headers
A Matsui | 1 Jun 2004 07:18
Picon
Picon

Re: init_FIQ

Thank you for advices.

> The 720T is v4T (Architecture version 4 with Thumb extensions) and it
> definitely does support high vectors.

I understood.

> However, looking at your original mail, you have the following:
> 
> | #define FIQ_VECTOR (vectors_base() + 0x1c)
> | #define vectors_base()  (0)
> | static unsigned long no_fiq_insn;

I thought 'vectors_base() (0)' was chosen but I found it was wrong and
the other one should be chosen as I had advices.

#ifdef __ARM_ARCH_4__
#define vectors_base()			((cr_alignment & CR_V) ? 0xffff0000 : 0)
#else
#define vectors_base()			(0)
#endif

> If your CPU supports high vectors (which it does) then your redefinition
> of vectors_base() is the central cause of the problem.  Keep the kernels
> own definition of this macro - it's correct for all cases.

After I've kept the kernels own definition like the above, I tried but
the result was not change. I'm reading data sheet of the CPU but I'm not
sure that the CPU supports High Vector or not.
(Continue reading)

Permalink | Reply | 10 comments |
headers
Russell King - ARM Linux | 1 Jun 2004 09:48
Picon

Re: init_FIQ

On Tue, Jun 01, 2004 at 02:18:39PM +0900, A Matsui wrote:
> After I've kept the kernels own definition like the above, I tried but
> the result was not change. I'm reading data sheet of the CPU but I'm not
> sure that the CPU supports High Vector or not.
> 
> Can I use Low vector with 720T in this kernels? If possible I would like
> to know what part of the kernel should be modified (ex. entry-common.S).

The 720T supports high vectors.  If it didn't the kernel wouldn't run at
all.

-------------------------------------------------------------------
Subscription options: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
FAQ:       http://www.arm.linux.org.uk/armlinux/mlfaq.php
Etiquette: http://www.arm.linux.org.uk/armlinux/mletiquette.php
Permalink | Reply | 9 comments |
headers
Lothar Wassmann | 1 Jun 2004 10:00
Picon
Favicon

Re: PXA+PCMCIA cache coherence

Hi Russell,

Sorry for the delay, there were some realtime threads (AKA customers)
demanding my attention and a prolonged weekend.

Russell King - ARM Linux writes:
> 
> Lothar, what's the story?  I've not heard anything since Tuesday.
> You gave the impression that you already have an analysis.
>
Ok, here is what I was doing and how far I got:
I have a test program that mmaps a file on an ext2 file system on an
CF card in TRUE IDE mode and starts reading it from the last page to
the first one (after unmounting and remounting the file system).
As a second test I copied /bin/bash to the CF card and
try to start it from there.
With the standard kernel (2.6.6 with bkpxa patches + KARO specific
additions) I consistently get data errors as soon as doing the first
read of any page in the file.
Execution of /bin/bash fails on the first try after mounting the file
system with various errors like segfault or inconsistencies in the ELF
binary. Subsequent calls work fine. Also copying the file off the CF
card before executing it (from CF card) helps.

With the call to flush_dcache_page() in function do_no_page() in
mm/memory.c where flush_icache_page() my test program doesn't find any
errors, but starting an ELF executable from CF card still doesn't work.

I traced the code path in case of an error as following:
The page fault when accessing a page of the mapped file for the first
(Continue reading)

Permalink | Reply | 7 comments |
headers
A Matsui | 1 Jun 2004 10:47
Picon
Picon

Re: init_FIQ

Thank you for comment.

> > Can I use Low vector with 720T in this kernels? If possible I would like
> > to know what part of the kernel should be modified (ex. entry-common.S).
> 
> The 720T supports high vectors.  If it didn't the kernel wouldn't run at
> all.

I see. I'm confirming the maker whether this CPU supports High vector or
not. When I get the reply I'll post to the mailing lists.

Regards,
Akira Matsui

-------------------------------------------------------------------
Subscription options: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
FAQ:       http://www.arm.linux.org.uk/armlinux/mlfaq.php
Etiquette: http://www.arm.linux.org.uk/armlinux/mletiquette.php
Permalink | Reply | 8 comments |
headers
Javier Álvarez | 1 Jun 2004 11:04
Picon

Re: init_FIQ

I don´t think you need to confirm this... Believe me. If Mr. King tell you
this, then it is true...

----- Original Message -----
From: "A Matsui" <matsui.akira <at> canon.co.jp>
To: <linux-arm-kernel <at> lists.arm.linux.org.uk>
Cc: "Philip Blundell" <pb <at> nexus.co.uk>
Sent: Tuesday, June 01, 2004 10:47 AM
Subject: Re: init_FIQ

> Thank you for comment.
>
> > > Can I use Low vector with 720T in this kernels? If possible I would
like
> > > to know what part of the kernel should be modified (ex.
entry-common.S).
> >
> > The 720T supports high vectors.  If it didn't the kernel wouldn't run at
> > all.
>
> I see. I'm confirming the maker whether this CPU supports High vector or
> not. When I get the reply I'll post to the mailing lists.
>
> Regards,
> Akira Matsui
>
> -------------------------------------------------------------------
> Subscription options:
http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
> FAQ:       http://www.arm.linux.org.uk/armlinux/mlfaq.php
(Continue reading)

Permalink | Reply | 7 comments |
headers
A Matsui | 1 Jun 2004 11:15
Picon
Picon

Re: init_FIQ

I believe you and Mr. king said. 720T support high vector but, is not
there the case that the CPU dose not support high vector? 

> I don´t think you need to confirm this... Believe me. If Mr. King tell you
> this, then it is true...

Regards,
Akira Matsui

-------------------------------------------------------------------
Subscription options: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
FAQ:       http://www.arm.linux.org.uk/armlinux/mlfaq.php
Etiquette: http://www.arm.linux.org.uk/armlinux/mletiquette.php
Permalink | Reply | 6 comments |
headers
Russell King - ARM Linux | 1 Jun 2004 12:00
Picon

Re: PXA+PCMCIA cache coherence

On Tue, Jun 01, 2004 at 10:00:23AM +0200, Lothar Wassmann wrote:
> The subsequent find_get_page() finds a page in the page cache that has
> been added by mpage_readpages() (called from do_page_cache_readahead()).
> Since the page isn't uptodate, execution continues at
> 'page_not_uptodate:'
> filemap_nopage() calls lock_page() which io_sleep()'s since the page
> has been added to the page cache with the PG_locked bit set.

Ok, this happens after we've called the address space operations ->readpage
method to actually read the page.

I commend you for your analysis, but you need to go to some deeper
levels, since its the deeper levels which are of concern.

> BTW this error is independent from the PCMCIA and IDE layer. It also
> occurs using a loop device and a file in tmpfs or mounted via NFS as
> file backed storage.

I've been using your disktest program this morning on a loop mount with
the backing file on ext2 on a real IDE, and also tried it on a NFS file,
both on a StrongARM-110 based machine.  So far, I can't get it to fail,
and I'd expect it to do so.

Are you using write allocate caches on PXA?

-------------------------------------------------------------------
Subscription options: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
FAQ:       http://www.arm.linux.org.uk/armlinux/mlfaq.php
Etiquette: http://www.arm.linux.org.uk/armlinux/mletiquette.php
(Continue reading)

Permalink | Reply | 6 comments |
headers
Erik Mouw | 1 Jun 2004 12:34
Picon
Favicon

Re: Instruction cycles

On Fri, May 28, 2004 at 05:02:05PM +0100, Russell King - ARM Linux wrote:
> If you want to create a small delay in kernel space, use udelay().  It
> will give you an accurate delay of known time no matter what CPU you're
> running on.

And no matter if the code is in cache or not.

Erik

--

-- 
----  Erik Mouw  ----  www.bitwizard.nl  ----  +31 15 2600 998  ----

-------------------------------------------------------------------
Subscription options: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
FAQ:       http://www.arm.linux.org.uk/armlinux/mlfaq.php
Etiquette: http://www.arm.linux.org.uk/armlinux/mletiquette.php
Permalink | Reply | 0 comments |
headers
Russell King - ARM Linux | 1 Jun 2004 13:00
Picon

Re: init_FIQ(Upgrade to 2.4.26)

On Mon, May 31, 2004 at 03:05:20PM +0900, A Matsui wrote:
> I found the size of input_data is deffrence using the bellow:
> 
> arm-linux-nm --size-sort arch/arm/boot/compressed/vmlinux

Have you checked the size of the main vmlinux ?

arm-linux-size vmlinux

-------------------------------------------------------------------
Subscription options: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
FAQ:       http://www.arm.linux.org.uk/armlinux/mlfaq.php
Etiquette: http://www.arm.linux.org.uk/armlinux/mletiquette.php
Permalink | Reply | 1 comment |
headers
Nicolas Pitre | 1 Jun 2004 16:25

Re: PXA+PCMCIA cache coherence

On Tue, 1 Jun 2004, Russell King - ARM Linux wrote:

> On Tue, Jun 01, 2004 at 10:00:23AM +0200, Lothar Wassmann wrote:
> > BTW this error is independent from the PCMCIA and IDE layer. It also
> > occurs using a loop device and a file in tmpfs or mounted via NFS as
> > file backed storage.
> 
> I've been using your disktest program this morning on a loop mount with
> the backing file on ext2 on a real IDE, and also tried it on a NFS file,
> both on a StrongARM-110 based machine.  So far, I can't get it to fail,
> and I'd expect it to do so.
> 
> Are you using write allocate caches on PXA?

People have already reported the error with write allocate _and_ write back 
caches.

But another difference between SA1110 and PXA2xx is the cache size:
SA1110 is 8kb + 16kb.  PXA/XScale is 32kb + 32kb.  Maybe the problem is
masked by earlier cache eviction on SA (although such explanation appears 
far fetched to me).

Nicolas

-------------------------------------------------------------------
Subscription options: http://lists.arm.linux.org.uk/mailman/listinfo/linux-arm-kernel
FAQ:       http://www.arm.linux.org.uk/armlinux/mlfaq.php
Etiquette: http://www.arm.linux.org.uk/armlinux/mletiquette.php
(Continue reading)

Permalink | Reply | 3 comments |

Gmane