* [Qemu-devel] [PATCH v8 1/2] move vm_start to cpus.c
2017-02-14 17:07 [Qemu-devel] [PATCH v8 0/2] Qemu: gdbstub: fix vCont Claudio Imbrenda
@ 2017-02-14 17:07 ` Claudio Imbrenda
2017-02-14 17:07 ` [Qemu-devel] [PATCH v8 2/2] gdbstub: Fix vCont behaviour Claudio Imbrenda
2017-02-14 17:10 ` [Qemu-devel] [PATCH v8 0/2] Qemu: gdbstub: fix vCont Paolo Bonzini
2 siblings, 0 replies; 4+ messages in thread
From: Claudio Imbrenda @ 2017-02-14 17:07 UTC (permalink / raw)
To: pbonzini; +Cc: borntraeger, qemu-devel, palves, alex.bennee
This patch:
* moves vm_start to cpus.c.
* exports qemu_vmstop_requested, since it's needed by vm_start.
* extracts vm_prepare_start from vm_start; it does what vm_start did,
except restarting the cpus.
* vm_start now calls vm_prepare_start and then restarts the cpus.
Signed-off-by: Claudio Imbrenda <imbrenda@linux.vnet.ibm.com>
---
cpus.c | 42 ++++++++++++++++++++++++++++++++++++++++++
include/sysemu/sysemu.h | 2 ++
vl.c | 30 +-----------------------------
3 files changed, 45 insertions(+), 29 deletions(-)
diff --git a/cpus.c b/cpus.c
index 71a82e5..0bcb5b5 100644
--- a/cpus.c
+++ b/cpus.c
@@ -1578,6 +1578,48 @@ int vm_stop(RunState state)
return do_vm_stop(state);
}
+/**
+ * Prepare for (re)starting the VM.
+ * Returns -1 if the vCPUs are not to be restarted (e.g. if they are already
+ * running or in case of an error condition), 0 otherwise.
+ */
+int vm_prepare_start(void)
+{
+ RunState requested;
+ int res = 0;
+
+ qemu_vmstop_requested(&requested);
+ if (runstate_is_running() && requested == RUN_STATE__MAX) {
+ return -1;
+ }
+
+ /* Ensure that a STOP/RESUME pair of events is emitted if a
+ * vmstop request was pending. The BLOCK_IO_ERROR event, for
+ * example, according to documentation is always followed by
+ * the STOP event.
+ */
+ if (runstate_is_running()) {
+ 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);
+ }
+
+ /* We are sending this now, but the CPUs will be resumed shortly later */
+ qapi_event_send_resume(&error_abort);
+ return res;
+}
+
+void vm_start(void)
+{
+ if (!vm_prepare_start()) {
+ resume_all_vcpus();
+ }
+}
+
/* does a state transition even if the VM is already stopped,
current state is forgotten forever */
int vm_stop_force_state(RunState state)
diff --git a/include/sysemu/sysemu.h b/include/sysemu/sysemu.h
index ff8ffb5..0bb0c40 100644
--- a/include/sysemu/sysemu.h
+++ b/include/sysemu/sysemu.h
@@ -37,6 +37,7 @@ void vm_state_notify(int running, RunState state);
#define VMRESET_REPORT true
void vm_start(void);
+int vm_prepare_start(void);
int vm_stop(RunState state);
int vm_stop_force_state(RunState state);
@@ -60,6 +61,7 @@ void qemu_register_powerdown_notifier(Notifier *notifier);
void qemu_system_debug_request(void);
void qemu_system_vmstop_request(RunState reason);
void qemu_system_vmstop_request_prepare(void);
+bool qemu_vmstop_requested(RunState *r);
int qemu_shutdown_requested_get(void);
int qemu_reset_requested_get(void);
void qemu_system_killed(int signal, pid_t pid);
diff --git a/vl.c b/vl.c
index abb0900..7a32a30 100644
--- a/vl.c
+++ b/vl.c
@@ -757,7 +757,7 @@ StatusInfo *qmp_query_status(Error **errp)
return info;
}
-static bool qemu_vmstop_requested(RunState *r)
+bool qemu_vmstop_requested(RunState *r)
{
qemu_mutex_lock(&vmstop_lock);
*r = vmstop_requested;
@@ -778,34 +778,6 @@ void qemu_system_vmstop_request(RunState state)
qemu_notify_event();
}
-void vm_start(void)
-{
- RunState requested;
-
- qemu_vmstop_requested(&requested);
- if (runstate_is_running() && requested == RUN_STATE__MAX) {
- return;
- }
-
- /* Ensure that a STOP/RESUME pair of events is emitted if a
- * vmstop request was pending. The BLOCK_IO_ERROR event, for
- * example, according to documentation is always followed by
- * the STOP event.
- */
- if (runstate_is_running()) {
- qapi_event_send_stop(&error_abort);
- } else {
- replay_enable_events();
- cpu_enable_ticks();
- runstate_set(RUN_STATE_RUNNING);
- vm_state_notify(1, RUN_STATE_RUNNING);
- resume_all_vcpus();
- }
-
- qapi_event_send_resume(&error_abort);
-}
-
-
/***********************************************************/
/* real time host monotonic timer */
--
2.7.4
^ permalink raw reply related [flat|nested] 4+ messages in thread* [Qemu-devel] [PATCH v8 2/2] gdbstub: Fix vCont behaviour
2017-02-14 17:07 [Qemu-devel] [PATCH v8 0/2] Qemu: gdbstub: fix vCont Claudio Imbrenda
2017-02-14 17:07 ` [Qemu-devel] [PATCH v8 1/2] move vm_start to cpus.c Claudio Imbrenda
@ 2017-02-14 17:07 ` Claudio Imbrenda
2017-02-14 17:10 ` [Qemu-devel] [PATCH v8 0/2] Qemu: gdbstub: fix vCont Paolo Bonzini
2 siblings, 0 replies; 4+ messages in thread
From: Claudio Imbrenda @ 2017-02-14 17:07 UTC (permalink / raw)
To: pbonzini; +Cc: borntraeger, qemu-devel, palves, alex.bennee
When GDB issues a "vCont", QEMU was not handling it correctly when
multiple VCPUs are active.
For vCont, for each thread (VCPU), it can be specified whether to
single step, continue or stop that thread. The default is to stop a
thread.
However, when (for example) "vCont;s:2" is issued, all VCPUs continue
to run, although all but VCPU nr 2 are to be stopped.
This patch completely rewrites the vCont parsing code.
Please note that this improvement only works in system emulation mode,
when in userspace emulation mode the old behaviour is preserved.
Signed-off-by: Claudio Imbrenda <imbrenda@linux.vnet.ibm.com>
---
gdbstub.c | 209 ++++++++++++++++++++++++++++++++++++++++++++++++--------------
1 file changed, 162 insertions(+), 47 deletions(-)
diff --git a/gdbstub.c b/gdbstub.c
index de9b62b..0068f0e 100644
--- a/gdbstub.c
+++ b/gdbstub.c
@@ -386,6 +386,60 @@ static inline void gdb_continue(GDBState *s)
#endif
}
+/*
+ * Resume execution, per CPU actions. For user-mode emulation it's
+ * equivalent to gdb_continue.
+ */
+static int gdb_continue_partial(GDBState *s, char *newstates)
+{
+ CPUState *cpu;
+ int res = 0;
+#ifdef CONFIG_USER_ONLY
+ /*
+ * This is not exactly accurate, but it's an improvement compared to the
+ * previous situation, where only one CPU would be single-stepped.
+ */
+ CPU_FOREACH(cpu) {
+ if (newstates[cpu->cpu_index] == 's') {
+ cpu_single_step(cpu, sstep_flags);
+ }
+ }
+ s->running_state = 1;
+#else
+ int flag = 0;
+
+ if (!runstate_needs_reset()) {
+ if (vm_prepare_start()) {
+ return 0;
+ }
+
+ CPU_FOREACH(cpu) {
+ switch (newstates[cpu->cpu_index]) {
+ case 0:
+ case 1:
+ break; /* nothing to do here */
+ case 's':
+ cpu_single_step(cpu, sstep_flags);
+ cpu_resume(cpu);
+ flag = 1;
+ break;
+ case 'c':
+ cpu_resume(cpu);
+ flag = 1;
+ break;
+ default:
+ res = -1;
+ break;
+ }
+ }
+ }
+ if (flag) {
+ qemu_clock_enable(QEMU_CLOCK_VIRTUAL, true);
+ }
+#endif
+ return res;
+}
+
static void put_buffer(GDBState *s, const uint8_t *buf, int len)
{
#ifdef CONFIG_USER_ONLY
@@ -784,6 +838,107 @@ static int is_query_packet(const char *p, const char *query, char separator)
(p[query_len] == '\0' || p[query_len] == separator);
}
+/**
+ * gdb_handle_vcont - Parses and handles a vCont packet.
+ * returns -ENOTSUP if a command is unsupported, -EINVAL or -ERANGE if there is
+ * a format error, 0 on success.
+ */
+static int gdb_handle_vcont(GDBState *s, const char *p)
+{
+ int res, idx, signal = 0;
+ char cur_action;
+ char *newstates;
+ unsigned long tmp;
+ CPUState *cpu;
+#ifdef CONFIG_USER_ONLY
+ int max_cpus = 1; /* global variable max_cpus exists only in system mode */
+
+ CPU_FOREACH(cpu) {
+ max_cpus = max_cpus <= cpu->cpu_index ? cpu->cpu_index + 1 : max_cpus;
+ }
+#endif
+ /* uninitialised CPUs stay 0 */
+ newstates = g_new0(char, max_cpus);
+
+ /* mark valid CPUs with 1 */
+ CPU_FOREACH(cpu) {
+ newstates[cpu->cpu_index] = 1;
+ }
+
+ /*
+ * res keeps track of what error we are returning, with -ENOTSUP meaning
+ * that the command is unknown or unsupported, thus returning an empty
+ * packet, while -EINVAL and -ERANGE cause an E22 packet, due to invalid,
+ * or incorrect parameters passed.
+ */
+ res = 0;
+ while (*p) {
+ if (*p++ != ';') {
+ res = -ENOTSUP;
+ goto out;
+ }
+
+ cur_action = *p++;
+ if (cur_action == 'C' || cur_action == 'S') {
+ cur_action = tolower(cur_action);
+ res = qemu_strtoul(p + 1, &p, 16, &tmp);
+ if (res) {
+ goto out;
+ }
+ signal = gdb_signal_to_target(tmp);
+ } else if (cur_action != 'c' && cur_action != 's') {
+ /* unknown/invalid/unsupported command */
+ res = -ENOTSUP;
+ goto out;
+ }
+ /* thread specification. special values: (none), -1 = all; 0 = any */
+ if ((p[0] == ':' && p[1] == '-' && p[2] == '1') || (p[0] != ':')) {
+ if (*p == ':') {
+ p += 3;
+ }
+ for (idx = 0; idx < max_cpus; idx++) {
+ if (newstates[idx] == 1) {
+ newstates[idx] = cur_action;
+ }
+ }
+ } else if (*p == ':') {
+ p++;
+ res = qemu_strtoul(p, &p, 16, &tmp);
+ if (res) {
+ goto out;
+ }
+ idx = tmp;
+ /* 0 means any thread, so we pick the first valid CPU */
+ if (!idx) {
+ idx = cpu_index(first_cpu);
+ }
+
+ /*
+ * If we are in user mode, the thread specified is actually a
+ * thread id, and not an index. We need to find the actual
+ * CPU first, and only then we can use its index.
+ */
+ cpu = find_cpu(idx);
+ /* invalid CPU/thread specified */
+ if (!idx || !cpu) {
+ res = -EINVAL;
+ goto out;
+ }
+ /* only use if no previous match occourred */
+ if (newstates[cpu->cpu_index] == 1) {
+ newstates[cpu->cpu_index] = cur_action;
+ }
+ }
+ }
+ s->signal = signal;
+ gdb_continue_partial(s, newstates);
+
+out:
+ g_free(newstates);
+
+ return res;
+}
+
static int gdb_handle_packet(GDBState *s, const char *line_buf)
{
CPUState *cpu;
@@ -829,60 +984,20 @@ static int gdb_handle_packet(GDBState *s, const char *line_buf)
return RS_IDLE;
case 'v':
if (strncmp(p, "Cont", 4) == 0) {
- int res_signal, res_thread;
-
p += 4;
if (*p == '?') {
put_packet(s, "vCont;c;C;s;S");
break;
}
- res = 0;
- res_signal = 0;
- res_thread = 0;
- while (*p) {
- int action, signal;
-
- if (*p++ != ';') {
- res = 0;
- break;
- }
- action = *p++;
- signal = 0;
- if (action == 'C' || action == 'S') {
- signal = gdb_signal_to_target(strtoul(p, (char **)&p, 16));
- if (signal == -1) {
- signal = 0;
- }
- } else if (action != 'c' && action != 's') {
- res = 0;
- break;
- }
- thread = 0;
- if (*p == ':') {
- thread = strtoull(p+1, (char **)&p, 16);
- }
- action = tolower(action);
- if (res == 0 || (res == 'c' && action == 's')) {
- res = action;
- res_signal = signal;
- res_thread = thread;
- }
- }
+
+ res = gdb_handle_vcont(s, p);
+
if (res) {
- if (res_thread != -1 && res_thread != 0) {
- cpu = find_cpu(res_thread);
- if (cpu == NULL) {
- put_packet(s, "E22");
- break;
- }
- s->c_cpu = cpu;
- }
- if (res == 's') {
- cpu_single_step(s->c_cpu, sstep_flags);
+ if ((res == -EINVAL) || (res == -ERANGE)) {
+ put_packet(s, "E22");
+ break;
}
- s->signal = res_signal;
- gdb_continue(s);
- return RS_IDLE;
+ goto unknown_command;
}
break;
} else {
--
2.7.4
^ permalink raw reply related [flat|nested] 4+ messages in thread* Re: [Qemu-devel] [PATCH v8 0/2] Qemu: gdbstub: fix vCont
2017-02-14 17:07 [Qemu-devel] [PATCH v8 0/2] Qemu: gdbstub: fix vCont Claudio Imbrenda
2017-02-14 17:07 ` [Qemu-devel] [PATCH v8 1/2] move vm_start to cpus.c Claudio Imbrenda
2017-02-14 17:07 ` [Qemu-devel] [PATCH v8 2/2] gdbstub: Fix vCont behaviour Claudio Imbrenda
@ 2017-02-14 17:10 ` Paolo Bonzini
2 siblings, 0 replies; 4+ messages in thread
From: Paolo Bonzini @ 2017-02-14 17:10 UTC (permalink / raw)
To: Claudio Imbrenda; +Cc: borntraeger, qemu-devel, palves, alex.bennee
Thanks, I hope to send a pull request this week, including this patch.
Paolo
On 14/02/2017 18:07, Claudio Imbrenda wrote:
> This small patchset fixes the incorrect behaviour of the vCont command
> in the gdb stub.
>
> The first patch, as suggested be Paolo, refactors some code. The most
> visible change is that it moves vm_start to cpus.c
>
> The second one fixes the incorrect behaviour of the vCont command.
> Previously, continuing or stepping a single thread (CPU) caused all
> other CPUs to be started too, whereas the GDB specification clearly
> states that without a default action all threads not explicitly
> mentioned in the command should stay stopped.
>
> So if the Qemu gdbstub receives a vCont;c:1 packet, no other CPU
> should be restarted except the first, and when a vCont;s:1 is
> received, the first CPU should be stepped without restarting the others.
> With this patchset Qemu now behaves as expected.
>
> See here for reference material about the packets:
> https://sourceware.org/gdb/current/onlinedocs/gdb/Packets.html
> https://sourceware.org/gdb/onlinedocs/gdb/Packets.html
>
> v7 -> v8
> * fixed and added some comments
> * fixed vCont a little bit for user mode
> * use cpu->cpu_index directly when possible
>
> v6 -> v7
> * fixed description of patch 1 to reflect what is actually happening
> and improved description of patch 2
> * removed leftover header declaration of resume_some_vcpus which had
> been removed a few versions ago
> * fixed a compilation issue when compiling userspace-mode only
> (global variable max_cpus is not defined when not in system-mode)
>
> v4 -> v6
> * rebased on master after target-s390x was moved
> * put qemu_clock_enable back into resume_all_vcpus
> * improved the parsing function of the vCont packet
> * added qemu_clock_enable to gdb_continue_partial
>
> v3 -> v4
> * rebased on v2.8.0-rc2, no changes needed
>
> v2 -> v3
> * removed resume_some_vcpus
> * cleared up the code and simplified the implementation in light of the
> clarification in the specification of the vCont packet
>
> Claudio Imbrenda (2):
> move vm_start to cpus.c
> gdbstub: Fix vCont behaviour
>
> cpus.c | 42 ++++++++++
> gdbstub.c | 209 +++++++++++++++++++++++++++++++++++++-----------
> include/sysemu/sysemu.h | 2 +
> vl.c | 30 +------
> 4 files changed, 207 insertions(+), 76 deletions(-)
>
^ permalink raw reply [flat|nested] 4+ messages in thread