[Crash-utility] [PATCH] vmware_vmss: read vCPUs regs and show them in 'bt'

Sergio Lopez slp at redhat.com
Wed Mar 21 16:10:08 UTC 2018


On Wed, Mar 21, 2018 at 10:44:47AM -0400, Dave Anderson wrote:
> 
> 
> ----- Original Message -----
> > VMSS dump files contain the state of each vCPU at the time of suspending
> > the VM. This change enables 'crash' to read some relevant registers from
> > each vCPU state and display them in 'bt'.
> 
> That's very helpful.  I'm not sure why VMware never did anything like that
> in the original patch.
> 
> And now that there finally is a VMSS_DUMPFILE() indicator, it would be helpful
> to at least display *something* w/respect to the dumpfile contents as viewed from
> "help -D".   If you bring up the session with "crash -d1" you will see a bunch
> of stuff from the dumpfile, although I'm not sure whether all of it would be
> useful during runtime.  But if nothing else, at least indicate what kind of 
> dumpfile it is, and anything else that might be useful.
>  
> Also, I just noticed that original patch from VMware does not display the
> VMWARE_VMSS bit in pc->flag as show by "help -p".  Can you add that too?

I didn't pay any attention to the "help" command. I'm going to implement
the VMWARE_VMSS bit in pc->flag, and also the "help -r" output, which
seems quite straightforward having all the registers already properly
placed in vmssregs64.

As for "help -D", I think we could implement a function that iterates
through groups and tags present in the VMSS, printing names, sizes and
values (for non-block tags). Basically, it'd be the same thing that
"vmware_vmss_init" does, but unconditionally printing the info without
touching anything else. Does this makes sense to you?

> > calculation for VMSS dump files.
> 
> Will that require the KASLR helper patch as a prerequisite?

Yes, otherwise we'd need to duplicate "calc_kaslr_offset" and all the
related functions.

Sergio.




More information about the Crash-utility mailing list