Re: [PATCH 0/3] sparc: port to copy_thread_tls() and struct kernel_clone_args

[Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>]

On 22/05/2020 01:05, Al Viro wrote:

> On Thu, May 21, 2020 at 09:23:50PM +0100, Al Viro wrote:
>> On Thu, May 21, 2020 at 08:42:34PM +0100, Mark Cave-Ayland wrote:
>>
>>>> Can you tell me a bit more about the host in terms of CPU and disk to help figure out
>>>> what's going on?
>>
>> phenom II X6 1100T (6-way 3.3GHz), 8Gb RAM (4Gb given to guest), WDC WD10EACS-00D
>> disk (hdparm -tT gives
>>  Timing cached reads:   6988 MB in  2.00 seconds = 3494.96 MB/sec
>>  Timing buffered disk reads: 280 MB in  3.02 seconds =  92.75 MB/sec
>> )
>>
>>> One other thought I had is that somehow the IVEC IRQs are managing to be overwritten
>>> on a faster host before being read by the guest. Does the following patch display the
>>> FATAL message at the point where things hang?
>>
>>> diff --git a/hw/pci-host/sabre.c b/hw/pci-host/sabre.c
>>> index fae20ee97c..618ebd1300 100644
>>> --- a/hw/pci-host/sabre.c
>>> +++ b/hw/pci-host/sabre.c
>>> @@ -63,6 +63,9 @@
>>>  static inline void sabre_set_request(SabreState *s, unsigned int irq_num)
>>>  {
>>>      trace_sabre_set_request(irq_num);
>>> +    if (s->irq_request != 0 && s->irq_request != NO_IRQ_REQUEST) {
>>> +        fprintf(stderr, "FATAL: still waiting for IRQ %x, now %x\n", s->irq_request,
>>> irq_num);
>>> +    }
>>>      s->irq_request = irq_num;
>>>      qemu_set_irq(s->ivec_irqs[irq_num], 1);
>>>  }
>>
>> I have to go AFK right now, will test when I get back (should be about an
>> hour or two)
> 
> Hang, nothing on stderr until killed, at which point it gave the expected
> qemu-system-sparc64: terminating on signal 15 from pid 15917 (-bash)
> IOW, stderr got flushed after hang - that fprintf simply has not triggered.

Okay thanks for giving that a go. I have one other possibility that I want to
eliminate: I see that you are using a very up-to-date 5.6.0-1-sparc64, whereas I was
testing with the stock CD image (and indeed, an older revision than the one that is
currently available in Debian ports).

When you have a moment, can you grab the latest qemu from git master since it has
some fixes for -kernel/-initrd, and then try booting your same sparc64 disk image but
 passing in explicit -kernel and -initrd parameters pointing to the same ones that I
am using. I'll send you a download link off-list.


ATB,

Mark.