Re: [Xen-users] [Xen-Users] Issues Booting Dom0 on ARM Cortex A15

[Ian Campbell <ian.campbell@xxxxxxxxxx>]

On Thu, 2015-06-18 at 10:41 +0100, Julien Grall wrote:
> Hi Chen,
> 
> On 18/06/2015 02:33, Chen Baozi wrote:
> 
> > On Wed, Jun 17, 2015 at 04:00:38PM -0400, Brandon Perez wrote:
> >> On 06/17/2015 03:15 PM, Julien Grall wrote:
> >>> Hello Brandon,
> >>>
> >>> On 17/06/2015 15:30, Brandon Perez wrote:
> >>>> The console and keyhandler are receiving 'q' and '0', but nothing is
> >>>> happening/being printed out. I suspect this is because they are
> >>>> scheduled as tasklets, and are never getting a chance to run.
> >>>>
> >>>> Opening the debugger, it seems like I'm stuck doing timer softirqs and
> >>>> handling timer interrupts continually. I'm going to double check the
> >>>> timer settings in my device tree.
> >>>
> >>> Looking to your first log [1], it seems that CNTFRQ is not set correctly:
> >>>
> >>> Generic Timer IRQ: phys=30 hyp=26 virt=27 Freq: 0 KHz
> >>>
> >>> The CNTFRQ register should be set by the firmware/bootloader. Does the
> >>> Linux kernel is booting on baremetal with the same firmware/bootloader?
> >>>
> >>> If yes, do you use the same device tree?
> >>>
> >>> Regards,
> >>>
> >>
> >> Hi Julien,
> >>
> >>      Good catch! That seems to have been the issue. To answer your 
> >> questions,
> >> yes the kernel is booting with same firmware, bootloader, and device tree 
> >> as
> >> Xen.
> >>
> >>      The real issue actually stems from the uBoot code, which was not
> >> properly setting the CNTFRQ register (which was indeed 0).
> >>
> >>      The CNTFRQ register being 0 lead to a whole slew of issues, the 
> >> primary
> >> of which being was that the timer interrupt ran extremely often, causing
> >> progress in preemptable sections of code to be extremely slow.
> >
> > This reminds me that I have had a similar issue when doing OMAP5432 
> > enablement.
> > IIRC, OMAP5 uses a different platform register for system timer, and CNTFRQ
> > is not actually used during platform initialisation in linux kernel. Maybe
> > it would be a good idea to have a check whether your SoC has the same
> > implementation?
> 
> How did you resolve it on OMAP5432? Does Xen get the timer frequency 
> from the property "clock-frequency"?
> 
> I'm wondering if we should get the frequency from the OMAP5 specific 
> timer register rather than modifying U-boot and/or the DT.

Independently from that we should complain very loudly if CNTFRQ is 0 on
any processor and a better value hasn't been supplied in the firmware
tables.

Ian.


_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users