[suspend/resume] memory_op hypercall failure: XENMEM_maximum_gpfn

[Jean-Yves Migeon <jean-yves.migeon%espci.fr@localhost>]

Hi!

This mail originated from the latest dom0/domU issue fixed by Manuel, 
relating to the cpu_idle() set to null (which caused an early panic).

I wanted to give a try to the "xm dump-core" functionality provided by 
Xen. Basically, this options pauses a domain, take a snapshot of the 
kernel and memory state, and unpause it. The obtained image is stored in 
a file, like a traditional core dump.

Things are not that easy though. Calls to xm dump-core fail with:

# xm dump-core 2 /root/core
Error: Failed to dump core: (1, 'Internal error', 'p2m_size < nr_pages 
-1 (0 < 1fff')

I tracked down the error. Basically, the p2m_size table (proportional to 
the size of memory allocated to a guest) is not reported as 0, whatever 
this error says. The privcmd ioctl does perform the hypercall to abtain 
such information, but returns -1. The code used to dump the guest, in 
xentools3 pkg (tools/libxc/xc_core_x86.c) just increments the return 
value by 1, without checking for errors:

static int nr_gpfns(int xc_handle, domid_t domid) {
   return xc_memory_op(xc_handle, XENMEM_maximum_gpfn, &domid) + 1;
}

Turning the error into a 0 sized p2m table.

Observations are the same in privcmd.c (xen/xen/privcmd.c): hypercall 
does return -1.

Questions are:
- hypercall APIs speak of "-ve errcode" on failure, but I can not manage 
to find which errcode they are referring to. Are they the same as the 
ones given in the mini-os from xentools? 
(extras/mini-os/include/errno-base.h). If yes, -1 indicates an EPERM 
error, which is weird for dom0.
- do the XENMEM_maximum_gpfn memory operation require some cooperation 
from the guest to obtain the proper value? I would say no, hypervisor 
should do it by itself alone. Am I missing something here?

While inspecting privcmd interface, I have found something weird too: 
xend is periodically polling hypervisor with a XEN_DOMCTL_getdomaininfo 
operation for a domain ID of value "latest_domain_id_started + 1". Of 
course, it returns an error, -3 (ESRCH, no such process, according to 
mini-os errno files). Is this normal behavior from xend? Some kind of 
regular polling to check that no other unmonitored domain has been 
created without xend supervision?

Lastly, all my debugging is done through a couple of printf()'s. Indeed, 
it works, but it is rather unpractical. Any advice for a more efficient 
method is gladly appreciated.

Note that save/restore operations dump the state of the kernel to a file 
too, so XENMEM_maximum_gpfn is necessary for suspend.

Thanks for your attention :)

Cheers,

--
Jean-Yves Migeon
jean-yves.migeon%espci.fr@localhost