Re: Multiword instruction support and memory poll PC reporting.

Kevin O'Connor wrote:
> On Thu, Apr 17, 2008 at 10:39:56PM +0100, Oliver Ford wrote:
>
>> I've uploaded some of the changes I've made to my copy of haret in
>> patch form: (I don't know if the list supports attachments)
>> http://www.oliford.co.uk/hpipaq214/haretChanges/
>>
>> 0001 and 0004 are for the multi word transfer support (ldmia etc).
>>
>
> Hi Oliver,
>
> I checked in 1 and 4 - Thanks! I did reformat it a little.
>
Ok, I'll switch to the CVS changes some time next week, and take a look.
Glad to be of some help.
>
>> 0003 is the code I used to develop and test them. It just uses a
>> series of them to write in different ways to the beginning of RAM on
>> the ipaq214 (unused by winCE) and 1 load to read some back. I just
>> checked that the area and the result looks the same with and without
>> the simulation on.
>>
>
> I'm assuming you aren't proposing to have this checked in. If you
> want to clean it up by making the test its own command, I'd be okay
> with checking it in. (It's a neat trick to trace in one haret session
> while running commands in another haret session.) If you want to do
> this, I'd use VRAM as the memory area to read/write to.
>
No, it wasn't intended to be an actual change, just wanted to give you
the code to test the other code with.
>
>> 0005 adds reporting of the (approximate) PC when checkPolls() detects
>> changes in the fault/IRQ handlers. I used it to find the rough area
>> of memory that is changing something when the mmutrace doesn't see it
>> (see my next email). It works quite well if lots of interrupts are
>> happening.
>>
>
> I have a couple of questions on this:
>
> * why report the address on 'IRQS' polling? Wouldn't external
> interrupts have essentially random pc values?
>
I see your point, this was just a quick addition that wasn't going to be
permanent but turned out to be quite useful. I guess I could pass 0 in
as faultaddr for the checkPolls(.., &data->irqpoll ..) line and then
change the reporting to not report 0s.
> * why report VA instead of MVA? Every other part of haret uses MVAs.
>
>
I was using it to find out which modules where changing the registers I
was watching. The LSMOD commands reports VAs and without knowing the PID
of each process, the MVAs weren't much use to be.
> I think its a great idea, but I see it only being useful for
> sophisticated users. I want to be careful not to confuse newer users.
>
>
Fair enough. This will probably go for the L1 table merging code as
well. Is there some way things like this can be added as options so if
anyone like me tries doing this, they have them available?

Oliver
Received on Fri Apr 25 2008 - 05:43:21 EDT