Re: [ PATCH 2.6.16-rc3-xen 3/3] sysfs: export Xen hypervisor attributes to sysfs

[Arjan van de Ven <arjan@infradead.org>]

On Wed, 2006-02-22 at 22:26 -0600, Anthony Liguori wrote:
> Arjan van de Ven wrote:
> > On Wed, 2006-02-22 at 08:43 -0500, Mike D. Day wrote:
> >   
> >> Heiko Carstens wrote:
> >>     
> >>> On Wed, Feb 22, 2006 at 08:06:12AM -0500, Mike D. Day wrote:
> >>>
> >>> If it's not needed, why include it at all?
> >>>       
> >> Sorry for not being clear. It *is* needed for control tools and agents 
> >> running in the privileged domain. 
> >>     
> >
> > but again those tools and agents *already* have a way of talking to the
> > hypervisor themselves. Why can't they just first ask this info? Why does
> > that need to be in the kernel, in unswappable memory?
> >   
> Hypercalls have to be done in ring 0 for security reasons)  There has to 
> be some kernel interface for making hypercalls.

sure but you need that ANYWAY; the management tools will want a generic
"THIS hypercall" thing

> 
> The current interface is a ioctl() on a /proc file (which is awful).  
> The ioctl just pretty much passes 5 word arguments to the hypervisor.  
> It was suggested previously here that a hypercall pass-through interface 
> isn't the right approach.  One suggestion that came up was a syscall 
> interface.

yeah a syscall sounds right for this



> Also, there are some kernel-level drivers, like the memory ballooning 
> driver, that only exist in the kernel.  Controlling the balloon driver 
> requires some sort of interface.  That was the original point of this 
> effort (since it's currently exposed as a /proc file).  I think it's 
> quite clear that the balloon driver should expose itself through sysfs 
> but I'm not personally convinced that this information (hypervisor 
> version information) ought to be exposed in sysfs.

that's a different thing; the ballooning driver obviously has to be in
kernel due to how it interacts with the VM, and having it's 1 or 2
controls in sysfs (hint: make it writable module parameters, so that the
defaults can be set at module load time, and you'll also get the rest
for free ;)

pure hypervisor info... no. "only" 8Kb of code + all data structures..
for something just as easily done in the management layer. In fact I
suspect the management layer will do the hypercalls anyway just because
it's easier to do than parsing sysfs !

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/