Skip to contentSkip to conversation

milstone reached: ia64 linux builds out of Linus' tree

DISCUSSION David Mosberger 9 participants 24 messages Aug 4, 2003 – Sep 10, 2003

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

As of this morning, Linus's current bk tree
(http://linux.bkbits.net:8080/linux-2.5) builds and works out of the box for ia64!

Thanks to everybody who helped make this happen. In particular, thanks to Andrew Morton and Christoph Hellwig for their efforts, and Andy Grover for a last-minute ACPI patch!

For maximum performance/stability, I'd still recommend to use the ia64-specific patches, but for someone who needs to build bleeding edge kernels for multiple architectures, being able to use Linus' tree should make it a lot easier to include ia64 in their regular testing etc.

Now that Linus' tree works for ia64, the next question is how we can keep it that way. I think it would be useful to have someone setup a cron job which does daily builds/automated tests off of Linus tree. If something breaks, the person doing this could perhaps come up with a minimal patch which gets Linus' tree to build again (and submit a patch to the appropriate maintainer, with cc to the linux-ia64 list). I plan on continuing to put out roughly monthly ia64-specific patches and during those normal cycles, I'd then integrate the "quick fix up" patches as needed. Does this sound reasonable? Anybody want to volunteer for this "Linus watchdog" role?

	--david
#2 lore raw
In reply to David Mosberger
Quoted 7 lines · H. J. Lu
Quoted 4 lines · David Mosberger

As of this morning, Linus's current bk tree
(http://linux.bkbits.net:8080/linux-2.5) builds and works out of the box for ia64!

Does it work on bigsur? Does it support kernel modules?

I just tried the latest on my big sur, and though I think modules work (at least they build for other machines), big sur is broken because non-ACPI based PCI enumeration has been removed from the tree.

Jesse

#3 lore raw
In reply to David Mosberger
Quoted 7 lines · Jesse Barnes
Quoted 6 lines · H. J. Lu
Quoted 4 lines · David Mosberger

As of this morning, Linus's current bk tree
(http://linux.bkbits.net:8080/linux-2.5) builds and works out of the box for ia64!

Quoted 6 lines
Quoted 1 line

Does it work on bigsur? Does it support kernel modules?

I just tried the latest on my big sur, and though I think modules work (at least they build for other machines), big sur is broken because non-ACPI based PCI enumeration has been removed from the tree.

Can you try this patch for bigsur?

H.J.
---

drivers/acpi/osl.c renamed from drivers/acpi/osl.c.acpi +6 -1
@@ -250,7 +250,12 @@ acpi_os_install_interrupt_handler(u32 ir
 	irq = acpi_fadt.sci_int;
 
 #ifdef CONFIG_IA64
-	irq = gsi_to_vector(irq);
+	irq = acpi_irq_to_vector (irq);
+	if (irq < 0) {
+		printk(KERN_ERR PREFIX "SCI (IRQ%d/%d) not registerd\n",
+		       irq, acpi_fadt.sci_int);
+		return AE_OK;
+	}
 #endif
 	acpi_irq_irq = irq;
 	acpi_irq_handler = handler;
#4 lore raw
In reply to David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · H. J. Lu

HJ> Does it work on bigsur?

It should, apart from a qla1280.c glitch (see the latest ia64 diff for the one-liner to get it to work; Jes Sorensen said he's going to cleanup qla1280 for real).

HJ> Does it support kernel modules?

Sure. Kernel modules have been working for a while. You do need module-init-tools, of course (Debian has the necessary package in the "unstable" tree).

	--david
#5 lore raw
In reply to David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · On Mon, 4 Aug 2003 11:00:16 -0700, [email protected] (Jesse Barnes) said:

Jesse> big sur is broken because non-ACPI based PCI enumeration has Jesse> been removed from the tree.

That shouldn't break big sur. Perhaps you need newer firmware, though?

	--david
#6 lore raw
In reply to David Mosberger
Quoted 3 lines · On Mon, 4 Aug 2003 10:37:39 -0700 David Mosberger <[email protected]>…

As of this morning, Linus's current bk tree
(http://linux.bkbits.net:8080/linux-2.5) builds and works out of the box for ia64!

I hate to rain on your parade, but for a platform that has so many companies paying people full time to maintain under Linux having it take this long is a big diasppointment for me :(

I do sparc64 as a hobby in my spare time, as a single individual, and it has been working daily throughout the entire 2.4.x and 2.5.x series.

The fact that there are still "external patches for performance" is even more disheartening.

My tree never lags Linus's by more than 24 or 48 hours, unless one of us is on vacation and not near computers. And if I can do this in my spare time, it should be doable by people getting paid full time to maintain the ia64 port, don't you think? 8-)

#7 lore raw
In reply to David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · David S. Miller

Dave> My tree never lags Linus's by more than 24 or 48 hours, unless Dave> one of us is on vacation and not near computers. And if I can Dave> do this in my spare time, it should be doable by people Dave> getting paid full time to maintain the ia64 port, don't you Dave> think? 8-)

Well, you're obviously just way better than everybody else!

For what it's worth, I spend on average about 1 day a week on ia64 linux maintenance and that's just about as much as I'd want to (though in recent months it has been a bit more, partly so that we can get into sync with Linus' tree).

	--david
#8 lore raw
In reply to David Mosberger
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

"David" = David Mosberger <[email protected]> writes:

On Mon, 4 Aug 2003 10:53:08 -0700, "H. J. Lu" <[email protected]> said:

HJ> Does it work on bigsur?

David> It should, apart from a qla1280.c glitch (see the latest ia64 David> diff for the one-liner to get it to work; Jes Sorensen said David> he's going to cleanup qla1280 for real).

Yes, I already have in fact. I submitted the updated driver to James Bottomley for inclusion in the tree. I'll post it to you shortly.

Cheers,
Jes

#9 lore raw
In reply to David Mosberger
Quoted 4 lines · On Mon, 4 Aug 2003 11:40:25 -0700 David Mosberger <[email protected]>…

For what it's worth, I spend on average about 1 day a week on ia64 linux maintenance and that's just about as much as I'd want to (though in recent months it has been a bit more, partly so that we can get into sync with Linus' tree).

I'm not addressing you specifically, but I will note how great you tend to say ia64 is when platform performance comparison discussions happen on the lists :-)

In general, I think some platform maintainence drifts way too easily into a tail-spin of local changes that take forever to get merged.

My point was that if some random single monkey like me can keep a loser platform like sparc64 still building in Linus's tree, a group of several trained professionals should be able to fare much better :)

#10 lore raw
In reply to David Mosberger
Quoted 7 lines · David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · On Mon, 4 Aug 2003 11:00:16 -0700, [email protected] (Jesse Barnes) said:

Jesse> big sur is broken because non-ACPI based PCI enumeration has Jesse> been removed from the tree.

That shouldn't break big sur. Perhaps you need newer firmware, though?

You're right, I thought that big sur didn't use ACPI to enumerate PCI, but I guess it does.

Jesse

#11 lore raw
In reply to David Mosberger

Thanks H.J., this patch works for me! I had to add one more for xfs (included), but after that, things seemed to work.

Thanks,
Jesse

Quoted 8 lines · H. J. Lu
Quoted 7 lines · Jesse Barnes
Quoted 6 lines · H. J. Lu
Quoted 4 lines · David Mosberger

As of this morning, Linus's current bk tree
(http://linux.bkbits.net:8080/linux-2.5) builds and works out of the box for ia64!

Quoted 8 lines
Quoted 6 lines
Quoted 1 line

Does it work on bigsur? Does it support kernel modules?

I just tried the latest on my big sur, and though I think modules work (at least they build for other machines), big sur is broken because non-ACPI based PCI enumeration has been removed from the tree.

Can you try this patch for bigsur?

fs/xfs/linux/xfs_vfs.h +2 -2
diff -Nru a/fs/xfs/linux/xfs_vfs.h b/fs/xfs/linux/xfs_vfs.h
@@ -44,8 +44,8 @@
 
 typedef struct vfs {
 	u_int			vfs_flag;	/* flags */
-	fsid_t			vfs_fsid;	/* file system ID */
-	fsid_t			*vfs_altfsid;	/* An ID fixed for life of FS */
+	__kernel_fsid_t			vfs_fsid;	/* file system ID */
+	__kernel_fsid_t			*vfs_altfsid;	/* An ID fixed for life of FS */
 	bhv_head_t		vfs_bh;		/* head of vfs behavior chain */
 	struct super_block	*vfs_super;	/* Linux superblock structure */
 	struct task_struct	*vfs_sync_task;
fs/xfs/xfs_mount.c +1 -1
diff -Nru a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
@@ -889,7 +889,7 @@
 	 *  File systems that don't support user level file handles (i.e.
 	 *  all of them except for XFS) will leave vfs_altfsid as NULL.
 	 */
-	vfsp->vfs_altfsid = (fsid_t *)mp->m_fixedfsid;
+	vfsp->vfs_altfsid = (__kernel_fsid_t *)mp->m_fixedfsid;
 	mp->m_dmevmask = 0;	/* not persistent; set after each mount */
 
 	/*
#12 lore raw
In reply to David Mosberger

Can someone look into/explain why this patch is needed for Big Sur? Either the old code is wrong for other machines, too, or the old code is broken and we just don't exercise it on the other ia64 machines?

	--david
Quoted 5 lines
Quoted 5 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line · H. J. Lu

I just tried the latest on my big sur, and though I think modules work (at least they build for other machines), big sur is broken because non-ACPI based PCI enumeration has been removed from the tree.

HJ> Can you try this patch for bigsur?

  HJ> H.J.
  HJ> ---
  HJ> --- linux/drivers/acpi/osl.c.acpi	Mon Jul 28 11:41:53 2003
  HJ> +++ linux/drivers/acpi/osl.c	Mon Jul 28 15:12:44 2003
  HJ> @@ -250,7 +250,12 @@ acpi_os_install_interrupt_handler(u32 ir
  HJ> irq = acpi_fadt.sci_int;
 
  HJ> #ifdef CONFIG_IA64
  HJ> -	irq = gsi_to_vector(irq);
  HJ> +	irq = acpi_irq_to_vector (irq);
  HJ> +	if (irq < 0) {
  HJ> +		printk(KERN_ERR PREFIX "SCI (IRQ%d/%d) not registerd\n",
  HJ> +		       irq, acpi_fadt.sci_int);
  HJ> +		return AE_OK;
  HJ> +	}
  HJ> #endif
  HJ> acpi_irq_irq = irq;
  HJ> acpi_irq_handler = handler;
#13 lore raw
In reply to David Mosberger
Quoted 3 lines · David S. Miller

My point was that if some random single monkey like me can keep a loser platform like sparc64 still building in Linus's tree, a group of several trained professionals should be able to fare much better :)

Right but ia64 - for better or worse started off as a hideously divergent tree that did really stupid things like redefine GFP_DMA. It may be that ia64 should have aimed in a different direction from day one but having finally ended up in the right place *is* a milestone.

#14 lore raw
In reply to David Mosberger
Quoted 1 line · On 04 Aug 2003 20:07:58 +0100 Alan Cox <[email protected]> wrote:

but having finally ended up in the right place *is* a milestone.

I agree.

I apologize, I really didn't want to diminsh what has been accomplished. I was trying to achieve something different :)

#15 lore raw
In reply to David Mosberger

[email protected] (Jesse Barnes) writes:

|> diff -Nru a/fs/xfs/linux/xfs_vfs.h b/fs/xfs/linux/xfs_vfs.h
|> --- a/fs/xfs/linux/xfs_vfs.h	Wed Jul 16 12:10:39 2003
|> +++ b/fs/xfs/linux/xfs_vfs.h	Wed Jul 16 12:10:39 2003
|> @@ -44,8 +44,8 @@
|>  
|>  typedef struct vfs {
|>  	u_int			vfs_flag;	/* flags */
|> -	fsid_t			vfs_fsid;	/* file system ID */
|> -	fsid_t			*vfs_altfsid;	/* An ID fixed for life of FS */
|> +	__kernel_fsid_t			vfs_fsid;	/* file system ID */
|> +	__kernel_fsid_t			*vfs_altfsid;	/* An ID fixed for life of FS */
|>  	bhv_head_t		vfs_bh;		/* head of vfs behavior chain */
|>  	struct super_block	*vfs_super;	/* Linux superblock structure */
|>  	struct task_struct	*vfs_sync_task;

Alternatively you could use this patch (following the other architectures):

include/asm-ia64/statfs.h renamed from include/asm-ia64/statfs.h.~1~ +5 -0
@@ -6,6 +6,11 @@
  * Copyright (C) 1998, 1999 David Mosberger-Tang <[email protected]>
  */
 
+#ifndef __KERNEL_STRICT_NAMES
+# include <linux/types.h>
+typedef __kernel_fsid_t	fsid_t;
+#endif
+
 /*
  * This is ugly --- we're already 64-bit, so just duplicate the definitions
  */

Andreas.

Signature · 4 lines
Andreas Schwab, SuSE Labs, [email protected]
SuSE Linux AG, Deutschherrnstr. 15-19, D-90429 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."
#16 lore raw
In reply to David Mosberger

This code is taken from 2.4 kernel, which is enabled for all ia64 machines. The only difference is

        if (has_8259 && irq < 16)
                return isa_irq_to_vector(irq);

	return gsi_to_vector(irq);

instead of just

	return gsi_to_vector(irq);

I think bigsur is one of those uses 8259 and ISA. I don't believe any new ia64 machines do that.

H.J.
---

Quoted 33 lines · David Mosberger

Can someone look into/explain why this patch is needed for Big Sur? Either the old code is wrong for other machines, too, or the old code is broken and we just don't exercise it on the other ia64 machines?

	--david
Quoted 5 lines
Quoted 5 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line · H. J. Lu

I just tried the latest on my big sur, and though I think modules work (at least they build for other machines), big sur is broken because non-ACPI based PCI enumeration has been removed from the tree.

HJ> Can you try this patch for bigsur?

  HJ> H.J.
  HJ> ---
  HJ> --- linux/drivers/acpi/osl.c.acpi	Mon Jul 28 11:41:53 2003
  HJ> +++ linux/drivers/acpi/osl.c	Mon Jul 28 15:12:44 2003
  HJ> @@ -250,7 +250,12 @@ acpi_os_install_interrupt_handler(u32 ir
  HJ> irq = acpi_fadt.sci_int;
 
  HJ> #ifdef CONFIG_IA64
  HJ> -	irq = gsi_to_vector(irq);
  HJ> +	irq = acpi_irq_to_vector (irq);
  HJ> +	if (irq < 0) {
  HJ> +		printk(KERN_ERR PREFIX "SCI (IRQ%d/%d) not registerd\n",
  HJ> +		       irq, acpi_fadt.sci_int);
  HJ> +		return AE_OK;
  HJ> +	}
  HJ> #endif
  HJ> acpi_irq_irq = irq;
  HJ> acpi_irq_handler = handler;
#17 lore raw
In reply to David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

"David" = David Mosberger <[email protected]> writes:

David> Now that Linus' tree works for ia64, the next question is how David> we can keep it that way. I think it would be useful to have David> someone setup a cron job which does daily builds/automated David> tests off of Linus tree. If something breaks, the person doing David> this could perhaps come up with a minimal patch which gets David> Linus' tree to build again (and submit a patch to the David> appropriate maintainer, with cc to the linux-ia64 list). I David> plan on continuing to put out roughly monthly ia64-specific David> patches and during those normal cycles, I'd then integrate the David> "quick fix up" patches as needed. Does this sound reasonable? David> Anybody want to volunteer for this "Linus watchdog" role?

We can do this. We're tracking Linus's tree anyway for the work we're doing.

We'd probably do daily automated builds to check that the kernel
still compiles cleanly for HPSIM, DIG, and ZX1, but test only weekly.

If you have anyu specific configuration options you think should be included, let us know.

--
Dr Peter Chubb http://www.gelato.unsw.edu.au peterc AT gelato.unsw.edu.au
You are lost in a maze of BitKeeper repositories, all slightly different.

#18 lore raw
In reply to David Mosberger

David, can you apply this one too? It's better than having it changed in XFS. Thanks Andreas.

Thanks,
Jesse

Quoted 17 lines · Andreas Schwab

Alternatively you could use this patch (following the other architectures):

include/asm-ia64/statfs.h renamed from include/asm-ia64/statfs.h.~1~ +5 -0
@@ -6,6 +6,11 @@
  * Copyright (C) 1998, 1999 David Mosberger-Tang <[email protected]>
  */
 
+#ifndef __KERNEL_STRICT_NAMES
+# include <linux/types.h>
+typedef __kernel_fsid_t	fsid_t;
+#endif
+
 /*
  * This is ugly --- we're already 64-bit, so just duplicate the definitions
  */
#19 lore raw
In reply to David Mosberger
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines · Peter Chubb

"David" = David Mosberger <[email protected]> writes:

David> Now that Linus' tree works for ia64, the next question is how David> we can keep it that way. I think it would be useful to have David> someone setup a cron job which does daily builds/automated David> tests off of Linus tree. If something breaks, the person doing David> this could perhaps come up with a minimal patch which gets David> Linus' tree to build again (and submit a patch to the David> appropriate maintainer, with cc to the linux-ia64 list). I David> plan on continuing to put out roughly monthly ia64-specific David> patches and during those normal cycles, I'd then integrate the David> "quick fix up" patches as needed. Does this sound reasonable? David> Anybody want to volunteer for this "Linus watchdog" role?

Peter> We can do this. We're tracking Linus's tree anyway for the work we're Peter> doing.

Excellent!

Peter> We'd probably do daily automated builds to check that the kernel Peter> still compiles cleanly for HPSIM, DIG, and ZX1, but test only weekly.

Sounds reasonable. Except doing a boot/halt cycle on the simulator should be easy to automate, no? The simulator can actually catch a surprising number of problems.

Peter> If you have anyu specific configuration options you think should be Peter> included, let us know.

Nothing in particular, though it would be good to cover UP, MP, 16, and 64KB page sizes (not in all permutations, of course).

Thanks,

	--david
#20 lore raw
In reply to David Mosberger

Sure.

	--david
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · On Mon, 4 Aug 2003 16:20:37 -0700, [email protected] (Jesse Barnes) said:

Jesse> David, can you apply this one too? It's better than having it changed Jesse> in XFS. Thanks Andreas.

Jesse> Thanks,
Jesse> Jesse

Quoted 17 lines · Andreas Schwab
Quoted 17 lines

Alternatively you could use this patch (following the other architectures):

include/asm-ia64/statfs.h renamed from include/asm-ia64/statfs.h.~1~ +5 -0
@@ -6,6 +6,11 @@
 * Copyright (C) 1998, 1999 David Mosberger-Tang <[email protected]>
 */
 
+#ifndef __KERNEL_STRICT_NAMES
+# include <linux/types.h>
+typedef __kernel_fsid_t	fsid_t;
+#endif
+
 /*
 * This is ugly --- we're already 64-bit, so just duplicate the definitions
 */
#21 lore raw
In reply to David Mosberger
Quoted 7 lines · David Mosberger

David> Now that Linus' tree works for ia64, the next question is how David> we can keep it that way. I think it would be useful to have David> someone setup a cron job which does daily builds/automated David> tests off of Linus tree.

Peter> We'd probably do daily automated builds to check that the kernel Peter> still compiles cleanly for HPSIM, DIG, and ZX1, but test only weekly.

It has been running OK for a few days now, so please feel free to check out

http://www.gelato.unsw.edu.au/kerncomp

for the status of daily IA64 builds. We will attempt to keep track of what is happening and fix anything that needs fixing or point to where it has been fixed.

We're still working on automated testing (i.e. booting on the simulator, maybe real hardware).

-i
[email protected]
http://www.gelato.unsw.edu.au

#23 lore raw
In reply to David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Ian Wienand
Quoted 2 lines · Ian Wienand
Quoted 2 lines

We're still working on automated testing (i.e. booting on the simulator, maybe real hardware).

Ian> For those who are interested, we now attempt to boot the autobuilt Ian> kernels on the simulator daily. See

Ian> http://www.gelato.unsw.edu.au/kerncomp/simboot.php

Very nice! (The test setup & web-page, not the fact that the boot failed... ;-).

As for perfmon under Ski: the latest perfmon should have the necessary code, but I don't remember whether it made it into test5 already (and I don't have CONFIG_PERFMON enabled in my Ski setup). If it's not there already, it should be there tomorrow, as Linus pulled on the ia64 repository a couple of hours ago.

	--david