Patchwork [[PATCH,v2] ] Add option to slow qemu down

login
register
mail settings
Submitter Wolfgang Mauerer
Date May 7, 2013, 9:12 a.m.
Message ID <5188C58A.3020203@siemens.com>
Download mbox | patch
Permalink /patch/242085/
State New
Headers show

Pull-request

git://github.com/siemens/qemu.git for-upstream

Comments

Wolfgang Mauerer - May 7, 2013, 9:12 a.m.
Hello,

On 17/01/13 21:21, Stefan Weil wrote:
> Am 17.01.2013 18:10, schrieb Wolfgang Mauerer:
>> For slow targets and fast hosts, the emulation may be faster
>> than the actual hardware, which can be undesirable for various
>> reasons. Add a run-time option to slow down the emulation
>> by sleeping in the CPU emulation.
>>
(...)  
> Please test your patch using scripts/checkpatch.pl before sending it.

all stylistic issues have already been addressed in the last iteration;
do you have additional comments on the code besides those raised in
http://lists.gnu.org/archive/html/qemu-devel/2013-01/msg03248.html?
Here's a pull request:

The following changes since commit b5803aa3583e82e5133f7621121bc15ee694f4a1:

  Merge remote-tracking branch 'qemu-kvm/uq/master' into staging (2013-05-06 15:45:08 -0500)

are available in the git repository at:


  git://github.com/siemens/qemu.git for-upstream

for you to fetch changes up to 3e3a2c7158139020aa9c172ba902694e6a531cbe:

  Add option to slow qemu down (2013-05-07 10:43:31 +0200)

----------------------------------------------------------------
Wolfgang Mauerer (1):
      Add option to slow qemu down

 cpus.c                |   30 ++++++++++++++++++++++++++++++
 include/qemu-common.h |    2 ++
 qemu-options.hx       |   14 ++++++++++++++
 vl.c                  |   10 ++++++++++
 4 files changed, 56 insertions(+)

I've also attached the patch below.

Thanks, Wolfgang

######################################################################
From 3e3a2c7158139020aa9c172ba902694e6a531cbe Mon Sep 17 00:00:00 2001
From: Wolfgang Mauerer <wolfgang.mauerer@siemens.com>
Date: Thu, 17 Jan 2013 15:53:39 +0100
Subject: [PATCH] Add option to slow qemu down

For slow targets and fast hosts, the emulation may be faster
than the actual hardware, which can be undesirable for various
reasons. Add a run-time option to slow down the emulation
by sleeping in the CPU emulation.

Signed-off-by: Wolfgang Mauerer <wolfgang.mauerer@siemens.com>
---
 cpus.c                |   30 ++++++++++++++++++++++++++++++
 include/qemu-common.h |    2 ++
 qemu-options.hx       |   14 ++++++++++++++
 vl.c                  |   10 ++++++++++
 4 files changed, 56 insertions(+)
Markus Armbruster - May 7, 2013, 10:56 a.m.
Wolfgang Mauerer <wolfgang.mauerer@siemens.com> writes:

> Hello,
>
> On 17/01/13 21:21, Stefan Weil wrote:
>> Am 17.01.2013 18:10, schrieb Wolfgang Mauerer:
>>> For slow targets and fast hosts, the emulation may be faster
>>> than the actual hardware, which can be undesirable for various
>>> reasons. Add a run-time option to slow down the emulation
>>> by sleeping in the CPU emulation.
>>>
> (...)  
>> Please test your patch using scripts/checkpatch.pl before sending it.
>
> all stylistic issues have already been addressed in the last iteration;
> do you have additional comments on the code besides those raised in
> http://lists.gnu.org/archive/html/qemu-devel/2013-01/msg03248.html?
> Here's a pull request:
>
> The following changes since commit b5803aa3583e82e5133f7621121bc15ee694f4a1:
>
>   Merge remote-tracking branch 'qemu-kvm/uq/master' into staging (2013-05-06 15:45:08 -0500)
>
> are available in the git repository at:
>
>
>   git://github.com/siemens/qemu.git for-upstream
>
> for you to fetch changes up to 3e3a2c7158139020aa9c172ba902694e6a531cbe:
>
>   Add option to slow qemu down (2013-05-07 10:43:31 +0200)
>
> ----------------------------------------------------------------
> Wolfgang Mauerer (1):
>       Add option to slow qemu down
>
>  cpus.c                |   30 ++++++++++++++++++++++++++++++
>  include/qemu-common.h |    2 ++
>  qemu-options.hx       |   14 ++++++++++++++
>  vl.c                  |   10 ++++++++++
>  4 files changed, 56 insertions(+)
>
> I've also attached the patch below.
>
> Thanks, Wolfgang
>
> ######################################################################
>>From 3e3a2c7158139020aa9c172ba902694e6a531cbe Mon Sep 17 00:00:00 2001
> From: Wolfgang Mauerer <wolfgang.mauerer@siemens.com>
> Date: Thu, 17 Jan 2013 15:53:39 +0100
> Subject: [PATCH] Add option to slow qemu down
>
> For slow targets and fast hosts, the emulation may be faster
> than the actual hardware, which can be undesirable for various
> reasons. Add a run-time option to slow down the emulation
> by sleeping in the CPU emulation.
>
> Signed-off-by: Wolfgang Mauerer <wolfgang.mauerer@siemens.com>
> ---
>  cpus.c                |   30 ++++++++++++++++++++++++++++++
>  include/qemu-common.h |    2 ++
>  qemu-options.hx       |   14 ++++++++++++++
>  vl.c                  |   10 ++++++++++
>  4 files changed, 56 insertions(+)
>
> diff --git a/cpus.c b/cpus.c
> index c232265..2cff5f3 100644
> --- a/cpus.c
> +++ b/cpus.c
> @@ -61,6 +61,7 @@
>  #endif /* CONFIG_LINUX */
>  
>  static CPUArchState *next_cpu;
> +static bool use_slowdown;
>  
>  static bool cpu_thread_is_idle(CPUArchState *env)
>  {
> @@ -106,6 +107,8 @@ static QEMUTimer *icount_warp_timer;
>  static int64_t vm_clock_warp_start;
>  static int64_t qemu_icount;
>  
> +static double slowdown_factor;
> +
>  typedef struct TimersState {
>      int64_t cpu_ticks_prev;
>      int64_t cpu_ticks_offset;
> @@ -385,6 +388,21 @@ void configure_icount(const char *option)
>                     qemu_get_clock_ns(vm_clock) + get_ticks_per_sec() / 10);
>  }
>  
> +void configure_slowdown(const char *option)
> +{
> +    if (!option) {
> +        return;
> +    }
> +
> +    slowdown_factor = strtod(option, NULL) / 100.0;
> +    /* We cannot provide speedups, obviously */
> +    if (slowdown_factor < 0) {
> +        slowdown_factor *= -1.0;
> +    }
> +
> +    use_slowdown = true;
> +}
> +
>  /***********************************************************/
>  void hw_error(const char *fmt, ...)
>  {
> @@ -1121,6 +1139,8 @@ void vm_stop_force_state(RunState state)
>  static int tcg_cpu_exec(CPUArchState *env)
>  {
>      int ret;
> +    int64_t ss = 0; /* Initialise to avoid bogous maybe-uninitialized error */
> +    uint64_t slowdown_sleep;
>  #ifdef CONFIG_PROFILER
>      int64_t ti;
>  #endif
> @@ -1141,7 +1161,17 @@ static int tcg_cpu_exec(CPUArchState *env)
>          env->icount_decr.u16.low = decr;
>          env->icount_extra = count;
>      }
> +    if (use_slowdown) {
> +        ss = qemu_get_clock_ns(rt_clock);
> +    }
> +
>      ret = cpu_exec(env);
> +
> +    if (use_slowdown) {
> +        slowdown_sleep = qemu_get_clock_ns(rt_clock) - ss;
> +        slowdown_sleep *= slowdown_factor;
> +        g_usleep(slowdown_sleep / 1024);
> +    }
>  #ifdef CONFIG_PROFILER
>      qemu_time += profile_getclock() - ti;
>  #endif
> diff --git a/include/qemu-common.h b/include/qemu-common.h
> index b399d85..219993f 100644
> --- a/include/qemu-common.h
> +++ b/include/qemu-common.h
> @@ -108,6 +108,8 @@ static inline char *realpath(const char *path, char *resolved_path)
>  void configure_icount(const char *option);
>  extern int use_icount;
>  
> +void configure_slowdown(const char *option);
> +
>  #include "qemu/osdep.h"
>  #include "qemu/bswap.h"
>  
> diff --git a/qemu-options.hx b/qemu-options.hx
> index fb62b75..c764e42 100644
> --- a/qemu-options.hx
> +++ b/qemu-options.hx
> @@ -2801,6 +2801,20 @@ order cores with complex cache hierarchies.  The number of instructions
>  executed often has little or no correlation with actual performance.
>  ETEXI
>  
> +DEF("slowdown", HAS_ARG, QEMU_OPTION_slowdown, \
> +    "-slowdown s\n" \
> +    "                Slow down the CPU emulation by (approximately) s percent\n",
> +    QEMU_ARCH_ALL)
> +STEXI
> +@item -slowdown @var{s}
> +@findex -slowdown
> +Slow down the virtual CPU by approximately s percent (i.e., for c time
> +units of execution time, sleep for c*s/100 time units). This makes it
> +possible to align the emulated machine's performance roughly with
> +the performance of physical entities, but does not provide identical
> +performance profiles since the emulation is not cycle accurate.
> +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 6e6225f..27acf8b 100644
> --- a/vl.c
> +++ b/vl.c
> @@ -2839,6 +2839,7 @@ int main(int argc, char **argv, char **envp)
>      int i;
>      int snapshot, linux_boot;
>      const char *icount_option = NULL;
> +    const char *slowdown_option = NULL;
>      const char *initrd_filename;
>      const char *kernel_filename, *kernel_cmdline;
>      char boot_devices[33] = "";
> @@ -3751,6 +3752,9 @@ int main(int argc, char **argv, char **envp)
>              case QEMU_OPTION_icount:
>                  icount_option = optarg;
>                  break;
> +            case QEMU_OPTION_slowdown:
> +                slowdown_option = optarg;
> +                break;
>              case QEMU_OPTION_incoming:
>                  incoming = optarg;
>                  runstate_set(RUN_STATE_INMIGRATE);
> @@ -4169,6 +4173,12 @@ int main(int argc, char **argv, char **envp)
>      /* clean up network at qemu process termination */
>      atexit(&net_cleanup);
>  
> +    if (slowdown_option && (kvm_enabled() || xen_enabled())) {
> +        fprintf(stderr, "-slowdown is not allowed with kvm or xen\n");
> +        exit(1);

Suggest to mention this in the commit message, and perhaps the manual
(STEXI..ETEXI section in qemu-options.hx).

> +    }
> +    configure_slowdown(slowdown_option);
> +
>      if (net_init_clients() < 0) {
>          exit(1);
>      }

Patch

diff --git a/cpus.c b/cpus.c
index c232265..2cff5f3 100644
--- a/cpus.c
+++ b/cpus.c
@@ -61,6 +61,7 @@ 
 #endif /* CONFIG_LINUX */
 
 static CPUArchState *next_cpu;
+static bool use_slowdown;
 
 static bool cpu_thread_is_idle(CPUArchState *env)
 {
@@ -106,6 +107,8 @@  static QEMUTimer *icount_warp_timer;
 static int64_t vm_clock_warp_start;
 static int64_t qemu_icount;
 
+static double slowdown_factor;
+
 typedef struct TimersState {
     int64_t cpu_ticks_prev;
     int64_t cpu_ticks_offset;
@@ -385,6 +388,21 @@  void configure_icount(const char *option)
                    qemu_get_clock_ns(vm_clock) + get_ticks_per_sec() / 10);
 }
 
+void configure_slowdown(const char *option)
+{
+    if (!option) {
+        return;
+    }
+
+    slowdown_factor = strtod(option, NULL) / 100.0;
+    /* We cannot provide speedups, obviously */
+    if (slowdown_factor < 0) {
+        slowdown_factor *= -1.0;
+    }
+
+    use_slowdown = true;
+}
+
 /***********************************************************/
 void hw_error(const char *fmt, ...)
 {
@@ -1121,6 +1139,8 @@  void vm_stop_force_state(RunState state)
 static int tcg_cpu_exec(CPUArchState *env)
 {
     int ret;
+    int64_t ss = 0; /* Initialise to avoid bogous maybe-uninitialized error */
+    uint64_t slowdown_sleep;
 #ifdef CONFIG_PROFILER
     int64_t ti;
 #endif
@@ -1141,7 +1161,17 @@  static int tcg_cpu_exec(CPUArchState *env)
         env->icount_decr.u16.low = decr;
         env->icount_extra = count;
     }
+    if (use_slowdown) {
+        ss = qemu_get_clock_ns(rt_clock);
+    }
+
     ret = cpu_exec(env);
+
+    if (use_slowdown) {
+        slowdown_sleep = qemu_get_clock_ns(rt_clock) - ss;
+        slowdown_sleep *= slowdown_factor;
+        g_usleep(slowdown_sleep / 1024);
+    }
 #ifdef CONFIG_PROFILER
     qemu_time += profile_getclock() - ti;
 #endif
diff --git a/include/qemu-common.h b/include/qemu-common.h
index b399d85..219993f 100644
--- a/include/qemu-common.h
+++ b/include/qemu-common.h
@@ -108,6 +108,8 @@  static inline char *realpath(const char *path, char *resolved_path)
 void configure_icount(const char *option);
 extern int use_icount;
 
+void configure_slowdown(const char *option);
+
 #include "qemu/osdep.h"
 #include "qemu/bswap.h"
 
diff --git a/qemu-options.hx b/qemu-options.hx
index fb62b75..c764e42 100644
--- a/qemu-options.hx
+++ b/qemu-options.hx
@@ -2801,6 +2801,20 @@  order cores with complex cache hierarchies.  The number of instructions
 executed often has little or no correlation with actual performance.
 ETEXI
 
+DEF("slowdown", HAS_ARG, QEMU_OPTION_slowdown, \
+    "-slowdown s\n" \
+    "                Slow down the CPU emulation by (approximately) s percent\n",
+    QEMU_ARCH_ALL)
+STEXI
+@item -slowdown @var{s}
+@findex -slowdown
+Slow down the virtual CPU by approximately s percent (i.e., for c time
+units of execution time, sleep for c*s/100 time units). This makes it
+possible to align the emulated machine's performance roughly with
+the performance of physical entities, but does not provide identical
+performance profiles since the emulation is not cycle accurate.
+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 6e6225f..27acf8b 100644
--- a/vl.c
+++ b/vl.c
@@ -2839,6 +2839,7 @@  int main(int argc, char **argv, char **envp)
     int i;
     int snapshot, linux_boot;
     const char *icount_option = NULL;
+    const char *slowdown_option = NULL;
     const char *initrd_filename;
     const char *kernel_filename, *kernel_cmdline;
     char boot_devices[33] = "";
@@ -3751,6 +3752,9 @@  int main(int argc, char **argv, char **envp)
             case QEMU_OPTION_icount:
                 icount_option = optarg;
                 break;
+            case QEMU_OPTION_slowdown:
+                slowdown_option = optarg;
+                break;
             case QEMU_OPTION_incoming:
                 incoming = optarg;
                 runstate_set(RUN_STATE_INMIGRATE);
@@ -4169,6 +4173,12 @@  int main(int argc, char **argv, char **envp)
     /* clean up network at qemu process termination */
     atexit(&net_cleanup);
 
+    if (slowdown_option && (kvm_enabled() || xen_enabled())) {
+        fprintf(stderr, "-slowdown is not allowed with kvm or xen\n");
+        exit(1);
+    }
+    configure_slowdown(slowdown_option);
+
     if (net_init_clients() < 0) {
         exit(1);
     }