Unable to mount the root fs on stable/8 r264339, GENERIC kernel, with MBR, FreeBSD slice, and UFS volume labels

[Claude Buisson]

On 04/16/2014 19:06, John Baldwin wrote:
> On Wednesday, April 16, 2014 11:54:20 am Claude Buisson wrote:
>> On 04/16/2014 16:38, John Baldwin wrote:
>>> On Wednesday, April 16, 2014 5:41:44 am Claude Buisson wrote:
>>>> On 04/16/2014 10:09, Trond Endrestøl wrote:
>>>>> On Tue, 15 Apr 2014 19:44+0200, Trond Endrestøl wrote:
>>>>>
>>>>>> On Fri, 11 Apr 2014 18:38+0200, Trond Endrestøl wrote:
>>>>>>
>>>>>>> Hi,
>>>>>>>
>>>>>>> I have a couple of uncritical systems running stable/8 r258344.
>>>>>>> Hardware is Dell OptiPlex GX260, BIOS A09, which is the latest rev.
>>>>>>>
>>>>>>> The r264339 GENERIC kernel are unable to mount the root fs from the
>>>>>>> hard drive using MBR, FreeBSD slice, and UFS volume labels.
>>>>>>>
>>>>>>> r258344 obviously can.
>>>>>>>
>>>>>>> I even tried regular device names like /dev/ad0s1a in /etc/fstab, and
>>>>>>> at the mountroot> prompt, i.e. ufs:/dev/ad0s1a. The kernel still
>>>>>>> cannot mount the root fs.
>>>>>>>
>>>>>>> The new kernel (r264339) does recognize the ad0 harddrive, and ad0 is
>>>>>>> listed as one of the GEOM managed disk devices; acd0 being the other
>>>>>>> one.
>>>>>>>
>>>>>>> Do I need to load additional geom modules, or is it a genuine bug?
>>>>>>>
>>>>>>> I have recreated the same conditions on a spare GX260, yes, I have
>>>>>>> plenty of them.
>>>>>>
>>>>>> I believe I have identified r262226 as the offending commit.
>>>>>>
>>>>>> Maybe the flags integer is set to a bad value before the calls to
>>>>>> resource_list_alloc().
>>>>>>
>>>>>> My spare system is currently recompiling r262221 of both world and
>>>>>> kernel, and I hope to confirm this assumption in a few hours.
>>>>>
>>>>> Confirmed.
>>>>>
>>>>>> If all goes well, I intend to move forward to the latest revision of
>>>>>> stable/8, back out the change done to sys/dev/pci/pci.c in r262226,
>>>>>> recompile world and kernel, install the kernel, and if successfully,
>>>>>> I'll install world as well.
>>>>>
>>>>> Confirmed, stable/8 r264519 with r262226 backed out does indeed work
>>>>> on a Dell OptiPlex GX260, BIOS A09.
>>>>>
>>>>
>>>> Just another data point related to r262226:
>>>>
>>>> Yesterday, I upgraded a stable/8 system from r260539 to r264426, which could not
>>>> finish booting. In my case, it stopped just after detecting an agp card (and
>>>> could boot if agp was disabled by device.hints).
>>>>
>>>> Reverting r262226 gave me back a fully usable system.
>>>>
>>>> This system is also a Dell, but a (oldish) Dimension 4550 BIOS A08.
>>>>
>>>>> Any chance anyone would like to dig deeper into the matter?
>>>>> Should I file a PR?
>>>>>
>>>>
>>>> I think so, if jhb (cced) keep being interested in stable/8..
>>>>
>>>> BTW, this morning I succeeded in booting a stable/10 snapshot r264194 USB stick
>>>> on this machine.
>>>
>>> Can you get a verbose dmesg with and without the change?
>>>
>>
>> If r262226 is not reverted, the only way to boot and have a verbose dmesg is to
>> disable agp. Is it OK ?
>
> Yes.  Hmmmm.  I may know what the problem is then.  Can you kldload agp after
> boot?
>

Here we are:

verbose dmesg, kernel with agp, r262226 reverted = verbose.pci.c_reverted

     http://dl.free.fr/hp7ORcx71

pciconf with the same kernel = pciconf.pci.c_reverted

     http://dl.free.fr/tSrfsnxhK

verbose dmesg, kernel without agp, not reverted = verbose.noagp

     http://dl.free.fr/gejndPCDZ

With this kernel, kldloading agp.ko has for only effect to lock the system 
(cannot switch vty, no answer to pings, ..) -> panic/power button.

Hope this help

CBu