Merging 64 bit changes to -HEAD

[Jayachandran C.]

On Thu, Jul 1, 2010 at 10:25 PM, Luiz Otavio O Souza <lists.br at gmail.com> wrote:
> On Jun 30, 2010, at 7:40 PM, Jayachandran C. wrote:
>> On Wed, Jun 30, 2010 at 10:38 PM, Luiz Otavio O Souza
>> <lists.br at gmail.com> wrote:
>>> On Jun 30, 2010, at 9:57 AM, Jayachandran C. wrote:
>>>
>>>> On Tue, Jun 29, 2010 at 10:32 PM, Luiz Otavio O Souza
>>>> <lists.br at gmail.com> wrote:
>>>>>
>>>>> On Jun 29, 2010, at 8:02 AM, Jayachandran C. wrote:
>>>>>
>>>>>> On Tue, Jun 29, 2010 at 2:28 AM, Luiz Otavio O Souza <lists.br at gmail.com> wrote:
>>>>>>>> Thanks for the the update. Looks like pmap_map for kernel is failing,
>>>>>>>> may be the new tlb_update code causes this.  Can you apply the
>>>>>>>> attached patch and see if the problem still persists, it replaces the
>>>>>>>> new tlb_update code with the older version.
>>>>>>>>
>>>>>>>> Obviously not a fix, but if we can narrow it down to this function,
>>>>>>>> fixing will be easier.
>>>>>>>>
>>>>>>>> JC.
>>>>>>>> <try.diff>
>>>>>>>
>>>>>>> JC,
>>>>>>>
>>>>>>> This fix the problem ! Thanks ! Now, at least, you know where to look :)
>>>>>>
>>>>>> The new tlb_update does not seem to update the tlb entry if the tlbp
>>>>>> fails.  Here's a patch that should make the new function behave like
>>>>>> the older one.  The patch is in attached file 'tlb-update.diff'.
>>>>>>
>>>>>> If that does not work, I'm not sure what the issue is.  You could also
>>>>>> try try the nop-change.diff attached. It tries to switch the ssnop
>>>>>> used for delay in the new code with 'nop' which was used by the old
>>>>>> code.
>>>>>>
>>>>>> Thanks,
>>>>>> JC.
>>>>>> <tlb-update.diff><nop-change.diff>
>>>>>
>>>>> JC,
>>>>>
>>>>> The nop-change seems to have no effect at all and with the tlb-update patch the kernel apparently crash at bzero(), here is the dmesg with TRAP_DEBUG enabled:
>>>>>
>>>>> http://mips.pastebin.com/jydPvJ20
>>>>>
>>>>> So hopefully you are on the right track and this may be something obvious to you.
>>>>
>>>> Not yet :) I really hoped the earlier change would fix it.  The number
>>>> of nop does not seem to be the issue as it is higher in the C code
>>>> than the assembly.
>>>>
>>>> Can you try the attached patch (try.diff) - this re-implements the
>>>> assembly code functionality almost in the same way in C.  This really
>>>> should work, given that the patch which made it assembly worked...
>>>>
>>>> If that works can you see if the second attached patch works, this
>>>> fixes a potential problem (ie, we should be masking 13bits for TLBHI).
>>>>
>>>> Both patches should apply directly to SVN (not dependent on each
>>>> other, or on previous patches)
>>>>
>>>> Thanks again,
>>>> JC.
>>>> <try.diff><pte.h-fix.diff>
>>>
>>>
>>> JC,
>>>
>>> The try.diff works with or without the pte.h change (at least for a simple boot) and the pte.h change does nothing without the try.diff.
>>
>> I've attached the final(final.diff) version I want to check-in, can
>> you please quickly test it?
>>
>> If that does not work, can you tell me if the attached alt1.diff or
>> alt2.diff works? The try.diff had three changes: handle case of
>> index>0, remove pagemask operation, restore full entryhi instead of
>> asid. So if the first does not work, this will help narrow down the
>> rest of the cases.
>>
>> Hopefully this is the last iteration :)
>>
>> JC.
>> <final.diff><alt1.diff><alt2.diff>
>
>
> JC,
>
> The final.diff and alt1.diff crash (into debugger) like the first try.diff (here is the dmesg: http://pastebin.com/r2uunZPZ).
>
> The alt2.diff works, i'll try it a little more (buildworld).
>
> Feel free to send more tests if needed.

Looks like I was on the wrong track, it now looks like the pagemask
maybe the cause.  My suspicion is that the bootloader setups a
pagemask and we never clear it because all the operations save and
restore the mask. As far as I can see the TLB exception handler will
not update pagemask.

Juli - any comments on this? Do you need to save/restore pagemask for
some reason, otherwise I will take out the part from tlb.c.  Will send
out a patch a bit later.

JC.