[Xen-devel] Re: [PATCH][RFC] to fix xm list memory reporting

[Anthony Liguori <aliguori@xxxxxxxxxx>]

McAfee, Tommie M wrote:
> Anthony,
>
> I'm using watches on the xenstore to set self.info['memory'].
>
>   
> > You should gracefully handle the case where the info isn't in XenStore.
> >     
> If the domain never writes to the store, Wether it is unable to, or if this balloon driver isn't in use, then functionality will be as it always was, I'm not relying on xen-store.  self.info['memory'] only gets overwritten if the balloon driver (or perhaps anyone) writes to the memory/xmtarget path. 
>
> However, I did not think of some of the other points that you have brought up.  
> Do you think that it is overkill at this point to address this, or should I 
> look further into the issue?
>   
If we're using the wrong entry from xc_dominfo_t then we ought to use 
the right one.  Should be a 1-line fix.
Regards,

Anthony Liguori

> Tommie McAfee
> Xen-Testing 
>
>
> -----Original Message-----
> From: Anthony Liguori [mailto:aliguori@xxxxxxxxxx]
> Sent: Wed 10/11/2006 5:34 PM
> To: McAfee, Tommie M
> Cc: xen-devel
> Subject: Re: [PATCH][RFC] to fix xm list memory reporting
>  
> This patch breaks the domU ABI.
> You can't assume that the driver will have the modified balloon driver 
> because it could be any 3.0.x guest.
> You should gracefully handle the case where the info isn't in XenStore.
>
> However, this patch concerns me in another way.  You're relying on the 
> guest to report how much memory it has?  What prevents a guest from 
> lying and claiming it has less memory than it really does?  Forget about 
> lying, what about buggy guests?
> Isn't this info available from the hypervisor?
>
> Regards,
>
> Anthony Liguori
>
> McAfee, Tommie M wrote:
>   
> > This patch addresses bugzilla #649, enabling `xm list' to display the actual amount of memory allocated by a guest.  This is done by checking the value that the balloon driver returns after memory requests.  As a result, I'm noticing that a guest that I've set to start with 128Mb will actually give the user 126Mb.  All other allocations perform normally, and attempting to over-allocate will simply return the amount of space that the domain was able to expand to.  
> >
> > Under the premises that xm list should report actual memory values, 
> >  2 tests in xm-test may need to be modified to verify the amount of memory that a guest is physically using, and not rely on the value in the config file as being the amount of memory that the guest will have.
> > xm-test reactions to patch as of changset 11376:
> >
> > REASON: Started domain with 128MB, but it got 126 MB
> > FAIL: 08_create_mem128_pos.test
> > REASON: Started domain with 256MB, but it got 254 MB
> > FAIL: 09_create_mem256_pos.test
> >
> > Running 'free' inside of each guest shows 126, and 254 respectively, not their config file #'s. 
> >
> > Signed-off-by: Tommie McAfee <tommie.mcafee@xxxxxxxxxx>
> >
> > Regards,
> > Tommie McAfee
> > Xen-Testing 
> >
> >
> >
> > ------------------------------------------------------------------------
> >
> > diff -r 593b5623a0d2 linux-2.6-xen-sparse/drivers/xen/balloon/balloon.c
> > --- a/linux-2.6-xen-sparse/drivers/xen/balloon/balloon.c        Fri Sep 29 
> > 15:40:35 2006 +0100
> > +++ b/linux-2.6-xen-sparse/drivers/xen/balloon/balloon.c        Wed Oct 04 
> > 14:46:33 2006 -0400
> > @@ -57,6 +57,8 @@
> >  #include <xen/xenbus.h>
> >  
> >  #define PAGES2KB(_p) ((_p)<<(PAGE_SHIFT-10))
> > +#define NOXENBUS 0
> > +#define XENBUSREADY 1
> >  
> >  #ifdef CONFIG_PROC_FS
> >  static struct proc_dir_entry *balloon_pde;
> > @@ -83,6 +85,9 @@ extern unsigned long totalram_pages;
> >  
> >  /* We may hit the hard limit in Xen. If we do then we remember it. */
> >  static unsigned long hard_limit;
> > +
> > +/* Aknowledge that xenstore is available */
> > +static int xenbus_status;
> >  
> >  /*
> >   * Drivers may alter the memory reservation independently, but they must
> > @@ -325,6 +330,22 @@ static int decrease_reservation(unsigned
> >         return need_sleep;
> >  }
> >  
> > +/* write currently allocated kbs to xenstore */
> > +static void xenbus_write_xmtarget(void){
> > +
> > +       struct xenbus_transaction xbt;
> > +       unsigned int xm_current_pages;
> > +       
> > +
> > +       if(likely(xenbus_status==XENBUSREADY)){
> > +               xm_current_pages = PAGES2KB(current_pages);
> > +               xenbus_transaction_start(&xbt);
> > +               xenbus_printf(xbt,"memory", "xmtarget", "%8u",xm_current_pages);
> > +               xenbus_transaction_end(xbt,0);
> > +       }
> > +
> > +}
> > +
> >  /*
> >   * We avoid multiple worker processes conflicting via the balloon mutex.
> >   * We may of course race updates of the target counts (which are protected
> > @@ -355,6 +376,8 @@ static void balloon_process(void *unused
> >         if (current_target() != current_pages)
> >                 mod_timer(&balloon_timer, jiffies + HZ);
> >  
> > +        xenbus_write_xmtarget();
> > +
> >         up(&balloon_mutex);
> >  }
> >  
> > @@ -384,6 +407,8 @@ static void watch_target(struct xenbus_w
> >                 /* This is ok (for domain0 at least) - so just return */
> >                 return;
> >         }
> > +
> > +        xenbus_status=XENBUSREADY;
> >  
> >  	/* The given memory/target value is in KiB, so it needs converting to
> >          * pages. PAGE_SHIFT converts bytes to pages, hence PAGE_SHIFT - 10.
> > @@ -462,6 +487,8 @@ static int __init balloon_init(void)
> >  {
> >         unsigned long pfn;
> >         struct page *page;
> > +
> > +       xenbus_status=NOXENBUS;
> >  
> >  	if (!is_running_on_xen())
> >                 return -ENODEV;
> > diff -r 593b5623a0d2 tools/python/xen/xend/XendDomainInfo.py
> > --- a/tools/python/xen/xend/XendDomainInfo.py   Fri Sep 29 15:40:35 2006 +0100
> > +++ b/tools/python/xen/xend/XendDomainInfo.py   Wed Oct 04 14:46:33 2006 -0400
> > @@ -459,6 +459,7 @@ class XendDomainInfo:
> >          self.console_mfn = None
> >  
> >          self.vmWatch = None
> > +        self.memWatch = None
> >          self.shutdownWatch = None
> >  
> >          self.shutdownStartTime = None
> > @@ -487,6 +488,14 @@ class XendDomainInfo:
> >              return []
> >  
> >  
> > +    def xsMemoryChanged(self, _):
> > +        """Get the memory target from xenstore of this domain.
> > +        """
> > +        xs_memory=int(self.readDom('memory/xmtarget'))/1024
> > +        self.info['memory']=xs_memory
> > +        self.storeVm("memory", xs_memory)
> > +       return 1
> > +      
> >      def storeChanged(self, _):
> >          log.trace("XendDomainInfo.storeChanged");
> >  
> > @@ -775,6 +784,9 @@ class XendDomainInfo:
> >          self.vmWatch = xswatch(self.vmpath, self.storeChanged)
> >          self.shutdownWatch = xswatch(self.dompath + '/control/shutdown',
> >                                       self.handleShutdownWatch)
> > +        self.memWatch = xswatch(self.dompath + '/memory/xmtarget',
> > +                                     self.xsMemoryChanged)
> > +
> >  
> >  
> >      def getDomid(self):
> > @@ -1015,6 +1027,7 @@ class XendDomainInfo:
> >          
> >          self.info['memory'] = target
> >          self.storeVm("memory", target)
> > +        self.storeDom('memory/xmtarget', target << 10)
> >          self.storeDom("memory/target", target << 10)
> >  
> >  
> > @@ -1372,6 +1385,7 @@ class XendDomainInfo:
> >          self.refresh_shutdown_lock.acquire()
> >          try:
> >              self.unwatchShutdown()
> > +            self.unwatchMemory()
> >  
> >              self.release_devices()
> >  
> > @@ -1439,6 +1453,18 @@ class XendDomainInfo:
> >                  self.shutdownWatch = None
> >          except:
> >              log.exception("Unwatching control/shutdown failed.")
> > +
> > +
> > +    def unwatchMemory(self):
> > +       """Remove the watch on the domain's control/shutdown node, if any."""
> > +       try:
> > +            try:
> > +                if self.memWatch:
> > +                    self.memWatch.unwatch()
> > +            finally:
> > +                self.memWatch= None
> > +        except:
> > +            log.exception("Unwatching memory/xmtarget failed.")
> >  
> >  
> >      ## public:
> > diff -r 593b5623a0d2 tools/python/xen/xend/image.py
> > --- a/tools/python/xen/xend/image.py    Fri Sep 29 15:40:35 2006 +0100
> > +++ b/tools/python/xen/xend/image.py    Wed Oct 04 14:46:33 2006 -0400
> > @@ -382,6 +382,7 @@ class HVMImageHandler(ImageHandler):
> >  
> >      def destroy(self):
> >          self.unregister_shutdown_watch();
> > +        self.unregister_memory_watch();
> >          import signal
> >          if not self.pid:
> >              return
> > @@ -406,6 +407,18 @@ class HVMImageHandler(ImageHandler):
> >              log.exception("Unwatching hvm shutdown watch failed.")
> >          self.shutdownWatch = None
> >          log.debug("hvm shutdown watch unregistered")
> > +       
> > +    def unregister_memory_watch(self):
> > +        """Remove the watch on the target/xmtarget, if any. Nothrow
> > +        guarantee."""
> > +        try:
> > +           if self.memWatch:
> > +                self.memWatch.unwatch()
> > +        except:
> > +            log.exception("Unwatching memory/xmtarget failed.")
> > +        self.memWatch = None
> > +        log.debug("hvm memory watch unregistered")
> > +
> >  
> >      def hvm_shutdown(self, _):
> >          """ watch call back on node control/shutdown,
> >
> >
> > ------------------------------------------------------------------------
> >
> > _______________________________________________
> > Xen-devel mailing list
> > Xen-devel@xxxxxxxxxxxxxxxxxxx
> > http://lists.xensource.com/xen-devel
> >     
>
>   

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel