replay additions
v7 cover replay additions
This set of patches moves replay lock upper in the function call tree. Now replay lock functions similar to BQL in older version and allows deterministic execution of the threads in icount mode.
It is also fixes some vmstate creation (and loading) issues in record/replay modes:
- VM start/stop fixes in replay mode
- overlay creation for blkreplay filter
- fixes for vmstate save/load in record/replay mode
- fixes for host clock vmstate
There is also a set of helper scripts written by Alex Bennée for debugging the record/replay code.
v6 patches with updates for v7 are available in the repository:
https://github.com/ispras/qemu/tree/rr-180207
v7 changes:
- updated record/replay documentation
- removed abort() from mutex stub functions
- fixed cpu_io_recompile function
v6 changes:
- removed BQL optimization at all
- refined replay lock patches
- removed lock/unlock from replay-audio
v5 changes:
- removed patch for narrowing BQL-protected code
- disabled bdrv_(drain/flush)_all for record/replay mode
v4 changes:
- removed upstreamed patches
- added patch for saving async queue state in replay
- minor fixes
v3 changes:
- removed upstreamed patches
- fixed bug with recursive checkpoints
- fixed bug with icount warp checkpoint
v2 changes:
- updated lock/unlock logic (as suggested by Paolo Bonzini)
- updated cpu execution loop to avoid races in setting/resetting exit request (as suggested by Paolo Bonzini)
- minor changes
---
Alex Bennée (5):
replay/replay.c: bump REPLAY_VERSION again replay/replay-internal.c: track holding of replay_lock
replay: make locking visible outside replay code
replay: push replay_mutex_lock up the call tree scripts/replay-dump.py: replay log dumper
Pavel Dovgalyuk (17):
cpu-exec: fix exception_index handling
block: implement bdrv_snapshot_goto for blkreplay
blkreplay: create temporary overlay for underlaying devices
replay: disable default snapshot for record/replay
replay: fix processing async events
replay: fixed replay_enable_events
replay: fix save/load vm for non-empty queue
replay: added replay log format description
replay: save prior value of the host clock
replay: don't destroy mutex at exit
replay: check return values of fwrite
replay: avoid recursive call of checkpoints
replay: don't process async events when warping the clock
replay: save vmstate of the asynchronous events
replay: don't drain/flush bdrv queue while RR is working
replay: update documentation
tcg: fix cpu_io_recompile
accel/tcg/cpu-exec.c | 5 + accel/tcg/translate-all.c | 18 ++- block/blkreplay.c | 75 +++++++++++ block/io.c | 22 +++ cpus.c | 26 +++- docs/replay.txt | 163 +++++++++++++++++++++--- include/qemu/timer.h | 14 ++ include/sysemu/replay.h | 18 +++ migration/savevm.c | 13 ++ replay/replay-audio.c | 14 +- replay/replay-char.c | 21 +-- replay/replay-events.c | 75 +++++------ replay/replay-internal.c | 47 ++++++- replay/replay-internal.h | 16 ++ replay/replay-snapshot.c | 12 ++ replay/replay-time.c | 10 + replay/replay.c | 62 ++++++--- scripts/replay-dump.py | 308 +++++++++++++++++++++++++++++++++++++++++++++ stubs/replay.c | 9 + util/main-loop.c | 15 ++ util/qemu-timer.c | 12 ++ vl.c | 12 +- 22 files changed, 831 insertions(+), 136 deletions(-) create mode 100755 scripts/replay-dump.py
Signature · 1 line
Pavel Dovgalyuk
PATCH v7 1/22 cpu-exec: fix exception_index handling
Function cpu_handle_interrupt calls cc->cpu_exec_interrupt to process pending hardware interrupts. Under the hood cpu_exec_interrupt uses cpu->exception_index to pass information to the internal function which is usually common for exception and interrupt processing.
But this value is not reset after return and may be processed again by cpu_handle_exception. This does not happen due to overwriting the exception_index at the end of cpu_handle_interrupt.
But this branch may also overwrite the valid exception_index in some cases. Therefore this patch:
1. resets exception_index just after the call to cpu_exec_interrupt
2. prevents overwriting the meaningful value of exception_index
--- accel/tcg/cpu-exec.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
accel/tcg/cpu-exec.c +4 -1
@@ -585,6 +585,7 @@ static inline bool cpu_handle_interrupt(CPUState *cpu,else {if (cc->cpu_exec_interrupt (cpu, interrupt_request)) {replay_interrupt (); + cpu->exception_index =-1 ; *last_tb =NULL ; }/* The target hook may have updated the 'cpu->interrupt_request';
@@ -606,7 +607,9 @@ static inline bool cpu_handle_interrupt(CPUState *cpu,if (unlikely (atomic_read (&cpu->exit_request) || (use_icount && cpu->icount_decr.u16.low + cpu->icount_extra ==0 ))) {atomic_set (&cpu->exit_request,0 ); - cpu->exception_index =EXCP_INTERRUPT ; +if (cpu->exception_index ==-1 ) { + cpu->exception_index =EXCP_INTERRUPT ; + }return true; }
PATCH v7 2/22 block: implement bdrv_snapshot_goto for blkreplay
In reply to Pavel DovgalyukFrom: Pavel Dovgalyuk <[email protected]>
This patch enables making snapshots with blkreplay used in block devices.
This function is required to make bdrv_snapshot_goto without calling .bdrv_open which is not implemented.
--- block/blkreplay.c | 8 ++++++++ 1 file changed, 8 insertions(+)
block/blkreplay.c +8 -0
@@ -127,6 +127,12 @@ static int coroutine_fn blkreplay_co_flush(BlockDriverState *bs)return ret; } +static int blkreplay_snapshot_goto (BlockDriverState *bs, +const char *snapshot_id) +{ +return bdrv_snapshot_goto (bs->file->bs, snapshot_id,NULL ); +} +static BlockDriver bdrv_blkreplay = { .format_name ="blkreplay" , .protocol_name ="blkreplay" ,
@@ -143,6 +149,8 @@ static BlockDriver bdrv_blkreplay = { .bdrv_co_pwrite_zeroes = blkreplay_co_pwrite_zeroes, .bdrv_co_pdiscard = blkreplay_co_pdiscard, .bdrv_co_flush = blkreplay_co_flush, + + .bdrv_snapshot_goto = blkreplay_snapshot_goto, };static void bdrv_blkreplay_init (void )
PATCH v7 3/22 blkreplay: create temporary overlay for underlaying devices
In reply to Pavel DovgalyukFrom: Pavel Dovgalyuk <[email protected]>
This patch allows using '-snapshot' behavior in record/replay mode. blkreplay layer creates temporary overlays on top of underlaying disk images. It is needed, because creating an overlay over blkreplay breaks the determinism.
This patch creates similar temporary overlay (when it is needed) under the blkreplay driver. Therefore all block operations are controlled by blkreplay.
--- block/blkreplay.c | 67 +++++++++++++++++++++++++++++++++++++++++++++++++++++ stubs/replay.c | 1 + vl.c | 2 +- 3 files changed, 69 insertions(+), 1 deletion(-)
block/blkreplay.c +67 -0
@@ -14,12 +14,71 @@#include "block/block_int.h" #include "sysemu/replay.h" #include "qapi/error.h" +#include "qapi/qmp/qstring.h" +#include "qapi/qmp/qdict.h" +#include "qemu/option.h" typedef struct Request {Coroutine *co;QEMUBH *bh; }Request ; +static BlockDriverState *blkreplay_append_snapshot (BlockDriverState *bs, +Error **errp) +{ +int ret; +BlockDriverState *bs_snapshot; +int64_t total_size; +QemuOpts *opts =NULL ; +char tmp_filename[PATH_MAX +1 ]; +QDict *snapshot_options =qdict_new (); + +/* Prepare options QDict for the overlay file */ +qdict_put (snapshot_options,"file.driver" ,qstring_from_str ("file" )); +qdict_put (snapshot_options,"driver" ,qstring_from_str ("qcow2" )); + +/* Create temporary file */ + ret =get_tmp_filename (tmp_filename,PATH_MAX +1 ); +if (ret <0 ) { +error_setg_errno (errp, -ret,"Could not get temporary filename" ); + goto out; + } +qdict_put (snapshot_options,"file.filename" , +qstring_from_str (tmp_filename)); + +/* Get the required size from the image */ + total_size =bdrv_getlength (bs); +if (total_size <0 ) { +error_setg_errno (errp, -total_size,"Could not get image size" ); + goto out; + } + + opts =qemu_opts_create (bdrv_qcow2.create_opts,NULL ,0 , &error_abort); +qemu_opt_set_number (opts,BLOCK_OPT_SIZE , total_size, &error_abort); + ret =bdrv_create (&bdrv_qcow2, tmp_filename, opts, errp); +qemu_opts_del (opts); +if (ret <0 ) { +error_prepend (errp,"Could not create temporary overlay '%s': " , + tmp_filename); + goto out; + } + + bs_snapshot =bdrv_open (NULL ,NULL , snapshot_options, +BDRV_O_RDWR |BDRV_O_TEMPORARY , errp); + snapshot_options =NULL ; +if (!bs_snapshot) { + goto out; + } + +bdrv_append (bs_snapshot, bs, errp); + +return bs_snapshot; + +out: +QDECREF (snapshot_options); +return NULL ; +} +static int blkreplay_open (BlockDriverState *bs,QDict *options,int flags,Error **errp) {
@@ -35,6 +94,14 @@ static int blkreplay_open(BlockDriverState *bs, QDict *options, int flags, goto fail; } +/* Add temporary snapshot to preserve the image */ +if (!replay_snapshot + && !blkreplay_append_snapshot (bs->file->bs, &local_err)) { + ret = -EINVAL ; +error_propagate (errp, local_err); + goto fail; + } + ret =0 ; fail:return ret;
stubs/replay.c +1 -0
@@ -3,6 +3,7 @@#include "sysemu/sysemu.h" ReplayMode replay_mode; +char *replay_snapshot;int64_t replay_save_clock (unsigned int kind,int64_t clock) {
vl.c +1 -1
@@ -4523,7 +4523,7 @@ int main(int argc, char **argv, char **envp)qapi_free_BlockdevOptions (bdo->bdo);g_free (bdo); } -if (snapshot || replay_mode !=REPLAY_MODE_NONE ) { +if (snapshot) {qemu_opts_foreach (qemu_find_opts ("drive" ), drive_enable_snapshot,NULL ,NULL ); }
PATCH v7 4/22 replay: disable default snapshot for record/replay
In reply to Pavel DovgalyukFrom: Pavel Dovgalyuk <[email protected]>
This patch disables setting '-snapshot' option on by default in record/replay mode. This is needed for creating vmstates in record and replay modes.
--- vl.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)
vl.c +7 -1
@@ -3221,7 +3221,13 @@ int main(int argc, char **argv, char **envp)drive_add (IF_PFLASH ,-1 , optarg,PFLASH_OPTS );break ;case QEMU_OPTION_snapshot: - snapshot =1 ; + { +Error *blocker =NULL ; + snapshot =1 ; +error_setg (&blocker,QERR_REPLAY_NOT_SUPPORTED , +"-snapshot" ); +replay_add_blocker (blocker); + }break ;case QEMU_OPTION_numa: opts = qemu_opts_parse_noisily(qemu_find_opts("numa"),
PATCH v7 5/22 replay: fix processing async events
In reply to Pavel DovgalyukAsynchronous events saved at checkpoints may invoke callbacks when processed. These callbacks may also generate/read new events (e.g. clock reads). Therefore event processing flag must be reset before callback invocation.
--- replay/replay-events.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
replay/replay-events.c +2 -2
@@ -295,13 +295,13 @@ void replay_read_events(int checkpoint)if (!event) {break ; } +replay_finish_event (); + read_event_kind =-1 ;replay_mutex_unlock ();replay_run_event (event);replay_mutex_lock ();g_free (event); -replay_finish_event (); - read_event_kind =-1 ; } }
PATCH v7 6/22 replay: fixed replay_enable_events
In reply to Pavel DovgalyukThis patch fixes assignment to internal events_enabled variable. Now it is set only in record/replay mode. This affects the behavior of the external functions that check this flag.
--- replay/replay-events.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
replay/replay-events.c +5 -3
@@ -67,7 +67,9 @@ static void replay_run_event(Event *event)void replay_enable_events (void ) { - events_enabled = true; +if (replay_mode !=REPLAY_MODE_NONE ) { + events_enabled = true; + } }bool replay_has_events (void )
@@ -141,7 +143,7 @@ void replay_add_event(ReplayAsyncEventKind event_kind,void replay_bh_schedule_event (QEMUBH *bh) { -if (replay_mode !=REPLAY_MODE_NONE && events_enabled) { +if (events_enabled) {uint64_t id =replay_get_current_step ();replay_add_event (REPLAY_ASYNC_EVENT_BH , bh,NULL , id); }else {
@@ -161,7 +163,7 @@ void replay_add_input_sync_event(void)void replay_block_event (QEMUBH *bh,uint64_t id) { -if (replay_mode !=REPLAY_MODE_NONE && events_enabled) { +if (events_enabled) {replay_add_event (REPLAY_ASYNC_EVENT_BLOCK , bh,NULL , id); }else {qemu_bh_schedule (bh);
PATCH v7 7/22 replay: fix save/load vm for non-empty queue
In reply to Pavel DovgalyukThis patch does not allows saving/loading vmstate when replay events queue is not empty. There is no reliable way to save events queue, because it describes internal coroutine state. Therefore saving and loading operations should be deferred to another record/replay step.
--
v2: fixed error_report calls
--- include/sysemu/replay.h | 3 +++ migration/savevm.c | 13 +++++++++++++ replay/replay-snapshot.c | 6 ++++++ 3 files changed, 22 insertions(+)
include/sysemu/replay.h +3 -0
@@ -164,5 +164,8 @@ void replay_audio_in(int *recorded, void *samples, int *wpos, int size);/*! Called at the start of execution. Loads or saves initial vmstate depending on execution mode. */ void replay_vmstate_init (void ); +/*! Called to ensure that replay state is consistent and VM snapshot +can be created */ +bool replay_can_snapshot (void );#endif
migration/savevm.c +13 -0
@@ -53,6 +53,7 @@#include "qemu/cutils.h" #include "io/channel-buffer.h" #include "io/channel-file.h" +#include "sysemu/replay.h" #ifndef ETH_P_RARP #define ETH_P_RARP 0x8035
@@ -2196,6 +2197,12 @@ int save_snapshot(const char *name, Error **errp)struct tm tm;AioContext *aio_context; +if (!replay_can_snapshot ()) { +error_report ("Record/replay does not allow making snapshot " +"right now. Try once more later." ); +return ret; + } +if (!bdrv_all_can_snapshot (&bs)) {error_setg (errp,"Device '%s' is writable but does not support " "snapshots" ,bdrv_get_device_name (bs));
@@ -2387,6 +2394,12 @@ int load_snapshot(const char *name, Error **errp)AioContext *aio_context;MigrationIncomingState *mis =migration_incoming_get_current (); +if (!replay_can_snapshot ()) { +error_report ("Record/replay does not allow loading snapshot " +"right now. Try once more later." ); +return -EINVAL ; + } +if (!bdrv_all_can_snapshot (&bs)) { error_setg(errp, "Device '%s' is writable but does not support snapshots",
replay/replay-snapshot.c +6 -0
@@ -83,3 +83,9 @@ void replay_vmstate_init(void) } } } + +bool replay_can_snapshot (void ) +{ +return replay_mode ==REPLAY_MODE_NONE + || !replay_has_events (); +}
PATCH v7 8/22 replay: added replay log format description
In reply to Pavel DovgalyukFrom: Pavel Dovgalyuk <[email protected]>
This patch adds description of the replay log file format into the docs/replay.txt.
--- docs/replay.txt | 69 +++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 69 insertions(+)
docs/replay.txt +69 -0
@@ -232,3 +232,72 @@ Audio devices Audio data is recorded and replay automatically. The command line for recording and replaying must contain identical specifications of audio hardware, e.g.: -soundhw ac97 + +Replay log format +----------------- + +Record/replay log consits of the header and the sequence of execution +events. The header includes 4-byte replay version id and 8-byte reserved +field. Version is updated every time replay log format changes to prevent +using replay log created by another build of qemu. + +The sequence of the events describes virtual machine state changes. +It includes all non-deterministic inputs of VM, synchronization marks and +instruction counts used to correctly inject inputs at replay. + +Synchronization marks (checkpoints) are used for synchronizing qemu threads +that perform operations with virtual hardware. These operations may change +system's state (e.g., change some register or generate interrupt) and +therefore should execute synchronously with CPU thread. + +Every event in the log includes 1-byte event id and optional arguments. +When argument is an array, it is stored as 4-byte array length +and corresponding number of bytes with data. +Here is the list of events that are written into the log: + + - EVENT_INSTRUCTION. Instructions executed since last event. + Argument: 4-byte number of executed instructions. + - EVENT_INTERRUPT. Used to synchronize interrupt processing. + - EVENT_EXCEPTION. Used to synchronize exception handling. + - EVENT_ASYNC. This is a group of events. They are always processed + together with checkpoints. When such an event is generated, it is + stored in the queue and processed only when checkpoint occurs. + Every such event is followed by 1-byte checkpoint id and 1-byte + async event id from the following list: + - REPLAY_ASYNC_EVENT_BH. Bottom-half callback. This event synchronizes + callbacks that affect virtual machine state, but normally called + asyncronously. + Argument: 8-byte operation id. + - REPLAY_ASYNC_EVENT_INPUT. Input device event. Contains + parameters of keyboard and mouse input operations + (key press/release, mouse pointer movement). + Arguments: 9-16 bytes depending of input event. + - REPLAY_ASYNC_EVENT_INPUT_SYNC. Internal input synchronization event. + - REPLAY_ASYNC_EVENT_CHAR_READ. Character (e.g., serial port) device input + initiated by the sender. + Arguments: 1-byte character device id. + Array with bytes were read. + - REPLAY_ASYNC_EVENT_BLOCK. Block device operation. Used to synchronize + operations with disk and flash drives with CPU. + Argument: 8-byte operation id. + - REPLAY_ASYNC_EVENT_NET. Incoming network packet. + Arguments: 1-byte network adapter id. + 4-byte packet flags. + Array with packet bytes. + - EVENT_SHUTDOWN. Occurs when user sends shutdown event to qemu, + e.g., by closing the window. + - EVENT_CHAR_WRITE. Used to synchronize character output operations. + Arguments: 4-byte output function return value. + 4-byte offset in the output array. + - EVENT_CHAR_READ_ALL. Used to synchronize character input operations, + initiated by qemu. + Argument: Array with bytes that were read. + - EVENT_CHAR_READ_ALL_ERROR. Unsuccessful character input operation, + initiated by qemu. + Argument: 4-byte error code. + - EVENT_CLOCK + clock_id. Group of events for host clock read operations. + Argument: 8-byte clock value. + - EVENT_CHECKPOINT + checkpoint_id. Checkpoint for synchronization of + CPU, internal threads, and asynchronous input events. May be followed + by one or more EVENT_ASYNC events. + - EVENT_END. Last event in the log.
PATCH v7 9/22 replay: save prior value of the host clock
In reply to Pavel DovgalyukThis patch adds saving/restoring of the host clock field 'last'. It is used in host clock calculation and therefore clock may become incorrect when using restored vmstate.
--- include/qemu/timer.h | 14 ++++++++++++++ replay/replay-internal.h | 2 ++ replay/replay-snapshot.c | 3 +++ util/qemu-timer.c | 12 ++++++++++++ 4 files changed, 31 insertions(+)
include/qemu/timer.h +14 -0
@@ -251,6 +251,20 @@ bool qemu_clock_run_timers(QEMUClockType type);*/ bool qemu_clock_run_all_timers (void ); +/** +* qemu_clock_get_last: +* +* Returns last clock query time. +*/ +uint64_t qemu_clock_get_last (QEMUClockType type); +/** +* qemu_clock_set_last: +* +* Sets last clock query time. +*/ +void qemu_clock_set_last (QEMUClockType type,uint64_t last); + +/* * QEMUTimerList */
replay/replay-internal.h +2 -0
@@ -78,6 +78,8 @@ typedef struct ReplayState {This counter is global, because requests from different block devices should not get overlapping ids. */ uint64_t block_request_id; +/*! Prior value of the host clock */ +uint64_t host_clock_last; } ReplayState;extern ReplayState replay_state;
replay/replay-snapshot.c +3 -0
@@ -25,6 +25,7 @@ static int replay_pre_save(void *opaque) {ReplayState *state = opaque; state->file_offset =ftell (replay_file); + state->host_clock_last =qemu_clock_get_last (QEMU_CLOCK_HOST );return 0 ; }
@@ -33,6 +34,7 @@ static int replay_post_load(void *opaque, int version_id) {ReplayState *state = opaque;fseek (replay_file, state->file_offset,SEEK_SET ); +qemu_clock_set_last (QEMU_CLOCK_HOST , state->host_clock_last);/* If this was a vmstate, saved in recording mode, we need to initialize replay data fields. */ replay_fetch_data_kind ();
@@ -54,6 +56,7 @@ static const VMStateDescription vmstate_replay = { VMSTATE_UINT32(has_unread_data, ReplayState), VMSTATE_UINT64(file_offset, ReplayState), VMSTATE_UINT64(block_request_id, ReplayState), + VMSTATE_UINT64(host_clock_last, ReplayState),VMSTATE_END_OF_LIST () }, };
util/qemu-timer.c +12 -0
@@ -622,6 +622,18 @@ int64_t qemu_clock_get_ns(QEMUClockType type) } } +uint64_t qemu_clock_get_last (QEMUClockType type) +{ +QEMUClock *clock =qemu_clock_ptr (type); +return clock->last; +} + +void qemu_clock_set_last (QEMUClockType type,uint64_t last) +{ +QEMUClock *clock =qemu_clock_ptr (type); + clock->last = last; +} +void qemu_clock_register_reset_notifier (QEMUClockType type,Notifier *notifier) {
PATCH v7 11/22 replay/replay-internal.c: track holding of replay_lock
In reply to Pavel DovgalyukFrom: Alex Bennée <[email protected]>
This is modelled after the iothread mutex lock. We keep a TLS flag to indicate when that thread has acquired the lock and assert we don't double-lock or release when we shouldn't have.
--- replay/replay-internal.c | 11 +++++++++++ 1 file changed, 11 insertions(+)
replay/replay-internal.c +11 -0
@@ -169,6 +169,8 @@ void replay_finish_event(void)replay_fetch_data_kind (); } +static __threadbool replay_locked; +void replay_mutex_init (void ) {qemu_mutex_init (&lock);
@@ -179,13 +181,22 @@ void replay_mutex_destroy(void)qemu_mutex_destroy (&lock); } +static bool replay_mutex_locked (void ) +{ +return replay_locked; +} +void replay_mutex_lock (void ) { +g_assert (!replay_mutex_locked ());qemu_mutex_lock (&lock); + replay_locked = true; }void replay_mutex_unlock (void ) { +g_assert (replay_mutex_locked ()); + replay_locked = false;qemu_mutex_unlock (&lock); }
PATCH v7 10/22 replay/replay.c: bump REPLAY_VERSION again
In reply to Pavel DovgalyukFrom: Alex Bennée <[email protected]>
This time commit 802f045a5f61b781df55e4492d896b4d20503ba7 broke the replay file format. Also add a comment about this to replay-internal.h.
--- replay/replay-internal.h | 2 +- replay/replay.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
replay/replay-internal.h +1 -1
@@ -12,7 +12,7 @@* */ - +/* Any changes to order/number of events will need to bump REPLAY_VERSION */ enum ReplayEvents {/* for instruction event */ EVENT_INSTRUCTION ,
replay/replay.c +1 -1
@@ -22,7 +22,7 @@/* Current version of the replay mechanism. Increase it when file format changes. */ -#define REPLAY_VERSION 0xe02006 +#define REPLAY_VERSION 0xe02007/* Size of replay log header */ #define HEADER_SIZE (sizeof(uint32_t) + sizeof(uint64_t))
PATCH v7 12/22 replay: make locking visible outside replay code
In reply to Pavel DovgalyukFrom: Alex Bennée <[email protected]>
The replay_mutex_lock/unlock/locked functions are now going to be used for ensuring lock-step behaviour between the two threads. Make them public API functions and also provide stubs for non-QEMU builds on common paths.
--
v7: - removed abort() from stubs to allow calling from utils
- removed exported replay_mutex_locked()
--- include/sysemu/replay.h | 13 +++++++++++++ replay/replay-internal.c | 2 +- replay/replay-internal.h | 6 +++--- stubs/replay.c | 8 ++++++++ 4 files changed, 25 insertions(+), 4 deletions(-)
include/sysemu/replay.h +13 -0
@@ -46,6 +46,19 @@ extern ReplayMode replay_mode;/* Name of the initial VM snapshot */ extern char *replay_snapshot; +/* Replay locking +* +* The locks are needed to protect the shared structures and log file +* when doing record/replay. They also are the main sync-point between +* the main-loop thread and the vCPU thread. This was a role +* previously filled by the BQL which has been busy trying to reduce +* its impact across the code. This ensures blocks of events stay +* sequential and reproducible. +*/ + +void replay_mutex_lock (void ); +void replay_mutex_unlock (void ); +/* Replay process control functions */ /*! Enables recording or saving event log with specified parameters */
replay/replay-internal.c +1 -1
@@ -181,7 +181,7 @@ void replay_mutex_destroy(void)qemu_mutex_destroy (&lock); } -static bool replay_mutex_locked (void ) +bool replay_mutex_locked (void ) {return replay_locked; }
replay/replay-internal.h +3 -3
@@ -100,12 +100,12 @@ int64_t replay_get_qword(void);void replay_get_array (uint8_t *buf,size_t *size);void replay_get_array_alloc (uint8_t **buf,size_t *size); -/* Mutex functions for protecting replay log file */ +/* Mutex functions for protecting replay log file and ensuring +* synchronisation between vCPU and main-loop threads. */ void replay_mutex_init (void );void replay_mutex_destroy (void ); -void replay_mutex_lock (void ); -void replay_mutex_unlock (void ); +bool replay_mutex_locked (void );/*! Checks error status of the file. */ void replay_check_error (void );
stubs/replay.c +8 -0
@@ -73,3 +73,11 @@ uint64_t blkreplay_next_id(void) {return 0 ; } + +void replay_mutex_lock (void ) +{ +} + +void replay_mutex_unlock (void ) +{ +}
PATCH v7 13/22 replay: push replay_mutex_lock up the call tree
In reply to Pavel DovgalyukFrom: Alex Bennée <[email protected]>
Now instead of using the replay_lock to guard the output of the log we now use it to protect the whole execution section. This replaces what the BQL used to do when it was held during TCG execution.
We also introduce some rules for locking order - mainly that you cannot take the replay_mutex while holding the BQL. This leads to some slight sophistry during start-up and extending the replay_mutex_destroy function to unlock the mutex without checking for the BQL condition so it can be cleanly dropped in the non-replay case.
--
v6: refined lock/unlock logic due to removing the BQL patches
removed replay lock/unlock from audio functions
v2: updated replay_mutex_lock/unlock functions as suggested by Paolo Bonzini
updated docs
--- cpus.c | 24 ++++++++++++++++++++++-- docs/replay.txt | 22 ++++++++++++++++++++++ include/sysemu/replay.h | 2 ++ replay/replay-audio.c | 14 ++++---------- replay/replay-char.c | 21 ++++++++------------- replay/replay-events.c | 20 +++++++------------- replay/replay-internal.c | 27 +++++++++++++++++++-------- replay/replay-time.c | 10 +++++----- replay/replay.c | 38 ++++++++++++++++++-------------------- util/main-loop.c | 15 +++++++++++---- vl.c | 2 ++ 11 files changed, 120 insertions(+), 75 deletions(-)
cpus.c +22 -2
@@ -1307,6 +1307,8 @@ static void prepare_icount_for_run(CPUState *cpu) insns_left =MIN (0xffff , cpu->icount_budget); cpu->icount_decr.u16.low = insns_left; cpu->icount_extra = cpu->icount_budget - insns_left; + +replay_mutex_lock (); } }
@@ -1322,6 +1324,8 @@ static void process_icount_data(CPUState *cpu) cpu->icount_budget =0 ;replay_account_executed_instructions (); + +replay_mutex_unlock (); } }
@@ -1336,11 +1340,9 @@ static int tcg_cpu_exec(CPUState *cpu)#ifdef CONFIG_PROFILER ti =profile_getclock ();#endif -qemu_mutex_unlock_iothread ();cpu_exec_start (cpu); ret =cpu_exec (cpu);cpu_exec_end (cpu); -qemu_mutex_lock_iothread ();#ifdef CONFIG_PROFILER tcg_time +=profile_getclock () - ti;#endif
@@ -1409,6 +1411,9 @@ static void *qemu_tcg_rr_cpu_thread_fn(void *arg) cpu->exit_request =1 ;while (1 ) { +qemu_mutex_unlock_iothread (); +replay_mutex_lock (); +qemu_mutex_lock_iothread ();/* Account partial waits to QEMU_CLOCK_VIRTUAL. */ qemu_account_warp_timer ();
@@ -1417,6 +1422,8 @@ static void *qemu_tcg_rr_cpu_thread_fn(void *arg)*/ handle_icount_deadline (); +replay_mutex_unlock (); +if (!cpu) { cpu = first_cpu; }
@@ -1432,11 +1439,13 @@ static void *qemu_tcg_rr_cpu_thread_fn(void *arg)if (cpu_can_run (cpu)) {int r; +qemu_mutex_unlock_iothread ();prepare_icount_for_run (cpu); r =tcg_cpu_exec (cpu);process_icount_data (cpu); +qemu_mutex_lock_iothread ();if (r ==EXCP_DEBUG ) {cpu_handle_guest_debug (cpu);
@@ -1626,7 +1635,9 @@ static void *qemu_tcg_cpu_thread_fn(void *arg)while (1 ) {if (cpu_can_run (cpu)) {int r; +qemu_mutex_unlock_iothread (); r =tcg_cpu_exec (cpu); +qemu_mutex_lock_iothread ();switch (r) {case EXCP_DEBUG :cpu_handle_guest_debug (cpu);
@@ -1773,12 +1784,21 @@ void pause_all_vcpus(void) } } +/* We need to drop the replay_lock so any vCPU threads woken up +* can finish their replay tasks +*/ +replay_mutex_unlock (); +while (!all_vcpus_paused ()) {qemu_cond_wait (&qemu_pause_cond, &qemu_global_mutex);CPU_FOREACH (cpu) {qemu_cpu_kick (cpu); } } + +qemu_mutex_unlock_iothread (); +replay_mutex_lock (); +qemu_mutex_lock_iothread (); }void cpu_resume (CPUState *cpu)
docs/replay.txt +22 -0
@@ -49,6 +49,28 @@ Modifications of qemu include: * recording/replaying user input (mouse and keyboard) * adding internal checkpoints for cpu and io synchronization +Locking and thread synchronisation +---------------------------------- + +Previously the synchronisation of the main thread and the vCPU thread +was ensured by the holding of the BQL. However the trend has been to +reduce the time the BQL was held across the system including under TCG +system emulation. As it is important that batches of events are kept +in sequence (e.g. expiring timers and checkpoints in the main thread +while instruction checkpoints are written by the vCPU thread) we need +another lock to keep things in lock-step. This role is now handled by +the replay_mutex_lock. It used to be held only for each event being +written but now it is held for a whole execution period. This results +in a deterministic ping-pong between the two main threads. + +As the BQL is now a finer grained lock than the replay_lock it is almost +certainly a bug, and a source of deadlocks, to take the +replay_mutex_lock while the BQL is held. This is enforced by an assert. +While the unlocks are usually in the reverse order, this is not +necessary; you can drop the replay_lock while holding the BQL, without +doing a more complicated unlock_iothread/replay_unlock/lock_iothread +sequence. + Non-deterministic events ------------------------
include/sysemu/replay.h +2 -0
@@ -61,6 +61,8 @@ void replay_mutex_unlock(void);/* Replay process control functions */ +/*! Enables and take replay locks (even if we don't use it) */ +void replay_init_locks (void );/*! Enables recording or saving event log with specified parameters */ void replay_configure (struct QemuOpts *opts);/*! Initializes timers used for snapshotting and enables events recording */
replay/replay-audio.c +4 -10
@@ -19,20 +19,17 @@void replay_audio_out (int *played) {if (replay_mode ==REPLAY_MODE_RECORD ) { +g_assert (replay_mutex_locked ());replay_save_instructions (); -replay_mutex_lock ();replay_put_event (EVENT_AUDIO_OUT );replay_put_dword (*played); -replay_mutex_unlock (); }else if (replay_mode ==REPLAY_MODE_PLAY ) { +g_assert (replay_mutex_locked ());replay_account_executed_instructions (); -replay_mutex_lock ();if (replay_next_event_is (EVENT_AUDIO_OUT )) { *played =replay_get_dword ();replay_finish_event (); -replay_mutex_unlock (); }else { -replay_mutex_unlock ();error_report ("Missing audio out event in the replay log" );abort (); }
@@ -44,8 +41,8 @@ void replay_audio_in(int *recorded, void *samples, int *wpos, int size)int pos;uint64_t left, right;if (replay_mode ==REPLAY_MODE_RECORD ) { +g_assert (replay_mutex_locked ());replay_save_instructions (); -replay_mutex_lock ();replay_put_event (EVENT_AUDIO_IN );replay_put_dword (*recorded);replay_put_dword (*wpos);
@@ -55,10 +52,9 @@ void replay_audio_in(int *recorded, void *samples, int *wpos, int size)replay_put_qword (left);replay_put_qword (right); } -replay_mutex_unlock (); }else if (replay_mode ==REPLAY_MODE_PLAY ) { +g_assert (replay_mutex_locked ());replay_account_executed_instructions (); -replay_mutex_lock ();if (replay_next_event_is (EVENT_AUDIO_IN )) { *recorded =replay_get_dword (); *wpos =replay_get_dword ();
@@ -69,9 +65,7 @@ void replay_audio_in(int *recorded, void *samples, int *wpos, int size)audio_sample_from_uint64 (samples, pos, left, right); }replay_finish_event (); -replay_mutex_unlock (); }else { -replay_mutex_unlock ();error_report ("Missing audio in event in the replay log" );abort (); }
replay/replay-char.c +8 -13
@@ -96,25 +96,24 @@ void *replay_event_char_read_load(void)void replay_char_write_event_save (int res,int offset) { +g_assert (replay_mutex_locked ()); +replay_save_instructions (); -replay_mutex_lock ();replay_put_event (EVENT_CHAR_WRITE );replay_put_dword (res);replay_put_dword (offset); -replay_mutex_unlock (); }void replay_char_write_event_load (int *res,int *offset) { +g_assert (replay_mutex_locked ()); +replay_account_executed_instructions (); -replay_mutex_lock ();if (replay_next_event_is (EVENT_CHAR_WRITE )) { *res =replay_get_dword (); *offset =replay_get_dword ();replay_finish_event (); -replay_mutex_unlock (); }else { -replay_mutex_unlock ();error_report ("Missing character write event in the replay log" );exit (1 ); }
@@ -122,23 +121,21 @@ void replay_char_write_event_load(int *res, int *offset)int replay_char_read_all_load (uint8_t *buf) { -replay_mutex_lock (); +g_assert (replay_mutex_locked ()); +if (replay_next_event_is (EVENT_CHAR_READ_ALL )) {size_t size;int res;replay_get_array (buf, &size);replay_finish_event (); -replay_mutex_unlock (); res = (int )size;assert (res >=0 );return res; }else if (replay_next_event_is (EVENT_CHAR_READ_ALL_ERROR )) {int res =replay_get_dword ();replay_finish_event (); -replay_mutex_unlock ();return res; }else { -replay_mutex_unlock ();error_report ("Missing character read all event in the replay log" );exit (1 ); }
@@ -146,19 +143,17 @@ int replay_char_read_all_load(uint8_t *buf)void replay_char_read_all_save_error (int res) { +g_assert (replay_mutex_locked ());assert (res <0 );replay_save_instructions (); -replay_mutex_lock ();replay_put_event (EVENT_CHAR_READ_ALL_ERROR );replay_put_dword (res); -replay_mutex_unlock (); }void replay_char_read_all_save_buf (uint8_t *buf,int offset) { +g_assert (replay_mutex_locked ());replay_save_instructions (); -replay_mutex_lock ();replay_put_event (EVENT_CHAR_READ_ALL );replay_put_array (buf, offset); -replay_mutex_unlock (); }
replay/replay-events.c +7 -13
@@ -79,16 +79,14 @@ bool replay_has_events(void)void replay_flush_events (void ) { -replay_mutex_lock (); +g_assert (replay_mutex_locked ()); +while (!QTAILQ_EMPTY (&events_list)) {Event *event =QTAILQ_FIRST (&events_list); -replay_mutex_unlock ();replay_run_event (event); -replay_mutex_lock ();QTAILQ_REMOVE (&events_list, event, events);g_free (event); } -replay_mutex_unlock (); }void replay_disable_events (void )
@@ -102,14 +100,14 @@ void replay_disable_events(void)void replay_clear_events (void ) { -replay_mutex_lock (); +g_assert (replay_mutex_locked ()); +while (!QTAILQ_EMPTY (&events_list)) {Event *event =QTAILQ_FIRST (&events_list);QTAILQ_REMOVE (&events_list, event, events);g_free (event); } -replay_mutex_unlock (); }/*! Adds specified async event to the queue */
@@ -136,9 +134,8 @@ void replay_add_event(ReplayAsyncEventKind event_kind, event->opaque2 = opaque2; event->id = id; -replay_mutex_lock (); +g_assert (replay_mutex_locked ());QTAILQ_INSERT_TAIL (&events_list, event, events); -replay_mutex_unlock (); }void replay_bh_schedule_event (QEMUBH *bh)
@@ -207,13 +204,11 @@ static void replay_save_event(Event *event, int checkpoint)/* Called with replay mutex locked */ void replay_save_events (int checkpoint) { +g_assert (replay_mutex_locked ());while (!QTAILQ_EMPTY (&events_list)) {Event *event =QTAILQ_FIRST (&events_list);replay_save_event (event, checkpoint); - -replay_mutex_unlock ();replay_run_event (event); -replay_mutex_lock ();QTAILQ_REMOVE (&events_list, event, events);g_free (event); }
@@ -292,6 +287,7 @@ static Event *replay_read_event(int checkpoint)/* Called with replay mutex locked */ void replay_read_events (int checkpoint) { +g_assert (replay_mutex_locked ());while (replay_state.data_kind ==EVENT_ASYNC ) {Event *event =replay_read_event (checkpoint);if (!event) {
@@ -299,9 +295,7 @@ void replay_read_events(int checkpoint) }replay_finish_event (); read_event_kind =-1 ; -replay_mutex_unlock ();replay_run_event (event); -replay_mutex_lock ();g_free (event); }
replay/replay-internal.c +19 -8
@@ -174,10 +174,16 @@ static __thread bool replay_locked;void replay_mutex_init (void ) {qemu_mutex_init (&lock); +/* Hold the mutex while we start-up */ +qemu_mutex_lock (&lock); + replay_locked = true; }void replay_mutex_destroy (void ) { +if (replay_mutex_locked ()) { +qemu_mutex_unlock (&lock); + }qemu_mutex_destroy (&lock); }
@@ -186,25 +192,31 @@ bool replay_mutex_locked(void)return replay_locked; } +/* Ordering constraints, replay_lock must be taken before BQL */ void replay_mutex_lock (void ) { -g_assert (!replay_mutex_locked ()); -qemu_mutex_lock (&lock); - replay_locked = true; +if (replay_mode !=REPLAY_MODE_NONE ) { +g_assert (!qemu_mutex_iothread_locked ()); +g_assert (!replay_mutex_locked ()); +qemu_mutex_lock (&lock); + replay_locked = true; + } }void replay_mutex_unlock (void ) { -g_assert (replay_mutex_locked ()); - replay_locked = false; -qemu_mutex_unlock (&lock); +if (replay_mode !=REPLAY_MODE_NONE ) { +g_assert (replay_mutex_locked ()); + replay_locked = false; +qemu_mutex_unlock (&lock); + } }/*! Saves cached instructions. */ void replay_save_instructions (void ) {if (replay_file && replay_mode ==REPLAY_MODE_RECORD ) { -replay_mutex_lock (); +g_assert (replay_mutex_locked ());int diff = (int )(replay_get_current_step () - replay_state.current_step);/* Time can only go forward */
@@ -215,6 +227,5 @@ void replay_save_instructions(void)replay_put_dword (diff); replay_state.current_step += diff; } -replay_mutex_unlock (); } }
replay/replay-time.c +5 -5
@@ -17,13 +17,13 @@int64_t replay_save_clock (ReplayClockKind kind,int64_t clock) { -replay_save_instructions ();if (replay_file) { -replay_mutex_lock (); +g_assert (replay_mutex_locked ()); + +replay_save_instructions ();replay_put_event (EVENT_CLOCK + kind);replay_put_qword (clock); -replay_mutex_unlock (); }return clock;
@@ -46,16 +46,16 @@ void replay_read_next_clock(ReplayClockKind kind)/*! Reads next clock event from the input. */ int64_t replay_read_clock (ReplayClockKind kind) { +g_assert (replay_file &&replay_mutex_locked ()); +replay_account_executed_instructions ();if (replay_file) {int64_t ret; -replay_mutex_lock ();if (replay_next_event_is (EVENT_CLOCK + kind)) {replay_read_next_clock (kind); } ret = replay_state.cached_clock[kind]; -replay_mutex_unlock ();return ret; }
replay/replay.c +18 -20
@@ -81,7 +81,7 @@ int replay_get_instructions(void)void replay_account_executed_instructions (void ) {if (replay_mode ==REPLAY_MODE_PLAY ) { -replay_mutex_lock (); +g_assert (replay_mutex_locked ());if (replay_state.instructions_count >0 ) {int count = (int )(replay_get_current_step () - replay_state.current_step);
@@ -100,24 +100,22 @@ void replay_account_executed_instructions(void)qemu_notify_event (); } } -replay_mutex_unlock (); } }bool replay_exception (void ) { +if (replay_mode ==REPLAY_MODE_RECORD ) { +g_assert (replay_mutex_locked ());replay_save_instructions (); -replay_mutex_lock ();replay_put_event (EVENT_EXCEPTION ); -replay_mutex_unlock ();return true; }else if (replay_mode ==REPLAY_MODE_PLAY ) { +g_assert (replay_mutex_locked ());bool res =replay_has_exception ();if (res) { -replay_mutex_lock ();replay_finish_event (); -replay_mutex_unlock (); }return res; }
@@ -129,10 +127,9 @@ bool replay_has_exception(void) {bool res = false;if (replay_mode ==REPLAY_MODE_PLAY ) { +g_assert (replay_mutex_locked ());replay_account_executed_instructions (); -replay_mutex_lock (); res =replay_next_event_is (EVENT_EXCEPTION ); -replay_mutex_unlock (); }return res;
@@ -141,17 +138,15 @@ bool replay_has_exception(void)bool replay_interrupt (void ) {if (replay_mode ==REPLAY_MODE_RECORD ) { +g_assert (replay_mutex_locked ());replay_save_instructions (); -replay_mutex_lock ();replay_put_event (EVENT_INTERRUPT ); -replay_mutex_unlock ();return true; }else if (replay_mode ==REPLAY_MODE_PLAY ) { +g_assert (replay_mutex_locked ());bool res =replay_has_interrupt ();if (res) { -replay_mutex_lock ();replay_finish_event (); -replay_mutex_unlock (); }return res; }
@@ -163,10 +158,9 @@ bool replay_has_interrupt(void) {bool res = false;if (replay_mode ==REPLAY_MODE_PLAY ) { +g_assert (replay_mutex_locked ());replay_account_executed_instructions (); -replay_mutex_lock (); res =replay_next_event_is (EVENT_INTERRUPT ); -replay_mutex_unlock (); }return res; }
@@ -174,9 +168,8 @@ bool replay_has_interrupt(void)void replay_shutdown_request (ShutdownCause cause) {if (replay_mode ==REPLAY_MODE_RECORD ) { -replay_mutex_lock (); +g_assert (replay_mutex_locked ());replay_put_event (EVENT_SHUTDOWN + cause); -replay_mutex_unlock (); } }
@@ -190,9 +183,9 @@ bool replay_checkpoint(ReplayCheckpoint checkpoint)return true; } -replay_mutex_lock ();if (replay_mode ==REPLAY_MODE_PLAY ) { +g_assert (replay_mutex_locked ());if (replay_next_event_is (EVENT_CHECKPOINT + checkpoint)) {replay_finish_event (); }else if (replay_state.data_kind !=EVENT_ASYNC ) {
@@ -205,15 +198,20 @@ bool replay_checkpoint(ReplayCheckpoint checkpoint)checkpoint were processed */ res = replay_state.data_kind !=EVENT_ASYNC ; }else if (replay_mode ==REPLAY_MODE_RECORD ) { +g_assert (replay_mutex_locked ());replay_put_event (EVENT_CHECKPOINT + checkpoint);replay_save_events (checkpoint); res = true; } out: -replay_mutex_unlock ();return res; } +void replay_init_locks (void ) +{ +replay_mutex_init (); +} +static void replay_enable (const char *fname,int mode) {const char *fmode =NULL ;
@@ -233,8 +231,6 @@ static void replay_enable(const char *fname, int mode)atexit (replay_finish); -replay_mutex_init (); - replay_file =fopen (fname, fmode);if (replay_file ==NULL ) {fprintf (stderr,"Replay: open %s: %s\n" , fname,strerror (errno));
@@ -274,6 +270,8 @@ void replay_configure(QemuOpts *opts)Location loc;if (!opts) { +/* we no longer need this lock */ +replay_mutex_destroy ();return ; }
util/main-loop.c +11 -4
@@ -29,6 +29,7 @@#include "qemu/sockets.h" // struct in_addr needed for libslirp.h #include "sysemu/qtest.h" #include "sysemu/cpus.h" +#include "sysemu/replay.h" #include "slirp/libslirp.h" #include "qemu/main-loop.h" #include "block/aio.h"
@@ -245,18 +246,19 @@ static int os_host_main_loop_wait(int64_t timeout) timeout =SCALE_MS ; } +if (timeout) { spin_counter =0 ; -qemu_mutex_unlock_iothread (); }else { spin_counter++; } +qemu_mutex_unlock_iothread (); +replay_mutex_unlock (); ret =qemu_poll_ns ((GPollFD *)gpollfds->data, gpollfds->len, timeout); -if (timeout) { -qemu_mutex_lock_iothread (); - } +replay_mutex_lock (); +qemu_mutex_lock_iothread ();glib_pollfds_poll ();
@@ -463,8 +465,13 @@ static int os_host_main_loop_wait(int64_t timeout) poll_timeout_ns =qemu_soonest_timeout (poll_timeout_ns, timeout);qemu_mutex_unlock_iothread (); + +replay_mutex_unlock (); + g_poll_ret =qemu_poll_ns (poll_fds, n_poll_fds + w->num, poll_timeout_ns); +replay_mutex_lock (); +qemu_mutex_lock_iothread ();if (g_poll_ret >0 ) {for (i =0 ; i < w->num; i++) {
vl.c +2 -0
@@ -3061,6 +3061,8 @@ int main(int argc, char **argv, char **envp)qemu_init_cpu_list ();qemu_init_cpu_loop (); + +replay_init_locks ();qemu_mutex_lock_iothread ();atexit (qemu_run_exit_notifiers);
PATCH v7 14/22 replay: don't destroy mutex at exit
In reply to Pavel DovgalyukReplay mutex is held by vCPU thread and destroy function is called from atexit of the main thread. Therefore we cannot destroy it safely.
--- replay/replay.c | 1 - 1 file changed, 1 deletion(-)
replay/replay.c +0 -1
@@ -356,7 +356,6 @@ void replay_finish(void) replay_snapshot =NULL ;replay_finish_events (); -replay_mutex_destroy (); }void replay_add_blocker (Error *reason)
PATCH v7 15/22 replay: check return values of fwrite
In reply to Pavel DovgalyukThis patch adds error reporting when fwrite cannot completely save the buffer to the file.
--
v3: also check putc() return value
--- replay/replay-internal.c | 17 +++++++++++++++-- 1 file changed, 15 insertions(+), 2 deletions(-)
replay/replay-internal.c +15 -2
@@ -24,12 +24,23 @@static QemuMutex lock;/* File for replay writing */ +static bool write_error;FILE *replay_file; +static void replay_write_error (void ) +{ +if (!write_error) { +error_report ("replay write error" ); + write_error = true; + } +} +void replay_put_byte (uint8_t byte) {if (replay_file) { -putc (byte, replay_file); +if (putc (byte, replay_file) ==EOF ) { +replay_write_error (); + } } }
@@ -62,7 +73,9 @@ void replay_put_array(const uint8_t *buf, size_t size) {if (replay_file) {replay_put_dword (size); -fwrite (buf,1 , size, replay_file); +if (fwrite (buf,1 , size, replay_file) != size) { +replay_write_error (); + } } }
PATCH v7 16/22 replay: avoid recursive call of checkpoints
In reply to Pavel DovgalyukThis patch adds a flag which denies recursive call of replay_checkpoint function. Checkpoints may be accompanied by the hardware events. When event is processed, virtual device may invoke timer modification functions that also invoke the checkpoint function. This leads to infinite loop.
--- replay/replay.c | 14 +++++++++++++- 1 file changed, 13 insertions(+), 1 deletion(-)
replay/replay.c +13 -1
@@ -176,13 +176,24 @@ void replay_shutdown_request(ShutdownCause cause)bool replay_checkpoint (ReplayCheckpoint checkpoint) {bool res = false; +static bool in_checkpoint;assert (EVENT_CHECKPOINT + checkpoint <=EVENT_CHECKPOINT_LAST ); -replay_save_instructions ();if (!replay_file) {return true; } +if (in_checkpoint) { +/* If we are already in checkpoint, then there is no need +for additional synchronization. +Recursion occurs when HW event modifies timers. +Timer modification may invoke the checkpoint and +proceed to recursion. */ +return true; + } + in_checkpoint = true; + +replay_save_instructions ();if (replay_mode ==REPLAY_MODE_PLAY ) {g_assert (replay_mutex_locked ());
@@ -204,6 +215,7 @@ bool replay_checkpoint(ReplayCheckpoint checkpoint) res = true; } out: + in_checkpoint = false;return res; }
PATCH v7 17/22 scripts/replay-dump.py: replay log dumper
In reply to Pavel DovgalyukFrom: Alex Bennée <[email protected]>
This script is a debugging tool for looking through the contents of a replay log file. It is incomplete but should fail gracefully at events it doesn't understand.
It currently understands two different log formats as the audio record/replay support was merged during since MTTCG. It was written to help debug what has caused the BQL changes to break replay support.
---
v2
- yet another update to the log format
--- scripts/replay-dump.py | 308 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 308 insertions(+) create mode 100755 scripts/replay-dump.py
scripts/replay-dump.py new file +308 -0
@@ -0,0 +1,308 @@ +#!/usr/bin/env python +# -*- coding: utf-8 -*- +# +# Dump the contents of a recorded execution stream +# +# Copyright (c) 2017 Alex Bennée <[email protected]> +# +# This library is free software; you can redistribute it and/or +# modify it under the terms of the GNU Lesser General Public +# License as published by the Free Software Foundation; either +# version 2 of the License, or (at your option) any later version. +# +# This library is distributed in the hope that it will be useful, +# but WITHOUT ANY WARRANTY; without even the implied warranty of +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +# Lesser General Public License for more details. +# +# You should have received a copy of the GNU Lesser General Public +# License along with this library; if not, see <http://www.gnu.org/licenses/>. + +import argparse +import struct +from collections import namedtuple + +# This mirrors some of the global replay state which some of the +# stream loading refers to. Some decoders may read the next event so +# we need handle that case. Calling reuse_event will ensure the next +# event is read from the cache rather than advancing the file. + +class ReplayState(object): + def __init__(self): + self.event = -1 + self.event_count = 0 + self.already_read = False + self.current_checkpoint = 0 + self.checkpoint = 0 + + def set_event(self, ev): + self.event = ev + self.event_count += 1 + + def get_event(self): + self.already_read = False + return self.event + + def reuse_event(self, ev): + self.event = ev + self.already_read = True + + def set_checkpoint(self): + self.checkpoint = self.event - self.checkpoint_start + + def get_checkpoint(self): + return self.checkpoint + +replay_state = ReplayState() + +# Simple read functions that mirror replay-internal.c +# The file-stream is big-endian and manually written out a byte at a time. + +def read_byte(fin): + "Read a single byte" + return struct.unpack('>B', fin.read(1))[0] + +def read_event(fin): + "Read a single byte event, but save some state" + if replay_state.already_read: + return replay_state.get_event() + else: + replay_state.set_event(read_byte(fin)) + return replay_state.event + +def read_word(fin): + "Read a 16 bit word" + return struct.unpack('>H', fin.read(2))[0] + +def read_dword(fin): + "Read a 32 bit word" + return struct.unpack('>I', fin.read(4))[0] + +def read_qword(fin): + "Read a 64 bit word" + return struct.unpack('>Q', fin.read(8))[0] + +# Generic decoder structure +Decoder = namedtuple("Decoder", "eid name fn") + +def call_decode(table, index, dumpfile): + "Search decode table for next step" + decoder = next((d for d in table if d.eid == index), None) + if not decoder: + print "Could not decode index: %d" % (index) + print "Entry is: %s" % (decoder) + print "Decode Table is:\n%s" % (table) + return False + else: + return decoder.fn(decoder.eid, decoder.name, dumpfile) + +# Print event +def print_event(eid, name, string=None, event_count=None): + "Print event with count" + if not event_count: + event_count = replay_state.event_count + + if string: + print "%d:%s(%d) %s" % (event_count, name, eid, string) + else: + print "%d:%s(%d)" % (event_count, name, eid) + + +# Decoders for each event type + +def decode_unimp(eid, name, _unused_dumpfile): + "Unimplimented decoder, will trigger exit" + print "%s not handled - will now stop" % (name) + return False + +# Checkpoint decoder +def swallow_async_qword(eid, name, dumpfile): + "Swallow a qword of data without looking at it" + step_id = read_qword(dumpfile) + print " %s(%d) @ %d" % (name, eid, step_id) + return True + +async_decode_table = [ Decoder(0, "REPLAY_ASYNC_EVENT_BH", swallow_async_qword), + Decoder(1, "REPLAY_ASYNC_INPUT", decode_unimp), + Decoder(2, "REPLAY_ASYNC_INPUT_SYNC", decode_unimp), + Decoder(3, "REPLAY_ASYNC_CHAR_READ", decode_unimp), + Decoder(4, "REPLAY_ASYNC_EVENT_BLOCK", decode_unimp), + Decoder(5, "REPLAY_ASYNC_EVENT_NET", decode_unimp), +] +# See replay_read_events/replay_read_event +def decode_async(eid, name, dumpfile): + """Decode an ASYNC event""" + + print_event(eid, name) + + async_event_kind = read_byte(dumpfile) + async_event_checkpoint = read_byte(dumpfile) + + if async_event_checkpoint != replay_state.current_checkpoint: + print " mismatch between checkpoint %d and async data %d" % ( + replay_state.current_checkpoint, async_event_checkpoint) + return True + + return call_decode(async_decode_table, async_event_kind, dumpfile) + + +def decode_instruction(eid, name, dumpfile): + ins_diff = read_dword(dumpfile) + print_event(eid, name, "0x%x" % (ins_diff)) + return True + +def decode_audio_out(eid, name, dumpfile): + audio_data = read_dword(dumpfile) + print_event(eid, name, "%d" % (audio_data)) + return True + +def decode_checkpoint(eid, name, dumpfile): + """Decode a checkpoint. + + Checkpoints contain a series of async events with their own specific data. + """ + replay_state.set_checkpoint() + # save event count as we peek ahead + event_number = replay_state.event_count + next_event = read_event(dumpfile) + + # if the next event is EVENT_ASYNC there are a bunch of + # async events to read, otherwise we are done + if next_event != 3: + print_event(eid, name, "no additional data", event_number) + else: + print_event(eid, name, "more data follows", event_number) + + replay_state.reuse_event(next_event) + return True + +def decode_checkpoint_init(eid, name, dumpfile): + print_event(eid, name) + return True + +def decode_interrupt(eid, name, dumpfile): + print_event(eid, name) + return True + +def decode_clock(eid, name, dumpfile): + clock_data = read_qword(dumpfile) + print_event(eid, name, "0x%x" % (clock_data)) + return True + + +# pre-MTTCG merge +v5_event_table = [Decoder(0, "EVENT_INSTRUCTION", decode_instruction), + Decoder(1, "EVENT_INTERRUPT", decode_interrupt), + Decoder(2, "EVENT_EXCEPTION", decode_unimp), + Decoder(3, "EVENT_ASYNC", decode_async), + Decoder(4, "EVENT_SHUTDOWN", decode_unimp), + Decoder(5, "EVENT_CHAR_WRITE", decode_unimp), + Decoder(6, "EVENT_CHAR_READ_ALL", decode_unimp), + Decoder(7, "EVENT_CHAR_READ_ALL_ERROR", decode_unimp), + Decoder(8, "EVENT_CLOCK_HOST", decode_clock), + Decoder(9, "EVENT_CLOCK_VIRTUAL_RT", decode_clock), + Decoder(10, "EVENT_CP_CLOCK_WARP_START", decode_checkpoint), + Decoder(11, "EVENT_CP_CLOCK_WARP_ACCOUNT", decode_checkpoint), + Decoder(12, "EVENT_CP_RESET_REQUESTED", decode_checkpoint), + Decoder(13, "EVENT_CP_SUSPEND_REQUESTED", decode_checkpoint), + Decoder(14, "EVENT_CP_CLOCK_VIRTUAL", decode_checkpoint), + Decoder(15, "EVENT_CP_CLOCK_HOST", decode_checkpoint), + Decoder(16, "EVENT_CP_CLOCK_VIRTUAL_RT", decode_checkpoint), + Decoder(17, "EVENT_CP_INIT", decode_checkpoint_init), + Decoder(18, "EVENT_CP_RESET", decode_checkpoint), +] + +# post-MTTCG merge, AUDIO support added +v6_event_table = [Decoder(0, "EVENT_INSTRUCTION", decode_instruction), + Decoder(1, "EVENT_INTERRUPT", decode_interrupt), + Decoder(2, "EVENT_EXCEPTION", decode_unimp), + Decoder(3, "EVENT_ASYNC", decode_async), + Decoder(4, "EVENT_SHUTDOWN", decode_unimp), + Decoder(5, "EVENT_CHAR_WRITE", decode_unimp), + Decoder(6, "EVENT_CHAR_READ_ALL", decode_unimp), + Decoder(7, "EVENT_CHAR_READ_ALL_ERROR", decode_unimp), + Decoder(8, "EVENT_AUDIO_OUT", decode_audio_out), + Decoder(9, "EVENT_AUDIO_IN", decode_unimp), + Decoder(10, "EVENT_CLOCK_HOST", decode_clock), + Decoder(11, "EVENT_CLOCK_VIRTUAL_RT", decode_clock), + Decoder(12, "EVENT_CP_CLOCK_WARP_START", decode_checkpoint), + Decoder(13, "EVENT_CP_CLOCK_WARP_ACCOUNT", decode_checkpoint), + Decoder(14, "EVENT_CP_RESET_REQUESTED", decode_checkpoint), + Decoder(15, "EVENT_CP_SUSPEND_REQUESTED", decode_checkpoint), + Decoder(16, "EVENT_CP_CLOCK_VIRTUAL", decode_checkpoint), + Decoder(17, "EVENT_CP_CLOCK_HOST", decode_checkpoint), + Decoder(18, "EVENT_CP_CLOCK_VIRTUAL_RT", decode_checkpoint), + Decoder(19, "EVENT_CP_INIT", decode_checkpoint_init), + Decoder(20, "EVENT_CP_RESET", decode_checkpoint), +] + +# Shutdown cause added +v7_event_table = [Decoder(0, "EVENT_INSTRUCTION", decode_instruction), + Decoder(1, "EVENT_INTERRUPT", decode_interrupt), + Decoder(2, "EVENT_EXCEPTION", decode_unimp), + Decoder(3, "EVENT_ASYNC", decode_async), + Decoder(4, "EVENT_SHUTDOWN", decode_unimp), + Decoder(5, "EVENT_SHUTDOWN_HOST_ERR", decode_unimp), + Decoder(6, "EVENT_SHUTDOWN_HOST_QMP", decode_unimp), + Decoder(7, "EVENT_SHUTDOWN_HOST_SIGNAL", decode_unimp), + Decoder(8, "EVENT_SHUTDOWN_HOST_UI", decode_unimp), + Decoder(9, "EVENT_SHUTDOWN_GUEST_SHUTDOWN", decode_unimp), + Decoder(10, "EVENT_SHUTDOWN_GUEST_RESET", decode_unimp), + Decoder(11, "EVENT_SHUTDOWN_GUEST_PANIC", decode_unimp), + Decoder(12, "EVENT_SHUTDOWN___MAX", decode_unimp), + Decoder(13, "EVENT_CHAR_WRITE", decode_unimp), + Decoder(14, "EVENT_CHAR_READ_ALL", decode_unimp), + Decoder(15, "EVENT_CHAR_READ_ALL_ERROR", decode_unimp), + Decoder(16, "EVENT_AUDIO_OUT", decode_audio_out), + Decoder(17, "EVENT_AUDIO_IN", decode_unimp), + Decoder(18, "EVENT_CLOCK_HOST", decode_clock), + Decoder(19, "EVENT_CLOCK_VIRTUAL_RT", decode_clock), + Decoder(20, "EVENT_CP_CLOCK_WARP_START", decode_checkpoint), + Decoder(21, "EVENT_CP_CLOCK_WARP_ACCOUNT", decode_checkpoint), + Decoder(22, "EVENT_CP_RESET_REQUESTED", decode_checkpoint), + Decoder(23, "EVENT_CP_SUSPEND_REQUESTED", decode_checkpoint), + Decoder(24, "EVENT_CP_CLOCK_VIRTUAL", decode_checkpoint), + Decoder(25, "EVENT_CP_CLOCK_HOST", decode_checkpoint), + Decoder(26, "EVENT_CP_CLOCK_VIRTUAL_RT", decode_checkpoint), + Decoder(27, "EVENT_CP_INIT", decode_checkpoint_init), + Decoder(28, "EVENT_CP_RESET", decode_checkpoint), +] + +def parse_arguments(): + "Grab arguments for script" + parser = argparse.ArgumentParser() + parser.add_argument("-f", "--file", help='record/replay dump to read from', + required=True) + return parser.parse_args() + +def decode_file(filename): + "Decode a record/replay dump" + dumpfile = open(filename, "rb") + + # read and throwaway the header + version = read_dword(dumpfile) + junk = read_qword(dumpfile) + + print "HEADER: version 0x%x" % (version) + + if version == 0xe02007: + event_decode_table = v7_event_table + replay_state.checkpoint_start = 12 + elif version == 0xe02006: + event_decode_table = v6_event_table + replay_state.checkpoint_start = 12 + else: + event_decode_table = v5_event_table + replay_state.checkpoint_start = 10 + + try: + decode_ok = True + while decode_ok: + event = read_event(dumpfile) + decode_ok = call_decode(event_decode_table, event, dumpfile) + finally: + dumpfile.close() + +if __name__ == "__main__": + args = parse_arguments() + decode_file(args.file)
PATCH v7 18/22 replay: don't process async events when warping the clock
In reply to Pavel DovgalyukVirtual clock is warped from iothread and vcpu thread. When the hardware events associated with warp checkpoint, then interrupt delivering may be non-deterministic if checkpoint is processed in different threads in record and replay.
This patch disables event processing for clock warp checkpoint and leaves all hardware events to other checkpoints (e.g., virtual clock).
--
v4: added assert for replay_save_events function
--- replay/replay-events.c | 1 + replay/replay.c | 7 ++++++- 2 files changed, 7 insertions(+), 1 deletion(-)
replay/replay-events.c +1 -0
@@ -205,6 +205,7 @@ static void replay_save_event(Event *event, int checkpoint)void replay_save_events (int checkpoint) {g_assert (replay_mutex_locked ()); +g_assert (checkpoint !=CHECKPOINT_CLOCK_WARP_START );while (!QTAILQ_EMPTY (&events_list)) {Event *event =QTAILQ_FIRST (&events_list);replay_save_event (event, checkpoint);
replay/replay.c +6 -1
@@ -211,7 +211,12 @@ bool replay_checkpoint(ReplayCheckpoint checkpoint) }else if (replay_mode ==REPLAY_MODE_RECORD ) {g_assert (replay_mutex_locked ());replay_put_event (EVENT_CHECKPOINT + checkpoint); -replay_save_events (checkpoint); +/* This checkpoint belongs to several threads. +Processing events from different threads is +non-deterministic */ +if (checkpoint !=CHECKPOINT_CLOCK_WARP_START ) { +replay_save_events (checkpoint); + } res = true; } out:
PATCH v7 19/22 replay: save vmstate of the asynchronous events
In reply to Pavel DovgalyukThis patch fixes saving and loading the snapshots in the replay mode. It is required for the snapshots created in the moment when the header of the asynchronous event is read. This information was not saved in the snapshot. After loading the vmstate replay continued with the file offset passed the event header. The event header is lost in this case and replay hangs.
--- replay/replay-events.c | 44 +++++++++++++++++++++----------------------- replay/replay-internal.h | 6 ++++++ replay/replay-snapshot.c | 3 +++ 3 files changed, 30 insertions(+), 23 deletions(-)
replay/replay-events.c +21 -23
@@ -27,10 +27,6 @@ typedef struct Event { } Event; static QTAILQ_HEAD(, Event) events_list = QTAILQ_HEAD_INITIALIZER(events_list); -static unsigned int read_event_kind =-1 ; -static uint64_t read_id =-1 ; -static int read_checkpoint =-1 ; -static bool events_enabled;/* Functions */
@@ -218,58 +214,60 @@ void replay_save_events(int checkpoint)static Event *replay_read_event (int checkpoint) {Event *event; -if (read_event_kind ==-1 ) { - read_checkpoint =replay_get_byte (); - read_event_kind =replay_get_byte (); - read_id =-1 ; +if (replay_state.read_event_kind ==-1 ) { + replay_state.read_event_checkpoint =replay_get_byte (); + replay_state.read_event_kind =replay_get_byte (); + replay_state.read_event_id =-1 ;replay_check_error (); } -if (checkpoint != read_checkpoint) { +if (checkpoint != replay_state.read_event_checkpoint) {return NULL ; }/* Events that has not to be in the queue */ -switch (read_event_kind) { +switch (replay_state.read_event_kind) {case REPLAY_ASYNC_EVENT_BH : -if (read_id ==-1 ) { - read_id =replay_get_qword (); +if (replay_state.read_event_id ==-1 ) { + replay_state.read_event_id =replay_get_qword (); }break ;case REPLAY_ASYNC_EVENT_INPUT : event =g_malloc0 (sizeof (Event)); - event->event_kind = read_event_kind; + event->event_kind = replay_state.read_event_kind; event->opaque =replay_read_input_event ();return event;case REPLAY_ASYNC_EVENT_INPUT_SYNC : event =g_malloc0 (sizeof (Event)); - event->event_kind = read_event_kind; + event->event_kind = replay_state.read_event_kind; event->opaque =0 ;return event;case REPLAY_ASYNC_EVENT_CHAR_READ : event =g_malloc0 (sizeof (Event)); - event->event_kind = read_event_kind; + event->event_kind = replay_state.read_event_kind; event->opaque =replay_event_char_read_load ();return event;case REPLAY_ASYNC_EVENT_BLOCK : -if (read_id ==-1 ) { - read_id =replay_get_qword (); +if (replay_state.read_event_id ==-1 ) { + replay_state.read_event_id =replay_get_qword (); }break ;case REPLAY_ASYNC_EVENT_NET : event =g_malloc0 (sizeof (Event)); - event->event_kind = read_event_kind; + event->event_kind = replay_state.read_event_kind; event->opaque =replay_event_net_load ();return event;default : -error_report ("Unknown ID %d of replay event" , read_event_kind); +error_report ("Unknown ID %d of replay event" , + replay_state.read_event_kind);exit (1 );break ; } QTAILQ_FOREACH(event, &events_list, events) { -if (event->event_kind == read_event_kind - && (read_id ==-1 || read_id == event->id)) { +if (event->event_kind == replay_state.read_event_kind + && (replay_state.read_event_id ==-1 + || replay_state.read_event_id == event->id)) {break ; } }
@@ -295,7 +293,7 @@ void replay_read_events(int checkpoint)break ; }replay_finish_event (); - read_event_kind =-1 ; + replay_state.read_event_kind =-1 ;replay_run_event (event);g_free (event);
@@ -304,7 +302,7 @@ void replay_read_events(int checkpoint)void replay_init_events (void ) { - read_event_kind =-1 ; + replay_state.read_event_kind =-1 ; }void replay_finish_events (void )
replay/replay-internal.h +6 -0
@@ -80,6 +80,12 @@ typedef struct ReplayState {uint64_t block_request_id;/*! Prior value of the host clock */ uint64_t host_clock_last; +/*! Asynchronous event type read from the log */ +int32_t read_event_kind; +/*! Asynchronous event id read from the log */ +uint64_t read_event_id; +/*! Asynchronous event checkpoint id read from the log */ +int32_t read_event_checkpoint; } ReplayState;extern ReplayState replay_state;
replay/replay-snapshot.c +3 -0
@@ -57,6 +57,9 @@ static const VMStateDescription vmstate_replay = { VMSTATE_UINT64(file_offset, ReplayState), VMSTATE_UINT64(block_request_id, ReplayState), VMSTATE_UINT64(host_clock_last, ReplayState), + VMSTATE_INT32(read_event_kind, ReplayState), + VMSTATE_UINT64(read_event_id, ReplayState), + VMSTATE_INT32(read_event_checkpoint, ReplayState),VMSTATE_END_OF_LIST () }, };
PATCH v7 20/22 replay: don't drain/flush bdrv queue while RR is working
In reply to Pavel DovgalyukIn record/replay mode bdrv queue is controlled by replay mechanism. It does not allow saving or loading the snapshots
when bdrv queue is not empty. Stopping the VM is not blocked by nonempty queue, but flushing the queue is still impossible there, because it may cause deadlocks in replay mode.
This patch disables bdrv_drain_all and bdrv_flush_all in record/replay mode.
--- block/io.c | 22 ++++++++++++++++++++++ cpus.c | 2 -- 2 files changed, 22 insertions(+), 2 deletions(-)
block/io.c +22 -0
@@ -31,6 +31,7 @@#include "qemu/cutils.h" #include "qapi/error.h" #include "qemu/error-report.h" +#include "sysemu/replay.h" #define NOT_DONE 0x7fffffff/* used while emulated sync operation in progress */
@@ -407,6 +408,13 @@ void bdrv_drain_all_begin(void)BdrvNextIterator it;GSList *aio_ctxs =NULL , *ctx; +/* bdrv queue is managed by record/replay, +waiting for finishing the I/O requests may +be infinite */ +if (replay_events_enabled ()) { +return ; + } +/* BDRV_POLL_WHILE() for a node can only be called from its own I/O thread * or the main loop AioContext. We potentially use BDRV_POLL_WHILE() on * nodes in several different AioContexts, so make sure we're in the main
@@ -458,6 +466,13 @@ void bdrv_drain_all_end(void)BlockDriverState *bs;BdrvNextIterator it; +/* bdrv queue is managed by record/replay, +waiting for finishing the I/O requests may +be endless */ +if (replay_events_enabled ()) { +return ; + } +for (bs =bdrv_first (&it); bs; bs =bdrv_next (&it)) {AioContext *aio_context =bdrv_get_aio_context (bs);
@@ -1839,6 +1854,13 @@ int bdrv_flush_all(void)BlockDriverState *bs =NULL ;int result =0 ; +/* bdrv queue is managed by record/replay, +creating new flush request for stopping +the VM may break the determinism */ +if (replay_events_enabled ()) { +return result; + } +for (bs =bdrv_first (&it); bs; bs =bdrv_next (&it)) {AioContext *aio_context =bdrv_get_aio_context (bs);int ret;
cpus.c +0 -2
@@ -1006,7 +1006,6 @@ static int do_vm_stop(RunState state) }bdrv_drain_all (); -replay_disable_events (); ret =bdrv_flush_all ();return ret;
@@ -2054,7 +2053,6 @@ int vm_prepare_start(void)qapi_event_send_stop (&error_abort); res =-1 ; }else { -replay_enable_events ();cpu_enable_ticks ();runstate_set (RUN_STATE_RUNNING );vm_state_notify (1 ,RUN_STATE_RUNNING );
PATCH v7 21/22 replay: update documentation
In reply to Pavel DovgalyukThis patch clarifies the description of the record/replay feature in docs/replay.txt
--- docs/replay.txt | 72 ++++++++++++++++++++++++++++++++++++++++--------------- 1 file changed, 52 insertions(+), 20 deletions(-)
docs/replay.txt +52 -20
@@ -7,14 +7,10 @@ See the COPYING file in the top-level directory. Record/replay ------------- -Record/replay functions are used for the reverse execution and deterministic -replay of qemu execution. This implementation of deterministic replay can -be used for deterministic debugging of guest code through a gdb remote -interface. - +Record/replay functions are used for the deterministic replay of qemu execution. Execution recording writes a non-deterministic events log, which can be later used for replaying the execution anywhere and for unlimited number of times. -It also supports checkpointing for faster rewinding during reverse debugging. +It also supports checkpointing for faster rewind to the specific replay moment. Execution replaying reads the log and replays all non-deterministic events including external input, hardware clocks, and interrupts.
@@ -28,16 +24,36 @@ Deterministic replay has the following features: input devices. Usage of the record/replay: - * First, record the execution, by adding the following arguments to the command line: - '-icount shift=7,rr=record,rrfile=replay.bin -net none'. - Block devices' images are not actually changed in the recording mode, + * First, record the execution with the following command line: + qemu-system-i386 \ + -icount shift=7,rr=record,rrfile=replay.bin \ + -drive file=disk.qcow2,if=none,id=img-direct \ + -drive driver=blkreplay,if=none,image=img-direct,id=img-blkreplay \ + -device ide-hd,drive=img-blkreplay \ + -netdev user,id=net1 -device rtl8139,netdev=net1 \ + -object filter-replay,id=replay,netdev=net1 + * After recording, you can replay it by using another command line: + qemu-system-i386 \ + -icount shift=7,rr=replay,rrfile=replay.bin \ + -drive file=disk.qcow2,if=none,id=img-direct \ + -drive driver=blkreplay,if=none,image=img-direct,id=img-blkreplay \ + -device ide-hd,drive=img-blkreplay \ + -netdev user,id=net1 -device rtl8139,netdev=net1 \ + -object filter-replay,id=replay,netdev=net1 + The only difference with recording is changing the rr option + from record to replay. + * Block device images are not actually changed in the recording mode, because all of the changes are written to the temporary overlay file. - * Then you can replay it by using another command - line option: '-icount shift=7,rr=replay,rrfile=replay.bin -net none' - * '-net none' option should also be specified if network replay patches - are not applied. - -Papers with description of deterministic replay implementation: + This behavior is enabled by using blkreplay driver. It should be used + for every enabled block device, as described in 'Block devices' section. + * '-net none' option should be specified when network is not used, + because QEMU adds network card by default. When network is needed, + it should be configured explicitly with replay filter, as described + in 'Network devices' section. + * Interaction with audio devices and serial ports are recorded and replayed + automatically when such devices are enabled. + +Academic papers with description of deterministic replay implementation: http://www.computer.org/csdl/proceedings/csmr/2012/4666/00/4666a553-abs.html http://dl.acm.org/citation.cfm?id=2786805.2803179
@@ -46,8 +62,11 @@ Modifications of qemu include: * saving different asynchronous events (e.g. system shutdown) into the log * synchronization of the bottom halves execution * synchronization of the threads from thread pool - * recording/replaying user input (mouse and keyboard) + * recording/replaying user input (mouse, keyboard, and microphone) * adding internal checkpoints for cpu and io synchronization + * network filter for recording and replaying the packets + * block driver for making block layer deterministic + * serial port input record and replay Locking and thread synchronisation ----------------------------------
@@ -77,12 +96,11 @@ Non-deterministic events Our record/replay system is based on saving and replaying non-deterministic events (e.g. keyboard input) and simulating deterministic ones (e.g. reading from HDD or memory of the VM). Saving only non-deterministic events makes -log file smaller, simulation faster, and allows using reverse debugging even -for realtime applications. +log file smaller and simulation faster. The following non-deterministic data from peripheral devices is saved into the log: mouse and keyboard input, network packets, audio controller input, -USB packets, serial port input, and hardware clocks (they are non-deterministic +serial port input, and hardware clocks (they are non-deterministic too, because their values are taken from the host machine). Inputs from simulated hardware, memory of VM, software interrupts, and execution of instructions are not saved into the log, because they are deterministic and
@@ -205,7 +223,7 @@ Block devices record/replay module intercepts calls of bdrv coroutine functions at the top of block drivers stack. To record and replay block operations the drive must be configured as following: - -drive file=disk.qcow,if=none,id=img-direct + -drive file=disk.qcow2,if=none,id=img-direct -drive driver=blkreplay,if=none,image=img-direct,id=img-blkreplay -device ide-hd,drive=img-blkreplay
@@ -234,6 +252,12 @@ This snapshot is created at start of recording and restored at start of replaying. It also can be loaded while replaying to roll back the execution. +Use QEMU monitor to create additional snapshots. 'savevm <name>' command +created the snapshot and 'loadvm <name>' restores it. To prevent corruption +of the original disk image, use overlay files linked to the original images. +Therefore all new snapshots (including the starting one) will be saved in +overlays and the original image remains unchanged. + Network devices ---------------
@@ -255,6 +279,14 @@ Audio data is recorded and replay automatically. The command line for recording and replaying must contain identical specifications of audio hardware, e.g.: -soundhw ac97 +Serial ports +------------ + +Serial ports input is recorded and replay automatically. The command lines +for recording and replaying must contain identical number of ports in record +and replay modes, but their backends may differ. +E.g., '-serial stdio' in record mode, and '-serial null' in replay mode. + Replay log format -----------------
PATCH v7 22/22 tcg: fix cpu_io_recompile
In reply to Pavel Dovgalyukcpu_io_recompile() function was broken by
the commit 9b990ee5a3cc6aa38f81266fb0c6ef37a36c45b9. Instead of regenerating the block starting from PC of the original block, it just set the instruction counter for TCG. In most cases this was unnoticed, but in icount mode there was an exception for incorrect usage of CF_LAST_IO flag. This patch recovers recompilation of the original block and also configures translation for executing single IO instruction which caused a recompilation.
--- accel/tcg/translate-all.c | 18 +++++++++++++++--- 1 file changed, 15 insertions(+), 3 deletions(-)
accel/tcg/translate-all.c +15 -3
@@ -1728,7 +1728,8 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)CPUArchState *env = cpu->env_ptr;#endif TranslationBlock *tb; -uint32_t n; +uint32_t n, flags; +target_ulong pc, cs_base;tb_lock (); tb =tb_find_pc (retaddr);
@@ -1766,8 +1767,14 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)cpu_abort (cpu,"TB too big during recompile" ); } -/* Adjust the execution state of the next TB. */ - cpu->cflags_next_tb =curr_cflags () |CF_LAST_IO | n; + pc = tb->pc; + cs_base = tb->cs_base; + flags = tb->flags; +tb_phys_invalidate (tb,-1 ); + +/* Execute one IO instruction without caching +instead of creating large TB. */ + cpu->cflags_next_tb =curr_cflags () |CF_LAST_IO |CF_NOCACHE |1 ;if (tb->cflags &CF_NOCACHE ) {if (tb->orig_tb) {
@@ -1778,6 +1785,11 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)tb_remove (tb); } +/* Generate new TB instead of the current one. */ +/* FIXME: In theory this could raise an exception. In practice +we have already translated the block once so it's probably ok. */ +tb_gen_code (cpu, pc, cs_base, flags,curr_cflags () |CF_LAST_IO | n); +/* TODO: If env->pc != tb->pc (i.e. the faulting instruction was not * the first in the TB) then we end up generating a whole new TB and * repeating the fault, which is horribly inefficient.
Ping.
Pavel Dovgalyuk
Quoted 113 lines
-----Original Message-----
From: Pavel Dovgalyuk [mailto:[email protected]]
Sent: Tuesday, February 27, 2018 12:52 PM
To: [email protected]
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: [ PATCH v7 00/22] replay additionsThis set of patches moves replay lock upper in the function call tree. Now replay lock functions similar to BQL in older version and allows deterministic execution of the threads in icount mode.
It is also fixes some vmstate creation (and loading) issues in record/replay modes:
- VM start/stop fixes in replay mode
- overlay creation for blkreplay filter
- fixes for vmstate save/load in record/replay mode
- fixes for host clock vmstateThere is also a set of helper scripts written by Alex Bennée for debugging the record/replay code.
v6 patches with updates for v7 are available in the repository:
https://github.com/ispras/qemu/tree/rr-180207v7 changes:
- updated record/replay documentation
- removedabort()from mutex stub functions
- fixedcpu_io_recompilefunctionv6 changes:
- removed BQL optimization at all
- refined replay lock patches
- removed lock/unlock from replay-audiov5 changes:
- removed patch for narrowing BQL-protected code
- disabled bdrv_(drain/flush)_all for record/replay modev4 changes:
- removed upstreamed patches
- added patch for saving async queue state in replay
- minor fixesv3 changes:
- removed upstreamed patches
- fixed bug with recursive checkpoints
- fixed bug with icount warp checkpointv2 changes:
- updated lock/unlock logic (as suggested by Paolo Bonzini)
- updated cpu execution loop to avoid races in setting/resetting exit request (as suggested
by Paolo Bonzini)
- minor changes---
Alex Bennée (5):
replay/replay.c: bumpREPLAY_VERSIONagainreplay/replay-internal.c: track holding ofreplay_lock
replay: make locking visible outside replay code
replay: pushreplay_mutex_lockup the call treescripts/replay-dump.py: replay log dumperPavel Dovgalyuk (17):
cpu-exec: fixexception_indexhandling
block: implementbdrv_snapshot_gotofor blkreplay
blkreplay: create temporary overlay for underlaying devices
replay: disable default snapshot for record/replay
replay: fix processing async events
replay: fixedreplay_enable_events
replay: fix save/load vm for non-empty queue
replay: added replay log format description
replay: save prior value of the host clock
replay: don't destroy mutex at exit
replay: check return values of fwrite
replay: avoid recursive call of checkpoints
replay: don't process async events when warping the clock
replay: save vmstate of the asynchronous events
replay: don't drain/flush bdrv queue while RR is working
replay: update documentation
tcg: fixcpu_io_recompileaccel/tcg/cpu-exec.c | 5 + accel/tcg/translate-all.c | 18 ++- block/blkreplay.c | 75 +++++++++++ block/io.c | 22 +++ cpus.c | 26 +++- docs/replay.txt | 163 +++++++++++++++++++++--- include/qemu/timer.h | 14 ++ include/sysemu/replay.h | 18 +++ migration/savevm.c | 13 ++ replay/replay-audio.c | 14 +- replay/replay-char.c | 21 +-- replay/replay-events.c | 75 +++++------ replay/replay-internal.c | 47 ++++++- replay/replay-internal.h | 16 ++ replay/replay-snapshot.c | 12 ++ replay/replay-time.c | 10 + replay/replay.c | 62 ++++++--- scripts/replay-dump.py | 308 +++++++++++++++++++++++++++++++++++++++++++++ stubs/replay.c | 9 + util/main-loop.c | 15 ++ util/qemu-timer.c | 12 ++ vl.c | 12 +- 22 files changed, 831 insertions(+), 136 deletions(-) create mode 100755 scripts/replay-dump.py--
Pavel Dovgalyuk
Just to re-affirm, I have ran this patch on x86 and arm, and it worked.
Quoted 119 lines · Pavel Dovgalyuk
Ping.
Pavel Dovgalyuk
Quoted 113 lines
-----Original Message-----
From: Pavel Dovgalyuk [mailto:[email protected]]
Sent: Tuesday, February 27, 2018 12:52 PM
To: [email protected]
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: [ PATCH v7 00/22] replay additionsThis set of patches moves replay lock upper in the function call tree. Now replay lock functions similar to BQL in older version and allows deterministic execution of the threads in icount mode.
It is also fixes some vmstate creation (and loading) issues in record/replay modes:
- VM start/stop fixes in replay mode
- overlay creation for blkreplay filter
- fixes for vmstate save/load in record/replay mode
- fixes for host clock vmstateThere is also a set of helper scripts written by Alex Bennée for debugging the record/replay code.
v6 patches with updates for v7 are available in the repository:
https://github.com/ispras/qemu/tree/rr-180207v7 changes:
- updated record/replay documentation
- removedabort()from mutex stub functions
- fixedcpu_io_recompilefunctionv6 changes:
- removed BQL optimization at all
- refined replay lock patches
- removed lock/unlock from replay-audiov5 changes:
- removed patch for narrowing BQL-protected code
- disabled bdrv_(drain/flush)_all for record/replay modev4 changes:
- removed upstreamed patches
- added patch for saving async queue state in replay
- minor fixesv3 changes:
- removed upstreamed patches
- fixed bug with recursive checkpoints
- fixed bug with icount warp checkpointv2 changes:
- updated lock/unlock logic (as suggested by Paolo Bonzini)
- updated cpu execution loop to avoid races in setting/resetting exit request (as suggested
by Paolo Bonzini)
- minor changes---
Alex Bennée (5):
replay/replay.c: bumpREPLAY_VERSIONagainreplay/replay-internal.c: track holding ofreplay_lock
replay: make locking visible outside replay code
replay: pushreplay_mutex_lockup the call treescripts/replay-dump.py: replay log dumperPavel Dovgalyuk (17):
cpu-exec: fixexception_indexhandling
block: implementbdrv_snapshot_gotofor blkreplay
blkreplay: create temporary overlay for underlaying devices
replay: disable default snapshot for record/replay
replay: fix processing async events
replay: fixedreplay_enable_events
replay: fix save/load vm for non-empty queue
replay: added replay log format description
replay: save prior value of the host clock
replay: don't destroy mutex at exit
replay: check return values of fwrite
replay: avoid recursive call of checkpoints
replay: don't process async events when warping the clock
replay: save vmstate of the asynchronous events
replay: don't drain/flush bdrv queue while RR is working
replay: update documentation
tcg: fixcpu_io_recompileaccel/tcg/cpu-exec.c | 5 + accel/tcg/translate-all.c | 18 ++- block/blkreplay.c | 75 +++++++++++ block/io.c | 22 +++ cpus.c | 26 +++- docs/replay.txt | 163 +++++++++++++++++++++--- include/qemu/timer.h | 14 ++ include/sysemu/replay.h | 18 +++ migration/savevm.c | 13 ++ replay/replay-audio.c | 14 +- replay/replay-char.c | 21 +-- replay/replay-events.c | 75 +++++------ replay/replay-internal.c | 47 ++++++- replay/replay-internal.h | 16 ++ replay/replay-snapshot.c | 12 ++ replay/replay-time.c | 10 + replay/replay.c | 62 ++++++--- scripts/replay-dump.py | 308 +++++++++++++++++++++++++++++++++++++++++++++ stubs/replay.c | 9 + util/main-loop.c | 15 ++ util/qemu-timer.c | 12 ++ vl.c | 12 +- 22 files changed, 831 insertions(+), 136 deletions(-) create mode 100755 scripts/replay-dump.py--
Pavel Dovgalyuk
Re: [Qemu-devel] [PATCH v7 13/22] replay: push replay_mutex_lock up the call tree
In reply to Pavel DovgalyukReview
Quoted 6 lines · Pavel Dovgalyuk
(quoted diff) +5
-0+void replay_init_locks(void) +{ + replay_mutex_init(); +} +
This should not be needed as a public function: until replay_mode is set, replay_lock and replay_unlock do nothing. I'm squashing this:
include/sysemu/replay.h +0 -2
@@ -63,8 +63,6 @@ void replay_mutex_unlock(void);/* Replay process control functions */ -/*! Enables and take replay locks (even if we don't use it) */ -void replay_init_locks (void );/*! Enables recording or saving event log with specified parameters */ void replay_configure (struct QemuOpts *opts);/*! Initializes timers used for snapshotting and enables events recording */
replay/replay.c +2 -6
@@ -207,11 +207,6 @@ out:return res; } -void replay_init_locks (void ) -{ -replay_mutex_init (); -} -static void replay_enable (const char *fname,int mode) {const char *fmode =NULL ;
@@ -238,8 +233,9 @@ static void replay_enable(const char *fname, int mode) } replay_filename =g_strdup (fname); - replay_mode = mode; +replay_mutex_init (); + replay_state.data_kind =-1 ; replay_state.instructions_count =0 ; replay_state.current_step =0 ;
vl.c +0 -1
@@ -3059,7 +3059,6 @@ int main(int argc, char **argv, char **envp)qemu_init_cpu_list ();qemu_init_cpu_loop (); -replay_init_locks ();qemu_mutex_lock_iothread ();atexit (qemu_run_exit_notifiers);
Re: [Qemu-devel] [PATCH v7 20/22] replay: don't drain/flush bdrv queue while RR is working
In reply to Pavel DovgalyukQuoted 9 lines · Pavel Dovgalyuk
In record/replay mode bdrv queue is controlled by replay mechanism. It does not allow saving or loading the snapshots
when bdrv queue is not empty. Stopping the VM is not blocked by nonempty queue, but flushing the queue is still impossible there, because it may cause deadlocks in replay mode.
This patch disablesbdrv_drain_allandbdrv_flush_allin record/replay mode.Signed-off-by Pavel Dovgalyuk <[email protected]>
A lot of monitor commands that use bdrv_drain_all are going to remain broken, so I'm not sure about this patch. (There have been many threads about bdrv_drain and bdrv_drain_all being a bad API...). I'm queuing everything except 2-3-4-20.
Thanks Pavel and Ciro.
Paolo
Quoted 80 lines
--- block/io.c | 22 ++++++++++++++++++++++ cpus.c | 2 -- 2 files changed, 22 insertions(+), 2 deletions(-)block/io.c +22
-0@@ -31,6 +31,7 @@#include "qemu/cutils.h" #include "qapi/error.h" #include "qemu/error-report.h" +#include "sysemu/replay.h" #define NOT_DONE 0x7fffffff/* used while emulated sync operation in progress */ @@ -407,6 +408,13 @@ void bdrv_drain_all_begin(void)BdrvNextIterator it;GSList *aio_ctxs =NULL , *ctx; +/* bdrv queue is managed by record/replay, +waiting for finishing the I/O requests may +be infinite */ +if (replay_events_enabled ()) { +return ; + } +/* BDRV_POLL_WHILE() for a node can only be called from its own I/O thread * or the main loop AioContext. We potentially use BDRV_POLL_WHILE() on * nodes in several different AioContexts, so make sure we're in the main @@ -458,6 +466,13 @@ void bdrv_drain_all_end(void)BlockDriverState *bs;BdrvNextIterator it; +/* bdrv queue is managed by record/replay, +waiting for finishing the I/O requests may +be endless */ +if (replay_events_enabled ()) { +return ; + } +for (bs =bdrv_first (&it); bs; bs =bdrv_next (&it)) {AioContext *aio_context =bdrv_get_aio_context (bs);@@ -1839,6 +1854,13 @@ int bdrv_flush_all(void)BlockDriverState *bs =NULL ;int result =0 ; +/* bdrv queue is managed by record/replay, +creating new flush request for stopping +the VM may break the determinism */ +if (replay_events_enabled ()) { +return result; + } +for (bs =bdrv_first (&it); bs; bs =bdrv_next (&it)) {AioContext *aio_context =bdrv_get_aio_context (bs);int ret;cpus.c +0
-2@@ -1006,7 +1006,6 @@ static int do_vm_stop(RunState state) }bdrv_drain_all (); -replay_disable_events (); ret =bdrv_flush_all ();return ret;@@ -2054,7 +2053,6 @@ int vm_prepare_start(void)qapi_event_send_stop (&error_abort); res =-1 ; }else { -replay_enable_events ();cpu_enable_ticks ();runstate_set (RUN_STATE_RUNNING );vm_state_notify (1 ,RUN_STATE_RUNNING );
Re: [Qemu-devel] [PATCH v7 22/22] tcg: fix cpu_io_recompile
In reply to Pavel DovgalyukQuoted 47 lines · Pavel Dovgalyuk
cpu_io_recompile()function was broken by
the commit9b990ee5a3cc6aa38f81266fb0c6ef37a36c45b9. Instead of regenerating the block starting from PC of the original block, it just set the instruction counter for TCG. In most cases this was unnoticed, but in icount mode there was an exception for incorrect usage ofCF_LAST_IOflag. This patch recovers recompilation of the original block and also configures translation for executing single IO instruction which caused a recompilation.Signed-off-by Pavel Dovgalyuk <[email protected]>--- accel/tcg/translate-all.c | 18 +++++++++++++++--- 1 file changed, 15 insertions(+), 3 deletions(-)accel/tcg/translate-all.c +10
-3@@ -1728,7 +1728,8 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)CPUArchState *env = cpu->env_ptr;#endif TranslationBlock *tb; -uint32_t n; +uint32_t n, flags; +target_ulong pc, cs_base;tb_lock (); tb =tb_find_pc (retaddr);@@ -1766,8 +1767,14 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)cpu_abort (cpu,"TB too big during recompile" ); } -/* Adjust the execution state of the next TB. */ - cpu->cflags_next_tb =curr_cflags () |CF_LAST_IO | n; + pc = tb->pc; + cs_base = tb->cs_base; + flags = tb->flags; +tb_phys_invalidate (tb,-1 ); + +/* Execute one IO instruction without caching +instead of creating large TB. */ + cpu->cflags_next_tb =curr_cflags () |CF_LAST_IO |CF_NOCACHE |1 ;if (tb->cflags &CF_NOCACHE ) {if (tb->orig_tb) {
accel/tcg/translate-all.c void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr) · line 1791
Quoted 7 lines
accel/tcg/translate-all.c +4
-0@@ -1778,6 +1785,11 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)tb_remove (tb); } +/* Generate new TB instead of the current one. */ +/* FIXME: In theory this could raise an exception. In practice +we have already translated the block once so it's probably ok. */ +tb_gen_code (cpu, pc, cs_base, flags,curr_cflags () |CF_LAST_IO | n);
This isn't right. The whole point of the patch that you reference as having broken
things is that calls to tb_gen_code which ignore their return value are by definition
relying on the side effect of altering the TB cache, and are therefore by definition racy.
That is exactly the point of cpu->cflags_next_tb, that when we next arrive in cpu_exec
we look up (or generate) the next TB with the given flags. At which point we will *not*
be relying on the TB cache, and we'll execute the generated TB right away.
I do not have enough context within this patch to determine what the proper solution is.
r~
Re: [Qemu-devel] [PATCH v7 22/22] tcg: fix cpu_io_recompile
Quoted 52 lines
From: Richard Henderson [mailto:[email protected]]
Quoted 50 lines · Pavel Dovgalyuk
cpu_io_recompile()function was broken by
the commit9b990ee5a3cc6aa38f81266fb0c6ef37a36c45b9. Instead of regenerating the block starting from PC of the original block, it just set the instruction counter for TCG. In most cases this was unnoticed, but in icount mode there was an exception for incorrect usage ofCF_LAST_IOflag. This patch recovers recompilation of the original block and also configures translation for executing single IO instruction which caused a recompilation.Signed-off-by Pavel Dovgalyuk <[email protected]>--- accel/tcg/translate-all.c | 18 +++++++++++++++--- 1 file changed, 15 insertions(+), 3 deletions(-)accel/tcg/translate-all.c +10
-3@@ -1728,7 +1728,8 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)CPUArchState *env = cpu->env_ptr;#endif TranslationBlock *tb; -uint32_t n; +uint32_t n, flags; +target_ulong pc, cs_base;tb_lock (); tb =tb_find_pc (retaddr);@@ -1766,8 +1767,14 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)cpu_abort (cpu,"TB too big during recompile" ); } -/* Adjust the execution state of the next TB. */ - cpu->cflags_next_tb =curr_cflags () |CF_LAST_IO | n; + pc = tb->pc; + cs_base = tb->cs_base; + flags = tb->flags; +tb_phys_invalidate (tb,-1 ); + +/* Execute one IO instruction without caching +instead of creating large TB. */ + cpu->cflags_next_tb =curr_cflags () |CF_LAST_IO |CF_NOCACHE |1 ;if (tb->cflags &CF_NOCACHE ) {if (tb->orig_tb) {@@ -1778,6 +1785,11 @@ void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr)tb_remove (tb); }
Quoted 11 lines
accel/tcg/translate-all.c void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr) · line 1791
Quoted 4 lines
accel/tcg/translate-all.c +4
-0+/* Generate new TB instead of the current one. */ +/* FIXME: In theory this could raise an exception. In practice +we have already translated the block once so it's probably ok. */ +tb_gen_code (cpu, pc, cs_base, flags,curr_cflags () |CF_LAST_IO | n);This isn't right. The whole point of the patch that you reference as having broken
things is that calls totb_gen_codewhich ignore their return value are by definition
relying on the side effect of altering the TB cache, and are therefore by definition racy.
I see.
Quoted 5 lines
That is exactly the point of
cpu->cflags_next_tb, that when we next arrive incpu_exec
we look up (or generate) the next TB with the given flags. At which point we will *not*
be relying on the TB cache, and we'll execute the generated TB right away.
Well, as a ineffective solution, we can just omit tb_gen_code, but still make
+ cpu->cflags_next_tb = curr_cflags() | CF_LAST_IO | CF_NOCACHE | 1;
Then the recompilation will occur every time, because the translation for the original address is not limited by some counter.
Quoted 2 lines
I do not have enough context within this patch to determine what the proper solution is.
The context is here:
https://lists.gnu.org/archive/html/qemu-devel/2018-02/msg04818.html
Pavel Dovgalyuk