Upgrade from xen-4.1-testing to xen-unstable report

Upgrade from xen-4.1-testing to xen-unstable report

[Teck Choon Giam]

Hi,

This is just my experience about issues I encountered when upgrade
from xen-4.1-testing changeset 23277:80130491806f to xen-unstable
changeset 25191:a95fc7decc83.

1. Immediately after upgrade, xl list show such error:

# xl list
libxl: error: libxl.c:506:libxl_list_domain: geting domain info list:
Permission denied
libxl_domain_infolist failed.

After a reboot, it is fine.  Any idea why such behaviour?  Imagine if
there are running domUs... this might cause issues to shutdown?  I
will downgrade and repeat such test to confirm.  Might be worth a note
in upgrading note about this if this is intended?

2. localtime setting not working.  Set to localtime=1 doesn't seems to
work whereby setting rtc_timeoffset works.  Any idea?

3. xl trigger hvmdomain power still never remove those related
iptables forward rules.  This is the same problem when I test in
xen-4.1-testing.  Refer to
http://lists.xen.org/archives/html/xen-devel/2012-02/msg00771.html
about past report.  When test in xen-4.1-testing, one funny behaviour
is if I start hvmdomain using xm then use xl trigger hvmdomain power
does remove those iptables related rule chain.

I shall do more testing when time permit and sorry if this report is
useless to anyone.

Thanks.

Kindest regards,
Giam Teck Choon

Re: Upgrade from xen-4.1-testing to xen-unstable report

[Ian Campbell]

On Wed, 2012-04-11 at 22:23 +0100, Teck Choon Giam wrote:
> Hi,
> 
> This is just my experience about issues I encountered when upgrade
> from xen-4.1-testing changeset 23277:80130491806f to xen-unstable
> changeset 25191:a95fc7decc83.
> 
> 1. Immediately after upgrade, xl list show such error:
> 
> # xl list
> libxl: error: libxl.c:506:libxl_list_domain: geting domain info list:
> Permission denied
> libxl_domain_infolist failed.
> 
> After a reboot, it is fine.  Any idea why such behaviour?  Imagine if
> there are running domUs... this might cause issues to shutdown?  I
> will downgrade and repeat such test to confirm.  Might be worth a note
> in upgrading note about this if this is intended?

The tools and the hypervisor are a matched pair so you would need to
reboot the system in order to use the new tools. This has always been
the case with Xen upgrades.

> 2. localtime setting not working.  Set to localtime=1 doesn't seems to
> work whereby setting rtc_timeoffset works.  Any idea?

I've CC'd Lin Ming who implemented both of those.

> 3. xl trigger hvmdomain power still never remove those related
> iptables forward rules.  This is the same problem when I test in
> xen-4.1-testing.  Refer to
> http://lists.xen.org/archives/html/xen-devel/2012-02/msg00771.html
> about past report.  When test in xen-4.1-testing, one funny behaviour
> is if I start hvmdomain using xm then use xl trigger hvmdomain power
> does remove those iptables related rule chain.

I think this is an issue with the hotplug scripts not being run at the
right time. There are some improvements to this in the pipeline which I
hope will help here.

> I shall do more testing when time permit and sorry if this report is
> useless to anyone.

On the contrary it is always useful to hear about these things, thank
you.

Ian.

Re: Upgrade from xen-4.1-testing to xen-unstable report

[Teck Choon Giam]

On Thu, Apr 12, 2012 at 8:08 AM, Ian Campbell <Ian.Campbell@citrix.com> wrote:
> On Wed, 2012-04-11 at 22:23 +0100, Teck Choon Giam wrote:
>> Hi,
>>
>> This is just my experience about issues I encountered when upgrade
>> from xen-4.1-testing changeset 23277:80130491806f to xen-unstable
>> changeset 25191:a95fc7decc83.
>>
>> 1. Immediately after upgrade, xl list show such error:
>>
>> # xl list
>> libxl: error: libxl.c:506:libxl_list_domain: geting domain info list:
>> Permission denied
>> libxl_domain_infolist failed.
>>
>> After a reboot, it is fine.  Any idea why such behaviour?  Imagine if
>> there are running domUs... this might cause issues to shutdown?  I
>> will downgrade and repeat such test to confirm.  Might be worth a note
>> in upgrading note about this if this is intended?
>
> The tools and the hypervisor are a matched pair so you would need to
> reboot the system in order to use the new tools. This has always been
> the case with Xen upgrades.

I think you mean minor version upgrade is fine but not for major
version upgrade which I can understand ;)

>
>> 2. localtime setting not working.  Set to localtime=1 doesn't seems to
>> work whereby setting rtc_timeoffset works.  Any idea?
>
> I've CC'd Lin Ming who implemented both of those.

Thanks.  I know Lin Ming implemented this.

>
>> 3. xl trigger hvmdomain power still never remove those related
>> iptables forward rules.  This is the same problem when I test in
>> xen-4.1-testing.  Refer to
>> http://lists.xen.org/archives/html/xen-devel/2012-02/msg00771.html
>> about past report.  When test in xen-4.1-testing, one funny behaviour
>> is if I start hvmdomain using xm then use xl trigger hvmdomain power
>> does remove those iptables related rule chain.
>
> I think this is an issue with the hotplug scripts not being run at the
> right time. There are some improvements to this in the pipeline which I
> hope will help here.

Hope so :)  Anyway, same problem if hvmdomain shutdown itself.

>
>> I shall do more testing when time permit and sorry if this report is
>> useless to anyone.
>
> On the contrary it is always useful to hear about these things, thank
> you.

My main frequent testing and production rely on xen-4.1-testing not
xen-unstable but I will try my best to do such testing in xen-unstable
once in a while until xen v4.2.0 officially released.

>
> Ian.
>

Thanks Ian for taking time to response ;)

Kindest regards,
Giam Teck Choon

Re: Upgrade from xen-4.1-testing to xen-unstable report

[Lin Ming]

On Thu, 2012-04-12 at 07:08 +0000, Ian Campbell wrote:
> On Wed, 2012-04-11 at 22:23 +0100, Teck Choon Giam wrote:
> > Hi,
> > 
> > This is just my experience about issues I encountered when upgrade
> > from xen-4.1-testing changeset 23277:80130491806f to xen-unstable
> > changeset 25191:a95fc7decc83.
> > 
> > 1. Immediately after upgrade, xl list show such error:
> > 
> > # xl list
> > libxl: error: libxl.c:506:libxl_list_domain: geting domain info list:
> > Permission denied
> > libxl_domain_infolist failed.
> > 
> > After a reboot, it is fine.  Any idea why such behaviour?  Imagine if
> > there are running domUs... this might cause issues to shutdown?  I
> > will downgrade and repeat such test to confirm.  Might be worth a note
> > in upgrading note about this if this is intended?
> 
> The tools and the hypervisor are a matched pair so you would need to
> reboot the system in order to use the new tools. This has always been
> the case with Xen upgrades.
> 
> > 2. localtime setting not working.  Set to localtime=1 doesn't seems to
> > work whereby setting rtc_timeoffset works.  Any idea?
> 
> I've CC'd Lin Ming who implemented both of those.

Just did a quick gdb debug, seems libxl__domain_build_info_setdefault
was called 3 times. So below statement was executed 3 times too.

b_info->rtc_timeoffset += tm->tm_gmtoff;

Will look at this issue later.

(gdb) bt
#0  libxl__domain_build_info_setdefault (gc=0x7fffffffdc40, b_info=0x7fffffffdd90) at libxl_create.c:70
#1  0x00007ffff7991fd4 in libxl_domain_need_memory (ctx=0x626050, b_info=0x7fffffffdd90, need_memkb=0x7fffffffdc84) at libxl.c:2696
#2  0x000000000040b0ca in freemem (b_info=0x7fffffffdd90) at xl_cmdimpl.c:1399
#3  0x000000000040bd6d in create_domain (dom_info=0x7fffffffe040) at xl_cmdimpl.c:1656
#4  0x000000000041142c in main_create (argc=1, argv=0x7fffffffe638) at xl_cmdimpl.c:3446
#5  0x00000000004069d8 in main (argc=2, argv=0x7fffffffe630) at xl.c:165
(gdb) 

(gdb) bt
#0  libxl__domain_build_info_setdefault (gc=0x7fffffffdc70, b_info=0x7fffffffdd90) at libxl_create.c:70
#1  0x00007ffff799739c in do_domain_create (gc=0x7fffffffdc70, d_config=0x7fffffffdd50, cb=0, priv=0x7fffffffdd34, domid_out=0x625dc8, restore_fd=-1)
    at libxl_create.c:571
#2  0x00007ffff7997cdd in libxl_domain_create_new (ctx=0x626050, d_config=0x7fffffffdd50, cb=0, priv=0x7fffffffdd34, domid=0x625dc8) at libxl_create.c:738
#3  0x000000000040be4a in create_domain (dom_info=0x7fffffffe040) at xl_cmdimpl.c:1679
#4  0x000000000041142c in main_create (argc=1, argv=0x7fffffffe638) at xl_cmdimpl.c:3446
#5  0x00000000004069d8 in main (argc=2, argv=0x7fffffffe630) at xl.c:165
(gdb)

(gdb) bt
#0  libxl__domain_build_info_setdefault (gc=0x7fffffffdaa0, b_info=0x7fffffffdd90) at libxl_create.c:70
#1  0x00007ffff79b2146 in libxl_run_bootloader (ctx=0x626050, info=0x7fffffffdd90, disk=0x626b20, domid=9) at libxl_bootloader.c:349
#2  0x00007ffff7997464 in do_domain_create (gc=0x7fffffffdc70, d_config=0x7fffffffdd50, cb=0, priv=0x7fffffffdd34, domid_out=0x625dc8, restore_fd=-1)
    at libxl_create.c:580
#3  0x00007ffff7997cdd in libxl_domain_create_new (ctx=0x626050, d_config=0x7fffffffdd50, cb=0, priv=0x7fffffffdd34, domid=0x625dc8) at libxl_create.c:738
#4  0x000000000040be4a in create_domain (dom_info=0x7fffffffe040) at xl_cmdimpl.c:1679
#5  0x000000000041142c in main_create (argc=1, argv=0x7fffffffe638) at xl_cmdimpl.c:3446
#6  0x00000000004069d8 in main (argc=2, argv=0x7fffffffe630) at xl.c:165
(gdb) 

Re: Upgrade from xen-4.1-testing to xen-unstable report

[Ian Campbell]

On Thu, 2012-04-12 at 13:50 +0100, Lin Ming wrote:
> On Thu, 2012-04-12 at 07:08 +0000, Ian Campbell wrote:
> > On Wed, 2012-04-11 at 22:23 +0100, Teck Choon Giam wrote:
> > > Hi,
> > > 
> > > This is just my experience about issues I encountered when upgrade
> > > from xen-4.1-testing changeset 23277:80130491806f to xen-unstable
> > > changeset 25191:a95fc7decc83.
> > > 
> > > 1. Immediately after upgrade, xl list show such error:
> > > 
> > > # xl list
> > > libxl: error: libxl.c:506:libxl_list_domain: geting domain info list:
> > > Permission denied
> > > libxl_domain_infolist failed.
> > > 
> > > After a reboot, it is fine.  Any idea why such behaviour?  Imagine if
> > > there are running domUs... this might cause issues to shutdown?  I
> > > will downgrade and repeat such test to confirm.  Might be worth a note
> > > in upgrading note about this if this is intended?
> > 
> > The tools and the hypervisor are a matched pair so you would need to
> > reboot the system in order to use the new tools. This has always been
> > the case with Xen upgrades.
> > 
> > > 2. localtime setting not working.  Set to localtime=1 doesn't seems to
> > > work whereby setting rtc_timeoffset works.  Any idea?
> > 
> > I've CC'd Lin Ming who implemented both of those.
> 
> Just did a quick gdb debug, seems libxl__domain_build_info_setdefault
> was called 3 times. So below statement was executed 3 times too.
> 
> b_info->rtc_timeoffset += tm->tm_gmtoff;

Oh, that sounds like the problem. The setdefault functions are supposed
to be idempotent, IOW calling it two or more times should give the same
result as calling it just once.

I think you need to move this logic from setdefault
    if (libxl_defbool_val(b_info->localtime)) {
        time_t t;
        struct tm *tm;

        t = time(NULL);
        tm = localtime(&t);

        b_info->rtc_timeoffset += tm->tm_gmtoff;
    }

into libxl__build_pre and do that calculation on a temporary variable
whjich you pass to xc_domain_set_time_offset rather than modifying
b_info.

Ian.

Re: Upgrade from xen-4.1-testing to xen-unstable report

[Lin Ming]

On Thu, 2012-04-12 at 15:03 +0100, Ian Campbell wrote:
> On Thu, 2012-04-12 at 13:50 +0100, Lin Ming wrote:
> > On Thu, 2012-04-12 at 07:08 +0000, Ian Campbell wrote:
> > > On Wed, 2012-04-11 at 22:23 +0100, Teck Choon Giam wrote:
> > > > Hi,
> > > > 
> > > > This is just my experience about issues I encountered when upgrade
> > > > from xen-4.1-testing changeset 23277:80130491806f to xen-unstable
> > > > changeset 25191:a95fc7decc83.
> > > > 
> > > > 1. Immediately after upgrade, xl list show such error:
> > > > 
> > > > # xl list
> > > > libxl: error: libxl.c:506:libxl_list_domain: geting domain info list:
> > > > Permission denied
> > > > libxl_domain_infolist failed.
> > > > 
> > > > After a reboot, it is fine.  Any idea why such behaviour?  Imagine if
> > > > there are running domUs... this might cause issues to shutdown?  I
> > > > will downgrade and repeat such test to confirm.  Might be worth a note
> > > > in upgrading note about this if this is intended?
> > > 
> > > The tools and the hypervisor are a matched pair so you would need to
> > > reboot the system in order to use the new tools. This has always been
> > > the case with Xen upgrades.
> > > 
> > > > 2. localtime setting not working.  Set to localtime=1 doesn't seems to
> > > > work whereby setting rtc_timeoffset works.  Any idea?
> > > 
> > > I've CC'd Lin Ming who implemented both of those.
> > 
> > Just did a quick gdb debug, seems libxl__domain_build_info_setdefault
> > was called 3 times. So below statement was executed 3 times too.
> > 
> > b_info->rtc_timeoffset += tm->tm_gmtoff;
> 
> Oh, that sounds like the problem. The setdefault functions are supposed
> to be idempotent, IOW calling it two or more times should give the same
> result as calling it just once.
> 
> I think you need to move this logic from setdefault
>     if (libxl_defbool_val(b_info->localtime)) {
>         time_t t;
>         struct tm *tm;
> 
>         t = time(NULL);
>         tm = localtime(&t);
> 
>         b_info->rtc_timeoffset += tm->tm_gmtoff;
>     }
> 
> into libxl__build_pre and do that calculation on a temporary variable
> whjich you pass to xc_domain_set_time_offset rather than modifying
> b_info.

How about below patch?

>From e06589cb5f1efd885f9589405d0f4ecc97427ad6 Mon Sep 17 00:00:00 2001
From: Lin Ming <mlin@ss.pku.edu.cn>
Date: Thu, 12 Apr 2012 22:36:24 +0800
Subject: [PATCH] libxl: fix rtc_timeoffset setting

libxl__domain_build_info_setdefault may be called several times, so
rtc_timeoffset can't be set in it.

Move rtc_timeoffset setting logic to libxl__build_pre.

Reported-by: Teck Choon Giam <giamteckchoon@gmail.com>
Signed-off-by: Lin Ming <mlin@ss.pku.edu.cn>
---
 tools/libxl/libxl_create.c |    9 ---------
 tools/libxl/libxl_dom.c    |   17 +++++++++++++++--
 2 files changed, 15 insertions(+), 11 deletions(-)

diff --git a/tools/libxl/libxl_create.c b/tools/libxl/libxl_create.c
index e63c7bd..e706124 100644
--- a/tools/libxl/libxl_create.c
+++ b/tools/libxl/libxl_create.c
@@ -127,15 +127,6 @@ int libxl__domain_build_info_setdefault(libxl__gc *gc,
         b_info->target_memkb = b_info->max_memkb;
 
     libxl_defbool_setdefault(&b_info->localtime, false);
-    if (libxl_defbool_val(b_info->localtime)) {
-        time_t t;
-        struct tm *tm;
-
-        t = time(NULL);
-        tm = localtime(&t);
-
-        b_info->rtc_timeoffset += tm->tm_gmtoff;
-    }
 
     libxl_defbool_setdefault(&b_info->disable_migrate, false);
 
diff --git a/tools/libxl/libxl_dom.c b/tools/libxl/libxl_dom.c
index 0bdd654..91c4bd8 100644
--- a/tools/libxl/libxl_dom.c
+++ b/tools/libxl/libxl_dom.c
@@ -65,6 +65,8 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
     libxl_ctx *ctx = libxl__gc_owner(gc);
     int tsc_mode;
     char *xs_domid, *con_domid;
+    uint32_t rtc_timeoffset;
+
     xc_domain_max_vcpus(ctx->xch, domid, info->max_vcpus);
     libxl_set_vcpuaffinity_all(ctx, domid, info->max_vcpus, &info->cpumap);
     xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb + LIBXL_MAXMEM_CONSTANT);
@@ -91,8 +93,19 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
     if (libxl_defbool_val(info->disable_migrate))
         xc_domain_disable_migrate(ctx->xch, domid);
 
-    if (info->rtc_timeoffset)
-        xc_domain_set_time_offset(ctx->xch, domid, info->rtc_timeoffset);
+    rtc_timeoffset = info->rtc_timeoffset;
+    if (libxl_defbool_val(info->localtime)) {
+        time_t t;
+        struct tm *tm;
+
+        t = time(NULL);
+        tm = localtime(&t);
+
+        rtc_timeoffset += tm->tm_gmtoff;
+    }
+
+    if (rtc_timeoffset)
+        xc_domain_set_time_offset(ctx->xch, domid, rtc_timeoffset);
 
     if (info->type == LIBXL_DOMAIN_TYPE_HVM) {
         unsigned long shadow;
-- 
1.7.2.5



> 
> Ian.
> 
> 

Re: Upgrade from xen-4.1-testing to xen-unstable report

[Ian Campbell]

On Thu, 2012-04-12 at 15:42 +0100, Lin Ming wrote:
> On Thu, 2012-04-12 at 15:03 +0100, Ian Campbell wrote:
> > On Thu, 2012-04-12 at 13:50 +0100, Lin Ming wrote:
> > > On Thu, 2012-04-12 at 07:08 +0000, Ian Campbell wrote:
> > > > On Wed, 2012-04-11 at 22:23 +0100, Teck Choon Giam wrote:
> > > > > Hi,
> > > > > 
> > > > > This is just my experience about issues I encountered when upgrade
> > > > > from xen-4.1-testing changeset 23277:80130491806f to xen-unstable
> > > > > changeset 25191:a95fc7decc83.
> > > > > 
> > > > > 1. Immediately after upgrade, xl list show such error:
> > > > > 
> > > > > # xl list
> > > > > libxl: error: libxl.c:506:libxl_list_domain: geting domain info list:
> > > > > Permission denied
> > > > > libxl_domain_infolist failed.
> > > > > 
> > > > > After a reboot, it is fine.  Any idea why such behaviour?  Imagine if
> > > > > there are running domUs... this might cause issues to shutdown?  I
> > > > > will downgrade and repeat such test to confirm.  Might be worth a note
> > > > > in upgrading note about this if this is intended?
> > > > 
> > > > The tools and the hypervisor are a matched pair so you would need to
> > > > reboot the system in order to use the new tools. This has always been
> > > > the case with Xen upgrades.
> > > > 
> > > > > 2. localtime setting not working.  Set to localtime=1 doesn't seems to
> > > > > work whereby setting rtc_timeoffset works.  Any idea?
> > > > 
> > > > I've CC'd Lin Ming who implemented both of those.
> > > 
> > > Just did a quick gdb debug, seems libxl__domain_build_info_setdefault
> > > was called 3 times. So below statement was executed 3 times too.
> > > 
> > > b_info->rtc_timeoffset += tm->tm_gmtoff;
> > 
> > Oh, that sounds like the problem. The setdefault functions are supposed
> > to be idempotent, IOW calling it two or more times should give the same
> > result as calling it just once.
> > 
> > I think you need to move this logic from setdefault
> >     if (libxl_defbool_val(b_info->localtime)) {
> >         time_t t;
> >         struct tm *tm;
> > 
> >         t = time(NULL);
> >         tm = localtime(&t);
> > 
> >         b_info->rtc_timeoffset += tm->tm_gmtoff;
> >     }
> > 
> > into libxl__build_pre and do that calculation on a temporary variable
> > whjich you pass to xc_domain_set_time_offset rather than modifying
> > b_info.
> 
> How about below patch?
> 
> From e06589cb5f1efd885f9589405d0f4ecc97427ad6 Mon Sep 17 00:00:00 2001
> From: Lin Ming <mlin@ss.pku.edu.cn>
> Date: Thu, 12 Apr 2012 22:36:24 +0800
> Subject: [PATCH] libxl: fix rtc_timeoffset setting
> 
> libxl__domain_build_info_setdefault may be called several times, so
> rtc_timeoffset can't be set in it.
> 
> Move rtc_timeoffset setting logic to libxl__build_pre.
> 
> Reported-by: Teck Choon Giam <giamteckchoon@gmail.com>
> Signed-off-by: Lin Ming <mlin@ss.pku.edu.cn>

Looks good to me.

Acked-by: Ian Campbell <ian.campbell@citrix.com>

> ---
>  tools/libxl/libxl_create.c |    9 ---------
>  tools/libxl/libxl_dom.c    |   17 +++++++++++++++--
>  2 files changed, 15 insertions(+), 11 deletions(-)
> 
> diff --git a/tools/libxl/libxl_create.c b/tools/libxl/libxl_create.c
> index e63c7bd..e706124 100644
> --- a/tools/libxl/libxl_create.c
> +++ b/tools/libxl/libxl_create.c
> @@ -127,15 +127,6 @@ int libxl__domain_build_info_setdefault(libxl__gc *gc,
>          b_info->target_memkb = b_info->max_memkb;
>  
>      libxl_defbool_setdefault(&b_info->localtime, false);
> -    if (libxl_defbool_val(b_info->localtime)) {
> -        time_t t;
> -        struct tm *tm;
> -
> -        t = time(NULL);
> -        tm = localtime(&t);
> -
> -        b_info->rtc_timeoffset += tm->tm_gmtoff;
> -    }
>  
>      libxl_defbool_setdefault(&b_info->disable_migrate, false);
>  
> diff --git a/tools/libxl/libxl_dom.c b/tools/libxl/libxl_dom.c
> index 0bdd654..91c4bd8 100644
> --- a/tools/libxl/libxl_dom.c
> +++ b/tools/libxl/libxl_dom.c
> @@ -65,6 +65,8 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
>      libxl_ctx *ctx = libxl__gc_owner(gc);
>      int tsc_mode;
>      char *xs_domid, *con_domid;
> +    uint32_t rtc_timeoffset;
> +
>      xc_domain_max_vcpus(ctx->xch, domid, info->max_vcpus);
>      libxl_set_vcpuaffinity_all(ctx, domid, info->max_vcpus, &info->cpumap);
>      xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb + LIBXL_MAXMEM_CONSTANT);
> @@ -91,8 +93,19 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
>      if (libxl_defbool_val(info->disable_migrate))
>          xc_domain_disable_migrate(ctx->xch, domid);
>  
> -    if (info->rtc_timeoffset)
> -        xc_domain_set_time_offset(ctx->xch, domid, info->rtc_timeoffset);
> +    rtc_timeoffset = info->rtc_timeoffset;
> +    if (libxl_defbool_val(info->localtime)) {
> +        time_t t;
> +        struct tm *tm;
> +
> +        t = time(NULL);
> +        tm = localtime(&t);
> +
> +        rtc_timeoffset += tm->tm_gmtoff;
> +    }
> +
> +    if (rtc_timeoffset)
> +        xc_domain_set_time_offset(ctx->xch, domid, rtc_timeoffset);
>  
>      if (info->type == LIBXL_DOMAIN_TYPE_HVM) {
>          unsigned long shadow;

Re: Upgrade from xen-4.1-testing to xen-unstable report

[Teck Choon Giam]

On Thu, Apr 12, 2012 at 10:50 PM, Ian Campbell <Ian.Campbell@citrix.com> wrote:
> On Thu, 2012-04-12 at 15:42 +0100, Lin Ming wrote:
>> On Thu, 2012-04-12 at 15:03 +0100, Ian Campbell wrote:
>> > On Thu, 2012-04-12 at 13:50 +0100, Lin Ming wrote:
>> > > On Thu, 2012-04-12 at 07:08 +0000, Ian Campbell wrote:
>> > > > On Wed, 2012-04-11 at 22:23 +0100, Teck Choon Giam wrote:
>> > > > > Hi,
>> > > > >
>> > > > > This is just my experience about issues I encountered when upgrade
>> > > > > from xen-4.1-testing changeset 23277:80130491806f to xen-unstable
>> > > > > changeset 25191:a95fc7decc83.
>> > > > >
>> > > > > 1. Immediately after upgrade, xl list show such error:
>> > > > >
>> > > > > # xl list
>> > > > > libxl: error: libxl.c:506:libxl_list_domain: geting domain info list:
>> > > > > Permission denied
>> > > > > libxl_domain_infolist failed.
>> > > > >
>> > > > > After a reboot, it is fine.  Any idea why such behaviour?  Imagine if
>> > > > > there are running domUs... this might cause issues to shutdown?  I
>> > > > > will downgrade and repeat such test to confirm.  Might be worth a note
>> > > > > in upgrading note about this if this is intended?
>> > > >
>> > > > The tools and the hypervisor are a matched pair so you would need to
>> > > > reboot the system in order to use the new tools. This has always been
>> > > > the case with Xen upgrades.
>> > > >
>> > > > > 2. localtime setting not working.  Set to localtime=1 doesn't seems to
>> > > > > work whereby setting rtc_timeoffset works.  Any idea?
>> > > >
>> > > > I've CC'd Lin Ming who implemented both of those.
>> > >
>> > > Just did a quick gdb debug, seems libxl__domain_build_info_setdefault
>> > > was called 3 times. So below statement was executed 3 times too.
>> > >
>> > > b_info->rtc_timeoffset += tm->tm_gmtoff;
>> >
>> > Oh, that sounds like the problem. The setdefault functions are supposed
>> > to be idempotent, IOW calling it two or more times should give the same
>> > result as calling it just once.
>> >
>> > I think you need to move this logic from setdefault
>> >     if (libxl_defbool_val(b_info->localtime)) {
>> >         time_t t;
>> >         struct tm *tm;
>> >
>> >         t = time(NULL);
>> >         tm = localtime(&t);
>> >
>> >         b_info->rtc_timeoffset += tm->tm_gmtoff;
>> >     }
>> >
>> > into libxl__build_pre and do that calculation on a temporary variable
>> > whjich you pass to xc_domain_set_time_offset rather than modifying
>> > b_info.
>>
>> How about below patch?

I am currently compiling xen-unstable changeset 25193:c6bde42c8845
with this patch.  Will test then report back.

Thanks.

Kindest regards,
Giam Teck Choon

>>
>> From e06589cb5f1efd885f9589405d0f4ecc97427ad6 Mon Sep 17 00:00:00 2001
>> From: Lin Ming <mlin@ss.pku.edu.cn>
>> Date: Thu, 12 Apr 2012 22:36:24 +0800
>> Subject: [PATCH] libxl: fix rtc_timeoffset setting
>>
>> libxl__domain_build_info_setdefault may be called several times, so
>> rtc_timeoffset can't be set in it.
>>
>> Move rtc_timeoffset setting logic to libxl__build_pre.
>>
>> Reported-by: Teck Choon Giam <giamteckchoon@gmail.com>
>> Signed-off-by: Lin Ming <mlin@ss.pku.edu.cn>
>
> Looks good to me.
>
> Acked-by: Ian Campbell <ian.campbell@citrix.com>
>
>> ---
>>  tools/libxl/libxl_create.c |    9 ---------
>>  tools/libxl/libxl_dom.c    |   17 +++++++++++++++--
>>  2 files changed, 15 insertions(+), 11 deletions(-)
>>
>> diff --git a/tools/libxl/libxl_create.c b/tools/libxl/libxl_create.c
>> index e63c7bd..e706124 100644
>> --- a/tools/libxl/libxl_create.c
>> +++ b/tools/libxl/libxl_create.c
>> @@ -127,15 +127,6 @@ int libxl__domain_build_info_setdefault(libxl__gc *gc,
>>          b_info->target_memkb = b_info->max_memkb;
>>
>>      libxl_defbool_setdefault(&b_info->localtime, false);
>> -    if (libxl_defbool_val(b_info->localtime)) {
>> -        time_t t;
>> -        struct tm *tm;
>> -
>> -        t = time(NULL);
>> -        tm = localtime(&t);
>> -
>> -        b_info->rtc_timeoffset += tm->tm_gmtoff;
>> -    }
>>
>>      libxl_defbool_setdefault(&b_info->disable_migrate, false);
>>
>> diff --git a/tools/libxl/libxl_dom.c b/tools/libxl/libxl_dom.c
>> index 0bdd654..91c4bd8 100644
>> --- a/tools/libxl/libxl_dom.c
>> +++ b/tools/libxl/libxl_dom.c
>> @@ -65,6 +65,8 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
>>      libxl_ctx *ctx = libxl__gc_owner(gc);
>>      int tsc_mode;
>>      char *xs_domid, *con_domid;
>> +    uint32_t rtc_timeoffset;
>> +
>>      xc_domain_max_vcpus(ctx->xch, domid, info->max_vcpus);
>>      libxl_set_vcpuaffinity_all(ctx, domid, info->max_vcpus, &info->cpumap);
>>      xc_domain_setmaxmem(ctx->xch, domid, info->target_memkb + LIBXL_MAXMEM_CONSTANT);
>> @@ -91,8 +93,19 @@ int libxl__build_pre(libxl__gc *gc, uint32_t domid,
>>      if (libxl_defbool_val(info->disable_migrate))
>>          xc_domain_disable_migrate(ctx->xch, domid);
>>
>> -    if (info->rtc_timeoffset)
>> -        xc_domain_set_time_offset(ctx->xch, domid, info->rtc_timeoffset);
>> +    rtc_timeoffset = info->rtc_timeoffset;
>> +    if (libxl_defbool_val(info->localtime)) {
>> +        time_t t;
>> +        struct tm *tm;
>> +
>> +        t = time(NULL);
>> +        tm = localtime(&t);
>> +
>> +        rtc_timeoffset += tm->tm_gmtoff;
>> +    }
>> +
>> +    if (rtc_timeoffset)
>> +        xc_domain_set_time_offset(ctx->xch, domid, rtc_timeoffset);
>>
>>      if (info->type == LIBXL_DOMAIN_TYPE_HVM) {
>>          unsigned long shadow;
>
>

Re: Upgrade from xen-4.1-testing to xen-unstable report

[Teck Choon Giam]

On Thu, Apr 12, 2012 at 11:25 PM, Teck Choon Giam
<giamteckchoon@gmail.com> wrote:
> On Thu, Apr 12, 2012 at 10:50 PM, Ian Campbell <Ian.Campbell@citrix.com> wrote:
>> On Thu, 2012-04-12 at 15:42 +0100, Lin Ming wrote:
>>> On Thu, 2012-04-12 at 15:03 +0100, Ian Campbell wrote:
>>> > On Thu, 2012-04-12 at 13:50 +0100, Lin Ming wrote:
>>> > > On Thu, 2012-04-12 at 07:08 +0000, Ian Campbell wrote:
>>> > > > On Wed, 2012-04-11 at 22:23 +0100, Teck Choon Giam wrote:
>>> > > > > Hi,
>>> > > > >
>>> > > > > This is just my experience about issues I encountered when upgrade
>>> > > > > from xen-4.1-testing changeset 23277:80130491806f to xen-unstable
>>> > > > > changeset 25191:a95fc7decc83.
>>> > > > >
>>> > > > > 1. Immediately after upgrade, xl list show such error:
>>> > > > >
>>> > > > > # xl list
>>> > > > > libxl: error: libxl.c:506:libxl_list_domain: geting domain info list:
>>> > > > > Permission denied
>>> > > > > libxl_domain_infolist failed.
>>> > > > >
>>> > > > > After a reboot, it is fine.  Any idea why such behaviour?  Imagine if
>>> > > > > there are running domUs... this might cause issues to shutdown?  I
>>> > > > > will downgrade and repeat such test to confirm.  Might be worth a note
>>> > > > > in upgrading note about this if this is intended?
>>> > > >
>>> > > > The tools and the hypervisor are a matched pair so you would need to
>>> > > > reboot the system in order to use the new tools. This has always been
>>> > > > the case with Xen upgrades.
>>> > > >
>>> > > > > 2. localtime setting not working.  Set to localtime=1 doesn't seems to
>>> > > > > work whereby setting rtc_timeoffset works.  Any idea?
>>> > > >
>>> > > > I've CC'd Lin Ming who implemented both of those.
>>> > >
>>> > > Just did a quick gdb debug, seems libxl__domain_build_info_setdefault
>>> > > was called 3 times. So below statement was executed 3 times too.
>>> > >
>>> > > b_info->rtc_timeoffset += tm->tm_gmtoff;
>>> >
>>> > Oh, that sounds like the problem. The setdefault functions are supposed
>>> > to be idempotent, IOW calling it two or more times should give the same
>>> > result as calling it just once.
>>> >
>>> > I think you need to move this logic from setdefault
>>> >     if (libxl_defbool_val(b_info->localtime)) {
>>> >         time_t t;
>>> >         struct tm *tm;
>>> >
>>> >         t = time(NULL);
>>> >         tm = localtime(&t);
>>> >
>>> >         b_info->rtc_timeoffset += tm->tm_gmtoff;
>>> >     }
>>> >
>>> > into libxl__build_pre and do that calculation on a temporary variable
>>> > whjich you pass to xc_domain_set_time_offset rather than modifying
>>> > b_info.
>>>
>>> How about below patch?
>
> I am currently compiling xen-unstable changeset 25193:c6bde42c8845
> with this patch.  Will test then report back.

Tested and it works!

Thanks.

Kindest regards,
Giam Teck Choon