[okl4-developer] Fast address-space switching on GUMSTIX?

[xi yang]

On Tue, May 27, 2008 at 12:59 AM, Geoffrey Lee <glee at ok-labs.com> wrote:
> On Tue, May 27, 2008 at 12:36:42AM +1000, xi yang wrote:
>
> Hi Xi
>
>> >
>> > The location of the stack and the heap are chosen by the build system
>> > when the boot image is built.  In 2.1, the allocator for the stack
>> > and heap locations do not take ARM domains into account, which leads
>> > to the issue you are seeing.
>> >
>> > This has since been fixed in the latest version.
>> The lastest version? you mean 2.1.
>> I am using 1.5.2. I modify the build.py, now it works for me.
>
> Unfortunately this issue is still present in 2.1.  We fixed this
> in our development branch some time ago.
>
> Just an aside, is there any reason why you cannot use 2.1?
> Do you have code or drivers which depend on 1.5.2?
I am doing my undergraduate thesis in ERTOS. "RTEMS on L4" is my
topic. 2.1 was not released when I came here, 3 month ago :).  RTEMSL4
doesn't depends on 1.5.2.
>
>> >
>> >> 2)Are there some constrains of memsection_create_fixed() system call
>> >> of iguana. I invoking this system call and always return NULL.
>> >
>> > Which platform are you running, and which address do you want to ask
>> > Iguana to return to you?
>> Gumstix, I want to get 6Mbytes map to 0xa5100000, it seems return NULL.
>> Regards.
>
> Just making sure, is 0xa5100000 a physical address or a virtual address?
> More generally, can you let us know what are you trying to achieve
> by mapping 6M?  We may be able to propose alternative solutions
> that better suit your needs.

0xa5100000 is a virtual address. When start up RTEMSL4, I have to get
a 6Mbytes space for RTEMS. If direct creating
a memsection in default method. It will cause domain fault then
flushing TLB and Cache since Iguana will map these 6Mbytes to
0x8000XXX. Flushing TLB and Caches will cause interrupt latency to a
huge number. Hence I have to avoid it. That's why I want to use
memsection_create_fixed function.  Now, I just put a big array to BSS
section, it works. Interrupt latency is absolutely good now!

Regards
>
>        -gl
>
>> >
>> >        -gl
>> >
>> >> Regards!
>> >> >>
>> >> >> Regards
>> >> >> Abi
>> >> >>
>> >> >
>> >>
>> >> _______________________________________________
>> >> Developer mailing list
>> >> Developer at okl4.org
>> >> https://lists.okl4.org/mailman/listinfo/developer
>> >>
>> >
>> > --
>> >
>> >
>>
>
> --
>
>