensaverestore.map
tools/libs/saverestore/common.c renamed from tools/libs/guest/xg_sr_common.c +1 -1
@@ -1,6 +1,6 @@#include <assert.h> -#include "xg_sr_common.h" +#include "common.h" #include <xen-tools/libs.h>
tools/libs/saverestore/common.h renamed from tools/libs/guest/xg_sr_common.h +14 -2
@@ -1,13 +1,25 @@#ifndef __COMMON__H#define __COMMON__H +#include <unistd.h> +#include <errno.h> #include <stdbool.h> +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <sys/mman.h> +#include <sys/types.h> +#include <sys/stat.h> + +#include "xc_private.h" +#include "xenguest.h" +#include "xensaverestore.h" #include "xg_private.h" -#include "xg_save_restore.h" +#include "save_restore.h" #include "xc_bitops.h" -#include "xg_sr_stream_format.h" +#include "stream_format.h" /* String representation of Domain Header types. */ const char *dhdr_type_to_str (uint32_t type);
tools/libs/saverestore/common_x86.c renamed from tools/libs/guest/xg_sr_common_x86.c +1 -1
@@ -1,4 +1,4 @@ -#include "xg_sr_common_x86.h" +#include "common_x86.h" int write_x86_tsc_info (struct xc_sr_context *ctx) {
tools/libs/saverestore/common_x86.h renamed from tools/libs/guest/xg_sr_common_x86.h +1 -1
@@ -1,7 +1,7 @@#ifndef __COMMON_X86__H#define __COMMON_X86__H -#include "xg_sr_common.h" +#include "common.h" /* * Obtains a domains TSC information from Xen and writes a X86_TSC_INFO record
tools/libs/saverestore/common_x86_pv.c renamed from tools/libs/guest/xg_sr_common_x86_pv.c +1 -1
@@ -1,6 +1,6 @@#include <assert.h> -#include "xg_sr_common_x86_pv.h" +#include "common_x86_pv.h" xen_pfn_t mfn_to_pfn (struct xc_sr_context *ctx,xen_pfn_t mfn) {
tools/libs/saverestore/common_x86_pv.h renamed from tools/libs/guest/xg_sr_common_x86_pv.h +1 -1
@@ -1,7 +1,7 @@#ifndef __COMMON_X86_PV_H#define __COMMON_X86_PV_H -#include "xg_sr_common_x86.h" +#include "common_x86.h" /* Virtual address ranges reserved for hypervisor. */ #define HYPERVISOR_VIRT_START_X86_64 0xFFFF800000000000ULL
tools/libs/saverestore/nomigrate.c renamed from tools/libs/guest/xg_nomigrate.c +0 -0
tools/libs/saverestore/restore.c renamed from tools/libs/guest/xg_sr_restore.c +1 -1
@@ -2,7 +2,7 @@#include <assert.h> -#include "xg_sr_common.h" +#include "common.h" /* * Read and validate the Image and Domain headers.
tools/libs/saverestore/restore_x86_hvm.c renamed from tools/libs/guest/xg_sr_restore_x86_hvm.c +1 -1
@@ -1,7 +1,7 @@#include <assert.h> #include <arpa/inet.h> -#include "xg_sr_common_x86.h" +#include "common_x86.h" /* * Process an HVM_CONTEXT record from the stream.
tools/libs/saverestore/restore_x86_pv.c renamed from tools/libs/guest/xg_sr_restore_x86_pv.c +1 -1
@@ -1,6 +1,6 @@#include <assert.h> -#include "xg_sr_common_x86_pv.h" +#include "common_x86_pv.h" static xen_pfn_t pfn_to_mfn (const struct xc_sr_context *ctx,xen_pfn_t pfn) {
tools/libs/saverestore/save.c renamed from tools/libs/guest/xg_sr_save.c +1 -1
@@ -1,7 +1,7 @@#include <assert.h> #include <arpa/inet.h> -#include "xg_sr_common.h" +#include "common.h" /* * Writes an Image header and Domain header into the stream.
tools/libs/saverestore/save_restore.h renamed from tools/libs/guest/xg_save_restore.h +0 -2
@@ -15,8 +15,6 @@* License along with this library; If not, see <http://www.gnu.org/licenses/>. */ -#include "xc_private.h" -#include <xen/foreign/x86_32.h> #include <xen/foreign/x86_64.h>
tools/libs/saverestore/save_x86_hvm.c renamed from tools/libs/guest/xg_sr_save_x86_hvm.c +1 -1
@@ -1,6 +1,6 @@#include <assert.h> -#include "xg_sr_common_x86.h" +#include "common_x86.h" #include <xen/hvm/params.h>
tools/libs/saverestore/save_x86_pv.c renamed from tools/libs/guest/xg_sr_save_x86_pv.c +1 -1
@@ -1,7 +1,7 @@#include <assert.h> #include <limits.h> -#include "xg_sr_common_x86_pv.h" +#include "common_x86_pv.h" /* Check a 64 bit virtual address for being canonical. */ static inline bool is_canonical_address (xen_vaddr_t vaddr)
tools/libs/saverestore/stream_format.h renamed from tools/libs/guest/xg_sr_stream_format.h +0 -0
tools/libs/uselibs.mk +3 -1
@@ -20,6 +20,8 @@ LIBS_LIBS += ctrl USELIBS_ctrl := toollog call evtchn gnttab foreignmemory devicemodel LIBS_LIBS += guest USELIBS_guest := evtchn ctrl +LIBS_LIBS += saverestore +USELIBS_saverestore := guest ctrl LIBS_LIBS += store USELIBS_store := toolcore LIBS_LIBS += vchan
@@ -27,7 +29,7 @@ USELIBS_vchan := toollog store gnttab evtchn LIBS_LIBS += stat USELIBS_stat := ctrl store LIBS_LIBS += light -USELIBS_light := toollog evtchn toolcore ctrl store hypfs guest +USELIBS_light := toollog evtchn toolcore ctrl store hypfs guest saverestore LIBS_LIBS += util USELIBS_util := light FILENAME_util := xlutil
Re: [PATCH v20210616 04/36] tools: create libxensaverestore
Am Wed, 16 Jun 2021 14:50:57 +0200
schrieb Olaf Hering <[email protected]>:
Quoted 2 lines
Move all save/restore related code from libxenguest.so into a separate library libxensaverestore.so.
This additional change is required to cover non-x86.
tools/libs/saverestore/nomigrate.c +1 -2
@@ -17,8 +17,7 @@#include <inttypes.h> #include <errno.h> -#include <xenctrl.h> -#include <xenguest.h> +#include <xensaverestore.h> int xc_domain_save(xc_interface *xch, int io_fd, uint32_t dom, uint32_t flags, struct save_callbacks *callbacks,
Olaf
PATCH v20210616 5/36 MAINTAINERS: add myself as saverestore maintainer
In reply to Olaf HeringI touched it last.
--- MAINTAINERS | 6 ++++++ 1 file changed, 6 insertions(+)
MAINTAINERS +6 -0
@@ -381,6 +381,12 @@ R: Juergen Gross <[email protected]> S: Supported F: tools/libs/ +LIBSAVERESTORE: +M: Olaf Hering <[email protected]> +S: Supported +F: tools/include/xensaverestore.h +F: tools/libs/saverestore/ + LIBXENLIGHT M: Ian Jackson <[email protected]> M: Wei Liu <[email protected]>
PATCH v20210616 6/36 tools: add readv_exact to libxenctrl
In reply to Olaf HeringRead a batch of iovec's.
Short reads are the common case, finish the trailing iov with read_exact.
v2:
- add comment to short-read handling
--- tools/libs/ctrl/xc_private.c | 57 +++++++++++++++++++++++++++++++++++- tools/libs/ctrl/xc_private.h | 1 + 2 files changed, 57 insertions(+), 1 deletion(-)
tools/libs/ctrl/xc_private.c +56 -1
@@ -659,8 +659,23 @@ int write_exact(int fd, const void *data, size_t size)#if defined(__MINIOS__)/* -* MiniOS's libc doesn't know about writev(). Implement it as multiple write()s. +* MiniOS's libc doesn't know about readv/writev(). +* Implement it as multiple read/write()s. */ +int readv_exact (int fd,const struct iovec *iov,int iovcnt) +{ +int rc, i; + +for ( i =0 ; i < iovcnt; ++i ) + { + rc =read_exact (fd, iov[i].iov_base, iov[i].iov_len); +if ( rc ) +return rc; + } + +return 0 ; +} +int writev_exact (int fd,const struct iovec *iov,int iovcnt) {int rc, i;
@@ -675,6 +690,46 @@ int writev_exact(int fd, const struct iovec *iov, int iovcnt)return 0 ; }#else +int readv_exact (int fd,const struct iovec *iov,int iovcnt) +{ +int rc =0 , idx =0 ; +ssize_t len; + +while ( idx < iovcnt ) + { + len =readv (fd, &iov[idx],min (iovcnt - idx,IOV_MAX )); +if ( len ==-1 && errno ==EINTR ) +continue ; +if ( len <=0 ) + { + rc =-1 ; + goto out; + } + +/* Finish a potential short read in the last iov */ +while ( len >0 && idx < iovcnt ) + { +if ( len >= iov[idx].iov_len ) + { + len -= iov[idx].iov_len; + } +else + { +void *p = iov[idx].iov_base + len; +size_t l = iov[idx].iov_len - len; + + rc =read_exact (fd, p, l); +if ( rc ) + goto out; + len =0 ; + } + idx++; + } + } +out: +return rc; +} +int writev_exact (int fd,const struct iovec *iov,int iovcnt) {struct iovec *local_iov =NULL ;
tools/libs/ctrl/xc_private.h +1 -0
@@ -410,6 +410,7 @@ int xc_flush_mmu_updates(xc_interface *xch, struct xc_mmu *mmu);/* Return 0 on success; -1 on error setting errno. */ int read_exact (int fd,void *data,size_t size);/* EOF => -1, errno=0 */ +int readv_exact (int fd,const struct iovec *iov,int iovcnt);int write_exact (int fd,const void *data,size_t size);int writev_exact (int fd,const struct iovec *iov,int iovcnt);
PATCH v20210616 7/36 tools: add sr_is_known_page_type to libsaverestore
In reply to Olaf HeringUsers of xc_get_pfn_type_batch may want to sanity check the data returned by Xen. Add a simple helper for this purpose.
v02:
- rename xc_is_known_page_type to sr_is_known_page_type
- move from ctrl/xc_private.h to saverestore/common.h (jgross)
--- tools/libs/saverestore/common.h | 33 +++++++++++++++++++++++++++++++++ 1 file changed, 33 insertions(+)
tools/libs/saverestore/common.h +33 -0
@@ -467,6 +467,39 @@ int populate_pfns(struct xc_sr_context *ctx, unsigned int count,/* Handle a STATIC_DATA_END record. */ int handle_static_data_end (struct xc_sr_context *ctx); +/* Sanitiy check for types returned by Xen */ +static inline bool sr_is_known_page_type (xen_pfn_t type) +{ +bool ret; + +switch (type) + { +case XEN_DOMCTL_PFINFO_NOTAB : + +case XEN_DOMCTL_PFINFO_L1TAB : +case XEN_DOMCTL_PFINFO_L1TAB |XEN_DOMCTL_PFINFO_LPINTAB : + +case XEN_DOMCTL_PFINFO_L2TAB : +case XEN_DOMCTL_PFINFO_L2TAB |XEN_DOMCTL_PFINFO_LPINTAB : + +case XEN_DOMCTL_PFINFO_L3TAB : +case XEN_DOMCTL_PFINFO_L3TAB |XEN_DOMCTL_PFINFO_LPINTAB : + +case XEN_DOMCTL_PFINFO_L4TAB : +case XEN_DOMCTL_PFINFO_L4TAB |XEN_DOMCTL_PFINFO_LPINTAB : + +case XEN_DOMCTL_PFINFO_XTAB : +case XEN_DOMCTL_PFINFO_XALLOC : +case XEN_DOMCTL_PFINFO_BROKEN : + ret = true; +break ; +default : + ret = false; +break ; + } +return ret; +} +#endif /* * Local variables:
PATCH v20210616 8/36 tools: use sr_is_known_page_type
In reply to Olaf HeringVerify pfn type on sending side, also verify incoming batch of pfns.
v02:
- use sr_is_known_page_type instead of xc_is_known_page_type
--- tools/libs/saverestore/restore.c | 3 +-- tools/libs/saverestore/save.c | 6 ++++++ 2 files changed, 7 insertions(+), 2 deletions(-)
tools/libs/saverestore/restore.c +1 -2
@@ -406,8 +406,7 @@ static int handle_page_data(struct xc_sr_context *ctx, struct xc_sr_record *rec) } type = (pages->pfn[i] &PAGE_DATA_TYPE_MASK ) >>32 ; -if ( ((type >>XEN_DOMCTL_PFINFO_LTAB_SHIFT ) >=5 ) && - ((type >>XEN_DOMCTL_PFINFO_LTAB_SHIFT ) <=8 ) ) +if (sr_is_known_page_type (type) == false ) {ERROR ("Invalid type %#" PRIx32" for pfn %#" PRIpfn" (index %u)" , type, pfn, i);
tools/libs/saverestore/save.c +6 -0
@@ -147,6 +147,12 @@ static int write_batch(struct xc_sr_context *ctx)for ( i =0 ; i < nr_pfns; ++i ) { +if (sr_is_known_page_type (types[i]) == false ) + { +ERROR ("Wrong type %#" PRIpfn" for pfn %#" PRIpfn, types[i], mfns[i]); + goto err; + } +switch ( types[i] ) {case XEN_DOMCTL_PFINFO_BROKEN :
PATCH v20210616 9/36 tools: unify type checking for data pfns in migration stream
In reply to Olaf HeringIntroduce a helper which decides if a given pfn type has data for the migration stream.
No change in behavior intended.
--- tools/libs/saverestore/common.h | 17 ++++++++++++++++ tools/libs/saverestore/restore.c | 34 +++++--------------------------- tools/libs/saverestore/save.c | 14 ++----------- 3 files changed, 24 insertions(+), 41 deletions(-)
tools/libs/saverestore/common.h +17 -0
@@ -500,6 +500,23 @@ static inline bool sr_is_known_page_type(xen_pfn_t type) return ret; } +static inline bool page_type_has_stream_data (uint32_t type) +{ +bool ret; + +switch (type) + { +case XEN_DOMCTL_PFINFO_XTAB : +case XEN_DOMCTL_PFINFO_XALLOC : +case XEN_DOMCTL_PFINFO_BROKEN : + ret = false; +break ; +default : + ret = true; +break ; + } +return ret; +}#endif /* * Local variables:
tools/libs/saverestore/restore.c +5 -29
@@ -152,9 +152,8 @@ int populate_pfns(struct xc_sr_context *ctx, unsigned int count,for ( i =0 ; i < count; ++i ) { -if ( (!types || (types && - (types[i] !=XEN_DOMCTL_PFINFO_XTAB && - types[i] !=XEN_DOMCTL_PFINFO_BROKEN ))) && +if ( (!types || + (types &&page_type_has_stream_data (types[i]) == true)) && !pfn_is_populated (ctx, original_pfns[i]) ) { rc =pfn_set_populated (ctx, original_pfns[i]);
@@ -233,25 +232,8 @@ static int process_page_data(struct xc_sr_context *ctx, unsigned int count, { ctx->restore.ops.set_page_type (ctx, pfns[i], types[i]); -switch ( types[i] ) - { -case XEN_DOMCTL_PFINFO_NOTAB : - -case XEN_DOMCTL_PFINFO_L1TAB : -case XEN_DOMCTL_PFINFO_L1TAB |XEN_DOMCTL_PFINFO_LPINTAB : - -case XEN_DOMCTL_PFINFO_L2TAB : -case XEN_DOMCTL_PFINFO_L2TAB |XEN_DOMCTL_PFINFO_LPINTAB : - -case XEN_DOMCTL_PFINFO_L3TAB : -case XEN_DOMCTL_PFINFO_L3TAB |XEN_DOMCTL_PFINFO_LPINTAB : - -case XEN_DOMCTL_PFINFO_L4TAB : -case XEN_DOMCTL_PFINFO_L4TAB |XEN_DOMCTL_PFINFO_LPINTAB : - +if (page_type_has_stream_data (types[i]) == true ) mfns[nr_pages++] = ctx->restore.ops.pfn_to_gfn (ctx, pfns[i]); -break ; - } }/* Nothing to do? */
@@ -271,14 +253,8 @@ static int process_page_data(struct xc_sr_context *ctx, unsigned int count,for ( i =0 , j =0 ; i < count; ++i ) { -switch ( types[i] ) - { -case XEN_DOMCTL_PFINFO_XTAB : -case XEN_DOMCTL_PFINFO_BROKEN : -case XEN_DOMCTL_PFINFO_XALLOC : -/* No page data to deal with. */ +if (page_type_has_stream_data (types[i]) == false )continue ; - }if ( map_errs[j] ) {
@@ -413,7 +389,7 @@ static int handle_page_data(struct xc_sr_context *ctx, struct xc_sr_record *rec) goto err; } -if ( type <XEN_DOMCTL_PFINFO_BROKEN ) +if (page_type_has_stream_data (type) == true )/* NOTAB and all L1 through L4 tables (including pinned) should * have a page worth of data in the record. */ pages_of_data++;
tools/libs/saverestore/save.c +2 -12
@@ -153,13 +153,8 @@ static int write_batch(struct xc_sr_context *ctx) goto err; } -switch ( types[i] ) - { -case XEN_DOMCTL_PFINFO_BROKEN : -case XEN_DOMCTL_PFINFO_XALLOC : -case XEN_DOMCTL_PFINFO_XTAB : +if (page_type_has_stream_data (types[i]) == false )continue ; - } mfns[nr_pages++] = mfns[i]; }
@@ -177,13 +172,8 @@ static int write_batch(struct xc_sr_context *ctx)for ( i =0 , p =0 ; i < nr_pfns; ++i ) { -switch ( types[i] ) - { -case XEN_DOMCTL_PFINFO_BROKEN : -case XEN_DOMCTL_PFINFO_XALLOC : -case XEN_DOMCTL_PFINFO_XTAB : +if (page_type_has_stream_data (types[i]) == false )continue ; - }if ( errors[p] ) {
PATCH v20210616 10/36 tools: show migration transfer rate in send_dirty_pages
In reply to Olaf HeringShow how fast domU pages are transferred in each iteration.
The relevant data is how fast the pfns travel, not so much how much protocol overhead exists. So the reported MiB/sec is just for pfns.
v02:
- rearrange MiB_sec calculation (jgross)
--- tools/libs/saverestore/common.h | 2 ++ tools/libs/saverestore/save.c | 46 +++++++++++++++++++++++++++++++++ 2 files changed, 48 insertions(+)
tools/libs/saverestore/common.h +2 -0
@@ -250,6 +250,8 @@ struct xc_sr_contextbool debug;unsigned long p2m_size; +size_t pages_sent; +size_t overhead_sent;struct precopy_stats stats;
tools/libs/saverestore/save.c +46 -0
78 lines not on this page · raw message
PATCH v20210616 11/36 tools: prepare to allocate saverestore arrays once
In reply to Olaf HeringThe hotpath 'send_dirty_pages' is supposed to do just one thing: sending. The other end 'handle_page_data' is supposed to do just receiving.
But instead both do other costly work like memory allocations and data moving. Do the allocations once, the array sizes are a compiletime constant. Avoid unneeded copying of data by receiving data directly into mapped guest memory.
This patch is just prepartion, subsequent changes will populate the arrays.
Once all changes are applied, migration of a busy HVM domU changes like that:
Without this series, from sr650 to sr950 (xen-4.15.20201027T173911.16a20963b3 xen_testing):
2020-10-29 10:23:10.711+0000: xc: show_transfer_rate: 23663128 bytes + 2879563 pages in 55.324905335 sec, 203 MiB/sec: Internal error 2020-10-29 10:23:35.115+0000: xc: show_transfer_rate: 16829632 bytes + 2097552 pages in 24.401179720 sec, 335 MiB/sec: Internal error 2020-10-29 10:23:59.436+0000: xc: show_transfer_rate: 16829032 bytes + 2097478 pages in 24.319025928 sec, 336 MiB/sec: Internal error 2020-10-29 10:24:23.844+0000: xc: show_transfer_rate: 16829024 bytes + 2097477 pages in 24.406992500 sec, 335 MiB/sec: Internal error 2020-10-29 10:24:48.292+0000: xc: show_transfer_rate: 16828912 bytes + 2097463 pages in 24.446489027 sec, 335 MiB/sec: Internal error 2020-10-29 10:25:01.816+0000: xc: show_transfer_rate: 16836080 bytes + 2098356 pages in 13.447091818 sec, 609 MiB/sec: Internal error
With this series, from sr650 to sr950 (xen-4.15.20201027T173911.16a20963b3 xen_unstable):
2020-10-28 21:26:05.074+0000: xc: show_transfer_rate: 23663128 bytes + 2879563 pages in 52.564054368 sec, 213 MiB/sec: Internal error 2020-10-28 21:26:23.527+0000: xc: show_transfer_rate: 16830040 bytes + 2097603 pages in 18.450592015 sec, 444 MiB/sec: Internal error 2020-10-28 21:26:41.926+0000: xc: show_transfer_rate: 16830944 bytes + 2097717 pages in 18.397862306 sec, 445 MiB/sec: Internal error 2020-10-28 21:27:00.339+0000: xc: show_transfer_rate: 16829176 bytes + 2097498 pages in 18.411973339 sec, 445 MiB/sec: Internal error 2020-10-28 21:27:18.643+0000: xc: show_transfer_rate: 16828592 bytes + 2097425 pages in 18.303326695 sec, 447 MiB/sec: Internal error 2020-10-28 21:27:26.289+0000: xc: show_transfer_rate: 16835952 bytes + 2098342 pages in 7.579846749 sec, 1081 MiB/sec: Internal error
Note: the performance improvement depends on the used network cards, wirespeed and the host:
- No improvement is expected with a 1G link.
- Improvement can be seen as shown above on a 10G link.
- Just a slight improvment can be seen on a 100G link.
This change also populates sr_save_arrays with "batch_pfns", and sr_restore_arrays with "pfns" to make sure malloc is always called with a non-zero value.
v02:
- rename xc_sr_save_arrays to sr_save_arrays
- rename xc_sr_restore_arrays to sr_restore_arrays
- merge handling of "batch_pfns" and "pfns" to make sure malloc is called with a non-zero size value (jgross)
--- tools/libs/saverestore/common.h | 12 +++++++++++- tools/libs/saverestore/restore.c | 14 ++++++++++---- tools/libs/saverestore/save.c | 27 +++++++++++++-------------- 3 files changed, 34 insertions(+), 19 deletions(-)
tools/libs/saverestore/common.h +11 -1
@@ -223,6 +223,15 @@ static inline int update_blob(struct xc_sr_blob *blob,return 0 ; } +struct sr_save_arrays { +xen_pfn_t batch_pfns[MAX_BATCH_SIZE ]; +}; + +struct sr_restore_arrays { +/* handle_page_data */ +xen_pfn_t pfns[MAX_BATCH_SIZE ]; +}; +struct xc_sr_context {xc_interface *xch;
@@ -255,11 +264,11 @@ struct xc_sr_contextstruct precopy_stats stats; -xen_pfn_t *batch_pfns;unsigned int nr_batch_pfns;unsigned long *deferred_pages;unsigned long nr_deferred_pages;xc_hypercall_buffer_t dirty_bitmap_hbuf; +struct sr_save_arrays *m; } save; struct /* Restore data. */
@@ -311,6 +320,7 @@ struct xc_sr_context/* Sender has invoked verify mode on the stream. */ bool verify; +struct sr_restore_arrays *m; } restore; };
tools/libs/saverestore/restore.c +10 -4
@@ -315,7 +315,7 @@ static int handle_page_data(struct xc_sr_context *ctx, struct xc_sr_record *rec)unsigned int i, pages_of_data =0 ;int rc =-1 ; -xen_pfn_t *pfns =NULL , pfn; +xen_pfn_t *pfns = ctx->restore.m->pfns, pfn;uint32_t *types =NULL , type;/*
@@ -363,9 +363,8 @@ static int handle_page_data(struct xc_sr_context *ctx, struct xc_sr_record *rec) goto err; } - pfns =malloc (pages->count *sizeof (*pfns)); types =malloc (pages->count *sizeof (*types)); -if ( !pfns || !types ) +if ( !types ) {ERROR ("Unable to allocate enough memory for %u pfns" , pages->count);
@@ -412,7 +411,6 @@ static int handle_page_data(struct xc_sr_context *ctx, struct xc_sr_record *rec) &pages->pfn[pages->count]); err:free (types); -free (pfns);return rc; }
@@ -739,6 +737,13 @@ static int setup(struct xc_sr_context *ctx) } ctx->restore.allocated_rec_num =DEFAULT_BUF_RECORDS ; + ctx->restore.m =malloc (sizeof (*ctx->restore.m)); +if ( !ctx->restore.m ) { +ERROR ("Unable to allocate memory for arrays" ); + rc =-1 ; + goto err; + } + err:return rc; }
@@ -757,6 +762,7 @@ static void cleanup(struct xc_sr_context *ctx)xc_hypercall_buffer_free_pages ( xch, dirty_bitmap,NRPAGES (bitmap_size (ctx->restore.p2m_size))); +free (ctx->restore.m);free (ctx->restore.buffered_records);free (ctx->restore.populated_pfns);
tools/libs/saverestore/save.c +13 -14
@@ -77,7 +77,7 @@ static int write_checkpoint_record(struct xc_sr_context *ctx)/* * Writes a batch of memory as a PAGE_DATA record into the stream. The batch -* is constructed in ctx->save.batch_pfns. +* is constructed in ctx->save.m->batch_pfns. * * This function: * - gets the types for each pfn in the batch.
@@ -128,12 +128,12 @@ static int write_batch(struct xc_sr_context *ctx)for ( i =0 ; i < nr_pfns; ++i ) { types[i] = mfns[i] = ctx->save.ops.pfn_to_gfn (ctx, - ctx->save.batch_pfns[i]); + ctx->save.m->batch_pfns[i]);/* Likely a ballooned page. */ if ( mfns[i] ==INVALID_MFN ) { -set_bit (ctx->save.batch_pfns[i], ctx->save.deferred_pages); +set_bit (ctx->save.m->batch_pfns[i], ctx->save.deferred_pages); ++ctx->save.nr_deferred_pages; } }
@@ -179,7 +179,7 @@ static int write_batch(struct xc_sr_context *ctx)if ( errors[p] ) {ERROR ("Mapping of pfn %#" PRIpfn" (mfn %#" PRIpfn") failed %d" , - ctx->save.batch_pfns[i], mfns[p], errors[p]); + ctx->save.m->batch_pfns[i], mfns[p], errors[p]); goto err; }
@@ -193,7 +193,7 @@ static int write_batch(struct xc_sr_context *ctx) {if ( rc ==-1 && errno ==EAGAIN ) { -set_bit (ctx->save.batch_pfns[i], ctx->save.deferred_pages); +set_bit (ctx->save.m->batch_pfns[i], ctx->save.deferred_pages); ++ctx->save.nr_deferred_pages; types[i] =XEN_DOMCTL_PFINFO_XTAB ; --nr_pages;
@@ -224,7 +224,7 @@ static int write_batch(struct xc_sr_context *ctx) rec.length += nr_pages *PAGE_SIZE ;for ( i =0 ; i < nr_pfns; ++i ) - rec_pfns[i] = ((uint64_t )(types[i]) <<32 ) | ctx->save.batch_pfns[i]; + rec_pfns[i] = ((uint64_t )(types[i]) <<32 ) | ctx->save.m->batch_pfns[i]; iov[0 ].iov_base = &rec.type; iov[0 ].iov_len =sizeof (rec.type);
@@ -296,9 +296,9 @@ static int flush_batch(struct xc_sr_context *ctx)if ( !rc ) { -VALGRIND_MAKE_MEM_UNDEFINED (ctx->save.batch_pfns, +VALGRIND_MAKE_MEM_UNDEFINED (ctx->save.m->batch_pfns,MAX_BATCH_SIZE * -sizeof (*ctx->save.batch_pfns)); +sizeof (*ctx->save.m->batch_pfns)); }return rc;
@@ -315,7 +315,7 @@ static int add_to_batch(struct xc_sr_context *ctx, xen_pfn_t pfn) rc =flush_batch (ctx);if ( rc ==0 ) - ctx->save.batch_pfns[ctx->save.nr_batch_pfns++] = pfn; + ctx->save.m->batch_pfns[ctx->save.nr_batch_pfns++] = pfn;return rc; }
@@ -849,13 +849,12 @@ static int setup(struct xc_sr_context *ctx) dirty_bitmap =xc_hypercall_buffer_alloc_pages ( xch, dirty_bitmap,NRPAGES (bitmap_size (ctx->save.p2m_size))); - ctx->save.batch_pfns =malloc (MAX_BATCH_SIZE * -sizeof (*ctx->save.batch_pfns)); ctx->save.deferred_pages =bitmap_alloc (ctx->save.p2m_size); + ctx->save.m =malloc (sizeof (*ctx->save.m)); -if ( !ctx->save.batch_pfns || !dirty_bitmap || !ctx->save.deferred_pages ) +if ( !ctx->save.m || !dirty_bitmap || !ctx->save.deferred_pages ) { -ERROR ("Unable to allocate memory for dirty bitmaps, batch pfns and" +ERROR ("Unable to allocate memory for dirty bitmaps and" " deferred pages" ); rc =-1 ; errno =ENOMEM ;
@@ -884,7 +883,7 @@ static void cleanup(struct xc_sr_context *ctx)xc_hypercall_buffer_free_pages (xch, dirty_bitmap,NRPAGES (bitmap_size (ctx->save.p2m_size)));free (ctx->save.deferred_pages); -free (ctx->save.batch_pfns); +free (ctx->save.m); }/*
PATCH v20210616 12/36 tools: save: move mfns array
In reply to Olaf HeringRemove allocation from hotpath, move mfns array into preallocated space.
--- tools/libs/saverestore/common.h | 2 ++ tools/libs/saverestore/save.c | 7 ++----- 2 files changed, 4 insertions(+), 5 deletions(-)
tools/libs/saverestore/common.h +2 -0
@@ -225,6 +225,8 @@ static inline int update_blob(struct xc_sr_blob *blob,struct sr_save_arrays {xen_pfn_t batch_pfns[MAX_BATCH_SIZE ]; +/* write_batch: Mfns of the batch pfns. */ +xen_pfn_t mfns[MAX_BATCH_SIZE ]; };struct sr_restore_arrays {
tools/libs/saverestore/save.c +2 -5
@@ -88,7 +88,7 @@ static int write_checkpoint_record(struct xc_sr_context *ctx)static int write_batch (struct xc_sr_context *ctx) {xc_interface *xch = ctx->xch; -xen_pfn_t *mfns =NULL , *types =NULL ; +xen_pfn_t *mfns = ctx->save.m->mfns, *types =NULL ;void *guest_mapping =NULL ;void **guest_data =NULL ;void **local_pages =NULL ;
@@ -105,8 +105,6 @@ static int write_batch(struct xc_sr_context *ctx)assert (nr_pfns !=0 ); -/* Mfns of the batch pfns. */ - mfns =malloc (nr_pfns *sizeof (*mfns));/* Types of the batch pfns. */ types =malloc (nr_pfns *sizeof (*types));/* Errors from attempting to map the gfns. */
@@ -118,7 +116,7 @@ static int write_batch(struct xc_sr_context *ctx)/* iovec[] for writev(). */ iov =malloc ((nr_pfns +4 ) *sizeof (*iov)); -if ( !mfns || !types || !errors || !guest_data || !local_pages || !iov ) +if ( !types || !errors || !guest_data || !local_pages || !iov ) {ERROR ("Unable to allocate arrays for a batch of %u pages" , nr_pfns);
@@ -277,7 +275,6 @@ static int write_batch(struct xc_sr_context *ctx)free (guest_data);free (errors);free (types); -free (mfns);return rc; }
PATCH v20210616 13/36 tools: save: move types array
In reply to Olaf HeringRemove allocation from hotpath, move types array into preallocated space.
--- tools/libs/saverestore/common.h | 2 ++ tools/libs/saverestore/save.c | 7 ++----- 2 files changed, 4 insertions(+), 5 deletions(-)
tools/libs/saverestore/common.h +2 -0
@@ -227,6 +227,8 @@ struct sr_save_arrays {xen_pfn_t batch_pfns[MAX_BATCH_SIZE ];/* write_batch: Mfns of the batch pfns. */ xen_pfn_t mfns[MAX_BATCH_SIZE ]; +/* write_batch: Types of the batch pfns. */ +xen_pfn_t types[MAX_BATCH_SIZE ]; };struct sr_restore_arrays {
tools/libs/saverestore/save.c +2 -5
@@ -88,7 +88,7 @@ static int write_checkpoint_record(struct xc_sr_context *ctx)static int write_batch (struct xc_sr_context *ctx) {xc_interface *xch = ctx->xch; -xen_pfn_t *mfns = ctx->save.m->mfns, *types =NULL ; +xen_pfn_t *mfns = ctx->save.m->mfns, *types = ctx->save.m->types;void *guest_mapping =NULL ;void **guest_data =NULL ;void **local_pages =NULL ;
@@ -105,8 +105,6 @@ static int write_batch(struct xc_sr_context *ctx)assert (nr_pfns !=0 ); -/* Types of the batch pfns. */ - types =malloc (nr_pfns *sizeof (*types));/* Errors from attempting to map the gfns. */ errors =malloc (nr_pfns *sizeof (*errors));/* Pointers to page data to send. Mapped gfns or local allocations. */
@@ -116,7 +114,7 @@ static int write_batch(struct xc_sr_context *ctx)/* iovec[] for writev(). */ iov =malloc ((nr_pfns +4 ) *sizeof (*iov)); -if ( !types || !errors || !guest_data || !local_pages || !iov ) +if ( !errors || !guest_data || !local_pages || !iov ) {ERROR ("Unable to allocate arrays for a batch of %u pages" , nr_pfns);
@@ -274,7 +272,6 @@ static int write_batch(struct xc_sr_context *ctx)free (local_pages);free (guest_data);free (errors); -free (types);return rc; }
PATCH v20210616 14/36 tools: save: move errors array
In reply to Olaf HeringRemove allocation from hotpath, move errors array into preallocated space.
--- tools/libs/saverestore/common.h | 2 ++ tools/libs/saverestore/save.c | 7 ++----- 2 files changed, 4 insertions(+), 5 deletions(-)
tools/libs/saverestore/common.h +2 -0
@@ -229,6 +229,8 @@ struct sr_save_arrays {xen_pfn_t mfns[MAX_BATCH_SIZE ];/* write_batch: Types of the batch pfns. */ xen_pfn_t types[MAX_BATCH_SIZE ]; +/* write_batch: Errors from attempting to map the gfns. */ +int errors[MAX_BATCH_SIZE ]; };struct sr_restore_arrays {
tools/libs/saverestore/save.c +2 -5
@@ -92,7 +92,7 @@ static int write_batch(struct xc_sr_context *ctx)void *guest_mapping =NULL ;void **guest_data =NULL ;void **local_pages =NULL ; -int *errors =NULL , rc =-1 ; +int *errors = ctx->save.m->errors, rc =-1 ;unsigned int i, p, nr_pages =0 , nr_pages_mapped =0 ;unsigned int nr_pfns = ctx->save.nr_batch_pfns;void *page, *orig_page;
@@ -105,8 +105,6 @@ static int write_batch(struct xc_sr_context *ctx)assert (nr_pfns !=0 ); -/* Errors from attempting to map the gfns. */ - errors =malloc (nr_pfns *sizeof (*errors));/* Pointers to page data to send. Mapped gfns or local allocations. */ guest_data =calloc (nr_pfns,sizeof (*guest_data));/* Pointers to locally allocated pages. Need freeing. */
@@ -114,7 +112,7 @@ static int write_batch(struct xc_sr_context *ctx)/* iovec[] for writev(). */ iov =malloc ((nr_pfns +4 ) *sizeof (*iov)); -if ( !errors || !guest_data || !local_pages || !iov ) +if ( !guest_data || !local_pages || !iov ) {ERROR ("Unable to allocate arrays for a batch of %u pages" , nr_pfns);
@@ -271,7 +269,6 @@ static int write_batch(struct xc_sr_context *ctx)free (iov);free (local_pages);free (guest_data); -free (errors);return rc; }
PATCH v20210616 15/36 tools: save: move iov array
In reply to Olaf HeringRemove allocation from hotpath, move iov array into preallocated space.
--- tools/libs/saverestore/common.h | 2 ++ tools/libs/saverestore/save.c | 7 ++----- 2 files changed, 4 insertions(+), 5 deletions(-)
tools/libs/saverestore/common.h +2 -0
@@ -231,6 +231,8 @@ struct sr_save_arrays {xen_pfn_t types[MAX_BATCH_SIZE ];/* write_batch: Errors from attempting to map the gfns. */ int errors[MAX_BATCH_SIZE ]; +/* write_batch: iovec[] for writev(). */ +struct iovec iov[MAX_BATCH_SIZE +4 ]; };struct sr_restore_arrays {
tools/libs/saverestore/save.c +2 -5
@@ -97,7 +97,7 @@ static int write_batch(struct xc_sr_context *ctx)unsigned int nr_pfns = ctx->save.nr_batch_pfns;void *page, *orig_page;uint64_t *rec_pfns =NULL ; -struct iovec *iov =NULL ;int iovcnt =0 ; +struct iovec *iov = ctx->save.m->iov;int iovcnt =0 ;struct xc_sr_rec_page_data_header hdr = {0 };struct xc_sr_record rec = { .type =REC_TYPE_PAGE_DATA ,
@@ -109,10 +109,8 @@ static int write_batch(struct xc_sr_context *ctx) guest_data =calloc (nr_pfns,sizeof (*guest_data));/* Pointers to locally allocated pages. Need freeing. */ local_pages =calloc (nr_pfns,sizeof (*local_pages)); -/* iovec[] for writev(). */ - iov =malloc ((nr_pfns +4 ) *sizeof (*iov)); -if ( !guest_data || !local_pages || !iov ) +if ( !guest_data || !local_pages ) {ERROR ("Unable to allocate arrays for a batch of %u pages" , nr_pfns);
@@ -266,7 +264,6 @@ static int write_batch(struct xc_sr_context *ctx)xenforeignmemory_unmap (xch->fmem, guest_mapping, nr_pages_mapped);for ( i =0 ; local_pages && i < nr_pfns; ++i )free (local_pages[i]); -free (iov);free (local_pages);free (guest_data);
PATCH v20210616 16/36 tools: save: move rec_pfns array
In reply to Olaf HeringRemove allocation from hotpath, move rec_pfns array into preallocated space.
--- tools/libs/saverestore/common.h | 2 ++ tools/libs/saverestore/save.c | 11 +---------- 2 files changed, 3 insertions(+), 10 deletions(-)
tools/libs/saverestore/common.h +2 -0
@@ -233,6 +233,8 @@ struct sr_save_arrays {int errors[MAX_BATCH_SIZE ];/* write_batch: iovec[] for writev(). */ struct iovec iov[MAX_BATCH_SIZE +4 ]; +/* write_batch */ +uint64_t rec_pfns[MAX_BATCH_SIZE ]; };struct sr_restore_arrays {
tools/libs/saverestore/save.c +1 -10
@@ -96,7 +96,7 @@ static int write_batch(struct xc_sr_context *ctx)unsigned int i, p, nr_pages =0 , nr_pages_mapped =0 ;unsigned int nr_pfns = ctx->save.nr_batch_pfns;void *page, *orig_page; -uint64_t *rec_pfns =NULL ; +uint64_t *rec_pfns = ctx->save.m->rec_pfns;struct iovec *iov = ctx->save.m->iov;int iovcnt =0 ;struct xc_sr_rec_page_data_header hdr = {0 };struct xc_sr_record rec = {
@@ -201,14 +201,6 @@ static int write_batch(struct xc_sr_context *ctx) } } - rec_pfns =malloc (nr_pfns *sizeof (*rec_pfns)); -if ( !rec_pfns ) - { -ERROR ("Unable to allocate %zu bytes of memory for page data pfn list" , - nr_pfns *sizeof (*rec_pfns)); - goto err; - } - hdr.count = nr_pfns; rec.length =sizeof (hdr);
@@ -259,7 +251,6 @@ static int write_batch(struct xc_sr_context *ctx) rc = ctx->save.nr_batch_pfns =0 ; err: -free (rec_pfns);if ( guest_mapping )xenforeignmemory_unmap (xch->fmem, guest_mapping, nr_pages_mapped);for ( i =0 ; local_pages && i < nr_pfns; ++i )
PATCH v20210616 17/36 tools: save: move guest_data array
In reply to Olaf HeringRemove allocation from hotpath, move guest_data array into preallocated space.
Because this was allocated with calloc:
Adjust the loop to clear unused entries as needed.
--- tools/libs/saverestore/common.h | 2 ++ tools/libs/saverestore/save.c | 11 ++++++----- 2 files changed, 8 insertions(+), 5 deletions(-)
tools/libs/saverestore/common.h +2 -0
@@ -235,6 +235,8 @@ struct sr_save_arrays {struct iovec iov[MAX_BATCH_SIZE +4 ];/* write_batch */ uint64_t rec_pfns[MAX_BATCH_SIZE ]; +/* write_batch: Pointers to page data to send. Mapped gfns or local allocations. */ +void *guest_data[MAX_BATCH_SIZE ]; };struct sr_restore_arrays {
tools/libs/saverestore/save.c +6 -5
@@ -90,7 +90,7 @@ static int write_batch(struct xc_sr_context *ctx)xc_interface *xch = ctx->xch;xen_pfn_t *mfns = ctx->save.m->mfns, *types = ctx->save.m->types;void *guest_mapping =NULL ; -void **guest_data =NULL ; +void **guest_data = ctx->save.m->guest_data;void **local_pages =NULL ;int *errors = ctx->save.m->errors, rc =-1 ;unsigned int i, p, nr_pages =0 , nr_pages_mapped =0 ;
@@ -105,12 +105,10 @@ static int write_batch(struct xc_sr_context *ctx)assert (nr_pfns !=0 ); -/* Pointers to page data to send. Mapped gfns or local allocations. */ - guest_data =calloc (nr_pfns,sizeof (*guest_data));/* Pointers to locally allocated pages. Need freeing. */ local_pages =calloc (nr_pfns,sizeof (*local_pages)); -if ( !guest_data || !local_pages ) +if ( !local_pages ) {ERROR ("Unable to allocate arrays for a batch of %u pages" , nr_pfns);
@@ -166,7 +164,10 @@ static int write_batch(struct xc_sr_context *ctx)for ( i =0 , p =0 ; i < nr_pfns; ++i ) {if (page_type_has_stream_data (types[i]) == false ) + { + guest_data[i] =NULL ;continue ; + }if ( errors[p] ) {
@@ -183,6 +184,7 @@ static int write_batch(struct xc_sr_context *ctx)if ( rc ) { + guest_data[i] =NULL ;if ( rc ==-1 && errno ==EAGAIN ) {set_bit (ctx->save.m->batch_pfns[i], ctx->save.deferred_pages);
@@ -256,7 +258,6 @@ static int write_batch(struct xc_sr_context *ctx)for ( i =0 ; local_pages && i < nr_pfns; ++i )free (local_pages[i]);free (local_pages); -free (guest_data);return rc; }
PATCH v20210616 18/36 tools: save: move local_pages array
In reply to Olaf HeringRemove allocation from hotpath, move local_pages array into preallocated space.
Adjust the code to use the src page as is in case of HVM. In case of PV the page may need to be normalised, use a private memory area for this purpose.
--- tools/libs/saverestore/common.h | 22 ++++++++++--------- tools/libs/saverestore/save.c | 25 +++------------------ tools/libs/saverestore/save_x86_hvm.c | 5 +++-- tools/libs/saverestore/save_x86_pv.c | 31 ++++++++++++++++++--------- 4 files changed, 39 insertions(+), 44 deletions(-)
tools/libs/saverestore/common.h +12 -10
@@ -45,16 +45,12 @@ struct xc_sr_save_ops* Optionally transform the contents of a page from being specific to the * sending environment, to being generic for the stream. * -* The page of data at the end of 'page' may be a read-only mapping of a -* running guest; it must not be modified. If no transformation is -* required, the callee should leave '*pages' untouched. +* The page of data '*src' may be a read-only mapping of a running guest; +* it must not be modified. If no transformation is required, the callee +* should leave '*src' untouched, and return it via '**ptr'. * -* If a transformation is required, the callee should allocate themselves -* a local page using malloc() and return it via '*page'. -* -* The caller shall free() '*page' in all cases. In the case that the -* callee encounters an error, it should *NOT* free() the memory it -* allocated for '*page'. +* If a transformation is required, the callee should provide the +* transformed page in a private buffer and return it via '**ptr'. * * It is valid to fail with EAGAIN if the transformation is not able to be * completed at this point. The page shall be retried later.
@@ -62,7 +58,7 @@ struct xc_sr_save_ops* @returns 0 for success, -1 for failure, with errno appropriately set. */ int (*normalise_page)(struct xc_sr_context *ctx,xen_pfn_t type, -void **page); +void *src,unsigned int idx,void **ptr);/** * Set up local environment to save a domain. (Typically querying
@@ -385,6 +381,12 @@ struct xc_sr_contextunion { +struct + { +/* Used by write_batch for modified pages. */ +void *normalised_pages; + } save; +struct {/* State machine for the order of received records. */
tools/libs/saverestore/save.c +3 -22
@@ -91,11 +91,10 @@ static int write_batch(struct xc_sr_context *ctx)xen_pfn_t *mfns = ctx->save.m->mfns, *types = ctx->save.m->types;void *guest_mapping =NULL ;void **guest_data = ctx->save.m->guest_data; -void **local_pages =NULL ;int *errors = ctx->save.m->errors, rc =-1 ;unsigned int i, p, nr_pages =0 , nr_pages_mapped =0 ;unsigned int nr_pfns = ctx->save.nr_batch_pfns; -void *page, *orig_page; +void *src;uint64_t *rec_pfns = ctx->save.m->rec_pfns;struct iovec *iov = ctx->save.m->iov;int iovcnt =0 ;struct xc_sr_rec_page_data_header hdr = {0 };
@@ -105,16 +104,6 @@ static int write_batch(struct xc_sr_context *ctx)assert (nr_pfns !=0 ); -/* Pointers to locally allocated pages. Need freeing. */ - local_pages =calloc (nr_pfns,sizeof (*local_pages)); - -if ( !local_pages ) - { -ERROR ("Unable to allocate arrays for a batch of %u pages" , - nr_pfns); - goto err; - } -for ( i =0 ; i < nr_pfns; ++i ) { types[i] = mfns[i] = ctx->save.ops.pfn_to_gfn(ctx,
@@ -176,11 +165,8 @@ static int write_batch(struct xc_sr_context *ctx) goto err; } - orig_page = page = guest_mapping + (p *PAGE_SIZE ); - rc = ctx->save.ops.normalise_page (ctx, types[i], &page); - -if ( orig_page != page ) - local_pages[i] = page; + src = guest_mapping + (p *PAGE_SIZE ); + rc = ctx->save.ops.normalise_page (ctx, types[i], src, i, &guest_data[i]);if ( rc ) {
@@ -195,8 +181,6 @@ static int write_batch(struct xc_sr_context *ctx)else goto err; } -else - guest_data[i] = page; rc =-1 ; ++p;
@@ -255,9 +239,6 @@ static int write_batch(struct xc_sr_context *ctx) err:if ( guest_mapping )xenforeignmemory_unmap (xch->fmem, guest_mapping, nr_pages_mapped); -for ( i =0 ; local_pages && i < nr_pfns; ++i ) -free (local_pages[i]); -free (local_pages);return rc; }
tools/libs/saverestore/save_x86_hvm.c +3 -2
@@ -129,9 +129,10 @@ static xen_pfn_t x86_hvm_pfn_to_gfn(const struct xc_sr_context *ctx,return pfn; } -static int x86_hvm_normalise_page (struct xc_sr_context *ctx, -xen_pfn_t type,void **page) +static int x86_hvm_normalise_page (struct xc_sr_context *ctx,xen_pfn_t type, +void *src,unsigned int idx,void **ptr) { + *ptr = src;return 0 ; }
tools/libs/saverestore/save_x86_pv.c +21 -10
@@ -999,29 +999,31 @@ static xen_pfn_t x86_pv_pfn_to_gfn(const struct xc_sr_context *ctx,* save_ops function. Performs pagetable normalisation on appropriate pages. */ static int x86_pv_normalise_page (struct xc_sr_context *ctx,xen_pfn_t type, -void **page) +void *src,unsigned int idx,void **ptr) {xc_interface *xch = ctx->xch; -void *local_page;int rc; +void *dst; type &=XEN_DOMCTL_PFINFO_LTABTYPE_MASK ;if ( type <XEN_DOMCTL_PFINFO_L1TAB || type >XEN_DOMCTL_PFINFO_L4TAB ) + { + *ptr = src;return 0 ; + } - local_page =malloc (PAGE_SIZE ); -if ( !local_page ) +if ( idx >=MAX_BATCH_SIZE ) { -ERROR ("Unable to allocate scratch page" ); - rc =-1 ; - goto out; +ERROR ("idx %u out of range" , idx); + errno =ERANGE ; +return -1 ; } - rc =normalise_pagetable (ctx, *page, local_page, type); - *page = local_page; + dst = ctx->x86.pv.save.normalised_pages + idx *PAGE_SIZE ; + rc =normalise_pagetable (ctx, src, dst, type); + *ptr = dst; - out:return rc; }
@@ -1031,8 +1033,16 @@ static int x86_pv_normalise_page(struct xc_sr_context *ctx, xen_pfn_t type,*/ static int x86_pv_setup (struct xc_sr_context *ctx) { +xc_interface *xch = ctx->xch;int rc; + ctx->x86.pv.save.normalised_pages =malloc (MAX_BATCH_SIZE *PAGE_SIZE ); +if ( !ctx->x86.pv.save.normalised_pages ) + { +PERROR ("Failed to allocate normalised_pages" ); +return -1 ; + } + rc =x86_pv_domain_info (ctx);if ( rc )return rc;
@@ -1118,6 +1128,7 @@ static int x86_pv_check_vm_state(struct xc_sr_context *ctx)static int x86_pv_cleanup (struct xc_sr_context *ctx) { +free (ctx->x86.pv.save.normalised_pages);free (ctx->x86.pv.p2m_pfns);if ( ctx->x86.pv.p2m )
PATCH v20210616 19/36 tools: restore: move types array
In reply to Olaf HeringRemove allocation from hotpath, move types array into preallocated space.
--- tools/libs/saverestore/common.h | 1 + tools/libs/saverestore/restore.c | 12 +----------- 2 files changed, 2 insertions(+), 11 deletions(-)
tools/libs/saverestore/common.h +1 -0
@@ -238,6 +238,7 @@ struct sr_save_arrays {struct sr_restore_arrays {/* handle_page_data */ xen_pfn_t pfns[MAX_BATCH_SIZE ]; +uint32_t types[MAX_BATCH_SIZE ]; };struct xc_sr_context
tools/libs/saverestore/restore.c +1 -11
@@ -316,7 +316,7 @@ static int handle_page_data(struct xc_sr_context *ctx, struct xc_sr_record *rec)int rc =-1 ;xen_pfn_t *pfns = ctx->restore.m->pfns, pfn; -uint32_t *types =NULL , type; +uint32_t *types = ctx->restore.m->types, type;/* * v2 compatibility only exists for x86 streams. This is a bit of a
@@ -363,14 +363,6 @@ static int handle_page_data(struct xc_sr_context *ctx, struct xc_sr_record *rec) goto err; } - types =malloc (pages->count *sizeof (*types)); -if ( !types ) - { -ERROR ("Unable to allocate enough memory for %u pfns" , - pages->count); - goto err; - } -for ( i =0 ; i < pages->count; ++i ) { pfn = pages->pfn[i] &PAGE_DATA_PFN_MASK ;
@@ -410,8 +402,6 @@ static int handle_page_data(struct xc_sr_context *ctx, struct xc_sr_record *rec) rc =process_page_data (ctx, pages->count, pfns, types, &pages->pfn[pages->count]); err: -free (types); -return rc; }
PATCH v20210616 20/36 tools: restore: move mfns array
In reply to Olaf HeringRemove allocation from hotpath, move mfns array into preallocated space.
--- tools/libs/saverestore/common.h | 2 ++ tools/libs/saverestore/restore.c | 5 ++--- 2 files changed, 4 insertions(+), 3 deletions(-)
tools/libs/saverestore/common.h +2 -0
@@ -239,6 +239,8 @@ struct sr_restore_arrays {/* handle_page_data */ xen_pfn_t pfns[MAX_BATCH_SIZE ];uint32_t types[MAX_BATCH_SIZE ]; +/* process_page_data */ +xen_pfn_t mfns[MAX_BATCH_SIZE ]; };struct xc_sr_context
tools/libs/saverestore/restore.c +2 -3
@@ -205,7 +205,7 @@ static int process_page_data(struct xc_sr_context *ctx, unsigned int count, xen_pfn_t *pfns, uint32_t *types, void *page_data) {xc_interface *xch = ctx->xch; -xen_pfn_t *mfns =malloc (count *sizeof (*mfns)); +xen_pfn_t *mfns = ctx->restore.m->mfns;int *map_errs =malloc (count *sizeof (*map_errs));int rc;void *mapping =NULL , *guest_page =NULL ;
@@ -213,7 +213,7 @@ static int process_page_data(struct xc_sr_context *ctx, unsigned int count, j,/* j indexes the subset of pfns we decide to map. */ nr_pages =0 ; -if ( !mfns || !map_errs ) +if ( !map_errs ) { rc =-1 ; ERROR("Failed to allocate %zu bytes to process page data",
@@ -299,7 +299,6 @@ static int process_page_data(struct xc_sr_context *ctx, unsigned int count,xenforeignmemory_unmap (xch->fmem, mapping, nr_pages);free (map_errs); -free (mfns);return rc; }
PATCH v20210616 21/36 tools: restore: move map_errs array
In reply to Olaf HeringRemove allocation from hotpath, move map_errs array into preallocated space.
--- tools/libs/saverestore/common.h | 1 + tools/libs/saverestore/restore.c | 12 +----------- 2 files changed, 2 insertions(+), 11 deletions(-)
tools/libs/saverestore/common.h +1 -0
@@ -241,6 +241,7 @@ struct sr_restore_arrays {uint32_t types[MAX_BATCH_SIZE ];/* process_page_data */ xen_pfn_t mfns[MAX_BATCH_SIZE ]; +int map_errs[MAX_BATCH_SIZE ]; };struct xc_sr_context
tools/libs/saverestore/restore.c +1 -11
@@ -206,21 +206,13 @@ static int process_page_data(struct xc_sr_context *ctx, unsigned int count, {xc_interface *xch = ctx->xch;xen_pfn_t *mfns = ctx->restore.m->mfns; -int *map_errs =malloc (count *sizeof (*map_errs)); +int *map_errs = ctx->restore.m->map_errs;int rc;void *mapping =NULL , *guest_page =NULL ;unsigned int i,/* i indexes the pfns from the record. */ j,/* j indexes the subset of pfns we decide to map. */ nr_pages =0 ; -if ( !map_errs ) - { - rc =-1 ; -ERROR ("Failed to allocate %zu bytes to process page data" , - count * (sizeof (*mfns) +sizeof (*map_errs))); - goto err; - } - rc =populate_pfns (ctx, count, pfns, types);if ( rc ) {
@@ -298,8 +290,6 @@ static int process_page_data(struct xc_sr_context *ctx, unsigned int count,if ( mapping )xenforeignmemory_unmap (xch->fmem, mapping, nr_pages); -free (map_errs); -return rc; }
PATCH v20210616 22/36 tools: restore: move mfns array in populate_pfns
In reply to Olaf HeringRemove allocation from hotpath, move populate_pfns mfns array into preallocated space. Use some prefix to avoid conflict with an array used in handle_page_data.
--- tools/libs/saverestore/common.h | 2 ++ tools/libs/saverestore/restore.c | 5 ++--- 2 files changed, 4 insertions(+), 3 deletions(-)
tools/libs/saverestore/common.h +2 -0
@@ -242,6 +242,8 @@ struct sr_restore_arrays {/* process_page_data */ xen_pfn_t mfns[MAX_BATCH_SIZE ];int map_errs[MAX_BATCH_SIZE ]; +/* populate_pfns */ +xen_pfn_t pp_mfns[MAX_BATCH_SIZE ]; };struct xc_sr_context
tools/libs/saverestore/restore.c +2 -3
@@ -138,12 +138,12 @@ int populate_pfns(struct xc_sr_context *ctx, unsigned int count, const xen_pfn_t *original_pfns, const uint32_t *types) { xc_interface *xch = ctx->xch; -xen_pfn_t *mfns =malloc (count *sizeof (*mfns)), +xen_pfn_t *mfns = ctx->restore.m->pp_mfns, *pfns =malloc (count *sizeof (*pfns));unsigned int i, nr_pfns =0 ;int rc =-1 ; -if ( !mfns || !pfns ) +if ( !pfns ) {ERROR ("Failed to allocate %zu bytes for populating the physmap" ,2 * count *sizeof (*mfns));
@@ -191,7 +191,6 @@ int populate_pfns(struct xc_sr_context *ctx, unsigned int count, err:free (pfns); -free (mfns);return rc; }
PATCH v20210616 23/36 tools: restore: move pfns array in populate_pfns
In reply to Olaf HeringRemove allocation from hotpath, move populate_pfns' pfns array into preallocated space. Use some prefix to avoid conflict with an array used in handle_page_data.
--- tools/libs/saverestore/common.h | 1 + tools/libs/saverestore/restore.c | 11 +---------- 2 files changed, 2 insertions(+), 10 deletions(-)
tools/libs/saverestore/common.h +1 -0
@@ -244,6 +244,7 @@ struct sr_restore_arrays {int map_errs[MAX_BATCH_SIZE ];/* populate_pfns */ xen_pfn_t pp_mfns[MAX_BATCH_SIZE ]; +xen_pfn_t pp_pfns[MAX_BATCH_SIZE ]; };struct xc_sr_context
tools/libs/saverestore/restore.c +1 -10
@@ -139,17 +139,10 @@ int populate_pfns(struct xc_sr_context *ctx, unsigned int count, {xc_interface *xch = ctx->xch;xen_pfn_t *mfns = ctx->restore.m->pp_mfns, - *pfns =malloc (count *sizeof (*pfns)); + *pfns = ctx->restore.m->pp_pfns;unsigned int i, nr_pfns =0 ;int rc =-1 ; -if ( !pfns ) - { -ERROR ("Failed to allocate %zu bytes for populating the physmap" , -2 * count *sizeof (*mfns)); - goto err; - } -for ( i =0 ; i < count; ++i ) { if ( (!types ||
@@ -190,8 +183,6 @@ int populate_pfns(struct xc_sr_context *ctx, unsigned int count, rc =0 ; err: -free (pfns); -return rc; }
PATCH v20210616 24/36 tools: restore: split record processing
In reply to Olaf Heringhandle_page_data must be able to read directly into mapped guest memory. This will avoid unneccesary memcpy calls for data which can be consumed verbatim.
Rearrange the code to allow decisions based on the incoming record.
This change is preparation for future changes in handle_page_data, no change in behavior is intended.
--- tools/libs/saverestore/common.c | 33 ++++++++++++--------- tools/libs/saverestore/common.h | 4 ++- tools/libs/saverestore/restore.c | 49 ++++++++++++++++++++++---------- tools/libs/saverestore/save.c | 7 ++++- 4 files changed, 63 insertions(+), 30 deletions(-)
tools/libs/saverestore/common.c +20 -13
@@ -91,26 +91,33 @@ int write_split_record(struct xc_sr_context *ctx, struct xc_sr_record *rec,return -1 ; } -int read_record (struct xc_sr_context *ctx,int fd,struct xc_sr_record *rec) +int read_record_header (struct xc_sr_context *ctx,int fd,struct xc_sr_rhdr *rhdr) {xc_interface *xch = ctx->xch; -struct xc_sr_rhdr rhdr; -size_t datasz; -if (read_exact (fd, &rhdr,sizeof (rhdr)) ) +if (read_exact (fd, rhdr,sizeof (*rhdr)) ) {PERROR ("Failed to read Record Header from stream" );return -1 ; } -if ( rhdr.length >REC_LENGTH_MAX ) +if ( rhdr->length >REC_LENGTH_MAX ) { -ERROR ("Record (0x%08x, %s) length %#x exceeds max (%#x)" , rhdr.type, -rec_type_to_str (rhdr.type), rhdr.length,REC_LENGTH_MAX ); +ERROR ("Record (0x%08x, %s) length %#x exceeds max (%#x)" , rhdr->type, +rec_type_to_str (rhdr->type), rhdr->length,REC_LENGTH_MAX );return -1 ; } - datasz =ROUNDUP (rhdr.length,REC_ALIGN_ORDER ); +return 0 ; +} + +int read_record_data (struct xc_sr_context *ctx,int fd,struct xc_sr_rhdr *rhdr, +struct xc_sr_record *rec) +{ +xc_interface *xch = ctx->xch; +size_t datasz; + + datasz =ROUNDUP (rhdr->length,REC_ALIGN_ORDER );if ( datasz ) {
@@ -119,7 +126,7 @@ int read_record(struct xc_sr_context *ctx, int fd, struct xc_sr_record *rec)if ( !rec->data ) {ERROR ("Unable to allocate %zu bytes for record data (0x%08x, %s)" , - datasz, rhdr.type,rec_type_to_str (rhdr.type)); + datasz, rhdr->type,rec_type_to_str (rhdr->type));return -1 ; }
@@ -128,18 +135,18 @@ int read_record(struct xc_sr_context *ctx, int fd, struct xc_sr_record *rec)free (rec->data); rec->data =NULL ;PERROR ("Failed to read %zu bytes of data for record (0x%08x, %s)" , - datasz, rhdr.type,rec_type_to_str (rhdr.type)); + datasz, rhdr->type,rec_type_to_str (rhdr->type));return -1 ; } }else rec->data =NULL ; - rec->type = rhdr.type; - rec->length = rhdr.length; + rec->type = rhdr->type; + rec->length = rhdr->length;return 0 ; -}; +}static void __attribute__((unused))build_assertions (void ) {
tools/libs/saverestore/common.h +3 -1
@@ -487,7 +487,9 @@ static inline int write_record(struct xc_sr_context *ctx,* * On failure, the contents of the record structure are undefined. */ -int read_record (struct xc_sr_context *ctx,int fd,struct xc_sr_record *rec); +int read_record_header (struct xc_sr_context *ctx,int fd,struct xc_sr_rhdr *rhdr); +int read_record_data (struct xc_sr_context *ctx,int fd,struct xc_sr_rhdr *rhdr, +struct xc_sr_record *rec);/* * This would ideally be private in restore.c, but is needed by
tools/libs/saverestore/restore.c +34 -15
125 lines not on this page · raw message
tools/libs/saverestore/save.c +6 -1
@@ -589,6 +589,7 @@ static int send_memory_live(struct xc_sr_context *ctx)static int colo_merge_secondary_dirty_bitmap (struct xc_sr_context *ctx) {xc_interface *xch = ctx->xch; +struct xc_sr_rhdr rhdr;struct xc_sr_record rec;uint64_t *pfns =NULL ;uint64_t pfn;
@@ -597,7 +598,11 @@ static int colo_merge_secondary_dirty_bitmap(struct xc_sr_context *ctx) DECLARE_HYPERCALL_BUFFER_SHADOW(unsigned long, dirty_bitmap, &ctx->save.dirty_bitmap_hbuf); - rc =read_record (ctx, ctx->save.recv_fd, &rec); + rc =read_record_header (ctx, ctx->save.recv_fd, &rhdr); +if ( rc ) + goto err; + + rc =read_record_data (ctx, ctx->save.recv_fd, &rhdr, &rec);if ( rc ) goto err;
PATCH v20210616 25/36 tools: restore: split handle_page_data
In reply to Olaf Heringhandle_page_data must be able to read directly into mapped guest memory. This will avoid unneccesary memcpy calls for data that can be consumed verbatim.
Split the various steps of record processing:
- move processing to handle_buffered_page_data
- adjust xenforeignmemory_map to set errno in case of failure
- adjust verify mode to set errno in case of failure
This change is preparation for future changes in handle_page_data, no change in behavior is intended.
--- tools/libs/saverestore/common.h | 9 + tools/libs/saverestore/restore.c | 343 ++++++++++++++++++++----------- 2 files changed, 231 insertions(+), 121 deletions(-)
tools/libs/saverestore/common.h +9 -0
@@ -242,9 +242,14 @@ struct sr_restore_arrays {/* process_page_data */ xen_pfn_t mfns[MAX_BATCH_SIZE ];int map_errs[MAX_BATCH_SIZE ]; +void *guest_data[MAX_BATCH_SIZE ]; +/* populate_pfns */ xen_pfn_t pp_mfns[MAX_BATCH_SIZE ];xen_pfn_t pp_pfns[MAX_BATCH_SIZE ]; + +/* Must be the last member */ +struct xc_sr_rec_page_data_header pages; };struct xc_sr_context
@@ -335,7 +340,11 @@ struct xc_sr_context/* Sender has invoked verify mode on the stream. */ bool verify; +void *verify_buf; +struct sr_restore_arrays *m; +void *guest_mapping; +uint32_t nr_mapped_pages; } restore; };
tools/libs/saverestore/restore.c +222 -121
412 lines not on this page · raw message
PATCH v20210616 26/36 tools: restore: write data directly into guest
In reply to Olaf HeringRead incoming migration stream directly into the guest memory. This avoids the memory allocation and copying, and the resulting performance penalty.
--- tools/libs/saverestore/common.h | 1 + tools/libs/saverestore/restore.c | 132 ++++++++++++++++++++++++++++++- 2 files changed, 129 insertions(+), 4 deletions(-)
tools/libs/saverestore/common.h +1 -0
@@ -243,6 +243,7 @@ struct sr_restore_arrays {xen_pfn_t mfns[MAX_BATCH_SIZE ];int map_errs[MAX_BATCH_SIZE ];void *guest_data[MAX_BATCH_SIZE ]; +struct iovec iov[MAX_BATCH_SIZE ];/* populate_pfns */ xen_pfn_t pp_mfns[MAX_BATCH_SIZE ];
tools/libs/saverestore/restore.c +128 -4
145 lines not on this page · raw message
PATCH v20210616 27/36 tools: recognize LIBXL_API_VERSION for 4.16
In reply to Olaf HeringThis is required by upcoming API changes.
--- tools/include/libxl.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)
tools/include/libxl.h +2 -1
@@ -729,7 +729,8 @@ typedef struct libxl__ctx libxl_ctx;#if LIBXL_API_VERSION !=0x040200 &&LIBXL_API_VERSION !=0x040300 && \LIBXL_API_VERSION !=0x040400 &&LIBXL_API_VERSION !=0x040500 && \LIBXL_API_VERSION !=0x040700 &&LIBXL_API_VERSION !=0x040800 && \ -LIBXL_API_VERSION !=0x041300 &&LIBXL_API_VERSION !=0x041400 +LIBXL_API_VERSION !=0x041300 &&LIBXL_API_VERSION !=0x041400 && \ +LIBXL_API_VERSION !=0x041600 #error Unknown LIBXL_API_VERSION#endif #endif
PATCH v20210616 28/36 tools: adjust libxl_domain_suspend to receive a struct props
In reply to Olaf HeringUpcoming changes will pass more knobs down to xc_domain_save. Adjust the libxl_domain_suspend API to allow easy adding of additional knobs.
No change in behavior intented.
--- tools/include/libxl.h | 26 +++++++++++++++++++++++--- tools/libs/light/libxl_domain.c | 7 ++++--- tools/ocaml/libs/xl/xenlight_stubs.c | 3 ++- tools/xl/xl_migrate.c | 9 ++++++--- tools/xl/xl_saverestore.c | 3 ++- 5 files changed, 37 insertions(+), 11 deletions(-)
tools/include/libxl.h +23 -3
@@ -1706,12 +1706,32 @@ static inline int libxl_retrieve_domain_configuration_0x041200( libxl_retrieve_domain_configuration_0x041200#endif +/* +* LIBXL_HAVE_DOMAIN_SUSPEND_PROPS indicates that the +* libxl_domain_suspend_props() function takes a props struct. +*/ +#define LIBXL_HAVE_DOMAIN_SUSPEND_PROPS 1 + +typedef struct { +uint32_t flags;/* LIBXL_SUSPEND_* */ +}libxl_domain_suspend_props ; +#define LIBXL_SUSPEND_DEBUG 1 +#define LIBXL_SUSPEND_LIVE 2 + int libxl_domain_suspend(libxl_ctx *ctx, uint32_t domid, int fd, - int flags, /* LIBXL_SUSPEND_* */ + libxl_domain_suspend_props *props, const libxl_asyncop_how *ao_how) LIBXL_EXTERNAL_CALLERS_ONLY; -#define LIBXL_SUSPEND_DEBUG 1 -#define LIBXL_SUSPEND_LIVE 2 +#if defined(LIBXL_API_VERSION ) &&LIBXL_API_VERSION <0x041600 +static inline int libxl_domain_suspend_0x041500 (libxl_ctx *ctx,uint32_t domid, +int fd,int flags,/* LIBXL_SUSPEND_* */ +const libxl_asyncop_how *ao_how) +{ +libxl_domain_suspend_props props = { .flags = flags, }; +return libxl_domain_suspend (ctx, domid, fd, &props, ao_how); +} +#define libxl_domain_suspend libxl_domain_suspend_0x041500 +#endif /* * Only suspend domain, do not save its state to file, do not destroy it.
tools/libs/light/libxl_domain.c +4 -3
@@ -505,7 +505,8 @@ static void domain_suspend_cb(libxl__egc *egc, } -int libxl_domain_suspend (libxl_ctx *ctx,uint32_t domid,int fd,int flags, +int libxl_domain_suspend (libxl_ctx *ctx,uint32_t domid,int fd, +libxl_domain_suspend_props *props,const libxl_asyncop_how *ao_how) {AO_CREATE (ctx, domid, ao_how);
@@ -526,8 +527,8 @@ int libxl_domain_suspend(libxl_ctx *ctx, uint32_t domid, int fd, int flags, dss->domid = domid; dss->fd = fd; dss->type = type; - dss->live = flags &LIBXL_SUSPEND_LIVE ; - dss->debug = flags &LIBXL_SUSPEND_DEBUG ; + dss->live = props->flags &LIBXL_SUSPEND_LIVE ; + dss->debug = props->flags &LIBXL_SUSPEND_DEBUG ; dss->checkpointed_stream =LIBXL_CHECKPOINTED_STREAM_NONE ; rc = libxl__fd_flags_modify_save(gc, dss->fd,
tools/ocaml/libs/xl/xenlight_stubs.c +2 -1
@@ -614,10 +614,11 @@ value stub_libxl_domain_suspend(value ctx, value domid, value fd, value async, vint ret;uint32_t c_domid =Int_val (domid);int c_fd =Int_val (fd); +libxl_domain_suspend_props props = {};libxl_asyncop_how *ao_how =aohow_val (async);caml_enter_blocking_section (); - ret =libxl_domain_suspend (CTX , c_domid, c_fd,0 , ao_how); + ret =libxl_domain_suspend (CTX , c_domid, c_fd, &props, ao_how);caml_leave_blocking_section ();free (ao_how);
tools/xl/xl_migrate.c +6 -3
@@ -186,7 +186,10 @@ static void migrate_domain(uint32_t domid, int preserve_domid,char *away_domname;char rc_buf;uint8_t *config_data; -int config_len, flags =LIBXL_SUSPEND_LIVE ; +int config_len; +libxl_domain_suspend_props props = { + .flags =LIBXL_SUSPEND_LIVE , + };save_domain_core_begin (domid, preserve_domid, override_config_file, &config_data, &config_len);
@@ -205,8 +208,8 @@ static void migrate_domain(uint32_t domid, int preserve_domid,xtl_stdiostream_adjust_flags (logger,XTL_STDIOSTREAM_HIDE_PROGRESS ,0 );if (debug) - flags |=LIBXL_SUSPEND_DEBUG ; - rc =libxl_domain_suspend (ctx, domid, send_fd, flags,NULL ); + props.flags |=LIBXL_SUSPEND_DEBUG ; + rc =libxl_domain_suspend (ctx, domid, send_fd, &props,NULL );if (rc) {fprintf (stderr,"migration sender: libxl_domain_suspend failed" " (rc=%d)\n" , rc);
tools/xl/xl_saverestore.c +2 -1
@@ -130,6 +130,7 @@ static int save_domain(uint32_t domid, int preserve_domid,int fd;uint8_t *config_data;int config_len; +libxl_domain_suspend_props props = {};save_domain_core_begin (domid, preserve_domid, override_config_file, &config_data, &config_len);
@@ -146,7 +147,7 @@ static int save_domain(uint32_t domid, int preserve_domid,save_domain_core_writeconfig (fd, filename, config_data, config_len); -int rc =libxl_domain_suspend (ctx, domid, fd,0 ,NULL ); +int rc =libxl_domain_suspend (ctx, domid, fd, &props,NULL );close (fd);if (rc <0 ) {
PATCH v20210616 29/36 tools: change struct precopy_stats to precopy_stats_t
In reply to Olaf HeringThis will help libxl_save_msgs_gen.pl to copy the struct as a region of memory.
No change in behavior intented.
--- tools/include/xensaverestore.h | 7 +++---- tools/libs/saverestore/common.h | 2 +- tools/libs/saverestore/save.c | 6 +++--- 3 files changed, 7 insertions(+), 8 deletions(-)
tools/include/xensaverestore.h +3 -4
@@ -23,18 +23,17 @@#define XCFLAGS_DEBUG (1 << 1)/* For save's precopy_policy(). */ -struct precopy_stats -{ +typedef struct {unsigned int iteration;unsigned long total_written;long dirty_count;/* -1 if unknown */ -}; +}precopy_stats_t ;/* * A precopy_policy callback may not be running in the same address * space as libxc an so precopy_stats is passed by value. */ -typedef int (*precopy_policy_t )(struct precopy_stats ,void *); +typedef int (*precopy_policy_t )(precopy_stats_t ,void *);/* callbacks provided by xc_domain_save */ struct save_callbacks {
tools/libs/saverestore/common.h +1 -1
@@ -283,7 +283,7 @@ struct xc_sr_contextsize_t pages_sent;size_t overhead_sent; -struct precopy_stats stats; +precopy_stats_t stats;unsigned int nr_batch_pfns;unsigned long *deferred_pages;
tools/libs/saverestore/save.c +3 -3
@@ -488,7 +488,7 @@ static int update_progress_string(struct xc_sr_context *ctx, char **str)#define SPP_MAX_ITERATIONS 5#define SPP_TARGET_DIRTY_COUNT 50 -static int simple_precopy_policy (struct precopy_stats stats,void *user) +static int simple_precopy_policy (precopy_stats_t stats,void *user) { return ((stats.dirty_count >= 0 && stats.dirty_count < SPP_TARGET_DIRTY_COUNT) ||
@@ -515,13 +515,13 @@ static int send_memory_live(struct xc_sr_context *ctx)precopy_policy_t precopy_policy = ctx->save.callbacks->precopy_policy;void *data = ctx->save.callbacks->data; -struct precopy_stats *policy_stats; +precopy_stats_t *policy_stats; rc =update_progress_string (ctx, &progress_str);if ( rc ) goto out; - ctx->save.stats = (struct precopy_stats ){ + ctx->save.stats = (precopy_stats_t ){ .dirty_count = ctx->save.p2m_size, }; policy_stats = &ctx->save.stats;
PATCH v20210616 30/36 tools: add callback to libxl for precopy_policy and precopy_stats_t
In reply to Olaf HeringThis duplicates simple_precopy_policy. To recap its purpose:
- do up to 5 iterations of copying dirty domU memory to target, including the initial copying of all domU memory, excluding the final copying while the domU is suspended
- do fewer iterations in case the domU dirtied less than 50 pages
Take the opportunity to also move xen_pfn_t into qw().
--- tools/libs/light/libxl_dom_save.c | 19 +++++++++++++++++++ tools/libs/light/libxl_internal.h | 2 ++ tools/libs/light/libxl_save_msgs_gen.pl | 3 ++- 3 files changed, 23 insertions(+), 1 deletion(-)
tools/libs/light/libxl_dom_save.c +19 -0
@@ -373,6 +373,24 @@ int libxl__save_emulator_xenstore_data(libxl__domain_save_state *dss,return rc; } +static int libxl__domain_save_precopy_policy (precopy_stats_t stats,void *user) +{ +libxl__save_helper_state *shs = user; +libxl__domain_save_state *dss = shs->caller_state; +STATE_AO_GC (dss->ao); + +LOGD (DEBUG , shs->domid,"iteration %u dirty_count %ld total_written %lu" , + stats.iteration, stats.dirty_count, stats.total_written); +if (stats.dirty_count >=0 && stats.dirty_count <LIBXL_XGS_POLICY_TARGET_DIRTY_COUNT ) + goto stop_copy; +if (stats.iteration >=LIBXL_XGS_POLICY_MAX_ITERATIONS ) + goto stop_copy; +return XGS_POLICY_CONTINUE_PRECOPY ; + +stop_copy: +return XGS_POLICY_STOP_AND_COPY ; +} +/*----- main code for saving, in order of execution -----*/ void libxl__domain_save (libxl__egc *egc,libxl__domain_save_state *dss)
@@ -430,6 +448,7 @@ void libxl__domain_save(libxl__egc *egc, libxl__domain_save_state *dss) callbacks->suspend = libxl__domain_suspend_callback; callbacks->switch_qemu_logdirty = libxl__domain_suspend_common_switch_qemu_logdirty; + callbacks->precopy_policy = libxl__domain_save_precopy_policy; dss->sws.ao = dss->ao; dss->sws.dss = dss;
tools/libs/light/libxl_internal.h +2 -0
@@ -125,6 +125,8 @@#define DOMID_XS_PATH "domid"#define PVSHIM_BASENAME "xen-shim"#define PVSHIM_CMDLINE "pv-shim console=xen,pv" +#define LIBXL_XGS_POLICY_MAX_ITERATIONS 5 +#define LIBXL_XGS_POLICY_TARGET_DIRTY_COUNT 50/* Size macros. */ #define __AC (X ,Y ) (X##Y)
tools/libs/light/libxl_save_msgs_gen.pl +2 -1
@@ -23,6 +23,7 @@ our @msgs = ( STRING doing_what), 'unsigned long', 'done', 'unsigned long', 'total'] ], + [ 'scxW', "precopy_policy", ['precopy_stats_t', 'stats'] ], [ 'srcxA', "suspend", [] ], [ 'srcxA', "postcopy", [] ], [ 'srcxA', "checkpoint", [] ],
@@ -142,7 +143,7 @@ static void bytes_put(unsigned char *const buf, int *len, END -foreach my $simpletype (qw(int uint16_t uint32_t unsigned), 'unsigned long', 'xen_pfn_t') { +foreach my $simpletype (qw(int uint16_t uint32_t unsigned precopy_stats_t xen_pfn_t), 'unsigned long') { my $typeid = typeid($simpletype); $out_body{'callout'} .= <<END; static int ${typeid}_get(const unsigned char **msg,
PATCH v20210616 31/36 tools: add --max_iters to libxl_domain_suspend
In reply to Olaf HeringMigrating a large, and potentially busy, domU will take more time than neccessary due to excessive number of copying iterations.
Allow to host admin to control the number of iterations which copy cumulated domU dirty pages to the target host.
The default remains 5, which means one initial iteration to copy the entire domU memory, and up to 4 additional iterations to copy dirty memory from the still running domU. After the given number of iterations the domU is suspended, remaining dirty memory is copied and the domU is finally moved to the target host.
This patch adjusts xl(1) and the libxl API.
External users check LIBXL_HAVE_DOMAIN_SUSPEND_PROPS for the availibility of the new .max_iters property.
--- docs/man/xl.1.pod.in | 4 ++++ tools/include/libxl.h | 1 + tools/libs/light/libxl_dom_save.c | 2 +- tools/libs/light/libxl_domain.c | 1 + tools/libs/light/libxl_internal.h | 1 + tools/xl/xl_cmdtable.c | 3 ++- tools/xl/xl_migrate.c | 10 +++++++++- 7 files changed, 19 insertions(+), 3 deletions(-)
docs/man/xl.1.pod.in +4 -0
@@ -494,6 +494,10 @@ such that it will be identical on the destination host, unless that configuration is overridden using the B<-C> option. Note that it is not possible to use this option for a 'localhost' migration. +=item B<--max_iters> I<iterations> + +Number of copy iterations before final suspend+move (default: 5) + =back =item B<remus> [I<OPTIONS>] I<domain-id> I<host>
tools/include/libxl.h +1 -0
@@ -1714,6 +1714,7 @@ static inline int libxl_retrieve_domain_configuration_0x041200(typedef struct {uint32_t flags;/* LIBXL_SUSPEND_* */ +uint32_t max_iters; }libxl_domain_suspend_props ;#define LIBXL_SUSPEND_DEBUG 1#define LIBXL_SUSPEND_LIVE 2
tools/libs/light/libxl_dom_save.c +1 -1
@@ -383,7 +383,7 @@ static int libxl__domain_save_precopy_policy(precopy_stats_t stats, void *user) stats.iteration, stats.dirty_count, stats.total_written);if (stats.dirty_count >=0 && stats.dirty_count <LIBXL_XGS_POLICY_TARGET_DIRTY_COUNT ) goto stop_copy; -if (stats.iteration >=LIBXL_XGS_POLICY_MAX_ITERATIONS ) +if (stats.iteration >= dss->max_iters) goto stop_copy;return XGS_POLICY_CONTINUE_PRECOPY ;
tools/libs/light/libxl_domain.c +1 -0
@@ -527,6 +527,7 @@ int libxl_domain_suspend(libxl_ctx *ctx, uint32_t domid, int fd, dss->domid = domid; dss->fd = fd; dss->type = type; + dss->max_iters = props->max_iters ?:LIBXL_XGS_POLICY_MAX_ITERATIONS ; dss->live = props->flags &LIBXL_SUSPEND_LIVE ; dss->debug = props->flags &LIBXL_SUSPEND_DEBUG ; dss->checkpointed_stream =LIBXL_CHECKPOINTED_STREAM_NONE ;
tools/libs/light/libxl_internal.h +1 -0
@@ -3649,6 +3649,7 @@ struct libxl__domain_save_state {int live;int debug;int checkpointed_stream; +uint32_t max_iters;const libxl_domain_remus_info *remus;/* private */ int rc;
tools/xl/xl_cmdtable.c +2 -1
@@ -174,7 +174,8 @@ const struct cmd_spec cmd_table[] = {" of the domain.\n" "--debug Ignored.\n" "-p Do not unpause domain after migrating it.\n" -"-D Preserve the domain id" +"-D Preserve the domain id\n" +"--max_iters N Number of copy iterations before final stop+move" }, { "restore", &main_restore, 0, 1,
tools/xl/xl_migrate.c +9 -1
@@ -178,6 +178,7 @@ static void migrate_do_preamble(int send_fd, int recv_fd, pid_t child,static void migrate_domain (uint32_t domid,int preserve_domid,const char *rune,int debug, +uint32_t max_iters,const char *override_config_file) {pid_t child =-1 ;
@@ -189,6 +190,7 @@ static void migrate_domain(uint32_t domid, int preserve_domid,int config_len;libxl_domain_suspend_props props = { .flags =LIBXL_SUSPEND_LIVE , + .max_iters = max_iters, }; save_domain_core_begin(domid, preserve_domid, override_config_file,
@@ -542,8 +544,10 @@ int main_migrate(int argc, char **argv)char *host;int opt, daemonize =1 , monitor =1 , debug =0 , pause_after_migration =0 ;int preserve_domid =0 ; +uint32_t max_iters =0 ;static struct option opts[] = { {"debug" ,0 ,0 ,0x100 }, + {"max_iters" ,1 ,0 ,0x101 }, {"live" ,0 ,0 ,0x200 },COMMON_LONG_OPTS };
@@ -571,6 +575,9 @@ int main_migrate(int argc, char **argv)case 0x100 :/* --debug */ debug =1 ;break ; +case 0x101 :/* --max_iters */ + max_iters =atoi (optarg); +break ;case 0x200 :/* --live */ /* ignored for compatibility with xm */ break ;
@@ -605,7 +612,8 @@ int main_migrate(int argc, char **argv) pause_after_migration ? " -p" : ""); } -migrate_domain (domid, preserve_domid, rune, debug, config_filename); +migrate_domain (domid, preserve_domid, rune, debug, + max_iters, config_filename);return EXIT_SUCCESS ; }
PATCH v20210616 32/36 tools: add --min_remaining to libxl_domain_suspend
In reply to Olaf HeringThe decision to stop+move a domU to the new host must be based on two factors:
- the available network bandwidth for the migration stream
- the maximum time a workload within a domU can be savely suspended
Both values define how many dirty pages a workload may produce prior the final stop+move.
The default value of 50 pages is much too low with todays network bandwidths. On an idle 1GiB link these 200K will be transferred within ~2ms.
Give the admin a knob to adjust the point when the final stop+move will be done, so he can base this decision on his own needs.
This patch adjusts xl(1) and the libxl API.
External users check LIBXL_HAVE_DOMAIN_SUSPEND_PROPS for the availibility of the new .min_remaining property.
--- docs/man/xl.1.pod.in | 8 ++++++++ tools/include/libxl.h | 1 + tools/libs/light/libxl_dom_save.c | 2 +- tools/libs/light/libxl_domain.c | 1 + tools/libs/light/libxl_internal.h | 1 + tools/xl/xl_cmdtable.c | 23 ++++++++++++----------- tools/xl/xl_migrate.c | 9 ++++++++- 7 files changed, 32 insertions(+), 13 deletions(-)
docs/man/xl.1.pod.in +8 -0
@@ -498,6 +498,14 @@ possible to use this option for a 'localhost' migration. Number of copy iterations before final suspend+move (default: 5) +=item B<--min_remaing> I<pages> + +Number of remaining dirty pages. If the number of dirty pages drops that +low, the guest is suspended and the domU will finally be moved to I<host>. + +This allows the host admin to control for how long the domU will likely +be suspended during transit. + =back =item B<remus> [I<OPTIONS>] I<domain-id> I<host>
tools/include/libxl.h +1 -0
@@ -1715,6 +1715,7 @@ static inline int libxl_retrieve_domain_configuration_0x041200(typedef struct {uint32_t flags;/* LIBXL_SUSPEND_* */ uint32_t max_iters; +uint32_t min_remaining; }libxl_domain_suspend_props ;#define LIBXL_SUSPEND_DEBUG 1#define LIBXL_SUSPEND_LIVE 2
tools/libs/light/libxl_dom_save.c +1 -1
@@ -381,7 +381,7 @@ static int libxl__domain_save_precopy_policy(precopy_stats_t stats, void *user)LOGD (DEBUG , shs->domid,"iteration %u dirty_count %ld total_written %lu" , stats.iteration, stats.dirty_count, stats.total_written); -if (stats.dirty_count >=0 && stats.dirty_count <LIBXL_XGS_POLICY_TARGET_DIRTY_COUNT ) +if (stats.dirty_count >=0 && stats.dirty_count < dss->min_remaining) goto stop_copy;if (stats.iteration >= dss->max_iters) goto stop_copy;
tools/libs/light/libxl_domain.c +1 -0
@@ -528,6 +528,7 @@ int libxl_domain_suspend(libxl_ctx *ctx, uint32_t domid, int fd, dss->fd = fd; dss->type = type; dss->max_iters = props->max_iters ?:LIBXL_XGS_POLICY_MAX_ITERATIONS ; + dss->min_remaining = props->min_remaining ?:LIBXL_XGS_POLICY_TARGET_DIRTY_COUNT ; dss->live = props->flags &LIBXL_SUSPEND_LIVE ; dss->debug = props->flags &LIBXL_SUSPEND_DEBUG ; dss->checkpointed_stream =LIBXL_CHECKPOINTED_STREAM_NONE ;
tools/libs/light/libxl_internal.h +1 -0
@@ -3650,6 +3650,7 @@ struct libxl__domain_save_state {int debug;int checkpointed_stream;uint32_t max_iters; +uint32_t min_remaining;const libxl_domain_remus_info *remus;/* private */ int rc;
tools/xl/xl_cmdtable.c +12 -11
@@ -165,17 +165,18 @@ const struct cmd_spec cmd_table[] = { &main_migrate,0 ,1 ,"Migrate a domain to another host" ,"[options] <Domain> <host>" , -"-h Print this help.\n" -"-C <config> Send <config> instead of config file from creation.\n" -"-s <sshcommand> Use <sshcommand> instead of ssh. String will be passed\n" -" to sh. If empty, run <host> instead of ssh <host> xl\n" -" migrate-receive [-d -e]\n" -"-e Do not wait in the background (on <host>) for the death\n" -" of the domain.\n" -"--debug Ignored.\n" -"-p Do not unpause domain after migrating it.\n" -"-D Preserve the domain id\n" -"--max_iters N Number of copy iterations before final stop+move" +"-h Print this help.\n" +"-C <config> Send <config> instead of config file from creation.\n" +"-s <sshcommand> Use <sshcommand> instead of ssh. String will be passed\n" +" to sh. If empty, run <host> instead of ssh <host> xl\n" +" migrate-receive [-d -e]\n" +"-e Do not wait in the background (on <host>) for the death\n" +" of the domain.\n" +"--debug Ignored.\n" +"-p Do not unpause domain after migrating it.\n" +"-D Preserve the domain id\n" +"--max_iters N Number of copy iterations before final stop+move\n" +"--min_remaining N Number of remaining dirty pages before final stop+move" }, { "restore", &main_restore, 0, 1,
tools/xl/xl_migrate.c +8 -1
@@ -179,6 +179,7 @@ static void migrate_do_preamble(int send_fd, int recv_fd, pid_t child,static void migrate_domain (uint32_t domid,int preserve_domid,const char *rune,int debug,uint32_t max_iters, +uint32_t min_remaining,const char *override_config_file) {pid_t child =-1 ;
@@ -191,6 +192,7 @@ static void migrate_domain(uint32_t domid, int preserve_domid,libxl_domain_suspend_props props = { .flags =LIBXL_SUSPEND_LIVE , .max_iters = max_iters, + .min_remaining = min_remaining, }; save_domain_core_begin(domid, preserve_domid, override_config_file,
@@ -545,9 +547,11 @@ int main_migrate(int argc, char **argv)int opt, daemonize =1 , monitor =1 , debug =0 , pause_after_migration =0 ;int preserve_domid =0 ;uint32_t max_iters =0 ; +uint32_t min_remaining =0 ;static struct option opts[] = { {"debug" ,0 ,0 ,0x100 }, {"max_iters" ,1 ,0 ,0x101 }, + {"min_remaining" ,1 ,0 ,0x102 }, {"live" ,0 ,0 ,0x200 },COMMON_LONG_OPTS };
@@ -578,6 +582,9 @@ int main_migrate(int argc, char **argv)case 0x101 :/* --max_iters */ max_iters =atoi (optarg);break ; +case 0x102 :/* --min_remaining */ + min_remaining =atoi (optarg); +break ;case 0x200 :/* --live */ /* ignored for compatibility with xm */ break ;
@@ -613,7 +620,7 @@ int main_migrate(int argc, char **argv) }migrate_domain (domid, preserve_domid, rune, debug, - max_iters, config_filename); + max_iters, min_remaining, config_filename);return EXIT_SUCCESS ; }
PATCH v20210616 33/36 tools: add --abort_if_busy to libxl_domain_suspend
In reply to Olaf HeringProvide a knob to the host admin to abort the live migration of a running domU if the downtime during final transit will be too long for the workload within domU.
Adjust error reporting. Add ERROR_MIGRATION_ABORTED to allow callers of libxl_domain_suspend to distinguish between errors and the requested constraint.
Adjust precopy_policy to simplify reporting of remaining dirty pages. The loop in send_memory_live populates ->dirty_count in a different place than ->iteration. Let it proceeed one more time to provide the desired information before leaving the loop.
This patch adjusts xl(1) and the libxl API.
External users check LIBXL_HAVE_DOMAIN_SUSPEND_PROPS for the availibility of the new .abort_if_busy property.
--- docs/man/xl.1.pod.in | 8 +++++++ tools/include/libxl.h | 1 + tools/libs/light/libxl_dom_save.c | 7 ++++++- tools/libs/light/libxl_domain.c | 1 + tools/libs/light/libxl_internal.h | 2 ++ tools/libs/light/libxl_stream_write.c | 9 +++++++- tools/libs/light/libxl_types.idl | 1 + tools/xl/xl_cmdtable.c | 6 +++++- tools/xl/xl_migrate.c | 30 ++++++++++++++++++++------- 9 files changed, 55 insertions(+), 10 deletions(-)
docs/man/xl.1.pod.in +8 -0
@@ -506,6 +506,14 @@ low, the guest is suspended and the domU will finally be moved to I<host>. This allows the host admin to control for how long the domU will likely be suspended during transit. +=item B<--abort_if_busy> + +Abort migration instead of doing final suspend/move/resume if the +guest produced more than I<min_remaining> dirty pages during th number +of I<max_iters> iterations. +This avoids long periods of time where the guest is suspended, which +may confuse the workload within domU. + =back =item B<remus> [I<OPTIONS>] I<domain-id> I<host>
tools/include/libxl.h +1 -0
@@ -1719,6 +1719,7 @@ typedef struct { } libxl_domain_suspend_props;#define LIBXL_SUSPEND_DEBUG 1#define LIBXL_SUSPEND_LIVE 2 +#define LIBXL_SUSPEND_ABORT_IF_BUSY 4 int libxl_domain_suspend(libxl_ctx *ctx, uint32_t domid, int fd, libxl_domain_suspend_props *props,
tools/libs/light/libxl_dom_save.c +6 -1
@@ -383,11 +383,16 @@ static int libxl__domain_save_precopy_policy(precopy_stats_t stats, void *user) stats.iteration, stats.dirty_count, stats.total_written);if (stats.dirty_count >=0 && stats.dirty_count < dss->min_remaining) goto stop_copy; -if (stats.iteration >= dss->max_iters) +if (stats.dirty_count >=0 && stats.iteration >= dss->max_iters) goto stop_copy;return XGS_POLICY_CONTINUE_PRECOPY ; stop_copy: +if (dss->abort_if_busy) + { + dss->remaining_dirty_pages = stats.dirty_count; +return XGS_POLICY_ABORT ; + }return XGS_POLICY_STOP_AND_COPY ; }
tools/libs/light/libxl_domain.c +1 -0
@@ -529,6 +529,7 @@ int libxl_domain_suspend(libxl_ctx *ctx, uint32_t domid, int fd, dss->type = type; dss->max_iters = props->max_iters ?:LIBXL_XGS_POLICY_MAX_ITERATIONS ; dss->min_remaining = props->min_remaining ?:LIBXL_XGS_POLICY_TARGET_DIRTY_COUNT ; + dss->abort_if_busy = props->flags &LIBXL_SUSPEND_ABORT_IF_BUSY ; dss->live = props->flags &LIBXL_SUSPEND_LIVE ; dss->debug = props->flags &LIBXL_SUSPEND_DEBUG ; dss->checkpointed_stream =LIBXL_CHECKPOINTED_STREAM_NONE ;
tools/libs/light/libxl_internal.h +2 -0
@@ -3648,9 +3648,11 @@ struct libxl__domain_save_state {libxl_domain_type type;int live;int debug; +int abort_if_busy;int checkpointed_stream;uint32_t max_iters;uint32_t min_remaining; +long remaining_dirty_pages;const libxl_domain_remus_info *remus;/* private */ int rc;
tools/libs/light/libxl_stream_write.c +8 -1
@@ -344,11 +344,18 @@ void libxl__xc_domain_save_done(libxl__egc *egc, void *dss_void, goto err;if (retval) { +if (dss->remaining_dirty_pages) { +LOGD (NOTICE , dss->domid,"saving domain: aborted," +" %ld remaining dirty pages." , dss->remaining_dirty_pages); + }else {LOGEVD (ERROR , errnoval, dss->domid,"saving domain: %s" , dss->dsps.guest_responded ?"domain responded to suspend request" :"domain did not respond to suspend request" ); -if (!dss->dsps.guest_responded) + } +if (dss->remaining_dirty_pages) + rc =ERROR_MIGRATION_ABORTED ; +else if (!dss->dsps.guest_responded) rc =ERROR_GUEST_TIMEDOUT ;else if (dss->rc) rc = dss->rc;
tools/libs/light/libxl_types.idl +1 -0
@@ -76,6 +76,7 @@ libxl_error = Enumeration("error", [ (-30, "QMP_DEVICE_NOT_ACTIVE"), # a device has failed to be become active (-31, "QMP_DEVICE_NOT_FOUND"), # the requested device has not been found (-32, "QEMU_API"), # QEMU's replies don't contains expected members + (-33, "MIGRATION_ABORTED"), ], value_namespace = "") libxl_domain_type = Enumeration("domain_type", [
tools/xl/xl_cmdtable.c +5 -1
@@ -176,7 +176,11 @@ const struct cmd_spec cmd_table[] = {"-p Do not unpause domain after migrating it.\n" "-D Preserve the domain id\n" "--max_iters N Number of copy iterations before final stop+move\n" -"--min_remaining N Number of remaining dirty pages before final stop+move" +"--min_remaining N Number of remaining dirty pages before final stop+move\n" +"--abort_if_busy Abort migration instead of doing final stop+move,\n" +" if the number of dirty pages is higher than <min_remaining>\n" +" after <max_iters> iterations. Otherwise the amount of memory\n" +" to be transfered would exceed maximum allowed domU downtime." }, { "restore", &main_restore, 0, 1,
tools/xl/xl_migrate.c +23 -7
@@ -177,7 +177,7 @@ static void migrate_do_preamble(int send_fd, int recv_fd, pid_t child, }static void migrate_domain (uint32_t domid,int preserve_domid, -const char *rune,int debug, +const char *rune,int debug,int abort_if_busy,uint32_t max_iters,uint32_t min_remaining,const char *override_config_file)
@@ -213,14 +213,20 @@ static void migrate_domain(uint32_t domid, int preserve_domid,if (debug) props.flags |=LIBXL_SUSPEND_DEBUG ; +if (abort_if_busy) + props.flags |=LIBXL_SUSPEND_ABORT_IF_BUSY ; rc =libxl_domain_suspend (ctx, domid, send_fd, &props,NULL );if (rc) {fprintf (stderr,"migration sender: libxl_domain_suspend failed" " (rc=%d)\n" , rc); -if (rc ==ERROR_GUEST_TIMEDOUT ) - goto failed_suspend; -else - goto failed_resume; +switch (rc) { +case ERROR_GUEST_TIMEDOUT : + goto failed_suspend; +case ERROR_MIGRATION_ABORTED : + goto failed_busy; +default : + goto failed_resume; + } }//fprintf(stderr, "migration sender: Transfer complete.\n");
@@ -302,6 +308,12 @@ static void migrate_domain(uint32_t domid, int preserve_domid,fprintf (stderr,"Migration failed, failed to suspend at sender.\n" );exit (EXIT_FAILURE ); + failed_busy: +close (send_fd); +migration_child_report (recv_fd); +fprintf (stderr,"Migration aborted as requested, domain is too busy.\n" ); +exit (EXIT_FAILURE ); + failed_resume:close (send_fd);migration_child_report (recv_fd);
@@ -545,13 +557,14 @@ int main_migrate(int argc, char **argv)char *rune =NULL ;char *host;int opt, daemonize =1 , monitor =1 , debug =0 , pause_after_migration =0 ; -int preserve_domid =0 ; +int preserve_domid =0 , abort_if_busy =0 ;uint32_t max_iters =0 ;uint32_t min_remaining =0 ;static struct option opts[] = { {"debug" ,0 ,0 ,0x100 }, {"max_iters" ,1 ,0 ,0x101 }, {"min_remaining" ,1 ,0 ,0x102 }, + {"abort_if_busy" ,0 ,0 ,0x103 }, {"live" ,0 ,0 ,0x200 },COMMON_LONG_OPTS };
@@ -585,6 +598,9 @@ int main_migrate(int argc, char **argv)case 0x102 :/* --min_remaining */ min_remaining =atoi (optarg);break ; +case 0x103 :/* --abort_if_busy */ + abort_if_busy =1 ; +break ;case 0x200 :/* --live */ /* ignored for compatibility with xm */ break ;
@@ -619,7 +635,7 @@ int main_migrate(int argc, char **argv) pause_after_migration ? " -p" : ""); } -migrate_domain (domid, preserve_domid, rune, debug, +migrate_domain (domid, preserve_domid, rune, debug, abort_if_busy, max_iters, min_remaining, config_filename);return EXIT_SUCCESS ; }
PATCH v20210616 34/36 tools: add API for expandable bitmaps
In reply to Olaf HeringSince the incoming migration stream lacks info about what the highest pfn will be, data structures can not be allocated upfront.
Add an API for expandable bitmaps, loosely based on pfn_set_populated.
v02:
- remove xg_ prefix from functions
--- tools/libs/saverestore/common.c | 40 ++++++++++++++++++++ tools/libs/saverestore/common.h | 67 +++++++++++++++++++++++++++++++++ 2 files changed, 107 insertions(+)
tools/libs/saverestore/common.c +40 -0
@@ -163,6 +163,46 @@ static void __attribute__((unused)) build_assertions(void)BUILD_BUG_ON (sizeof (struct xc_sr_rec_hvm_params ) !=8 ); } +/* +* Expand the tracking structures as needed. +* To avoid realloc()ing too excessively, the size increased to the nearest +* power of two large enough to contain the required number of bits. +*/ +bool _sr_bitmap_expand (struct sr_bitmap *bm,unsigned long bits) +{ +size_t new_max; +size_t old_sz, new_sz; +void *p; + +if (bits <= bm->bits) +return true; + +/* Round up to the nearest power of two larger than bit, less 1. */ + new_max = bits; + new_max |= new_max >>1 ; + new_max |= new_max >>2 ; + new_max |= new_max >>4 ; + new_max |= new_max >>8 ; + new_max |= new_max >>16 ; +if (sizeof (unsigned long ) >4 ) + new_max |= new_max >>32 ; + +/* Allocate units of unsigned long */ + new_max = (new_max +BITS_PER_LONG -1 ) & ~(BITS_PER_LONG -1 ); + + old_sz =bitmap_size (bm->bits); + new_sz =bitmap_size (new_max); + p =realloc (bm->p, new_sz); +if (!p) +return false; + +memset (p + old_sz,0 , new_sz - old_sz); + bm->p = p; + bm->bits = new_max; + +return true; +} +/* * Local variables: * mode: C
tools/libs/saverestore/common.h +67 -0
73 lines not on this page · raw message
PATCH v20210616 35/36 tools: use sr_bitmap for populated_pfns
In reply to Olaf Heringv02:
- remove xg_ prefix from called functions
--- tools/libs/saverestore/common.h | 21 +++++++- tools/libs/saverestore/restore.c | 69 ------------------------ tools/libs/saverestore/restore_x86_hvm.c | 9 ++++ tools/libs/saverestore/restore_x86_pv.c | 7 +++ 4 files changed, 35 insertions(+), 71 deletions(-)
tools/libs/saverestore/common.h +19 -2
@@ -403,8 +403,7 @@ struct xc_sr_contextuint32_t xenstore_domid, console_domid;/* Bitmap of currently populated PFNs during restore. */ -unsigned long *populated_pfns; -xen_pfn_t max_populated_pfn; +struct sr_bitmap populated_pfns;/* Sender has invoked verify mode on the stream. */ bool verify;
@@ -629,6 +628,24 @@ static inline bool page_type_has_stream_data(uint32_t type) }return ret; } + +static inline bool pfn_is_populated (struct xc_sr_context *ctx,xen_pfn_t pfn) +{ +return sr_test_bit (pfn, &ctx->restore.populated_pfns); +} + +static inline int pfn_set_populated (struct xc_sr_context *ctx,xen_pfn_t pfn) +{ +xc_interface *xch = ctx->xch; + +if (sr_set_bit (pfn, &ctx->restore.populated_pfns) == false ) + { +PERROR ("Failed to realloc populated_pfns bitmap" ); + errno =ENOMEM ; +return -1 ; + } +return 0 ; +}#endif /* * Local variables:
tools/libs/saverestore/restore.c +0 -69
87 lines not on this page · raw message
tools/libs/saverestore/restore_x86_hvm.c +9 -0
@@ -136,6 +136,7 @@ static int x86_hvm_localise_page(struct xc_sr_context *ctx,static int x86_hvm_setup (struct xc_sr_context *ctx) {xc_interface *xch = ctx->xch; +unsigned long max_pfn;if ( ctx->restore.guest_type !=DHDR_TYPE_X86_HVM ) {
@@ -161,6 +162,13 @@ static int x86_hvm_setup(struct xc_sr_context *ctx) }#endif + max_pfn =max (ctx->restore.p2m_size, ctx->dominfo.max_memkb >> (PAGE_SHIFT -10 )); +if ( !sr_bitmap_expand (&ctx->restore.populated_pfns, max_pfn) ) + { +PERROR ("Unable to allocate memory for populated_pfns bitmap" ); +return -1 ; + } +return 0 ; }
@@ -241,6 +249,7 @@ static int x86_hvm_stream_complete(struct xc_sr_context *ctx)static int x86_hvm_cleanup (struct xc_sr_context *ctx) { +sr_bitmap_free (&ctx->restore.populated_pfns);free (ctx->x86.hvm.restore.context.ptr);free (ctx->x86.restore.cpuid.ptr);
tools/libs/saverestore/restore_x86_pv.c +7 -0
@@ -1060,6 +1060,12 @@ static int x86_pv_setup(struct xc_sr_context *ctx)if ( rc )return rc; +if ( !sr_bitmap_expand (&ctx->restore.populated_pfns,32 *1024 /4 ) ) + { +PERROR ("Unable to allocate memory for populated_pfns bitmap" ); +return -1 ; + } + ctx->x86.pv.restore.nr_vcpus = ctx->dominfo.max_vcpu_id +1 ; ctx->x86.pv.restore.vcpus =calloc (sizeof (struct xc_sr_x86_pv_restore_vcpu ), ctx->x86.pv.restore.nr_vcpus);
@@ -1153,6 +1159,7 @@ static int x86_pv_stream_complete(struct xc_sr_context *ctx)*/ static int x86_pv_cleanup (struct xc_sr_context *ctx) { +sr_bitmap_free (&ctx->restore.populated_pfns);free (ctx->x86.pv.p2m);free (ctx->x86.pv.p2m_pfns);
PATCH v20210616 36/36 tools: use superpages during restore of HVM guest
In reply to Olaf HeringDuring creating of a HVM domU meminit_hvm() tries to map superpages. After save/restore or migration this mapping is lost, everything is allocated in single pages. This causes a performance degradation after migration.
Add neccessary code to preallocate a superpage for an incoming chunk of pfns. In case a pfn was not populated on the sending side, it must be freed on the receiving side to avoid over-allocation.
The existing code for x86_pv is moved unmodified into its own file.
v02:
- remove xg_ prefix from called functions
--- tools/libs/guest/xg_dom_x86.c | 5 - tools/libs/guest/xg_private.h | 5 + tools/libs/saverestore/common.c | 1 - tools/libs/saverestore/common.h | 28 +- tools/libs/saverestore/restore.c | 62 +--- tools/libs/saverestore/restore_x86_hvm.c | 370 ++++++++++++++++++++++- tools/libs/saverestore/restore_x86_pv.c | 61 +++- 7 files changed, 455 insertions(+), 77 deletions(-)
tools/libs/guest/xg_dom_x86.c +0 -5
@@ -44,11 +44,6 @@#define SUPERPAGE_BATCH_SIZE 512 -#define SUPERPAGE_2MB_SHIFT 9 -#define SUPERPAGE_2MB_NR_PFNS (1UL << SUPERPAGE_2MB_SHIFT) -#define SUPERPAGE_1GB_SHIFT 18 -#define SUPERPAGE_1GB_NR_PFNS (1UL << SUPERPAGE_1GB_SHIFT) -#define X86_CR0_PE 0x01#define X86_CR0_ET 0x10
tools/libs/guest/xg_private.h +5 -0
@@ -179,4 +179,9 @@ struct xc_cpu_policy { };#endif /* x86 */ +#define SUPERPAGE_2MB_SHIFT 9 +#define SUPERPAGE_2MB_NR_PFNS (1UL << SUPERPAGE_2MB_SHIFT) +#define SUPERPAGE_1GB_SHIFT 18 +#define SUPERPAGE_1GB_NR_PFNS (1UL << SUPERPAGE_1GB_SHIFT) +#endif /* XG_PRIVATE_H */
tools/libs/saverestore/common.c +0 -1
@@ -1,5 +1,4 @@#include <assert.h> -#include "common.h" #include <xen-tools/libs.h>
tools/libs/saverestore/common.h +20 -8
@@ -219,6 +219,16 @@ struct xc_sr_restore_ops*/ int (*setup)(struct xc_sr_context *ctx); +/** +* Populate PFNs +* +* Given a set of pfns, obtain memory from Xen to fill the physmap for the +* unpopulated subset. +*/ +int (*populate_pfns)(struct xc_sr_context *ctx,unsigned count, +const xen_pfn_t *original_pfns,const uint32_t *types); + +/** * Process an individual record from the stream. The caller shall take * care of processing common records (e.g. END, PAGE_DATA).
@@ -366,6 +376,8 @@ struct xc_sr_contextint send_back_fd;unsigned long p2m_size; +unsigned long max_pages; +unsigned long tot_pages;xc_hypercall_buffer_t dirty_bitmap_hbuf;/* From Image Header. */
@@ -503,6 +515,14 @@ struct xc_sr_context {/* HVM context blob. */ struct xc_sr_blob context; + +/* Bitmap of currently allocated PFNs during restore. */ +struct sr_bitmap attempted_1g; +struct sr_bitmap attempted_2m; +struct sr_bitmap allocated_pfns; +xen_pfn_t prev_populated_pfn; +xen_pfn_t iteration_tracker_pfn; +unsigned long iteration; } restore; }; } hvm;
@@ -567,14 +587,6 @@ int read_record_header(struct xc_sr_context *ctx, int fd, struct xc_sr_rhdr *rhdint read_record_data (struct xc_sr_context *ctx,int fd,struct xc_sr_rhdr *rhdr,struct xc_sr_record *rec); -/* -* This would ideally be private in restore.c, but is needed by -* x86_pv_localise_page() if we receive pagetables frames ahead of the -* contents of the frames they point at. -*/ -int populate_pfns (struct xc_sr_context *ctx,unsigned int count, -const xen_pfn_t *original_pfns,const uint32_t *types); -/* Handle a STATIC_DATA_END record. */ int handle_static_data_end (struct xc_sr_context *ctx);
tools/libs/saverestore/restore.c +4 -58
80 lines not on this page · raw message
tools/libs/saverestore/restore_x86_hvm.c +366 -4
402 lines not on this page · raw message
tools/libs/saverestore/restore_x86_pv.c +60 -1
79 lines not on this page · raw message
Quoted 1 line · Olaf Hering
Various unreviewed changes, rebase to
4bcf6433ee.
General CI run:
https://gitlab.com/xen-project/patchew/xen/-/pipelines/322032419
Some specific failures.
https://gitlab.com/xen-project/patchew/xen/-/jobs/1351977567 (32bit toolstack build):
common.c: In function '_sr_bitmap_expand':
common.c:187:9: error: right shift count >= width of type [-Werror] new_max |= new_max >> 32;
^
https://gitlab.com/xen-project/patchew/xen/-/jobs/1351977708 (arm64)
nomigrate.c:25:20: error: unknown type name 'xc_stream_type_t' 25 | xc_stream_type_t stream_type, int recv_fd) | ^~~~~~~~~~~~~~~~
I haven't looked through all the failures in the general run, but be aware that there might still be some clang fallout in dom0_build.c in Xen, and PV32 fallout for the smoke tests, which won't be from your series.
~Andrew
Am Wed, 16 Jun 2021 15:50:24 +0100
schrieb Andrew Cooper <[email protected]>:
Quoted 1 line
new_max|=new_max>> 32;
Lazy compiler? I hoped this is a compile-time constant, which evaluates to zero in 32bit builds.
if ( sizeof(unsigned long) > 4 )
I guess a #ifdef as it is done in old code must be done.
Olaf
Hi Olaf,
Quoted 6 lines · Olaf Hering
Am Wed, 16 Jun 2021 15:50:24 +0100
schrieb Andrew Cooper <[email protected]>:Quoted 1 line
new_max|=new_max>> 32;Lazy compiler? I hoped this is a compile-time constant, which evaluates to zero in 32bit builds.
See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=4210, this seems to be a known issue on GCC.
Quoted 7 lines
if ( sizeof(unsigned long) > 4 )
I guess a #ifdef as it is done in old code must be done.
Olaf
Cheers,
Signature · 1 line
Julien Grall
Am Thu, 17 Jun 2021 13:02:39 +0200
schrieb Julien Grall <[email protected]>:
Quoted 1 line
Thanks, comments in this bug suggest a workaround like this:
new_max |= sizeof(unsigned long) > 4 ? new_max >> 32 : 0;
This triggers no warning.
Olaf
Am Wed, 16 Jun 2021 15:50:24 +0100
schrieb Andrew Cooper <[email protected]>:
Quoted 1 line
32bit toolstack build
as in i386?
How is this used in practice?
I guess such build should be marked as CONFIG_MIGRATE=n in config/x86_32.mk?
Olaf
Quoted 6 lines · Olaf Hering
Am Wed, 16 Jun 2021 15:50:24 +0100
schrieb Andrew Cooper <[email protected]>:Quoted 1 line
32bit toolstack build
as in i386?
How is this used in practice?
Every OSSTest run. Also, arm32 is absolutely a thing (the only reason ARM can't migrate right now is because there is no logdirty support in Xen yet).
Quoted 1 line
I guess such build should be marked as
CONFIG_MIGRATE=ninconfig/x86_32.mk?
Migration (v2) very definitely works for i386 toolstacks. Part of the testing process during development was migrating a VM between a 32bit and 64bit dom0's, specifically to check that we'd got rid of all of the bitness problems in the stream format.
~Andrew
Am Thu, 17 Jun 2021 12:24:22 +0100
schrieb Andrew Cooper <[email protected]>:
Quoted 8 lines · Olaf Hering
Quoted 6 lines
Am Wed, 16 Jun 2021 15:50:24 +0100
schrieb Andrew Cooper <[email protected]>:Quoted 1 line
32bit toolstack build
as in i386?
How is this used in practice?Every OSSTest run.
This is not what I mean.
I think there is a 32bit xen-tools, a 32bit dom0 kernel and a 64bit Xen? Is 32bit xen-tools, 64bit dom0 kernel and 64bit Xen expected to work?
Olaf
Quoted 7 lines · Olaf Hering
Am Thu, 17 Jun 2021 12:24:22 +0100
schrieb Andrew Cooper <[email protected]>:Quoted 4 lines · Olaf Hering
Quoted 3 lines
Am Wed, 16 Jun 2021 15:50:24 +0100
schrieb Andrew Cooper <[email protected]>:
Quoted 6 lines
Quoted 4 lines
Quoted 3 lines
Quoted 1 line
32bit toolstack build
as in i386?
How is this used in practice?Every OSSTest run.
This is not what I mean.
I think there is a 32bit xen-tools, a 32bit dom0 kernel and a 64bit Xen?
Yes - this exists.
Quoted 1 line
Is 32bit xen-tools, 64bit dom0 kernel and 64bit Xen expected to work?
In an ideal world, yes. In reality, no.
Lots of hypercalls have embedded pointers (every GUEST_HANDLE(), to a first approximation), and dom0's ABI with Xen is 64bit, which is not the ABI that 32bit userspace speaks.
This is one of many errors in the hypercall design intending to be addressed by the ABIv2 plans.
~Andrew