* [Qemu-devel] [RFC PATCH v4 01/13] icount: put icount variables into TimerState.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 02/13] migration: migrate icount fields fred.konrad
` (11 subsequent siblings)
12 siblings, 0 replies; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
This puts qemu_icount and qemu_icount_bias into TimerState structure to allow
them to be migrated.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
---
cpus.c | 29 ++++++++++++++++-------------
1 file changed, 16 insertions(+), 13 deletions(-)
diff --git a/cpus.c b/cpus.c
index 5e7f2cf..127de1c 100644
--- a/cpus.c
+++ b/cpus.c
@@ -102,17 +102,12 @@ static bool all_cpu_threads_idle(void)
/* Protected by TimersState seqlock */
-/* Compensate for varying guest execution speed. */
-static int64_t qemu_icount_bias;
static int64_t vm_clock_warp_start;
/* Conversion factor from emulated instructions to virtual clock ticks. */
static int icount_time_shift;
/* Arbitrarily pick 1MIPS as the minimum allowable speed. */
#define MAX_ICOUNT_SHIFT 10
-/* Only written by TCG thread */
-static int64_t qemu_icount;
-
static QEMUTimer *icount_rt_timer;
static QEMUTimer *icount_vm_timer;
static QEMUTimer *icount_warp_timer;
@@ -129,6 +124,11 @@ typedef struct TimersState {
int64_t cpu_clock_offset;
int32_t cpu_ticks_enabled;
int64_t dummy;
+
+ /* Compensate for varying guest execution speed. */
+ int64_t qemu_icount_bias;
+ /* Only written by TCG thread */
+ int64_t qemu_icount;
} TimersState;
static TimersState timers_state;
@@ -139,14 +139,14 @@ static int64_t cpu_get_icount_locked(void)
int64_t icount;
CPUState *cpu = current_cpu;
- icount = qemu_icount;
+ icount = timers_state.qemu_icount;
if (cpu) {
if (!cpu_can_do_io(cpu)) {
fprintf(stderr, "Bad clock read\n");
}
icount -= (cpu->icount_decr.u16.low + cpu->icount_extra);
}
- return qemu_icount_bias + (icount << icount_time_shift);
+ return timers_state.qemu_icount_bias + (icount << icount_time_shift);
}
int64_t cpu_get_icount(void)
@@ -284,7 +284,8 @@ static void icount_adjust(void)
icount_time_shift++;
}
last_delta = delta;
- qemu_icount_bias = cur_icount - (qemu_icount << icount_time_shift);
+ timers_state.qemu_icount_bias = cur_icount
+ - (timers_state.qemu_icount << icount_time_shift);
seqlock_write_unlock(&timers_state.vm_clock_seqlock);
}
@@ -333,7 +334,7 @@ static void icount_warp_rt(void *opaque)
int64_t delta = cur_time - cur_icount;
warp_delta = MIN(warp_delta, delta);
}
- qemu_icount_bias += warp_delta;
+ timers_state.qemu_icount_bias += warp_delta;
}
vm_clock_warp_start = -1;
seqlock_write_unlock(&timers_state.vm_clock_seqlock);
@@ -351,7 +352,7 @@ void qtest_clock_warp(int64_t dest)
int64_t deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
int64_t warp = qemu_soonest_timeout(dest - clock, deadline);
seqlock_write_lock(&timers_state.vm_clock_seqlock);
- qemu_icount_bias += warp;
+ timers_state.qemu_icount_bias += warp;
seqlock_write_unlock(&timers_state.vm_clock_seqlock);
qemu_clock_run_timers(QEMU_CLOCK_VIRTUAL);
@@ -1250,7 +1251,8 @@ static int tcg_cpu_exec(CPUArchState *env)
int64_t count;
int64_t deadline;
int decr;
- qemu_icount -= (cpu->icount_decr.u16.low + cpu->icount_extra);
+ timers_state.qemu_icount -= (cpu->icount_decr.u16.low
+ + cpu->icount_extra);
cpu->icount_decr.u16.low = 0;
cpu->icount_extra = 0;
deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
@@ -1265,7 +1267,7 @@ static int tcg_cpu_exec(CPUArchState *env)
}
count = qemu_icount_round(deadline);
- qemu_icount += count;
+ timers_state.qemu_icount += count;
decr = (count > 0xffff) ? 0xffff : count;
count -= decr;
cpu->icount_decr.u16.low = decr;
@@ -1278,7 +1280,8 @@ static int tcg_cpu_exec(CPUArchState *env)
if (use_icount) {
/* Fold pending instructions back into the
instruction counter, and clear the interrupt flag. */
- qemu_icount -= (cpu->icount_decr.u16.low + cpu->icount_extra);
+ timers_state.qemu_icount -= (cpu->icount_decr.u16.low
+ + cpu->icount_extra);
cpu->icount_decr.u32 = 0;
cpu->icount_extra = 0;
}
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 02/13] migration: migrate icount fields.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 01/13] icount: put icount variables into TimerState fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 12:17 ` Juan Quintela
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 03/13] migration: make qemu_savevm_state public fred.konrad
` (10 subsequent siblings)
12 siblings, 1 reply; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
This fixes a bug where qemu_icount and qemu_icount_bias are not migrated.
It adds a subsection "timer/icount" to vmstate_timers so icount is migrated only
when needed.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
---
cpus.c | 28 ++++++++++++++++++++++++++++
1 file changed, 28 insertions(+)
diff --git a/cpus.c b/cpus.c
index 127de1c..8d2ddf0 100644
--- a/cpus.c
+++ b/cpus.c
@@ -429,6 +429,26 @@ void qemu_clock_warp(QEMUClockType type)
}
}
+static bool icount_state_needed(void *opaque)
+{
+ return use_icount;
+}
+
+/*
+ * This is a subsection for icount migration.
+ */
+static const VMStateDescription icount_vmstate_timers = {
+ .name = "timer/icount",
+ .version_id = 1,
+ .minimum_version_id = 1,
+ .minimum_version_id_old = 1,
+ .fields = (VMStateField[]) {
+ VMSTATE_INT64(qemu_icount_bias, TimersState),
+ VMSTATE_INT64(qemu_icount, TimersState),
+ VMSTATE_END_OF_LIST()
+ }
+};
+
static const VMStateDescription vmstate_timers = {
.name = "timer",
.version_id = 2,
@@ -438,6 +458,14 @@ static const VMStateDescription vmstate_timers = {
VMSTATE_INT64(dummy, TimersState),
VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2),
VMSTATE_END_OF_LIST()
+ },
+ .subsections = (VMStateSubsection[]) {
+ {
+ .vmsd = &icount_vmstate_timers,
+ .needed = icount_state_needed,
+ }, {
+ /* empty */
+ }
}
};
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [RFC PATCH v4 02/13] migration: migrate icount fields.
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 02/13] migration: migrate icount fields fred.konrad
@ 2014-06-25 12:17 ` Juan Quintela
2014-06-25 15:17 ` Frederic Konrad
0 siblings, 1 reply; 17+ messages in thread
From: Juan Quintela @ 2014-06-25 12:17 UTC (permalink / raw)
To: fred.konrad
Cc: peter.maydell, mark.burton, qemu-devel, dgilbert, amit.shah,
pbonzini, vilanova
fred.konrad@greensocs.com wrote:
> From: KONRAD Frederic <fred.konrad@greensocs.com>
>
> This fixes a bug where qemu_icount and qemu_icount_bias are not migrated.
> It adds a subsection "timer/icount" to vmstate_timers so icount is migrated only
> when needed.
>
> Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
> Reviewed-by: Amit Shah <amit.shah@redhat.com>
> ---
> cpus.c | 28 ++++++++++++++++++++++++++++
> 1 file changed, 28 insertions(+)
>
> diff --git a/cpus.c b/cpus.c
> index 127de1c..8d2ddf0 100644
> --- a/cpus.c
> +++ b/cpus.c
> @@ -429,6 +429,26 @@ void qemu_clock_warp(QEMUClockType type)
> }
> }
>
> +static bool icount_state_needed(void *opaque)
> +{
> + return use_icount;
> +}
> +
> +/*
> + * This is a subsection for icount migration.
> + */
> +static const VMStateDescription icount_vmstate_timers = {
> + .name = "timer/icount",
> + .version_id = 1,
> + .minimum_version_id = 1,
> + .minimum_version_id_old = 1,
Drop this line, not needed anymore.
> + .fields = (VMStateField[]) {
Remove spaces
> + VMSTATE_INT64(qemu_icount_bias, TimersState),
> + VMSTATE_INT64(qemu_icount, TimersState),
> + VMSTATE_END_OF_LIST()
> + }
> +};
> +
> static const VMStateDescription vmstate_timers = {
> .name = "timer",
> .version_id = 2,
> @@ -438,6 +458,14 @@ static const VMStateDescription vmstate_timers = {
> VMSTATE_INT64(dummy, TimersState),
> VMSTATE_INT64_V(cpu_clock_offset, TimersState, 2),
> VMSTATE_END_OF_LIST()
> + },
> + .subsections = (VMStateSubsection[]) {
> + {
> + .vmsd = &icount_vmstate_timers,
> + .needed = icount_state_needed,
> + }, {
> + /* empty */
> + }
> }
> };
Rest is ok.
Thanks, Juan.
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [Qemu-devel] [RFC PATCH v4 02/13] migration: migrate icount fields.
2014-06-25 12:17 ` Juan Quintela
@ 2014-06-25 15:17 ` Frederic Konrad
0 siblings, 0 replies; 17+ messages in thread
From: Frederic Konrad @ 2014-06-25 15:17 UTC (permalink / raw)
To: quintela
Cc: peter.maydell, mark.burton, qemu-devel, dgilbert, amit.shah,
pbonzini, vilanova
On 25/06/2014 14:17, Juan Quintela wrote:
> + .minimum_version_id_old = 1,
> Drop this line, not needed anymore.
>
>> + .fields = (VMStateField[]) {
> Remove spaces
oops didn't see that sorry.
>
>> [...]
> Rest is ok.
>
> Thanks, Juan.
I will resent with this fix.
Thanks,
Fred.
^ permalink raw reply [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 03/13] migration: make qemu_savevm_state public.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 01/13] icount: put icount variables into TimerState fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 02/13] migration: migrate icount fields fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 12:18 ` Juan Quintela
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 04/13] icount: introduce icount timer fred.konrad
` (9 subsequent siblings)
12 siblings, 1 reply; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
This makes qemu_savevm_state public for reverse-execution.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
Reviewed-by: Amit Shah <amit.shah@redhat.com>
---
include/sysemu/sysemu.h | 1 +
savevm.c | 2 +-
2 files changed, 2 insertions(+), 1 deletion(-)
diff --git a/include/sysemu/sysemu.h b/include/sysemu/sysemu.h
index 285c45b..c2fb786 100644
--- a/include/sysemu/sysemu.h
+++ b/include/sysemu/sysemu.h
@@ -81,6 +81,7 @@ void do_info_snapshots(Monitor *mon, const QDict *qdict);
void qemu_announce_self(void);
+int qemu_savevm_state(QEMUFile *f);
bool qemu_savevm_state_blocked(Error **errp);
void qemu_savevm_state_begin(QEMUFile *f,
const MigrationParams *params);
diff --git a/savevm.c b/savevm.c
index ba900d3..77fe794 100644
--- a/savevm.c
+++ b/savevm.c
@@ -782,7 +782,7 @@ void qemu_savevm_state_cancel(void)
}
}
-static int qemu_savevm_state(QEMUFile *f)
+int qemu_savevm_state(QEMUFile *f)
{
int ret;
MigrationParams params = {
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 04/13] icount: introduce icount timer.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
` (2 preceding siblings ...)
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 03/13] migration: make qemu_savevm_state public fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 05/13] icount: check for icount clock deadline when cpu loop exits fred.konrad
` (8 subsequent siblings)
12 siblings, 0 replies; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
This introduces a new timer based only on instruction counter and without any
compensation.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
---
cpus.c | 19 ++++++++++++-------
include/qemu/timer.h | 9 ++++++++-
qemu-timer.c | 8 +++++++-
stubs/cpu-get-icount.c | 2 +-
4 files changed, 28 insertions(+), 10 deletions(-)
diff --git a/cpus.c b/cpus.c
index 8d2ddf0..7f99d5e 100644
--- a/cpus.c
+++ b/cpus.c
@@ -134,7 +134,7 @@ typedef struct TimersState {
static TimersState timers_state;
/* Return the virtual CPU time, based on the instruction counter. */
-static int64_t cpu_get_icount_locked(void)
+static int64_t cpu_get_icount_locked(int with_bias)
{
int64_t icount;
CPUState *cpu = current_cpu;
@@ -146,17 +146,22 @@ static int64_t cpu_get_icount_locked(void)
}
icount -= (cpu->icount_decr.u16.low + cpu->icount_extra);
}
- return timers_state.qemu_icount_bias + (icount << icount_time_shift);
+
+ if (with_bias) {
+ return timers_state.qemu_icount_bias + (icount << icount_time_shift);
+ } else {
+ return icount << icount_time_shift;
+ }
}
-int64_t cpu_get_icount(void)
+int64_t cpu_get_icount(int with_bias)
{
int64_t icount;
unsigned start;
do {
start = seqlock_read_begin(&timers_state.vm_clock_seqlock);
- icount = cpu_get_icount_locked();
+ icount = cpu_get_icount_locked(with_bias);
} while (seqlock_read_retry(&timers_state.vm_clock_seqlock, start));
return icount;
@@ -169,7 +174,7 @@ int64_t cpu_get_ticks(void)
int64_t ticks;
if (use_icount) {
- return cpu_get_icount();
+ return cpu_get_icount(true);
}
ticks = timers_state.cpu_ticks_offset;
@@ -267,7 +272,7 @@ static void icount_adjust(void)
seqlock_write_lock(&timers_state.vm_clock_seqlock);
cur_time = cpu_get_clock_locked();
- cur_icount = cpu_get_icount_locked();
+ cur_icount = cpu_get_icount_locked(true);
delta = cur_icount - cur_time;
/* FIXME: This is a very crude algorithm, somewhat prone to oscillation. */
@@ -330,7 +335,7 @@ static void icount_warp_rt(void *opaque)
* far ahead of real time.
*/
int64_t cur_time = cpu_get_clock_locked();
- int64_t cur_icount = cpu_get_icount_locked();
+ int64_t cur_icount = cpu_get_icount_locked(true);
int64_t delta = cur_time - cur_icount;
warp_delta = MIN(warp_delta, delta);
}
diff --git a/include/qemu/timer.h b/include/qemu/timer.h
index 7f9a074..6204cab 100644
--- a/include/qemu/timer.h
+++ b/include/qemu/timer.h
@@ -36,12 +36,19 @@
* is suspended, and it will reflect system time changes the host may
* undergo (e.g. due to NTP). The host clock has the same precision as
* the virtual clock.
+ *
+ * @QEMU_CLOCK_ICOUNT: icount clock
+ *
+ * The icount clock is based on instruction counter without any compensation for
+ * speed. It will run only when instruction are executed and make only sense in
+ * icount mode.
*/
typedef enum {
QEMU_CLOCK_REALTIME = 0,
QEMU_CLOCK_VIRTUAL = 1,
QEMU_CLOCK_HOST = 2,
+ QEMU_CLOCK_ICOUNT = 3,
QEMU_CLOCK_MAX
} QEMUClockType;
@@ -743,7 +750,7 @@ static inline int64_t get_clock(void)
#endif
/* icount */
-int64_t cpu_get_icount(void);
+int64_t cpu_get_icount(int with_bias);
int64_t cpu_get_clock(void);
/*******************************************/
diff --git a/qemu-timer.c b/qemu-timer.c
index 00a5d35..3be80b1 100644
--- a/qemu-timer.c
+++ b/qemu-timer.c
@@ -554,7 +554,7 @@ int64_t qemu_clock_get_ns(QEMUClockType type)
default:
case QEMU_CLOCK_VIRTUAL:
if (use_icount) {
- return cpu_get_icount();
+ return cpu_get_icount(true);
} else {
return cpu_get_clock();
}
@@ -566,6 +566,12 @@ int64_t qemu_clock_get_ns(QEMUClockType type)
notifier_list_notify(&clock->reset_notifiers, &now);
}
return now;
+ case QEMU_CLOCK_ICOUNT:
+ if (use_icount) {
+ return cpu_get_icount(false);
+ } else {
+ return -1;
+ }
}
}
diff --git a/stubs/cpu-get-icount.c b/stubs/cpu-get-icount.c
index d685859..1968de7 100644
--- a/stubs/cpu-get-icount.c
+++ b/stubs/cpu-get-icount.c
@@ -3,7 +3,7 @@
int use_icount;
-int64_t cpu_get_icount(void)
+int64_t cpu_get_icount(int with_bias)
{
abort();
}
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 05/13] icount: check for icount clock deadline when cpu loop exits.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
` (3 preceding siblings ...)
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 04/13] icount: introduce icount timer fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 06/13] icount: make icount extra computed on icount clock as well fred.konrad
` (7 subsequent siblings)
12 siblings, 0 replies; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
Notify events on icount clock when CPU loop exits.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
---
cpus.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/cpus.c b/cpus.c
index 7f99d5e..76de504 100644
--- a/cpus.c
+++ b/cpus.c
@@ -993,6 +993,11 @@ static void *qemu_tcg_cpu_thread_fn(void *arg)
if (deadline == 0) {
qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
}
+
+ deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_ICOUNT);
+ if (deadline == 0) {
+ qemu_clock_notify(QEMU_CLOCK_ICOUNT);
+ }
}
qemu_tcg_wait_io_event();
}
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 06/13] icount: make icount extra computed on icount clock as well.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
` (4 preceding siblings ...)
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 05/13] icount: check for icount clock deadline when cpu loop exits fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 07/13] timer: add cpu_icount_to_ns function fred.konrad
` (6 subsequent siblings)
12 siblings, 0 replies; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
This takes icount clock in account for icount extra computation so icount
clock's timers will be triggered at the exact time.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
---
cpus.c | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/cpus.c b/cpus.c
index 76de504..4b551f1 100644
--- a/cpus.c
+++ b/cpus.c
@@ -1288,6 +1288,7 @@ static int tcg_cpu_exec(CPUArchState *env)
if (use_icount) {
int64_t count;
int64_t deadline;
+ int64_t icount_deadline;
int decr;
timers_state.qemu_icount -= (cpu->icount_decr.u16.low
+ cpu->icount_extra);
@@ -1304,6 +1305,15 @@ static int tcg_cpu_exec(CPUArchState *env)
deadline = INT32_MAX;
}
+ /*
+ * Take icount clock deadline in account too, and keep the nearest
+ * deadline.
+ */
+ icount_deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_ICOUNT);
+ if ((icount_deadline >= 0) && (icount_deadline < deadline)) {
+ deadline = icount_deadline;
+ }
+
count = qemu_icount_round(deadline);
timers_state.qemu_icount += count;
decr = (count > 0xffff) ? 0xffff : count;
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 07/13] timer: add cpu_icount_to_ns function.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
` (5 preceding siblings ...)
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 06/13] icount: make icount extra computed on icount clock as well fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 08/13] trace-events: add reverse-execution events fred.konrad
` (5 subsequent siblings)
12 siblings, 0 replies; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
This adds cpu_icount_to_ns function which is needed for reverse execution.
It returns the time for a specific instruction.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
---
cpus.c | 9 +++++++--
include/qemu/timer.h | 1 +
2 files changed, 8 insertions(+), 2 deletions(-)
diff --git a/cpus.c b/cpus.c
index 4b551f1..89c7a1a 100644
--- a/cpus.c
+++ b/cpus.c
@@ -148,9 +148,9 @@ static int64_t cpu_get_icount_locked(int with_bias)
}
if (with_bias) {
- return timers_state.qemu_icount_bias + (icount << icount_time_shift);
+ return timers_state.qemu_icount_bias + cpu_icount_to_ns(icount);
} else {
- return icount << icount_time_shift;
+ return cpu_icount_to_ns(icount);
}
}
@@ -167,6 +167,11 @@ int64_t cpu_get_icount(int with_bias)
return icount;
}
+int64_t cpu_icount_to_ns(int64_t icount)
+{
+ return icount << icount_time_shift;
+}
+
/* return the host CPU cycle counter and handle stop/restart */
/* Caller must hold the BQL */
int64_t cpu_get_ticks(void)
diff --git a/include/qemu/timer.h b/include/qemu/timer.h
index 6204cab..0ae7f28 100644
--- a/include/qemu/timer.h
+++ b/include/qemu/timer.h
@@ -752,6 +752,7 @@ static inline int64_t get_clock(void)
/* icount */
int64_t cpu_get_icount(int with_bias);
int64_t cpu_get_clock(void);
+int64_t cpu_icount_to_ns(int64_t icount);
/*******************************************/
/* host CPU ticks (if available) */
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 08/13] trace-events: add reverse-execution events.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
` (6 preceding siblings ...)
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 07/13] timer: add cpu_icount_to_ns function fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 09/13] introduce reverse execution mechanism fred.konrad
` (4 subsequent siblings)
12 siblings, 0 replies; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
This adds some trace-events for reverse execution.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
---
trace-events | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/trace-events b/trace-events
index ba01ad5..c1423e0 100644
--- a/trace-events
+++ b/trace-events
@@ -1292,3 +1292,9 @@ mhp_pc_dimm_assigned_address(uint64_t addr) "0x%"PRIx64
# target-s390x/kvm.c
kvm_enable_cmma(int rc) "CMMA: enabling with result code %d"
kvm_clear_cmma(int rc) "CMMA: clearing with result code %d"
+
+# reverse-execution.c
+snap_callback(uint64_t time, const char *filename) "snapshot @%li -> %s"
+stop_callback(uint64_t time) "stopping @%li"
+cexe_stop_reverse_continue(void) "stop reverse continue."
+cexe_step_backward(uint64_t time, uint64_t stop) "stepping backward @%li stop @%li"
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 09/13] introduce reverse execution mechanism.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
` (7 preceding siblings ...)
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 08/13] trace-events: add reverse-execution events fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 10/13] gdbstub: allow reverse execution in gdb stub fred.konrad
` (3 subsequent siblings)
12 siblings, 0 replies; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
This introduces the basic reverse-execution mechanism.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
---
Makefile.target | 1 +
cpus.c | 6 +
include/reverse-execution.h | 41 ++++++
reverse-execution.c | 306 ++++++++++++++++++++++++++++++++++++++++++++
vl.c | 7 +-
5 files changed, 360 insertions(+), 1 deletion(-)
create mode 100644 include/reverse-execution.h
create mode 100644 reverse-execution.c
diff --git a/Makefile.target b/Makefile.target
index 6089d29..7cb5cae 100644
--- a/Makefile.target
+++ b/Makefile.target
@@ -120,6 +120,7 @@ endif #CONFIG_BSD_USER
# System emulator target
ifdef CONFIG_SOFTMMU
obj-y += arch_init.o cpus.o monitor.o gdbstub.o balloon.o ioport.o numa.o
+obj-y += reverse-execution.o
obj-y += qtest.o
obj-y += hw/
obj-$(CONFIG_FDT) += device_tree.o
diff --git a/cpus.c b/cpus.c
index 89c7a1a..e86045e 100644
--- a/cpus.c
+++ b/cpus.c
@@ -63,6 +63,8 @@
#endif /* CONFIG_LINUX */
+#include "reverse-execution.h"
+
static CPUState *next_cpu;
bool cpu_is_stopped(CPUState *cpu)
@@ -593,7 +595,11 @@ static bool cpu_can_run(CPUState *cpu)
static void cpu_handle_guest_debug(CPUState *cpu)
{
+ if (cexe_is_continuing_backward()) {
+ cexe_step_done();
+ }
gdb_set_stop_cpu(cpu);
+ cexe_stop_stepping_back_mode();
qemu_system_debug_request();
cpu->stopped = true;
}
diff --git a/include/reverse-execution.h b/include/reverse-execution.h
new file mode 100644
index 0000000..bf42003
--- /dev/null
+++ b/include/reverse-execution.h
@@ -0,0 +1,41 @@
+/*
+ * reverse execution.
+ *
+ * Copyright (C) 2014 : GreenSocs Ltd
+ * http://www.greensocs.com/ , email: info@greensocs.com
+ *
+ * Developed by :
+ * Frederic Konrad <fred.konrad@greensocs.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#ifndef REVERSE_EXECUTION
+#define REVERSE_EXECUTION
+
+void cexe_setup(void);
+void cexe_step_backward(CPUState *cpu, uint64_t steps);
+void cexe_stop_stepping_back_mode(void);
+void cexe_continue_backward(CPUState *cpu);
+int cexe_is_continuing_backward(void);
+void cexe_next_reverse_continue_step(void);
+void cexe_stop_reverse_continue(void);
+void cexe_step_done(void);
+bool cexe_is_step_done(void);
+bool cexe_is_enabled(void);
+void cexe_cleanup(void);
+bool cexe_dbg_requested(void);
+
+#endif /* REVERSE_EXECUTION */
diff --git a/reverse-execution.c b/reverse-execution.c
new file mode 100644
index 0000000..9bf2ee8
--- /dev/null
+++ b/reverse-execution.c
@@ -0,0 +1,306 @@
+/*
+ * reverse execution.
+ *
+ * Copyright (C) 2014 : GreenSocs Ltd
+ * http://www.greensocs.com/ , email: info@greensocs.com
+ *
+ * Developed by :
+ * Frederic Konrad <fred.konrad@greensocs.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#include "qemu/timer.h"
+#include "sysemu/sysemu.h"
+#include "migration/qemu-file.h"
+
+#include "reverse-execution.h"
+
+#include "trace.h" /* needed for trace event prototype */
+
+typedef struct snapshot_entry {
+ uint32_t id;
+ int64_t time;
+ QLIST_ENTRY(snapshot_entry) next;
+} snapshot_entry;
+
+static QLIST_HEAD(, snapshot_entry) snapshot = QLIST_HEAD_INITIALIZER(snapshot);
+
+QEMUTimer *snap_timer;
+QEMUTimer *stop_timer;
+
+struct cexe_state {
+ int stepping_back;
+ int continue_backward_mode;
+ int singlestep_was_enabled;
+ bool step_done;
+ bool stop_requested;
+};
+
+static bool cexe_enabled;
+struct cexe_state cexe_state;
+
+static snapshot_entry *new_snapshot(void)
+{
+ snapshot_entry *snap = NULL;
+ snap = g_malloc(sizeof(snapshot_entry));
+ assert(snap);
+
+ if (QLIST_FIRST(&snapshot) != NULL) {
+ snap->id = QLIST_FIRST(&snapshot)->id + 1;
+ } else {
+ snap->id = 0;
+ }
+
+ QLIST_INSERT_HEAD(&snapshot, snap, next);
+ return snap;
+}
+
+/*
+ * Timer callback called when a snapshot must be done.
+ */
+static void snap_callback(void *opaque)
+{
+ QEMUFile *file = NULL;
+ int saved_vm_running;
+ snapshot_entry *snap = NULL;
+ CPUArchState *cpu = NULL;
+ char filename[20];
+
+ cpu = qemu_get_cpu(0)->env_ptr;
+ assert(cpu != NULL);
+
+ if (!cexe_state.stepping_back) {
+ snap = new_snapshot();
+
+ saved_vm_running = runstate_is_running();
+ vm_stop(RUN_STATE_SAVE_VM);
+ snap->time = qemu_clock_get_ns(QEMU_CLOCK_ICOUNT);
+ sprintf(filename, ".save%04u", snap->id);
+
+ trace_snap_callback(snap->time, filename);
+
+ file = qemu_fopen(filename, "wb");
+ qemu_savevm_state(file);
+ qemu_fclose(file);
+
+ if (saved_vm_running) {
+ vm_start();
+ }
+ timer_mod_ns(snap_timer, snap->time + 100000000);
+ }
+}
+
+/*
+ * Timer callback called when the VM have to stop.
+ */
+static void stop_callback(void *opaque)
+{
+ trace_stop_callback(qemu_clock_get_ns(QEMU_CLOCK_ICOUNT));
+ cexe_state.stop_requested = true;
+}
+
+void cexe_setup(void)
+{
+ snap_timer = timer_new_ns(QEMU_CLOCK_ICOUNT, snap_callback, NULL);
+ stop_timer = timer_new_ns(QEMU_CLOCK_ICOUNT, stop_callback, NULL);
+
+ timer_mod_ns(snap_timer, qemu_clock_get_ns(QEMU_CLOCK_ICOUNT));
+ cexe_enabled = true;
+ cexe_state.stepping_back = 0;
+ cexe_state.continue_backward_mode = 0;
+ cexe_state.stop_requested = false;
+}
+
+void cexe_stop_stepping_back_mode(void)
+{
+ if (cexe_state.stepping_back) {
+ singlestep = cexe_state.singlestep_was_enabled;
+ cexe_state.stepping_back = 0;
+ }
+ cexe_state.stop_requested = false;
+}
+
+static void cexe_start_stepping_back_mode(CPUState *cpu)
+{
+ assert(!cexe_state.stepping_back);
+ /*
+ * Flushing tb.
+ * FIXME: might not be necessary with counter.
+ */
+ tb_flush(cpu->env_ptr);
+
+ /*
+ * Single step to the right PC.
+ */
+ cexe_state.singlestep_was_enabled = singlestep;
+ singlestep = 1;
+
+ cexe_state.stepping_back = 1;
+}
+
+/**
+ * \func cexe_step_backward
+ * \param cpu GDBStub's cpu.
+ * \param steps Number of steps to step back.
+ * \brief Steps backward: "reverse-step" in GDB.
+ *
+ */
+void cexe_step_backward(CPUState *cpu, uint64_t steps)
+{
+ QEMUFile *file = NULL;
+ char filename[20];
+ snapshot_entry *snap = QLIST_FIRST(&snapshot);
+
+ int64_t stop_time = qemu_clock_get_ns(QEMU_CLOCK_ICOUNT)
+ - cpu_icount_to_ns(steps);
+
+ /*
+ * FIXME: Remove the file?
+ */
+ while ((stop_time > 0) && ((snap = QLIST_FIRST(&snapshot)) != NULL)
+ && (snap->time >= stop_time)) {
+ /*
+ * Remove the snapshot from the list and mod the snapshot timer to its
+ * time. This will cause the snapshot to be taken at the same value in
+ * case of a forward execution.
+ */
+ QLIST_REMOVE(snap, next);
+ timer_mod_ns(snap_timer, snap->time);
+ g_free(snap);
+ }
+
+ if ((stop_time <= 0) || (snap == NULL)) {
+ /*
+ * This happens when an instruction behind the first snapshot is asked.
+ * Just trigger a debug event so it won't move.
+ */
+ cexe_state.stop_requested = true;
+ vm_start();
+ return;
+ }
+
+ sprintf(filename, ".save%04u", snap->id);
+
+ /*
+ * Load the previous state.
+ */
+ vm_stop(RUN_STATE_RESTORE_VM);
+
+ trace_cexe_step_backward(qemu_clock_get_ns(QEMU_CLOCK_ICOUNT), stop_time);
+
+ file = qemu_fopen(filename, "rb");
+ qemu_loadvm_state(file);
+ qemu_fclose(file);
+
+ /*
+ * Mod the timer so it will stop at the exact instruction.
+ */
+ timer_mod_ns(stop_timer, stop_time);
+
+ cexe_start_stepping_back_mode(cpu);
+ /*
+ * Restart the vm.
+ */
+ vm_start();
+}
+
+/**
+ * \func cexe_continue_backward
+ * \brief Continue execution backward.
+ * \param cpu GDB's stub cpu.
+ *
+ */
+void cexe_continue_backward(CPUState *cpu)
+{
+ cexe_state.continue_backward_mode = 1;
+ cexe_state.step_done = false;
+ cexe_step_backward(cpu, 1);
+}
+
+/**
+ * \func cexe_is_continuing_backward
+ * \brief Check if we are continuing backward.
+ * \return Return true if we are continuing backward.
+ *
+ */
+int cexe_is_continuing_backward(void)
+{
+ return cexe_state.continue_backward_mode;
+}
+
+void cexe_next_reverse_continue_step(void)
+{
+ CPUState *cpu = qemu_get_cpu(0);
+
+ assert(cpu != NULL);
+ cexe_state.step_done = false;
+
+ /*
+ * FIXME:
+ * - Stop at breakpoint in reverse order.
+ * - The reverse execution speed is not constant as the snapshot
+ * replay is not constant.
+ */
+ cexe_step_backward(cpu, 10000000);
+}
+
+void cexe_stop_reverse_continue(void)
+{
+ if (cexe_state.continue_backward_mode) {
+ trace_cexe_stop_reverse_continue();
+ cexe_state.continue_backward_mode = false;
+ cexe_state.step_done = false;
+ cexe_stop_stepping_back_mode();
+ }
+}
+
+void cexe_step_done(void)
+{
+ cexe_state.step_done = true;
+}
+
+bool cexe_is_step_done(void)
+{
+ return cexe_state.step_done;
+}
+
+bool cexe_is_enabled(void)
+{
+ return cexe_enabled;
+}
+
+void cexe_cleanup(void)
+{
+ snapshot_entry *snap = QLIST_FIRST(&snapshot);
+
+ /*
+ * FIXME: Remove the file?
+ */
+ while ((snap = QLIST_FIRST(&snapshot)) != NULL) {
+ /*
+ * Remove the snapshot from the list and mod the snapshot timer to its
+ * time. This will cause the snapshot to be taken at the same value in
+ * case of a forward execution.
+ */
+ QLIST_REMOVE(snap, next);
+ g_free(snap);
+ }
+}
+
+bool cexe_dbg_requested(void)
+{
+ return cexe_state.stop_requested;
+}
diff --git a/vl.c b/vl.c
index a1686ef..d0352e3 100644
--- a/vl.c
+++ b/vl.c
@@ -119,6 +119,8 @@ int main(int argc, char **argv)
#include "qom/object_interfaces.h"
#include "qapi-event.h"
+#include "reverse-execution.h"
+
#define DEFAULT_RAM_SIZE 128
#define MAX_VIRTIO_CONSOLES 1
@@ -1950,7 +1952,7 @@ void qemu_system_debug_request(void)
static bool main_loop_should_exit(void)
{
RunState r;
- if (qemu_debug_requested()) {
+ if (qemu_debug_requested() && !cexe_is_continuing_backward()) {
vm_stop(RUN_STATE_DEBUG);
}
if (qemu_suspend_requested()) {
@@ -2000,6 +2002,9 @@ static void main_loop(void)
int64_t ti;
#endif
do {
+ if (cexe_is_continuing_backward() && cexe_is_step_done()) {
+ cexe_next_reverse_continue_step();
+ }
nonblocking = !kvm_enabled() && !xen_enabled() && last_io > 0;
#ifdef CONFIG_PROFILER
ti = profile_getclock();
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 10/13] gdbstub: allow reverse execution in gdb stub.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
` (8 preceding siblings ...)
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 09/13] introduce reverse execution mechanism fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 11/13] cpu-exec: trigger a debug request when rexec stops fred.konrad
` (2 subsequent siblings)
12 siblings, 0 replies; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
This allows gdb to reverse step QEMU: reverse-stepi and reverse-cont commands
are allowed.
When step_backward is called, QEMU restores a snapshot before the actual
instruction and stops (with a debug exit) when the previous instruction is
reached.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
---
gdbstub.c | 31 ++++++++++++++++++++++++++++++-
1 file changed, 30 insertions(+), 1 deletion(-)
diff --git a/gdbstub.c b/gdbstub.c
index 8afe0b7..99769af 100644
--- a/gdbstub.c
+++ b/gdbstub.c
@@ -33,6 +33,7 @@
#include "sysemu/char.h"
#include "sysemu/sysemu.h"
#include "exec/gdbstub.h"
+#include "reverse-execution.h"
#endif
#define MAX_PACKET_LENGTH 4096
@@ -1113,6 +1114,17 @@ static int gdb_handle_packet(GDBState *s, const char *line_buf)
if (cc->gdb_core_xml_file != NULL) {
pstrcat(buf, sizeof(buf), ";qXfer:features:read+");
}
+
+ #ifndef CONFIG_USER_ONLY
+ /*
+ * When reverse execution is enabled those additional features must
+ * be set so GDB allows reverse-stepi and reverse-continue command.
+ */
+ if (cexe_is_enabled()) {
+ pstrcat(buf, sizeof(buf), ";ReverseStep+;ReverseContinue+");
+ }
+ #endif /* !CONFIG_USER_ONLY */
+
put_packet(s, buf);
break;
}
@@ -1161,7 +1173,23 @@ static int gdb_handle_packet(GDBState *s, const char *line_buf)
}
/* Unrecognised 'q' command. */
goto unknown_command;
-
+ #ifndef CONFIG_USER_ONLY
+ case 'b':
+ /* Reverse execution. */
+ switch (*p) {
+ case 's':
+ cexe_step_backward(s->c_cpu, 1);
+ break;
+ case 'c':
+ cexe_continue_backward(s->c_cpu);
+ break;
+ default:
+ buf[0] = '\0';
+ put_packet(s, buf);
+ break;
+ }
+ break;
+ #endif /* !CONFIG_USER_ONLY */
default:
unknown_command:
/* put empty packet */
@@ -1221,6 +1249,7 @@ static void gdb_vm_state_change(void *opaque, int running, RunState state)
ret = GDB_SIGNAL_TRAP;
break;
case RUN_STATE_PAUSED:
+ cexe_stop_reverse_continue();
ret = GDB_SIGNAL_INT;
break;
case RUN_STATE_SHUTDOWN:
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 11/13] cpu-exec: trigger a debug request when rexec stops.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
` (9 preceding siblings ...)
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 10/13] gdbstub: allow reverse execution in gdb stub fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 12/13] cexe: synchronize icount on the next event fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 13/13] cexe: allow to enable reverse execution fred.konrad
12 siblings, 0 replies; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
This allows QEMU to trigger a debug exception when cexe_dbg_requested
is set.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
---
cpu-exec.c | 13 +++++++++++++
1 file changed, 13 insertions(+)
diff --git a/cpu-exec.c b/cpu-exec.c
index 38e5f02..c92cfd1 100644
--- a/cpu-exec.c
+++ b/cpu-exec.c
@@ -22,6 +22,7 @@
#include "tcg.h"
#include "qemu/atomic.h"
#include "sysemu/qtest.h"
+#include "reverse-execution.h"
void cpu_loop_exit(CPUState *cpu)
{
@@ -230,6 +231,18 @@ int cpu_exec(CPUArchState *env)
/* This must be volatile so it is not trashed by longjmp() */
volatile bool have_tb_lock = false;
+ #ifndef CONFIG_USER_ONLY
+ if (cexe_is_enabled() && cexe_dbg_requested()) {
+ /*
+ * Reverse execution need to stop right now.
+ * So just generate a EXCP_DEBUG.
+ */
+ cpu->exception_index = EXCP_DEBUG;
+ cpu_handle_debug_exception(env);
+ return EXCP_DEBUG;
+ }
+ #endif /* !CONFIG_USER_ONLY */
+
if (cpu->halted) {
if (!cpu_has_work(cpu)) {
return EXCP_HALTED;
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 12/13] cexe: synchronize icount on the next event.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
` (10 preceding siblings ...)
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 11/13] cpu-exec: trigger a debug request when rexec stops fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 13/13] cexe: allow to enable reverse execution fred.konrad
12 siblings, 0 replies; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
We don't want to warp on host clock as it is not deterministic for replay.
So this patch warp icount on the next QEMU_VIRTUAL_CLOCK event if reverse
execution is enabled.
The normal behaviour is kept when reverse execution is disabled.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
---
cpus.c | 19 +++++++++++++++++--
include/qemu/timer.h | 8 ++++++++
include/reverse-execution.h | 2 ++
main-loop.c | 10 ++++++++++
stubs/Makefile.objs | 1 +
stubs/cexe-stub.c | 32 ++++++++++++++++++++++++++++++++
stubs/cpu-get-icount.c | 8 ++++++++
7 files changed, 78 insertions(+), 2 deletions(-)
create mode 100644 stubs/cexe-stub.c
diff --git a/cpus.c b/cpus.c
index e86045e..77d4700 100644
--- a/cpus.c
+++ b/cpus.c
@@ -321,8 +321,10 @@ static int64_t qemu_icount_round(int64_t count)
return (count + (1 << icount_time_shift) - 1) >> icount_time_shift;
}
-static void icount_warp_rt(void *opaque)
+void icount_warp_rt(void *opaque)
{
+ int64_t next_vm_deadline = -1;
+
/* The icount_warp_timer is rescheduled soon after vm_clock_warp_start
* changes from -1 to another value, so the race here is okay.
*/
@@ -330,6 +332,13 @@ static void icount_warp_rt(void *opaque)
return;
}
+ if (cexe_is_enabled()) {
+ /*
+ * We need this because the standard warp_delta is not deterministic.
+ */
+ next_vm_deadline = qemu_clock_deadline_ns_all(QEMU_CLOCK_VIRTUAL);
+ }
+
seqlock_write_lock(&timers_state.vm_clock_seqlock);
if (runstate_is_running()) {
int64_t clock = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
@@ -346,7 +355,13 @@ static void icount_warp_rt(void *opaque)
int64_t delta = cur_time - cur_icount;
warp_delta = MIN(warp_delta, delta);
}
- timers_state.qemu_icount_bias += warp_delta;
+ if (cexe_is_enabled()) {
+ if (next_vm_deadline > 0) {
+ timers_state.qemu_icount_bias += next_vm_deadline;
+ }
+ } else {
+ timers_state.qemu_icount_bias += warp_delta;
+ }
}
vm_clock_warp_start = -1;
seqlock_write_unlock(&timers_state.vm_clock_seqlock);
diff --git a/include/qemu/timer.h b/include/qemu/timer.h
index 0ae7f28..de2641a 100644
--- a/include/qemu/timer.h
+++ b/include/qemu/timer.h
@@ -754,6 +754,14 @@ int64_t cpu_get_icount(int with_bias);
int64_t cpu_get_clock(void);
int64_t cpu_icount_to_ns(int64_t icount);
+/**
+ * void icount_warp_rt:
+ *
+ * Move icount to the realtime clock or to the next QEMU_VIRTUAL_CLOCK event
+ * when reverse execution is enabled.
+ */
+void icount_warp_rt(void *opaque);
+
/*******************************************/
/* host CPU ticks (if available) */
diff --git a/include/reverse-execution.h b/include/reverse-execution.h
index bf42003..739572e 100644
--- a/include/reverse-execution.h
+++ b/include/reverse-execution.h
@@ -25,6 +25,8 @@
#ifndef REVERSE_EXECUTION
#define REVERSE_EXECUTION
+#include "qom/cpu.h"
+
void cexe_setup(void);
void cexe_step_backward(CPUState *cpu, uint64_t steps);
void cexe_stop_stepping_back_mode(void);
diff --git a/main-loop.c b/main-loop.c
index 8a85493..6130438 100644
--- a/main-loop.c
+++ b/main-loop.c
@@ -34,6 +34,8 @@
#include "qemu/compatfd.h"
+#include "reverse-execution.h"
+
/* If we have signalfd, we mask out the signals we want to handle and then
* use signalfd to listen for them. We rely on whatever the current signal
* handler is to dispatch the signals when we receive them.
@@ -489,6 +491,14 @@ int main_loop_wait(int nonblocking)
qemu_clock_run_all_timers();
+ /*
+ * Sometimes deadlock can appears because there is no pending event on
+ * virtual clock.
+ */
+ if (cexe_is_enabled()) {
+ icount_warp_rt(NULL);
+ }
+
return ret;
}
diff --git a/stubs/Makefile.objs b/stubs/Makefile.objs
index 528e161..5e362f6 100644
--- a/stubs/Makefile.objs
+++ b/stubs/Makefile.objs
@@ -39,3 +39,4 @@ stub-obj-$(CONFIG_WIN32) += fd-register.o
stub-obj-y += cpus.o
stub-obj-y += kvm.o
stub-obj-y += qmp_pc_dimm_device_list.o
+stub-obj-y += cexe-stub.o
diff --git a/stubs/cexe-stub.c b/stubs/cexe-stub.c
new file mode 100644
index 0000000..7723998
--- /dev/null
+++ b/stubs/cexe-stub.c
@@ -0,0 +1,32 @@
+/*
+ * cexe-stub.c
+ *
+ * Copyright (C) 2014 : GreenSocs Ltd
+ * http://www.greensocs.com/ , email: info@greensocs.com
+ *
+ * Developed by :
+ * Frederic Konrad <fred.konrad@greensocs.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License along
+ * with this program; if not, see <http://www.gnu.org/licenses/>.
+ *
+ */
+
+#include <stdbool.h>
+
+bool cexe_is_enabled(void);
+
+bool cexe_is_enabled(void)
+{
+ return false;
+}
diff --git a/stubs/cpu-get-icount.c b/stubs/cpu-get-icount.c
index 1968de7..231e659 100644
--- a/stubs/cpu-get-icount.c
+++ b/stubs/cpu-get-icount.c
@@ -7,3 +7,11 @@ int64_t cpu_get_icount(int with_bias)
{
abort();
}
+
+void icount_warp_rt(void *opaque)
+{
+ /*
+ * Should not happen, as cexe_is_enabled() always return false.
+ */
+ abort();
+}
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [Qemu-devel] [RFC PATCH v4 13/13] cexe: allow to enable reverse execution.
2014-06-25 8:26 [Qemu-devel] [RFC PATCH v4 00/13] Reverse execution fred.konrad
` (11 preceding siblings ...)
2014-06-25 8:26 ` [Qemu-devel] [RFC PATCH v4 12/13] cexe: synchronize icount on the next event fred.konrad
@ 2014-06-25 8:26 ` fred.konrad
12 siblings, 0 replies; 17+ messages in thread
From: fred.konrad @ 2014-06-25 8:26 UTC (permalink / raw)
To: qemu-devel
Cc: peter.maydell, quintela, mark.burton, dgilbert, amit.shah,
pbonzini, vilanova, fred.konrad
From: KONRAD Frederic <fred.konrad@greensocs.com>
This creates QEMU options for reverse execution.
Signed-off-by: KONRAD Frederic <fred.konrad@greensocs.com>
---
qemu-options.hx | 9 +++++++++
vl.c | 16 ++++++++++++++++
2 files changed, 25 insertions(+)
diff --git a/qemu-options.hx b/qemu-options.hx
index ff76ad4..2afb85d 100644
--- a/qemu-options.hx
+++ b/qemu-options.hx
@@ -2946,6 +2946,15 @@ order cores with complex cache hierarchies. The number of instructions
executed often has little or no correlation with actual performance.
ETEXI
+DEF("cexe", 0, QEMU_OPTION_cexe, \
+ "-cexe\n" \
+ " enable reverse execution\n", QEMU_ARCH_ALL)
+STEXI
+@item -cexe
+@findex -cexe
+Enable reverse execution.
+ETEXI
+
DEF("watchdog", HAS_ARG, QEMU_OPTION_watchdog, \
"-watchdog i6300esb|ib700\n" \
" enable virtual hardware watchdog [default=none]\n",
diff --git a/vl.c b/vl.c
index d0352e3..a94563a 100644
--- a/vl.c
+++ b/vl.c
@@ -228,6 +228,7 @@ static int default_floppy = 1;
static int default_cdrom = 1;
static int default_sdcard = 1;
static int default_vga = 1;
+static int default_cexe;
static struct {
const char *driver;
@@ -3821,6 +3822,9 @@ int main(int argc, char **argv, char **envp)
case QEMU_OPTION_icount:
icount_option = optarg;
break;
+ case QEMU_OPTION_cexe:
+ default_cexe = true;
+ break;
case QEMU_OPTION_incoming:
incoming = optarg;
runstate_set(RUN_STATE_INMIGRATE);
@@ -4520,6 +4524,15 @@ int main(int argc, char **argv, char **envp)
vm_start();
}
+ if (default_cexe) {
+ if (!icount_option) {
+ fprintf(stderr, "Reverse execution requires icount.\n");
+ exit(1);
+ }
+
+ cexe_setup();
+ }
+
os_setup_post();
if (is_daemonized()) {
@@ -4536,5 +4549,8 @@ int main(int argc, char **argv, char **envp)
tpm_cleanup();
#endif
+ if (cexe_is_enabled()) {
+ cexe_cleanup();
+ }
return 0;
}
--
1.9.0
^ permalink raw reply related [flat|nested] 17+ messages in thread