Re: kbuild 2.5.26 - arch/alpha

Re: kbuild 2.5.26 - arch/alpha

[Martin Brulisauer]

On 23 Jul 2002, at 18:29, George France wrote:

> On Tuesday 23 July 2002 14:48, Oliver Pitzeier wrote:
> >
> > [ ... ]
> >
> > > You have made me aware that we have unintentionally created a
> > > private sort of club.  I apologize. This will have to be corrected.
> >
> > That's not what I expected to read...
> > I think that this "private club" is not wrong at all... It just would
> > be nicer if there would be some kind of batch every week where all
> > alpha users/developers get a mail...
> 
> I agree. We should send a weekly e-mail with the current status.

I did not see any news on the alpha/linux topic in lkml lately.

What is the way to keep in touch with the "private club" to help/
assist in getting further to a running 2.5.x kernel on alpha? I
am still on 2.4.18 on my test system.

Did anybody use gcc-3.0.x or gcc-3.1? With gcc-3.0.4 I 
successfully built 2.4.18 but some applications don't run
correctly (eg. MySQL -> Parser). Is the kernel compilable
with gcc-3.1? Today I am using gcc-2.95.3 and I think is
ok; better than egcs (generates less unaligned traps at
runtime without changing the source).


Greetings,
Martin

Re: kbuild 2.5.26 - arch/alpha

[George France]

Hello Martin,

I am glad to have received your e-mail.  I was just thinking about
you.  For a quick update, as I am attempting to get demos out the
door for LWE:

Bryce has convinced me to setup a daily diary on advantgo.

   http://www.advogato.org/person/France

which I plan on updating this afternoon.  In short for me on alpha this
week:

Monday: I worked on the usb-uhci.c driver which has a few
               32 bitisms.  I will not have time to submit a tested
               patch until after LWE. The problem appears to be

                   unsigned int io_addr = pci_resource_start()

               and possibly

                   unsigned int io_size = pci_resource_len()

              putting a 64bit address into a 32bit slot just does not
              work very well. :-)  Even after changes the 'int' to 'long',
              the USB device worked extremely well, but upon inserting or
              removing a USB device, the SCSI controller on my system
              hangs for about 30 secs while it resets due to receiving
              an invalid instruction.  I suspect that there is corruption
              of some kind on the PCI bus, but I do not have time this 
             week to track this down.

Tuesday: My Binutils patch was accepted for adding
               -mev67, -mev68, -m21264a and -m21264b
               Chatted with Bryce on several Alpha related issues.

Wed: Meet with Jay, Jeff, John and Harry in Nashua.
         Jeff has 2.5.x (x=29 IIRC) working with smp and non smp systems.
         after some testing the patches should make it to the kernel soon.
         I pushed out updates for RH7.2 (Alpha) for  gcc, util-linux, glibc
         and openssl.
         I worked on the RSS patches for the autobuild system.

Thur: I chatted (e-mail) with Peter Petrakis today.  He would like me to upgrade
         the system which hosts alphanews.net and linuxalpha.org to use RH7.2
         I hope to have time on Friday (today, yikes!), before I leave for LWE.

As for gcc 3.1 in the kernel, most Alpha kernel hackers use egcs or 2.95. 
Personally I tend to use 2.95 for my kernels.

I have UNH students that build several complete toolchains everyday
for alpha including 2.95.x, 3.0.x, 3.1.x and gcc-head.  By complete tool
chain I mean binutils, gcc, gdb, glibc and all the support programs.

   http://handhelds.org/projects/toolchain/autobuild/build-results.php3

I am certain that they would like to chat with you in great detail about
toolchain issues related to the Alpha architecture.

I agree that communications in regard to Alpha could and should be better.
We should probably setup a wiki or webpage to help keep track of  Alpha
issues or maybe just use one of the existing alpha mailing lists or I could
setup something on alpha.crl.dec.com next week.

I hope this helps.

Best Regards,


--George

On Friday 09 August 2002 06:00, Martin Brulisauer wrote:
> On 23 Jul 2002, at 18:29, George France wrote:
> > On Tuesday 23 July 2002 14:48, Oliver Pitzeier wrote:
> > > [ ... ]
> > >
> > > > You have made me aware that we have unintentionally created a
> > > > private sort of club.  I apologize. This will have to be corrected.
> > >
> > > That's not what I expected to read...
> > > I think that this "private club" is not wrong at all... It just would
> > > be nicer if there would be some kind of batch every week where all
> > > alpha users/developers get a mail...
> >
> > I agree. We should send a weekly e-mail with the current status.
>
> I did not see any news on the alpha/linux topic in lkml lately.
>
> What is the way to keep in touch with the "private club" to help/
> assist in getting further to a running 2.5.x kernel on alpha? I
> am still on 2.4.18 on my test system.
>
> Did anybody use gcc-3.0.x or gcc-3.1? With gcc-3.0.4 I
> successfully built 2.4.18 but some applications don't run
> correctly (eg. MySQL -> Parser). Is the kernel compilable
> with gcc-3.1? Today I am using gcc-2.95.3 and I think is
> ok; better than egcs (generates less unaligned traps at
> runtime without changing the source).
>

Re: kbuild 2.5.26 - arch/alpha

[Måns Rullgård]

"Martin Brulisauer" <martin@bruli.net> writes:

> Did anybody use gcc-3.0.x or gcc-3.1? With gcc-3.0.4 I 
> successfully built 2.4.18 but some applications don't run
> correctly (eg. MySQL -> Parser). Is the kernel compilable
> with gcc-3.1? Today I am using gcc-2.95.3 and I think is
> ok; better than egcs (generates less unaligned traps at
> runtime without changing the source).

I've been using gcc 3.1 to compile kernel for a while (since it was
released).  I don't have any problems with 2.4.18 and upward.  I don't
remember which compiler I used previously.

-- 
Måns Rullgård
mru@users.sf.net

Re: kbuild 2.5.26 - arch/alpha

[Thunder from the hill]

Hi,

On Fri, 9 Aug 2002, Martin Brulisauer wrote:
> Is the kernel compilable with gcc-3.1?

Works with me. I didn't have problems on Alpha for now...

Well, yes, I did have problems compiling the stock kernels. But that's a 
different page.

			Thunder
-- 
--./../...-/. -.--/---/..-/.-./..././.-../..-. .---/..-/.../- .-
--/../-./..-/-/./--..-- ../.----./.-../.-.. --./../...-/. -.--/---/..-
.- -/---/--/---/.-./.-./---/.--/.-.-.-
--./.-/-.../.-./.././.-../.-.-.-

RE: kbuild 2.5.26 - arch/alpha

[Martin Brulisauer]

On 24 Jul 2002, at 0:52, Oliver Pitzeier wrote:
> (Maybe an CVS tree for alpha-kernel developers. Those people
> must be trusted you can imagine...)
As I can see in the kernel howto, there is only one
linux version _all_ patches have to go to at least -
that's the one Linus maintains. And this makes
all sense to me - it keeps the kernel alive and
avoids falling apart over the time.

Martin

RE: kbuild 2.5.26 - arch/alpha

[Martin Brulisauer]

On 24 Jul 2002, at 1:05, Oliver Pitzeier wrote:
> > On 23 Jul 2002, at 18:29, George France wrote:
> > > site or mailing list.  It is probably better to utilize 
> > what exist.  Either 
> > > the debian-alpha@lists.debian.org, axp-list@redhat.com,   
> > > axp-hardware@talisman.alphalinux.org or suse-axp@suse.com.  In the 
> > > worst
> > 
> > As long as these lists are kernel (?) and not user oriented 
> > and not too distribution specific.
> 
> That's why I think it would be the best to setup a new one. But you'll
> never be able to lock out distribution specify questions and so
> replies... :o(
> 

All these mailing lists are used to have kind of a "virtual" team,
are they? If we move the linux/alpha discussions away from
linux-kernel and/or vger.kernel.org how big is the risk that we
solve problems in this list others have already done?

In my oppinion linux/alpha should not be treated as a separate
task, offline from other kernel discussions. Too much information
could get lost - and as we can see now - linux/alpha is way
behind the current tree Linus maintains.

If Linus will participate in a linux/alpha mailing list, then it would
be a good idea - If not ... (guess if he will ;-}).

Regards,
Martin

Re: kbuild 2.5.26 - arch/alpha

[Martin Brulisauer]

>>On Sat, 20 Jul 2002, Martin Brulisauer wrote:
>> Is the kernel arch tree for alphas not maintained anymore? If I download
>> the vanilla 2.5.26 I can't build it at all. Even a make clean fails
>> due to missing directives in arch/alpha/kernel/Makefile.
>
>On Sat, 20 Jul 2002, Thunder from the hill wrote:
>What exactly are you experiencing?
>

make mrproper:	ok.
make defconfig:	ok.
make dep:

make[1]: Entering directory `/usr/src/linux-2.5.27'
make[1]: Nothing to be done for `include/linux/modversions.h'.
make[1]: Leaving directory `/usr/src/linux-2.5.27'

make boot:

make[1]: Entering directory `/usr/src/linux-2.5.27/scripts'
  gcc -Wp,-MD,./.split-include.d -Wall -Wstrict-prototypes -O2 -fomit-frame-pointer   -o split-include split-include.c
In file included from /usr/include/linux/errno.h:4,
                 from /usr/include/bits/errno.h:25,
                 from /usr/include/errno.h:36,
                 from split-include.c:26:
/usr/include/asm/errno.h:4: asm-generic/errno-base.h: No such file or directory
make[1]: *** [split-include] Error 1
make[1]: Leaving directory `/usr/src/linux-2.5.27/scripts'
make: *** [scripts] Error 2


make clean:

make[1]: Entering directory `/usr/src/linux-2.5.27/arch/alpha/kernel'
make[1]: *** No rule to make target `clean'.  Stop.
make[1]: Leaving directory `/usr/src/linux-2.5.27/arch/alpha/kernel'
make: *** [archclean] Error 2


Looks to me like noone ever tried to compile this
kernel on this platform. That is why I asked my
silly question.

Regards,
Martin

RE: kbuild 2.5.26 - arch/alpha

[Martin Brulisauer]

On 21 Jul 2002, at 18:57, Oliver Pitzeier wrote:
> Oh... I did... :o( :o) But it's currently a mission
> impossible. The last kernel running fine for my alpha
> is 2.5.18 (with a lot of patches...)
>
> I'm also currently not sure that kernel 2.6.X will
> ever run on alpha. There are not very much alpha-users.
> And there are lesser alpha kernel maintainers.
> Ivan Kokshaysky and "Thunder from the hill" are two
> persons who often work an the Alpha Code. And me
> as well (a bit....). But it's currently not easy
> to fix the new errors (for alpha) in every kernel
> release, because they are growing...

Do you think it's worth the time to patch the current
version? Will Linus apply the patch so we will hopefully
have a 2.6.x kernel that compiles (at least) on alpha's?

Is there anybody who is willing to test such a patch
on different alpha's (I only have some XLT's, an AS800
and one AS250, so all alcor based systems with
ISA and PCI but without EISA and all are using sys_alcor.c)?
Further I can't test SMP with this _very_ old hardware.


Regards,
Martin Brulisauer

Re: kbuild 2.5.26 - arch/alpha

[Jan-Benedict Glaw]

[-- Attachment #1: Type: text/plain, Size: 1080 bytes --]

On Tue, 2002-07-23 14:42:03 +0200, Martin Brulisauer <martin@uceb.org>
wrote in message <3D3D6B3B.25754.1392D3FD@localhost>:
> On 21 Jul 2002, at 18:57, Oliver Pitzeier wrote:
> > Oh... I did... :o( :o) But it's currently a mission
> > impossible. The last kernel running fine for my alpha
> > is 2.5.18 (with a lot of patches...)

There was a quite good patch sent to l-k some weeks ago which
(basically) still applies. I'm using this one (with watering eyes
waiting for a compileable Linus-Kernel...).

> Is there anybody who is willing to test such a patch
> on different alpha's (I only have some XLT's, an AS800
> and one AS250, so all alcor based systems with
> ISA and PCI but without EISA and all are using sys_alcor.c)?
> Further I can't test SMP with this _very_ old hardware.

I cannot test SMP either (I've not got a SMP alpha), but I can test on
Miata, Avanit and NoName.

MfG, JBG

-- 
Jan-Benedict Glaw   .   jbglaw@lug-owl.de   .   +49-172-7608481
	 -- New APT-Proxy written in shell script --
	   http://lug-owl.de/~jbglaw/software/ap2/

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

RE: kbuild 2.5.26 - arch/alpha

[Oliver Pitzeier]

Jan-Benedict Glaw wrote:
> On Tue, 2002-07-23 14:42:03 +0200, Martin Brulisauer 
> <martin@uceb.org> wrote in message 
> <3D3D6B3B.25754.1392D3FD@localhost>:
> > On 21 Jul 2002, at 18:57, Oliver Pitzeier wrote:
> > > Oh... I did... :o( :o) But it's currently a mission 
> impossible. The 
> > > last kernel running fine for my alpha is 2.5.18 (with a lot of 
> > > patches...)
> 
> There was a quite good patch sent to l-k some weeks ago which
> (basically) still applies. I'm using this one (with watering 
> eyes waiting for a compileable Linus-Kernel...).

I go and search it...
 
> > Is there anybody who is willing to test such a patch
> > on different alpha's (I only have some XLT's, an AS800
> > and one AS250, so all alcor based systems with
> > ISA and PCI but without EISA and all are using 
> sys_alcor.c)? Further I 
> > can't test SMP with this _very_ old hardware.
> 
> I cannot test SMP either (I've not got a SMP alpha), but I 
> can test on Miata, Avanit and NoName.

I've got a AS1000. Noritake. But I do only have _one_ processor.
I don't believe that anybody buys me a second... :o(

Our Dual-Processor Machine (a DS20e) has been moved back to
Compaq a few month ago... On our seconds DS20e are running postgresql
databases which cannot be stopped. Not even for a few minutes while
rebooting...

-Oliver 

RE: kbuild 2.5.26 - arch/alpha

[Thunder from the hill]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 604 bytes --]

Hi,

On Wed, 24 Jul 2002, Oliver Pitzeier wrote:
> > There was a quite good patch sent to l-k some weeks ago which
> > (basically) still applies. I'm using this one (with watering 
> > eyes waiting for a compileable Linus-Kernel...).
> 
> I go and search it...

Possibly this one?

							Regards,
							Thunder
-- 
(Use http://www.ebb.org/ungeek if you can't decode)
------BEGIN GEEK CODE BLOCK------
Version: 3.12
GCS/E/G/S/AT d- s++:-- a? C++$ ULAVHI++++$ P++$ L++++(+++++)$ E W-$
N--- o?  K? w-- O- M V$ PS+ PE- Y- PGP+ t+ 5+ X+ R- !tv b++ DI? !D G
e++++ h* r--- y- 
------END GEEK CODE BLOCK------

[-- Attachment #2: Type: APPLICATION/octet-stream, Size: 2454 bytes --]

RE: kbuild 2.5.26 - arch/alpha

[Thunder from the hill]

Hi,

On Wed, 24 Jul 2002, Thunder from the hill wrote:
> Possibly this one?

No, I can't find it any more, either. It was a big fat file labeled 
alpha-2.5.20.bz2 or such, IIRC. I've used it until 2.5.24 or so. 
Unfortunately, it never ended up in the patches directories (wonder why).

							Regards,
							Thunder
-- 
(Use http://www.ebb.org/ungeek if you can't decode)
------BEGIN GEEK CODE BLOCK------
Version: 3.12
GCS/E/G/S/AT d- s++:-- a? C++$ ULAVHI++++$ P++$ L++++(+++++)$ E W-$
N--- o?  K? w-- O- M V$ PS+ PE- Y- PGP+ t+ 5+ X+ R- !tv b++ DI? !D G
e++++ h* r--- y- 
------END GEEK CODE BLOCK------

RE: kbuild 2.5.26 - arch/alpha

[Martin Brulisauer]

On 24 Jul 2002, at 0:53, Thunder from the hill wrote:
> Hi,
> 
> On Wed, 24 Jul 2002, Oliver Pitzeier wrote:
> > > There was a quite good patch sent to l-k some weeks ago which
> > > (basically) still applies. I'm using this one (with watering 
> > > eyes waiting for a compileable Linus-Kernel...).
> > 
> > I go and search it...
> 
> Possibly this one?
Don't waste our time. If you have to say anything
constructive do it. If not - keep out of the thread.

Best Regards,
Martin

RE: kbuild 2.5.26 - arch/alpha

[Thunder from the hill]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 583 bytes --]

Hi,

On Wed, 24 Jul 2002, Martin Brulisauer wrote:
> Don't waste our time. If you have to say anything
> constructive do it. If not - keep out of the thread.

Instead of complaining, could you please try out this one? I'm seriously 
trying to be constructive.

							Regards,
							Thunder
-- 
(Use http://www.ebb.org/ungeek if you can't decode)
------BEGIN GEEK CODE BLOCK------
Version: 3.12
GCS/E/G/S/AT d- s++:-- a? C++$ ULAVHI++++$ P++$ L++++(+++++)$ E W-$
N--- o?  K? w-- O- M V$ PS+ PE- Y- PGP+ t+ 5+ X+ R- !tv b++ DI? !D G
e++++ h* r--- y- 
------END GEEK CODE BLOCK------

[-- Attachment #2: Type: APPLICATION/octet-stream, Size: 4579 bytes --]

Re: kbuild 2.5.26 - arch/alpha

[George France]

On Tuesday 23 July 2002 08:42, Martin Brulisauer wrote:
> On 21 Jul 2002, at 18:57, Oliver Pitzeier wrote:
> > I'm also currently not sure that kernel 2.6.X will
> > ever run on alpha. There are not very much alpha-users.
> > And there are lesser alpha kernel maintainers.
> > Ivan Kokshaysky and "Thunder from the hill" are two
> > persons who often work an the Alpha Code. And me
> > as well (a bit....). But it's currently not easy
> > to fix the new errors (for alpha) in every kernel
> > release, because they are growing...

2.6.x will run on alpha.  There are still a handfull of people that still 
activly maintian Linux on Alpha.   Since there is only a few people that 
activly work on Alpha, they tend to chose a kernel versions, then work with 
that version for a while until it is stable.  In the past few months most of 
the efforts have been spent on 2.4.9.  Currently there have been discussions 
in regard to:

1) porting all those patches for 2.4.9 forward to 2.4.[18-19] and 2.5.x.  
2) taking a look at the latest 2.5.x in the next few weeks, as we are aware 
that 2.5.x does not compile on Alpha.

>
> Do you think it's worth the time to patch the current
> version? Will Linus apply the patch so we will hopefully
> have a 2.6.x kernel that compiles (at least) on alpha's?

It is certainly worth the time.  It is not too difficult to get any sane 
patch applied to the kernel for Alpha.  

>
> Is there anybody who is willing to test such a patch
> on different alpha's (I only have some XLT's, an AS800
> and one AS250, so all alcor based systems with
> ISA and PCI but without EISA and all are using sys_alcor.c)?
> Further I can't test SMP with this _very_ old hardware.

I have access to a Lab that is filled with Alpha systems.  I would be happy 
to test any patch as I have time.

Please let me know if I can be of further assistence.

Best Regards,


--George.

Re: kbuild 2.5.26 - arch/alpha

[Ghozlane Toumi]

On Tuesday 23 July 2002 11:05, George France wrote:

> 2.6.x will run on alpha.  There are still a handfull of people that still
> activly maintian Linux on Alpha.   Since there is only a few people that
> activly work on Alpha, they tend to chose a kernel versions, then work with
> that version for a while until it is stable.  In the past few months most
> of the efforts have been spent on 2.4.9.  Currently there have been
> discussions in regard to:

Just out of curiosity, on what mailing lists discussions are taking place ?
I'm subscribed to some alpha mailing lists, but still, most of the kernel 
patches for alpha I see are comming from LKML ...


Thanks, 

ghoz

Re: kbuild 2.5.26 - arch/alpha

[Martin Brulisauer]

On 23 Jul 2002, at 11:05, George France wrote:
> that version for a while until it is stable.  In the past few months most of 
> the efforts have been spent on 2.4.9.  Currently there have been discussions 
> in regard to:
> 
> 1) porting all those patches for 2.4.9 forward to 2.4.[18-19] and 2.5.x.  

I am currently running 2.4.18 from SuSE without any (major) 
problems. I found it here:
ftp://ftp.suse.com/pub/people/sf/axp/7.1/RPMS/kernel-source-
2.4.18.SuSE-0.alpha.rpm.
Then I took arch/alpha/kernel/core_cia.c from version 2.4.12
(the current version does not run on XLT's booting with MILO;
the latest one is 2.4.12).

> 2) taking a look at the latest 2.5.x in the next few weeks, as we are aware 
> that 2.5.x does not compile on Alpha.

Hopefully I can fix core_cia.c to run on XLT's (it's hard to find any 
documentation on this toppic) and arch/alpha/kernel/setup.c for 
machines booting with linload.exe/MILO because the hwrpb
struct is built by MILO and does not match the one booting from
SRM (eg. empty percpu struct resulting in a cpucount of zero
in /proc/cpuinfo).

Regards,
Martin

Re: kbuild 2.5.26 - arch/alpha

[George France]

On Tuesday 23 July 2002 13:01, Martin Brulisauer wrote:
> On 23 Jul 2002, at 11:05, George France wrote:
> > that version for a while until it is stable.  In the past few months most
> > of the efforts have been spent on 2.4.9.  Currently there have been
> > discussions in regard to:
> >
> > 1) porting all those patches for 2.4.9 forward to 2.4.[18-19] and 2.5.x.
>
> I am currently running 2.4.18 from SuSE without any (major)
> problems. I found it here:
> ftp://ftp.suse.com/pub/people/sf/axp/7.1/RPMS/kernel-source-
> 2.4.18.SuSE-0.alpha.rpm.
> Then I took arch/alpha/kernel/core_cia.c from version 2.4.12
> (the current version does not run on XLT's booting with MILO;
> the latest one is 2.4.12).

I have not tried Stefan's 2.4.18 kernel.  I am glad to hear that it works for 
you.  I will give it a try.

> > 2) taking a look at the latest 2.5.x in the next few weeks, as we are
> > aware that 2.5.x does not compile on Alpha.
>
> Hopefully I can fix core_cia.c to run on XLT's (it's hard to find any
> documentation on this toppic) and arch/alpha/kernel/setup.c for
> machines booting with linload.exe/MILO because the hwrpb
> struct is built by MILO and does not match the one booting from
> SRM (eg. empty percpu struct resulting in a cpucount of zero
> in /proc/cpuinfo).

I am not very familiar with the XLT systems. Maybe Jay can help.  He has been 
working on Alpha systems for a very long time.

Jay, do you have any suggestions???

Best Regards,


--George

Re: kbuild 2.5.26 - arch/alpha

[Sven Koch]

On Tue, 23 Jul 2002, George France wrote:

> On Tuesday 23 July 2002 13:01, Martin Brulisauer wrote:
> > Hopefully I can fix core_cia.c to run on XLT's (it's hard to find any
> > documentation on this toppic) and arch/alpha/kernel/setup.c for
> > machines booting with linload.exe/MILO because the hwrpb
> > struct is built by MILO and does not match the one booting from
> > SRM (eg. empty percpu struct resulting in a cpucount of zero
> > in /proc/cpuinfo).
>
> I am not very familiar with the XLT systems. Maybe Jay can help.  He has been
> working on Alpha systems for a very long time.

I am using Stock 2.4.19-rc2 with the following simple patch on an xl-300
with milo:
(without it, it breaks while initalizing the ncr-scsi-controller)

--- linux/arch/alpha/kernel/core_cia.c.orig	Sun Oct 21 19:30:58 2001
+++ linux/arch/alpha/kernel/core_cia.c	Fri Jul 19 16:11:46 2002
@@ -382,10 +382,10 @@
 	for (i = 0; i < CIA_BROKEN_TBIA_SIZE / sizeof(unsigned long); ++i)
 		ppte[i] = pte;

-	*(vip)CIA_IOC_PCI_W1_BASE = CIA_BROKEN_TBIA_BASE | 3;
-	*(vip)CIA_IOC_PCI_W1_MASK = (CIA_BROKEN_TBIA_SIZE*1024 - 1)
+	*(vip)CIA_IOC_PCI_W3_BASE = CIA_BROKEN_TBIA_BASE | 3;
+	*(vip)CIA_IOC_PCI_W3_MASK = (CIA_BROKEN_TBIA_SIZE*1024 - 1)
 				    & 0xfff00000;
-	*(vip)CIA_IOC_PCI_T1_BASE = virt_to_phys(ppte) >> 2;
+	*(vip)CIA_IOC_PCI_T3_BASE = virt_to_phys(ppte) >> 2;
 }

 static void __init


I've got the patch from Alexander Stokman, who was kind to send it to me
~3 month after sending my question to lkml


c'ya
sven

-- 

The Internet treats censorship as a routing problem, and routes around it.
(John Gilmore on http://www.cygnus.com/~gnu/)

Re: kbuild 2.5.26 - arch/alpha

[Jay Estabrook]

On Tue, Jul 23, 2002 at 09:08:26PM +0200, Sven Koch wrote:
> 
> I am using Stock 2.4.19-rc2 with the following simple patch on an xl-300
> with milo:
> (without it, it breaks while initalizing the ncr-scsi-controller)
> 
> --- linux/arch/alpha/kernel/core_cia.c.orig	Sun Oct 21 19:30:58 2001
> +++ linux/arch/alpha/kernel/core_cia.c	Fri Jul 19 16:11:46 2002
> @@ -382,10 +382,10 @@
>  	for (i = 0; i < CIA_BROKEN_TBIA_SIZE / sizeof(unsigned long); ++i)
>  		ppte[i] = pte;
> 
> -	*(vip)CIA_IOC_PCI_W1_BASE = CIA_BROKEN_TBIA_BASE | 3;
> -	*(vip)CIA_IOC_PCI_W1_MASK = (CIA_BROKEN_TBIA_SIZE*1024 - 1)
> +	*(vip)CIA_IOC_PCI_W3_BASE = CIA_BROKEN_TBIA_BASE | 3;
> +	*(vip)CIA_IOC_PCI_W3_MASK = (CIA_BROKEN_TBIA_SIZE*1024 - 1)
>  				    & 0xfff00000;
> -	*(vip)CIA_IOC_PCI_T1_BASE = virt_to_phys(ppte) >> 2;
> +	*(vip)CIA_IOC_PCI_T3_BASE = virt_to_phys(ppte) >> 2;
>  }
> 
>  static void __init

Yes, this will help on XLT-300.

What this patch does is revert the code to an older version which uses
PCI DMA window #3 for the scatter/gather operations, thus avoiding the
use of window #1 for that operation.

On older machines like the XLT-300 and EB164, their core logic, CIA
Rev 1, appears to have a bug in that PCI DMA windows #1 and #2 cannot
be used for scatter/gather. Boxes based on CIA rev 2, and PYXIS, do
not have the same problem.

The patches I attached to an earlier posting essentially do this for
the rev 1 CIA machines, but leave active the code for dual-address
cycle (DAC) support for the CIA rev 2  and PYXIS based machines.

 --Jay++

-----------------------------------------------------------------------------
Jay A Estabrook                            HPTC - LINUX support
Hewlett-Packard Company - MRO1-2/K15       (508) 467-2080
200 Forest Street, Marlboro MA 01752       Jay.Estabrook@hp.com
-----------------------------------------------------------------------------

Re: kbuild 2.5.26 - arch/alpha

[Jay Estabrook]

On Tue, Jul 23, 2002 at 03:00:50PM -0400, George France wrote:
> On Tuesday 23 July 2002 13:01, Martin Brulisauer wrote:
> >
> > Hopefully I can fix core_cia.c to run on XLT's (it's hard to find any
> > documentation on this toppic) and arch/alpha/kernel/setup.c for
> > machines booting with linload.exe/MILO because the hwrpb
> > struct is built by MILO and does not match the one booting from
> > SRM (eg. empty percpu struct resulting in a cpucount of zero
> > in /proc/cpuinfo).
> 
> I am not very familiar with the XLT systems. Maybe Jay can help.  He
> has been working on Alpha systems for a very long time.
> 
> Jay, do you have any suggestions???

Yup, use the following patches, based on pre8-2.4.19, to fix the
DMA windowing problem with the early (read: 300MHz) XLT boxes.

As for the CPU count being zero, well, aside from looking bad, what
seems to be the problem? ;-}

 --Jay++

-----------------------------------------------------------------------------
Jay A Estabrook                            HPTC - LINUX support
Hewlett-Packard Company - MRO1-2/K15       (508) 467-2080
200 Forest Street, Marlboro MA 01752       Jay.Estabrook@hp.com
-----------------------------------------------------------------------------

RE: kbuild 2.5.26 - arch/alpha

[Oliver Pitzeier]

George France wrote:
[ ... ]
> > I am currently running 2.4.18 from SuSE without any (major)
> > problems. I found it here:
> > ftp://ftp.suse.com/pub/people/sf/axp/7.1/RPMS/kernel-source-
> > 2.4.18.SuSE-0.alpha.rpm.
> > Then I took arch/alpha/kernel/core_cia.c from version 2.4.12
> > (the current version does not run on XLT's booting with MILO;
> > the latest one is 2.4.12).
> 
> I have not tried Stefan's 2.4.18 kernel.  I am glad to hear 
> that it works for you. I will give it a try.

For 2.4.18 I have a very good experience. It's running on
our mailserver (as well an alpha-machine [ds10]) with ext3.
Without patches it worked well. Now I have a few patches
(most of them are from l-k list)...

> > > 2) taking a look at the latest 2.5.x in the next few 
> weeks, as we are
> > > aware that 2.5.x does not compile on Alpha.

[ ... ]

Just a comment:
2.5.26/27 doesn't create a modversions.h on my machine... So
make dep/clean and so on doesn't work for me...

-Oliver

RE: kbuild 2.5.26 - arch/alpha

[Oliver Pitzeier]

Martin Brulisauer wrote:
[ ... ]
> Do you think it's worth the time to patch the current
> version? Will Linus apply the patch so we will hopefully
> have a 2.6.x kernel that compiles (at least) on alpha's?

I hope linus joins this thread some day... :o)
 
[ ... ]

-Oliver

RE: kbuild 2.5.26 - arch/alpha

[Oliver Pitzeier]

Martin Brulisauer wrote:
> >>On Sat, 20 Jul 2002, Martin Brulisauer wrote:
> >> Is the kernel arch tree for alphas not maintained anymore? If I 
> >>download  the vanilla 2.5.26 I can't build it at all. Even a make 
> >>clean fails  due to missing directives in 
> arch/alpha/kernel/Makefile.

As I saw it... There is no modversions.h created while trying to
compile the kernel on an alpha... I'll take a look at this in
a few hours (after sleeping... :-) ).

> >On Sat, 20 Jul 2002, Thunder from the hill wrote:
> >What exactly are you experiencing?

[ ... ]
 
> Looks to me like noone ever tried to compile this
> kernel on this platform. That is why I asked my
> silly question.

Have you ever ended this discussion??? I only mean
you two. Because I havn't found a reply from Thunder
to Martin...

However... Let me know...

-Oliver

RE: kbuild 2.5.26 - arch/alpha

[Thunder from the hill]

Hi,

On Wed, 24 Jul 2002, Oliver Pitzeier wrote:
> Have you ever ended this discussion??? I only mean
> you two. Because I havn't found a reply from Thunder
> to Martin...

IIRC, no.

							Regards,
							Thunder
-- 
(Use http://www.ebb.org/ungeek if you can't decode)
------BEGIN GEEK CODE BLOCK------
Version: 3.12
GCS/E/G/S/AT d- s++:-- a? C++$ ULAVHI++++$ P++$ L++++(+++++)$ E W-$
N--- o?  K? w-- O- M V$ PS+ PE- Y- PGP+ t+ 5+ X+ R- !tv b++ DI? !D G
e++++ h* r--- y- 
------END GEEK CODE BLOCK------

kbuild 2.5.26 - arch/alpha

[Martin Brulisauer]

Is the kernel arch tree for alphas not maintained anymore? If I download
the vanilla 2.5.26 I can't build it at all. Even a make clean fails
due to missing directives in arch/alpha/kernel/Makefile.

Has anybody any idea?

Martin