[PATCH] xl: introduce specific VCPU to PCPU mapping in config file

classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|

[PATCH] xl: introduce specific VCPU to PCPU mapping in config file

Dario Faggioli
xm supports the following syntax (in the config file) for
specific VCPU to PCPU mapping:

cpus = ["2", "3"] # VCPU0 runs on CPU2, VCPU1 runs on CPU3

Allow for the same to be done in xl.

Signed-off-by: Dario Faggioli <[hidden email]>

diff --git a/tools/libxl/xl_cmdimpl.c b/tools/libxl/xl_cmdimpl.c
--- a/tools/libxl/xl_cmdimpl.c
+++ b/tools/libxl/xl_cmdimpl.c
@@ -71,6 +71,8 @@ static uint32_t domid;
 static const char *common_domname;
 static int fd_lock = -1;
 
+/* Stash for specific vcpu to pcpu mappping */
+static int *vcpu_to_pcpu;
 
 static const char savefileheader_magic[32]=
     "Xen saved domain, xl format\n \0 \r";
@@ -630,6 +632,21 @@ static void parse_config_data(const char
             exit(1);
         }
 
+        /* Prepare the array for single vcpu to pcpu mappings */
+        vcpu_to_pcpu = xmalloc(sizeof(int) * b_info->max_vcpus);
+        memset(vcpu_to_pcpu, -1, sizeof(int) * b_info->max_vcpus);
+
+        /*
+         * Idea here is to let libxl think all the domain's vcpus
+         * have cpu affinity with all the pcpus on the list.
+         * It is then us, here in xl, that matches each single vcpu
+         * to its pcpu (and that's why we need to stash such info in
+         * the vcpu_to_pcpu array now) after the domain has been created.
+         * Doing it like this saves the burden of passing to libxl
+         * some big array hosting the single mappings. Also, using
+         * the cpumap derived from the list ensures memory is being
+         * allocated on the proper nodes anyway.
+         */
         libxl_cpumap_set_none(&b_info->cpumap);
         while ((buf = xlu_cfg_get_listitem(cpus, n_cpus)) != NULL) {
             i = atoi(buf);
@@ -638,6 +655,8 @@ static void parse_config_data(const char
                 exit(1);
             }
             libxl_cpumap_set(&b_info->cpumap, i);
+            if (n_cpus < b_info->max_vcpus)
+                vcpu_to_pcpu[n_cpus] = i;
             n_cpus++;
         }
     }
@@ -1709,6 +1728,31 @@ start:
     if ( ret )
         goto error_out;
 
+    /* If single vcpu to pcpu mapping was requested, honour it */
+    if (vcpu_to_pcpu) {
+        libxl_cpumap vcpu_cpumap;
+
+        libxl_cpumap_alloc(ctx, &vcpu_cpumap);
+        for (i = 0; i < d_config.b_info.max_vcpus; i++) {
+
+            if (vcpu_to_pcpu[i] != -1) {
+                libxl_cpumap_set_none(&vcpu_cpumap);
+                libxl_cpumap_set(&vcpu_cpumap, vcpu_to_pcpu[i]);
+            } else {
+                libxl_cpumap_set_any(&vcpu_cpumap);
+            }
+            if (libxl_set_vcpuaffinity(ctx, domid, i, &vcpu_cpumap)) {
+                fprintf(stderr, "setting affinity failed on vcpu `%d'.\n", i);
+                libxl_cpumap_dispose(&vcpu_cpumap);
+                free(vcpu_to_pcpu);
+                ret = ERROR_FAIL;
+                goto error_out;
+            }
+        }
+        libxl_cpumap_dispose(&vcpu_cpumap);
+        free(vcpu_to_pcpu);
+    }
+
     ret = libxl_userdata_store(ctx, domid, "xl",
                                     config_data, config_len);
     if (ret) {

_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl: introduce specific VCPU to PCPU mapping in config file

Dario Faggioli
On Sat, 2012-05-12 at 01:20 +0200, Dario Faggioli wrote:
> xm supports the following syntax (in the config file) for
> specific VCPU to PCPU mapping:
>
> cpus = ["2", "3"] # VCPU0 runs on CPU2, VCPU1 runs on CPU3
>
> Allow for the same to be done in xl.
>
And yes, I'm proposing this for 4.2, as not having it is a feature
parity between xm and xl bug someone reported on @xen-users.

Thanks and Regards,
Dario

--
<<This happens because I choose it to happen!>> (Raistlin Majere)
-----------------------------------------------------------------
Dario Faggioli, Ph.D, http://retis.sssup.it/people/faggioli
Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)



_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel

signature.asc (205 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl: introduce specific VCPU to PCPU mapping in config file

Dario Faggioli
On Sat, 2012-05-12 at 01:23 +0200, Dario Faggioli wrote:

> On Sat, 2012-05-12 at 01:20 +0200, Dario Faggioli wrote:
> > xm supports the following syntax (in the config file) for
> > specific VCPU to PCPU mapping:
> >
> > cpus = ["2", "3"] # VCPU0 runs on CPU2, VCPU1 runs on CPU3
> >
> > Allow for the same to be done in xl.
> >
> And yes, I'm proposing this for 4.2, as not having it is a feature
> parity between xm and xl bug someone reported on @xen-users.
>
Damn! I forgot to update the doc (xl manual) accordingly!

Feel free to provide any comments but, please, wait for a new and
complete version before applying.

Thanks and Regards,
Dario

--
<<This happens because I choose it to happen!>> (Raistlin Majere)
-----------------------------------------------------------------
Dario Faggioli, Ph.D, http://retis.sssup.it/people/faggioli
Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)



_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel

signature.asc (205 bytes) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl: introduce specific VCPU to PCPU mapping in config file

Ian Campbell-10
On Sun, 2012-05-13 at 09:35 +0100, Dario Faggioli wrote:

> On Sat, 2012-05-12 at 01:23 +0200, Dario Faggioli wrote:
> > On Sat, 2012-05-12 at 01:20 +0200, Dario Faggioli wrote:
> > > xm supports the following syntax (in the config file) for
> > > specific VCPU to PCPU mapping:
> > >
> > > cpus = ["2", "3"] # VCPU0 runs on CPU2, VCPU1 runs on CPU3
> > >
> > > Allow for the same to be done in xl.
> > >
> > And yes, I'm proposing this for 4.2, as not having it is a feature
> > parity between xm and xl bug someone reported on @xen-users.
> >
> Damn! I forgot to update the doc (xl manual) accordingly!

That was going to be my first comment ;-)

It would also be useful if the commit message mentions the impact on the
existing syntax -- I'm not sure if this is changing the behaviour of an
existing xl syntax or adding a whole new one (maybe the docs will answer
that).

Ian.

> Feel free to provide any comments but, please, wait for a new and
> complete version before applying.
>
> Thanks and Regards,
> Dario
>



_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl: introduce specific VCPU to PCPU mapping in config file

Ian Campbell-10
In reply to this post by Dario Faggioli
On Sat, 2012-05-12 at 00:20 +0100, Dario Faggioli wrote:

> xm supports the following syntax (in the config file) for
> specific VCPU to PCPU mapping:
>
> cpus = ["2", "3"] # VCPU0 runs on CPU2, VCPU1 runs on CPU3
>
> Allow for the same to be done in xl.
>
> Signed-off-by: Dario Faggioli <[hidden email]>
>
> diff --git a/tools/libxl/xl_cmdimpl.c b/tools/libxl/xl_cmdimpl.c
> --- a/tools/libxl/xl_cmdimpl.c
> +++ b/tools/libxl/xl_cmdimpl.c
> @@ -71,6 +71,8 @@ static uint32_t domid;
>  static const char *common_domname;
>  static int fd_lock = -1;
>  
> +/* Stash for specific vcpu to pcpu mappping */
> +static int *vcpu_to_pcpu;
>  
>  static const char savefileheader_magic[32]=
>      "Xen saved domain, xl format\n \0 \r";
> @@ -630,6 +632,21 @@ static void parse_config_data(const char
>              exit(1);
>          }
>  
> +        /* Prepare the array for single vcpu to pcpu mappings */
> +        vcpu_to_pcpu = xmalloc(sizeof(int) * b_info->max_vcpus);
> +        memset(vcpu_to_pcpu, -1, sizeof(int) * b_info->max_vcpus);
> +
> +        /*
> +         * Idea here is to let libxl think all the domain's vcpus
> +         * have cpu affinity with all the pcpus on the list.
> +         * It is then us, here in xl, that matches each single vcpu
> +         * to its pcpu (and that's why we need to stash such info in
> +         * the vcpu_to_pcpu array now) after the domain has been created.
> +         * Doing it like this saves the burden of passing to libxl
> +         * some big array hosting the single mappings. Also, using
> +         * the cpumap derived from the list ensures memory is being
> +         * allocated on the proper nodes anyway.

So effectively we create the domain pinned to the right nodes and then
rebind all the CPUS later to be mapped to specific phys-processors? This
means that the memory which is allocated is from the correct nodes, even
though we appear to do the pinning later. Clever.


> +         */
>          libxl_cpumap_set_none(&b_info->cpumap);
>          while ((buf = xlu_cfg_get_listitem(cpus, n_cpus)) != NULL) {
>              i = atoi(buf);
> @@ -638,6 +655,8 @@ static void parse_config_data(const char
>                  exit(1);
>              }
>              libxl_cpumap_set(&b_info->cpumap, i);
> +            if (n_cpus < b_info->max_vcpus)
> +                vcpu_to_pcpu[n_cpus] = i;
>              n_cpus++;
>          }
>      }
> @@ -1709,6 +1728,31 @@ start:
>      if ( ret )
>          goto error_out;
>  
> +    /* If single vcpu to pcpu mapping was requested, honour it */
> +    if (vcpu_to_pcpu) {

This is always allocated above, isn't it? I'm concerned that this might
break the non-1-1 case.

> +        libxl_cpumap vcpu_cpumap;
> +
> +        libxl_cpumap_alloc(ctx, &vcpu_cpumap);
> +        for (i = 0; i < d_config.b_info.max_vcpus; i++) {
> +
> +            if (vcpu_to_pcpu[i] != -1) {
> +                libxl_cpumap_set_none(&vcpu_cpumap);
> +                libxl_cpumap_set(&vcpu_cpumap, vcpu_to_pcpu[i]);
> +            } else {
> +                libxl_cpumap_set_any(&vcpu_cpumap);
> +            }
> +            if (libxl_set_vcpuaffinity(ctx, domid, i, &vcpu_cpumap)) {
> +                fprintf(stderr, "setting affinity failed on vcpu `%d'.\n", i);
> +                libxl_cpumap_dispose(&vcpu_cpumap);
> +                free(vcpu_to_pcpu);
> +                ret = ERROR_FAIL;
> +                goto error_out;
> +            }
> +        }
> +        libxl_cpumap_dispose(&vcpu_cpumap);
> +        free(vcpu_to_pcpu);

vpuc_to_pcpu = NULL, in case you go back around again...

For that reason it might be preferable to put vcpu_to_pcpu struct
dom_info and pass that to parse_config -- I think Goncalo is doing
something similar for the vncviewer option.

> +    }
> +
>      ret = libxl_userdata_store(ctx, domid, "xl",
>                                      config_data, config_len);
>      if (ret) {



_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel
Reply | Threaded
Open this post in threaded view
|

Re: [PATCH] xl: introduce specific VCPU to PCPU mapping in config file

Dario Faggioli
In reply to this post by Ian Campbell-10
On Mon, 2012-05-14 at 09:14 +0100, Ian Campbell wrote:
> > Damn! I forgot to update the doc (xl manual) accordingly!
>
> That was going to be my first comment ;-)
>
:-)

> It would also be useful if the commit message mentions the impact on the
> existing syntax -- I'm not sure if this is changing the behaviour of an
> existing xl syntax or adding a whole new one (maybe the docs will answer
> that).
>
From since when I added the `cpus=xxx` support to xl (back in February
and righ because it was a missing feature wrt to xm) it was using both:

cpus="2, 3"

and

cpus=["2", "3"]

to do the same thing, i.e., bind all the vcpus of the guest to the pcpus
#2 and #3 of the host.

On the other hand xm/xend do/did the following:

cpus="2, 3" --> bind all vcpus to pcpus #2 and #3

cpus=["2", "3"] --> bind vcpu #0 to pcpu #3 and vcpu #1 to pcpu #3

So, yes and no. :-) The point is using both syntax for he same thing in
xl hasn't been a good choice at all at that time, and this changest is
fixing that, other than improving xl-xm compatibility.

I hope I've clarified that here, and yes, I'll add something about this
in the commit message.

Thanks and Regards,
Dario

--
<<This happens because I choose it to happen!>> (Raistlin Majere)
-----------------------------------------------------------------
Dario Faggioli, Ph.D, http://retis.sssup.it/people/faggioli
Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)


_______________________________________________
Xen-devel mailing list
[hidden email]
http://lists.xen.org/xen-devel

signature.asc (205 bytes) Download Attachment