OS fails to notice MP CPU

classic Classic list List threaded Threaded
9 messages Options
Reply | Threaded
Open this post in threaded view
|

OS fails to notice MP CPU

Dr.Boutari
System specs
Toshiba Satellite c850 laptop
CPU i5-3230M
Amd vga 7610M
Latest BIOS updated.

1 months ago I installed Openbsd 6.3 amd64.
The cd and hdd boot will do a shutdown panic without providing informative messages. Leading me to randomly try to disable acpi on kernel boot and continue setup which worked. I soon figured out afterwards I was only running one core of my multicore CPU and I thought maybe I did press something wrong during install. Firmware updated post install then I moved to stable then current brach for another reason(another kernel panic due to radeondrm which was fixed on current branch).

I knew 6.4 stable was coming so I waited to reinstall.
Yesterday tried booting 6.4 amd64 cd and it still shutdown crashes. Disabled acpi like before and continued setup leading me to the sets. Im not offered bsd.mp set so my cpu isnt well recognised.
As a final trial I tried the 6.4 i386 cd boot. It crashes aswell but doesnt shutdown and Im provided with an error message this time so Im providing that as an attachment.Bypassing the crash with disable acpi also leads me to the base sets install without offering the bsd.mp kernel.

Thanks for any provided help and if I could do anything to make this issue more clear im ready to do that,so just give me suggestions.

Sent from ProtonMail mobile

1539943506233.png (635K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: OS fails to notice MP CPU

Philip Guenther
On Fri, 19 Oct 2018, Dr.Boutari wrote:
> 1 months ago I installed Openbsd 6.3 amd64.
> The cd and hdd boot will do a shutdown panic without providing
> informative messages. Leading me to randomly try to disable acpi on
> kernel boot and continue setup which worked. I soon figured out
> afterwards I was only running one core of my multicore CPU and I thought
> maybe I did press something wrong during install.

On many modern machines, ACPI _is_ how the OS can find the other CPUs.


> Firmware updated post install then I moved to stable then current brach
> for another reason(another kernel panic due to radeondrm which was fixed
> on current branch).
>
> I knew 6.4 stable was coming so I waited to reinstall.

If no one is supplying the information to fix the issues with this
hardware, it will only get fixed by luck.  You were not lucky.

> Yesterday tried booting 6.4 amd64 cd and it still shutdown crashes.
> Disabled acpi like before and continued setup leading me to the sets. Im
> not offered bsd.mp set so my cpu isnt well recognised.

Totally expected: see above.

> As a final trial I tried the 6.4 i386 cd boot. It crashes aswell but
> doesnt shutdown and Im provided with an error message this time so Im
> providing that as an attachment.Bypassing the crash with disable acpi
> also leads me to the base sets install without offering the bsd.mp
> kernel.

Totally expected: see above.

> Thanks for any provided help and if I could do anything to make this
> issue more clear im ready to do that,so just give me suggestions.

At a _minimum_, you should generate a full bug report by running sendbug
*as root*, verifying it includes
 - the full dmesg (albeit with acpi disabled)
 - pcidump(8) output
 - acpidump(8) output

If it includes all of those, then email it to <[hidden email]>

If it doesn't include those, then don't bother emailing it in, as without
those we can only guess what the problem is...and no one is going to waste
their time trying to guess what the problem is.


Philip Guenther

Reply | Threaded
Open this post in threaded view
|

Re: OS fails to notice MP CPU

Dr.Boutari
-------- Original Message --------
On 20 Oct 2018, 10:09, Philip Guenther wrote:

>On many modern machines, ACPI _is_ how the OS can find the other CPUs.

–--------------

This makes sense however the data I collected experimenting seems to contradict this. I will explain with reports.

---------------

>If no one is supplying the information to fix the issues with this
hardware, it will only get fixed by luck. You were not lucky.
>At a _minimum_, you should generate a full bug report by running sendbug
*as root*, verifying it includes
- the full dmesg (albeit with acpi disabled)
- pcidump(8) output
- acpidump(8) output

>Philip Guenther

---------

Again thanks for your work, im not a programmer so forgive my ignorance at times.
I attached the required reports in addition to others.

First of all I gathered acpidump pcidump and dmesg(of successful kernel boot with acpi disabled) for my system.
Later I booted again with acpi enabled so I can record the dmesg of the crash boot and compare it to the dmesg with acpi disabled. This showed acpi warnings that my other 3 cores are not configured so I thought probably as you stated acpi is required to recognise them and the warnings are expected since im booting bsd.sp. Comparing to acpi-disabled dmesg it shows that when acpi is disabled Intel MP Specification 1.4 from bios is loaded instead however no mention of other cores is noted. I provided an .odt document in attachments showing head to head differences between the 2 dmesgs.
With acpi enabled the crash occurs at a later point so it doesnt seem to be necessarily related.

Later to make sure if the problem with unrecognised cores is due to disabling acpi I booted again from the 6.4 amd64 install cd without disabling acpi.
This however showed warnings that even in this case the cpu cores are not recognised, so this basicly rules out the theory disabling acpi is the cause. The crash occurs at a later stage after swap/dump messages. I included screenshots of both stages for documentation.

I believe the bios provides different methods for cpu running whether acpi is disabled or enabled,however some required software is missing or bugged. But im no expert so hope you could make something out of this data.

acpidump.tar (234K) Download Attachment
dmesg head to head comparison.odt (25K) Download Attachment
dmesg with disabled acpi.txt (5K) Download Attachment
install cd-boot shows uncofiguered cpu msg.png (1M) Download Attachment
last msg b4 install-cd boot crash.png (1M) Download Attachment
pcidump.txt (26K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: OS fails to notice MP CPU

Philip Guenther
On Sat, 20 Oct 2018, Dr.Boutari wrote:
> Again thanks for your work, im not a programmer so forgive my ignorance
> at times. I attached the required reports in addition to others.

Thank you.

> First of all I gathered acpidump pcidump and dmesg(of successful kernel
> boot with acpi disabled) for my system. Later I booted again with acpi
> enabled so I can record the dmesg of the crash boot and compare it to
> the dmesg with acpi disabled. This showed acpi warnings that my other 3
> cores are not configured so I thought probably as you stated acpi is
> required to recognise them and the warnings are expected since im
> booting bsd.sp.

Correct.

> Comparing to acpi-disabled dmesg it shows that when acpi is disabled
> Intel MP Specification 1.4 from bios is loaded instead however no
> mention of other cores is noted.  <...>
>
> With acpi enabled the crash occurs at a later point so it doesnt seem to
> be necessarily related.

I don't know what you mean by that last statement.  The information
provided by the ACPI subsystem is used the entire time the OS is running.  
Indeed, the crash information from your original report showed it faulting
in acpi_get_table_with_size() _after_ the "root on sd0a ..." line.


> Later to make sure if the problem with unrecognised cores is due to
> disabling acpi I booted again from the 6.4 amd64 install cd without
> disabling acpi. This however showed warnings that even in this case the
> cpu cores are not recognised, so this basicly rules out the theory
> disabling acpi is the cause. The crash occurs at a later stage after
> swap/dump messages. I included screenshots of both stages for
> documentation.

That is not a good conclusion to draw.


> I believe the bios provides different methods for cpu running whether
> acpi is disabled or enabled,however some required software is missing or
> bugged. But im no expert so hope you could make something out of this
> data.

OpenBSD has supported discovery of additional cpus using the mpbios tables
since 2004, before it supported ACPI at all.  This machine just doesn't
appear to report all its CPUs in its mpbios table.  Frankly, trying to
improve the behavior of this machine when ACPI is not used is a waste of
programmer time.  If the machine has ACPI, then practically the only
reason to run it _without_ using ACPI is to debug ACPI issues!

Looking at your previous messages, they seem to mix information from
different attempts, with different kernels (amd64 vs i386) and different
devices disabled.  Mixing those makes it impossible to know what problem
is actually being diagnosed and is simply frustrating.


IMHO, if you want assistance from the OpenBSD developers in making this
machine function better, you need to go back to the original problem: when
booting a _stock_ amd64 GENERIC.MP kernel with _nothing_ disabled, what
exactly happens:
 - what is the sequence of events leading to the crash (boot?  shutdown?)
 - what is the output before the crash?
 - if it drops into ddb, then gather the crash or panic message and the
   other information requested at https://www.openbsd.org/ddb.html


Philip Guenther

Reply | Threaded
Open this post in threaded view
|

Re: OS fails to notice MP CPU

Dr.Boutari
-------- Original Message --------
On 21 Oct 2018, 01:55, Philip Guenther wrote:

IMHO, if you want assistance from the OpenBSD developers in making this
machine function better, you need to go back to the original problem: when
booting a _stock_ amd64 GENERIC.MP kernel with _nothing_ disabled, what
exactly happens:
- what is the sequence of events leading to the crash (boot? shutdown?)
- what is the output before the crash?
- if it drops into ddb, then gather the crash or panic message and the
other information requested

–------------

Thanks again. You were right without acpi the system wont recognise the other cores.Still my system is crashing if I dont disable acpi. I provide dmseg acpidump and pcidump after booting bsd.mp with acpi disabled.

Steps I did to reach this conclusion

I downloaded bsd.mp from a mirror and booted it first with acpi enabled. It crashed as usual at about the same point where the crash occured previously and still displaying no error messages then and not hanging at ddb prompt. The laptop just shuts down.However upon reviewing the dmseg I noticed this time that the other cores were recognised.
I later disabled acpi and booted but the other cores arent recognised after checking.

Thanks for the help. I guess I still have to figure out whats crashing my system or hope someone finds out whats the problem and fix it.

On another note. Just wondering why doesnt the amd64 install cd recognise the cores? Does it default to loading an sp kernel for the install which is the bsd.rd ?

acpidump.tar (179K) Download Attachment
dmesg-BSD.MP.txt (5K) Download Attachment
pcidump-BSD.MP.txt (26K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: OS fails to notice MP CPU

Philip Guenther
On Sun, 21 Oct 2018, Dr.Boutari wrote:

> -------- Original Message --------
> On 21 Oct 2018, 01:55, Philip Guenther wrote:
>
> IMHO, if you want assistance from the OpenBSD developers in making this
> machine function better, you need to go back to the original problem: when
> booting a _stock_ amd64 GENERIC.MP kernel with _nothing_ disabled, what
> exactly happens:
> - what is the sequence of events leading to the crash (boot? shutdown?)
> - what is the output before the crash?
> - if it drops into ddb, then gather the crash or panic message and the
> other information requested
> –------------
...
> I downloaded bsd.mp from a mirror and booted it first with acpi enabled.
> It crashed as usual at about the same point where the crash occured
> previously

Unfortunately, you haven't told us what that "same point is" or what the
output leading up to it is, so we're at an impasse.


...
> On another note. Just wondering why doesnt the amd64 install cd
> recognise the cores? Does it default to loading an sp kernel for the
> install which is the bsd.rd ?

Yes, because that makes bsd.rd smaller and more robust and you should need
MP when doing install/upgrade/recovery/etc.


Philip Guenther

Reply | Threaded
Open this post in threaded view
|

Re: OS fails to notice MP CPU

Dr.Boutari
-------- Original Message --------
On 21 Oct 2018, 22:52, Philip Guenther wrote:

> I downloaded bsd.mp from a mirror and booted it first with acpi enabled.
> It crashed as usual at about the same point where the crash occured
> previously

>Unfortunately, you haven't told us what that "same point is" or what the output leading up to it is, so we're at an impasse.

Its the same point where it crashes with the bsd.sp kernel with acpi enabled.

I already attached an .odt file of dmesg comparison in my previous email pin pointing the time it happens.

Thanks
Reply | Threaded
Open this post in threaded view
|

Re: OS fails to notice MP CPU

Philip Guenther
On Sun, 21 Oct 2018, Dr.Boutari wrote:

> -------- Original Message --------
> On 21 Oct 2018, 22:52, Philip Guenther wrote:
>
> > I downloaded bsd.mp from a mirror and booted it first with acpi enabled.
> > It crashed as usual at about the same point where the crash occured
> > previously
>
> >Unfortunately, you haven't told us what that "same point is" or what
> >the output leading up to it is, so we're at an impasse.
>
> Its the same point where it crashes with the bsd.sp kernel with acpi
> enabled.
>
> I already attached an .odt file of dmesg comparison in my previous email
> pin pointing the time it happens.

My apologies, but I'm not going to expose my computer to the complexity of
OpenOffice in order to see what could be captured at worst as a screenshot
and at best as plain text.

You've sent screenshots of crashes when booted with acpi (but not radeon)
disabled, and of i386 install crashes, and of amd64 bsd.rd dmesgs, and
you've sent textual dmesgs of boots when both acpi and radeon are
disabled.  I don't understand why capturing the output of booting amd64
bsd.mp with nothing disabled requires OpenOffice.


Philip Guenther

Reply | Threaded
Open this post in threaded view
|

Re: OS fails to notice MP CPU

Dr.Boutari
-------- Original Message --------
On 21 Oct 2018, 23:40, Philip Guenther wrote:

On Sun, 21 Oct 2018, Dr.Boutari wrote:
> -------- Original Message --------
> On 21 Oct 2018, 22:52, Philip Guenther wrote:
>

>I don't understand why capturing the output of booting amd64 bsd.mp with nothing disabled requires OpenOffice.

Easier formating with colors to notice differences.

Anyway here is a screencap.

On another note,dunno if it helps but I suspected one of my usb ports could be the problem after noticing the crash occuring around uhub message so I disabled uhub and it took longer to crash this time. It actually loaded the vga display before crashing since probably the vga and other devices need uhub.

scrot.png (180K) Download Attachment