c/s 20384

c/s 20384

[Jan Beulich]

Keir,

I'm having the impression that this c/s unintentionally modifies behavior in
certain ways:

- map_vcpu_info() no longer sets evtchn_upcall_mask for a newly mapped
info struct
- VCPUOP_initialise no longer fails when a vCPU didn't have a proper info
struct installed
- the changes to xen/common/event_channel.c make it so that
dummy_vcpu_info can be written to, and hence subsequently initialized
vcpu_info structs would have unpredictable initial state

Jan

Re: c/s 20384

[Keir Fraser]

On 02/11/2009 08:17, "Jan Beulich" <JBeulich@novell.com> wrote:

> - map_vcpu_info() no longer sets evtchn_upcall_mask for a newly mapped
> info struct
> - VCPUOP_initialise no longer fails when a vCPU didn't have a proper info
> struct installed
> - the changes to xen/common/event_channel.c make it so that
> dummy_vcpu_info can be written to, and hence subsequently initialized
> vcpu_info structs would have unpredictable initial state

Thanks, I've applied fixes as c/s 20390. The first issue is a genuine bug.
The second I think is not a critical issue, but since no good comes of
starting a vcpu without its own info structure, we may as well
check-and-fail in this case. The third issue I think does not matter, since
the first two fixes ensure that a vcpu will have a cleanly-initialised info
structure before it ever runs.

 -- Keir

Re: c/s 20384

[Jeremy Fitzhardinge]

On 11/02/09 00:17, Jan Beulich wrote:
> Keir,
>
> I'm having the impression that this c/s unintentionally modifies behavior in
> certain ways:
>
> - map_vcpu_info() no longer sets evtchn_upcall_mask for a newly mapped
> info struct
> - VCPUOP_initialise no longer fails when a vCPU didn't have a proper info
> struct installed
> - the changes to xen/common/event_channel.c make it so that
> dummy_vcpu_info can be written to, and hence subsequently initialized
> vcpu_info structs would have unpredictable initial state
>   

What's the real changeset id you're referring to?  The small-integer
numbers are only locally valid (and globally by coincidence), and 20384
here doesn't seem at all relevant to your points.

I you're referring to f74f0c1ae8ab, which is  21773 in my tree...

    J

Re: c/s 20384

[Keir Fraser]

On 02/11/2009 21:08, "Jeremy Fitzhardinge" <jeremy@goop.org> wrote:

> What's the real changeset id you're referring to?  The small-integer
> numbers are only locally valid (and globally by coincidence), and 20384
> here doesn't seem at all relevant to your points.
> 
> I you're referring to f74f0c1ae8ab, which is  21773 in my tree...

Yes, that's the one.

 K.

Re: c/s 20384

[Jan Beulich]

>>> Jeremy Fitzhardinge <jeremy@goop.org> 02.11.09 22:08 >>>
>What's the real changeset id you're referring to?  The small-integer
>numbers are only locally valid (and globally by coincidence), and 20384
>here doesn't seem at all relevant to your points.

Hmm, I think http://xenbits.xensource.com/xen-unstable.hg can
reasonably be considered a canonical reference, and hence I would
assume that using the small number ids is uniquely identifying a c/s.
Furthermore, using the full ids doesn't allow easily judging how long
ago the referred to c/s was committed, including immediately knowing
which releases it may have been part of.

Jan

Re: c/s 20384

[Jeremy Fitzhardinge]

On 11/03/09 02:09, Jan Beulich wrote:
> Hmm, I think http://xenbits.xensource.com/xen-unstable.hg can
> reasonably be considered a canonical reference, and hence I would
> assume that using the small number ids is uniquely identifying a c/s.
>   

I have a repo here which is what I'm going to refer to if you highlight
a particular changeset for some reason.  If I have to go back to xenbits
to map a local change number to a changeset id then that's pretty
awkward (particularly if I'm not online at the time).

> Furthermore, using the full ids doesn't allow easily judging how long
> ago the referred to c/s was committed, including immediately knowing
> which releases it may have been part of.
>   

That's pretty easy to determine once you look it up.  I don't know that
I have a good idea about how change numbers relate to releases anyway.

    J