On 08/24/2012 05:07 AM, Adam Hraska wrote:
> On Thu, Aug 23, 2012 at 1:21 PM, Jakub Jermar <[email protected]> wrote:
>> On 23.8.2012 9:56, Adam Hraska wrote:
>>> On Thu, Aug 23, 2012 at 12:45 AM, Jakub Jermar <[email protected]> wrote:
>>>> Tonight, I did some testing and unfortunately it appears the branch is
>>>> not that scrubbed or even tested as one would have hoped. These are the
>>>> issues I noticed:
>>>>
>>>> - test cht panics the kernel when run on real amd64 or ia32 hardware
>>>> (4-core Intel i5, 1.4GHz, 4G memory)
>>>> - test cht panics the kernel when run on the above machine in QEMU (-smp
>>>> 4 --enable-kvm)
>>>> - unfortunately the panic message is interleaved with some other
>>>> concurrent output so it is completely unreadable (some synchronization
>>>> should be probably added back to printf)
>>>
>>> That is very unfortunate. While I expected there to be some latent
>>> bugs in the implementation (as in any new software) I would not have
>>> expected "test cht" would fail (eg leak memory) and definitely not to
>>> panic.
>>>
>>> It is however quite difficult to respond to this point since I have no
>>> idea what went wrong.
>>
>> I believe this is the same thing as the QEMU panic on UP. The same
>> assertion is just hit by each CPU at about the same time, which would
>> explain the blended panic message.
> 
> Hmm, good observation. That very well explains the panic messages.
> If you by any chance had some spare time and were in the proper
> mood could you, please, remove the assert in adt/cht.c:872 and give
> it one last try? Without any holding back, eg SMP and real hw.
> 
> The assert is not really needed and I already regret adding it.
> 
> If it passes the test without any errors it means the panics are a
> result of a silly mistake (the overly protective assert). If not, well,
> then there are actual bugs in the code.

I have some good news for you, without the assert, test cht does not crash.

Jakub

_______________________________________________
HelenOS-devel mailing list
[email protected]
http://lists.modry.cz/cgi-bin/listinfo/helenos-devel

Reply via email to