[PATCH] tools/xl: correctly shows split eventchannel for netfront

[PATCH] tools/xl: correctly shows split eventchannel for netfront

[Annie Li]

From: Annie Li <annie.li@oracle.com>

After split eventchannel feature was supported by netback/netfront,
"xl network-list" does not show eventchannel correctly. Add tx-/rx-evt-ch
to show tx/rx eventchannel correctly.

Signed-off-by: Annie Li <annie.li@oracle.com>
---
 tools/libxl/libxl.c         |   11 ++++++++++-
 tools/libxl/libxl_types.idl |    3 ++-
 tools/libxl/xl_cmdimpl.c    |   12 ++++++------
 3 files changed, 18 insertions(+), 8 deletions(-)

diff --git a/tools/libxl/libxl.c b/tools/libxl/libxl.c
index 2845ca4..4222687 100644
--- a/tools/libxl/libxl.c
+++ b/tools/libxl/libxl.c
@@ -3125,6 +3125,7 @@ int libxl_device_nic_getinfo(libxl_ctx *ctx, uint32_t domid,
     GC_INIT(ctx);
     char *dompath, *nicpath;
     char *val;
+    int  evtch;
 
     dompath = libxl__xs_get_dompath(gc, domid);
     nicinfo->devid = nic->devid;
@@ -3141,7 +3142,15 @@ int libxl_device_nic_getinfo(libxl_ctx *ctx, uint32_t domid,
     val = libxl__xs_read(gc, XBT_NULL, libxl__sprintf(gc, "%s/state", nicpath));
     nicinfo->state = val ? strtoul(val, NULL, 10) : -1;
     val = libxl__xs_read(gc, XBT_NULL, libxl__sprintf(gc, "%s/event-channel", nicpath));
-    nicinfo->evtch = val ? strtoul(val, NULL, 10) : -1;
+    evtch = val ? strtoul(val, NULL, 10) : -1;
+    if(evtch > 0)
+        nicinfo->evtch_tx = nicinfo->evtch_rx = evtch;
+    else {
+        val = libxl__xs_read(gc, XBT_NULL, libxl__sprintf(gc, "%s/event-channel-tx", nicpath));
+        nicinfo->evtch_tx = val ? strtoul(val, NULL, 10) : -1;
+        val = libxl__xs_read(gc, XBT_NULL, libxl__sprintf(gc, "%s/event-channel-rx", nicpath));
+        nicinfo->evtch_rx = val ? strtoul(val, NULL, 10) : -1;
+    }
     val = libxl__xs_read(gc, XBT_NULL, libxl__sprintf(gc, "%s/tx-ring-ref", nicpath));
     nicinfo->rref_tx = val ? strtoul(val, NULL, 10) : -1;
     val = libxl__xs_read(gc, XBT_NULL, libxl__sprintf(gc, "%s/rx-ring-ref", nicpath));
diff --git a/tools/libxl/libxl_types.idl b/tools/libxl/libxl_types.idl
index 649ce50..e6368c7 100644
--- a/tools/libxl/libxl_types.idl
+++ b/tools/libxl/libxl_types.idl
@@ -488,7 +488,8 @@ libxl_nicinfo = Struct("nicinfo", [
     ("frontend_id", uint32),
     ("devid", libxl_devid),
     ("state", integer),
-    ("evtch", integer),
+    ("evtch_tx", integer),
+    ("evtch_rx", integer),
     ("rref_tx", integer),
     ("rref_rx", integer),
     ], dir=DIR_OUT)
diff --git a/tools/libxl/xl_cmdimpl.c b/tools/libxl/xl_cmdimpl.c
index c30f495..7353187 100644
--- a/tools/libxl/xl_cmdimpl.c
+++ b/tools/libxl/xl_cmdimpl.c
@@ -5842,9 +5842,9 @@ int main_networklist(int argc, char **argv)
         /* No options */
     }
 
-    /*      Idx  BE   MAC   Hdl  Sta  evch txr/rxr  BE-path */
-    printf("%-3s %-2s %-17s %-6s %-5s %-6s %5s/%-5s %-30s\n",
-           "Idx", "BE", "Mac Addr.", "handle", "state", "evt-ch", "tx-", "rx-ring-ref", "BE-path");
+    /*      Idx  BE   MAC   Hdl  Sta  txev/rxev txr/rxr  BE-path */
+    printf("%-3s %-2s %-17s %-6s %-5s %6s/%-6s %5s/%-5s %-30s\n",
+           "Idx", "BE", "Mac Addr.", "handle", "state", "tx-", "rx-evt-ch", "tx-", "rx-ring-ref", "BE-path");
     for (argv += optind, argc -= optind; argc > 0; --argc, ++argv) {
         uint32_t domid = find_domain(*argv);
         nics = libxl_device_nic_list(ctx, domid, &nb);
@@ -5857,9 +5857,9 @@ int main_networklist(int argc, char **argv)
                 printf("%-3d %-2d ", nicinfo.devid, nicinfo.backend_id);
                 /* MAC */
                 printf(LIBXL_MAC_FMT, LIBXL_MAC_BYTES(nics[i].mac));
-                /* Hdl  Sta  evch txr/rxr  BE-path */
-                printf("%6d %5d %6d %5d/%-11d %-30s\n",
-                       nicinfo.devid, nicinfo.state, nicinfo.evtch,
+                /* Hdl  Sta  txev/rxev txr/rxr  BE-path */
+                printf(" %6d %5d %6d/%-9d %5d/%-11d %-30s\n",
+                       nicinfo.devid, nicinfo.state, nicinfo.evtch_tx, nicinfo.evtch_rx,
                        nicinfo.rref_tx, nicinfo.rref_rx, nicinfo.backend);
                 libxl_nicinfo_dispose(&nicinfo);
             }
-- 
1.7.3.4

Re: [PATCH] tools/xl: correctly shows split eventchannel for netfront

[Ian Campbell]

On Wed, 2014-01-15 at 02:33 +0800, Annie Li wrote:
> From: Annie Li <annie.li@oracle.com>
> 
> After split eventchannel feature was supported by netback/netfront,
> "xl network-list" does not show eventchannel correctly. Add tx-/rx-evt-ch
> to show tx/rx eventchannel correctly.
> 
> Signed-off-by: Annie Li <annie.li@oracle.com>

How critical is this for 4.4?

Please consider
http://wiki.xen.org/wiki/Xen_Roadmap/4.4#Exception_guidelines_for_after_the_code_freeze and make a case for it if you think it should go in.

> diff --git a/tools/libxl/libxl_types.idl b/tools/libxl/libxl_types.idl
> index 649ce50..e6368c7 100644
> --- a/tools/libxl/libxl_types.idl
> +++ b/tools/libxl/libxl_types.idl
> @@ -488,7 +488,8 @@ libxl_nicinfo = Struct("nicinfo", [
>      ("frontend_id", uint32),
>      ("devid", libxl_devid),
>      ("state", integer),
> -    ("evtch", integer),
> +    ("evtch_tx", integer),
> +    ("evtch_rx", integer),

This needs backwards compatibility handling, see the big comment at the
head of libxl.h and the other examples in that file. I'm doubtful that
you will be able to remove the evtch field without breaking the API, so
it probably needs to stay even if it is explicitly invalid under some
circumstances.

It also needs a suitable LIBXL_HAVE_ #define, again see libxl.h.

Ian.

Re: [PATCH] tools/xl: correctly shows split eventchannel for netfront

[annie li]

On 2014-1-15 21:13, Ian Campbell wrote:
> On Wed, 2014-01-15 at 02:33 +0800, Annie Li wrote:
>> From: Annie Li <annie.li@oracle.com>
>>
>> After split eventchannel feature was supported by netback/netfront,
>> "xl network-list" does not show eventchannel correctly. Add tx-/rx-evt-ch
>> to show tx/rx eventchannel correctly.
>>
>> Signed-off-by: Annie Li <annie.li@oracle.com>
> How critical is this for 4.4?

I think it can wait. This issue only happens with split event channel 
feature implemented in latest netback/netfront, "xl network-list" works 
OK for old netback/netfront.

>
> Please consider
> http://wiki.xen.org/wiki/Xen_Roadmap/4.4#Exception_guidelines_for_after_the_code_freeze and make a case for it if you think it should go in.
>
>> diff --git a/tools/libxl/libxl_types.idl b/tools/libxl/libxl_types.idl
>> index 649ce50..e6368c7 100644
>> --- a/tools/libxl/libxl_types.idl
>> +++ b/tools/libxl/libxl_types.idl
>> @@ -488,7 +488,8 @@ libxl_nicinfo = Struct("nicinfo", [
>>       ("frontend_id", uint32),
>>       ("devid", libxl_devid),
>>       ("state", integer),
>> -    ("evtch", integer),
>> +    ("evtch_tx", integer),
>> +    ("evtch_rx", integer),
> This needs backwards compatibility handling, see the big comment at the
> head of libxl.h and the other examples in that file. I'm doubtful that
> you will be able to remove the evtch field without breaking the API, so
> it probably needs to stay even if it is explicitly invalid under some
> circumstances.
>
> It also needs a suitable LIBXL_HAVE_ #define, again see libxl.h.

Yes, this patch does not handle backwards compatibility,  and probably 
breaks the API. Let me fix them, thanks!

Thanks
Annie
>
> Ian.
>
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel

Re: [PATCH] tools/xl: correctly shows split eventchannel for netfront

[David Vrabel]

On 14/01/14 18:33, Annie Li wrote:
> From: Annie Li <annie.li@oracle.com>
> 
> After split eventchannel feature was supported by netback/netfront,
> "xl network-list" does not show eventchannel correctly. Add tx-/rx-evt-ch
> to show tx/rx eventchannel correctly.
[...]
> --- a/tools/libxl/libxl_types.idl
> +++ b/tools/libxl/libxl_types.idl
> @@ -488,7 +488,8 @@ libxl_nicinfo = Struct("nicinfo", [
>      ("frontend_id", uint32),
>      ("devid", libxl_devid),
>      ("state", integer),
> -    ("evtch", integer),
> +    ("evtch_tx", integer),
> +    ("evtch_rx", integer),

Does this break libxl's API or ABI?

David

Re: [PATCH] tools/xl: correctly shows split eventchannel for netfront

[Ian Campbell]

On Wed, 2014-01-15 at 15:31 +0000, David Vrabel wrote:
> On 14/01/14 18:33, Annie Li wrote:
> > From: Annie Li <annie.li@oracle.com>
> > 
> > After split eventchannel feature was supported by netback/netfront,
> > "xl network-list" does not show eventchannel correctly. Add tx-/rx-evt-ch
> > to show tx/rx eventchannel correctly.
> [...]
> > --- a/tools/libxl/libxl_types.idl
> > +++ b/tools/libxl/libxl_types.idl
> > @@ -488,7 +488,8 @@ libxl_nicinfo = Struct("nicinfo", [
> >      ("frontend_id", uint32),
> >      ("devid", libxl_devid),
> >      ("state", integer),
> > -    ("evtch", integer),
> > +    ("evtch_tx", integer),
> > +    ("evtch_rx", integer),
> 
> Does this break libxl's API or ABI?

Both, but we only guarantee API stability, not ABI stability (IOW we'd
just bump the SONAME).

Ian.