[PATCH] set altix preferred console

[PATCH] set altix preferred console

[Greg Edwards]

Set the preferred console on Altix, if no console=<console> was passed
on the boot line.

Bjorn, the hunk I removed in pcdp.c broke the vga console on Altix and
the virtual terminal if any console= was passed in.  For example, if
you booted with console=ttySG0 (the L1 serial console), there would be
no login prompt on tty1 when you got to multiuser, even though a getty
was running.  Is removing this ok?


Signed-off-by: Greg Edwards <edwardsg@sgi.com>


 arch/ia64/sn/kernel/setup.c |   12 +++++++++---
 drivers/firmware/pcdp.c     |    3 ---
 2 files changed, 9 insertions(+), 6 deletions(-)

Index: git-linus/arch/ia64/sn/kernel/setup.c
=================================--- git-linus.orig/arch/ia64/sn/kernel/setup.c	2005-09-12 11:34:36.342472611 -0500
+++ git-linus/arch/ia64/sn/kernel/setup.c	2005-09-30 14:49:50.941056190 -0500
@@ -281,15 +281,21 @@ void __init sn_setup(char **cmdline_p)
 #if defined(CONFIG_VT) && defined(CONFIG_VGA_CONSOLE)
 	/*
 	 * If there was a primary vga adapter identified through the
-	 * EFI PCDP table, make it the preferred console.  Otherwise
+	 * EFI PCDP table, and no console was specified on the boot
+	 * line, make vga the preferred console.
+	 *
+	 * If no vga adapter was found, and no console was specified on
+	 * the boot line, make the L1 console the preferred console and
 	 * zero out conswitchp.
 	 */
 
 	if (vga_console_membase) {
-		/* usable vga ... make tty0 the preferred default console */
-		add_preferred_console("tty", 0, NULL);
+		if (!strstr(*cmdline_p, "console="))
+			add_preferred_console("tty", 0, NULL);
 	} else {
 		printk(KERN_DEBUG "SGI: Disabling VGA console\n");
+		if (!strstr(*cmdline_p, "console="))
+			add_preferred_console("ttySG", 0, NULL);
 #ifdef CONFIG_DUMMY_CONSOLE
 		conswitchp = &dummy_con;
 #else
Index: git-linus/drivers/firmware/pcdp.c
=================================--- git-linus.orig/drivers/firmware/pcdp.c	2005-08-24 16:06:59.911622221 -0500
+++ git-linus/drivers/firmware/pcdp.c	2005-09-30 14:49:50.998667555 -0500
@@ -99,9 +99,6 @@ efi_setup_pcdp_console(char *cmdline)
 	if (strstr(cmdline, "console=hcdp")) {
 		if (pcdp->rev < 3)
 			serial = 1;
-	} else if (strstr(cmdline, "console=")) {
-		printk(KERN_INFO "Explicit \"console=\"; ignoring PCDP\n");
-		return -ENODEV;
 	}
 
 	if (pcdp->rev < 3 && efi_uart_console_only())

Re: [PATCH] set altix preferred console

[Bjorn Helgaas]

On Friday 30 September 2005 2:04 pm, Greg Edwards wrote:
> Set the preferred console on Altix, if no console=<console> was passed
> on the boot line.
> 
> Bjorn, the hunk I removed in pcdp.c broke the vga console on Altix and
> the virtual terminal if any console= was passed in.  For example, if
> you booted with console=ttySG0 (the L1 serial console), there would be
> no login prompt on tty1 when you got to multiuser, even though a getty
> was running.  Is removing this ok?

I don't think removing the pcdp.c hunk is the right fix.

The PCDP not a device discovery interface.  The OS must be able to
discover console devices some other way, and the PCDP is only a
mechanism for firmware to tell the OS which of those devices to
use.  That's why we ignore the PCDP if the user explicitly uses
"console=" -- he's saying "ignore what the firmware told you,
and use *this* device as the console."

So the intent is that pcdp.c should be completely optional, i.e.,
it basically gives you a pointer to a device that is completely
described elsewhere, either in ACPI or by standard PCI enumeration.
If you turn off "CONFIG_EFI_PCDP", everything should work (except
that now you have to use "console=" if you want something other
than the default console).

But I think that if you turn off CONFIG_EFI_PCDP, your VGA
device won't work at all, because you're relying on the PCDP
to set vga_console_{iobase,membase}, and there's currently no
other way those get set.

I think what *should* happen is that we should discover VGA
devices by PCI enumeration, and the iobase/membase information
would come from the ACPI description of the PCI root bridges.
Then we could support multiple VGA devices, and the PCDP could
point to one of them to be used as the console.

The VGA code in the kernel is a long ways from that.  But your
firmware should provide the vga_console_{iobase,membase} data
somewhere besides the PCDP.  Can you do something in ia64/sn
to dig it out?

Re: [PATCH] set altix preferred console

[Greg Edwards]

On Fri, Sep 30, 2005 at 03:20:11PM -0600, Bjorn Helgaas wrote:
| I think what *should* happen is that we should discover VGA
| devices by PCI enumeration, and the iobase/membase information
| would come from the ACPI description of the PCI root bridges.
| Then we could support multiple VGA devices, and the PCDP could
| point to one of them to be used as the console.
| 
| The VGA code in the kernel is a long ways from that.  But your
| firmware should provide the vga_console_{iobase,membase} data
| somewhere besides the PCDP.  Can you do something in ia64/sn
| to dig it out?

Thanks for the info and suggestion, Bjorn.  I think we should be able to
come up with something in sn-specific code.  I'll talk to our I/O guys
and see what we can come up with.

Greg

Re: [PATCH] set altix preferred console

[Mark Maule]

On Mon, Oct 03, 2005 at 09:44:08AM -0500, Greg Edwards wrote:
> On Fri, Sep 30, 2005 at 03:20:11PM -0600, Bjorn Helgaas wrote:
> | I think what *should* happen is that we should discover VGA
> | devices by PCI enumeration, and the iobase/membase information
> | would come from the ACPI description of the PCI root bridges.
> | Then we could support multiple VGA devices, and the PCDP could
> | point to one of them to be used as the console.
> | 
> | The VGA code in the kernel is a long ways from that.  But your
> | firmware should provide the vga_console_{iobase,membase} data
> | somewhere besides the PCDP.  Can you do something in ia64/sn
> | to dig it out?
> 
> Thanks for the info and suggestion, Bjorn.  I think we should be able to
> come up with something in sn-specific code.  I'll talk to our I/O guys
> and see what we can come up with.
> 
> Greg

Our prom currently does not build up ACPI tables describing PCI (or any IO)
devices.  We are working toward that goal, but it is still
work in progress and will be some time yet before this is done.

It is because of this shortcoming that we have to rely on the PCDP table
to provide the io addresses needed for VGA console.

Outside of PCDP, our earliest opportunity to get information about VGA
devices is through the call sequence initiated with
subsys_initcall(sn_pci_init), which would seem way too late for early boot
messages.

Out of curiousity, what are the mmio_tra/ioport_tra fields in the PCDP PCI
interface structure intended for if not for this purpose?

Mark

Re: [PATCH] set altix preferred console

[Bjorn Helgaas]

> Our prom currently does not build up ACPI tables describing PCI (or any
> IO)
> devices.  We are working toward that goal, but it is still
> work in progress and will be some time yet before this is done.
>
> It is because of this shortcoming that we have to rely on the PCDP table
> to provide the io addresses needed for VGA console.
>
> Outside of PCDP, our earliest opportunity to get information about VGA
> devices is through the call sequence initiated with
> subsys_initcall(sn_pci_init), which would seem way too late for early boot
> messages.

It's the same for us.  We don't find out about our built-in serial
ports until the 8250 driver initializes, which is pretty late.  So
we have "8250_early" that uses information from the PCDP to talk
to the serial port before the normal 8250 driver discovers it.
The important thing is that the PCDP doesn't influence the naming
of the device (and there isn't really a "name" for VGA devices,
AFAIK).

> Out of curiousity, what are the mmio_tra/ioport_tra fields in the PCDP PCI
> interface structure intended for if not for this purpose?

You *can* use the _tra fields in the PCDP -- that's what they're
there for.  But the PCDP should be *optional*, and I think it's
not for your current setup.

If we don't build pcdp.o on an HP box, the console still works
just fine; it just doesn't work until the 8250 driver discovers
the port, which is fairly late.

So I think all you need to do is add some SN-specific code
somewhere that discovers the VGA device.  It sounds like the
only way to do that right now is to get the information from
the PCDP, so I guess that's what you need to do.  But having
this duplicated in the SN code means that you won't need the
change to pcdp.c.

Re: [PATCH] set altix preferred console

[Mark Maule]

On Tue, Oct 04, 2005 at 01:07:14PM -0600, Bjorn Helgaas wrote:
> > Our prom currently does not build up ACPI tables describing PCI (or any
> > IO)
> > devices.  We are working toward that goal, but it is still
> > work in progress and will be some time yet before this is done.
> >
> > It is because of this shortcoming that we have to rely on the PCDP table
> > to provide the io addresses needed for VGA console.
> >
> > Outside of PCDP, our earliest opportunity to get information about VGA
> > devices is through the call sequence initiated with
> > subsys_initcall(sn_pci_init), which would seem way too late for early boot
> > messages.
> 
> It's the same for us.  We don't find out about our built-in serial
> ports until the 8250 driver initializes, which is pretty late.  So
> we have "8250_early" that uses information from the PCDP to talk
> to the serial port before the normal 8250 driver discovers it.
> The important thing is that the PCDP doesn't influence the naming
> of the device (and there isn't really a "name" for VGA devices,
> AFAIK).

Ok.  I was assuming that tty0 implied a vga device, but I guess tty0 is
really the VT driver which drives whatever conswitchp is pointing to ...

> 
> > Out of curiousity, what are the mmio_tra/ioport_tra fields in the PCDP PCI
> > interface structure intended for if not for this purpose?
> 
> You *can* use the _tra fields in the PCDP -- that's what they're
> there for.  But the PCDP should be *optional*, and I think it's
> not for your current setup.
> 
> If we don't build pcdp.o on an HP box, the console still works
> just fine; it just doesn't work until the 8250 driver discovers
> the port, which is fairly late.
> 
> So I think all you need to do is add some SN-specific code
> somewhere that discovers the VGA device.  It sounds like the
> only way to do that right now is to get the information from
> the PCDP, so I guess that's what you need to do.  But having
> this duplicated in the SN code means that you won't need the
> change to pcdp.c.
> 

Ok.  One other hitch is that the sn prom doesn't guarantee that all
VGA cards will have their firmware executed.  The only vga cards guaranteed
to be init'd are the ones present in the pcdp table which could be a
subset of the vga cards in a system.  Were it not for this, we could
just pick one (say lowest segment/bus number or some such thing) during
sn pci discovery.  But we'll have to scan the pcdp in order to see what
cards are available (unless there's some other standard way to tell if a
vga card is initialized).

Mark

Re: [PATCH] set altix preferred console

[Mark Maule]

On Tue, Oct 04, 2005 at 01:07:14PM -0600, Bjorn Helgaas wrote:
> > Our prom currently does not build up ACPI tables describing PCI (or any
> > IO)
> > devices.  We are working toward that goal, but it is still
> > work in progress and will be some time yet before this is done.
> >
> > It is because of this shortcoming that we have to rely on the PCDP table
> > to provide the io addresses needed for VGA console.
> >
> > Outside of PCDP, our earliest opportunity to get information about VGA
> > devices is through the call sequence initiated with
> > subsys_initcall(sn_pci_init), which would seem way too late for early boot
> > messages.
> 
> It's the same for us.  We don't find out about our built-in serial
> ports until the 8250 driver initializes, which is pretty late.  So
> we have "8250_early" that uses information from the PCDP to talk
> to the serial port before the normal 8250 driver discovers it.
> The important thing is that the PCDP doesn't influence the naming
> of the device (and there isn't really a "name" for VGA devices,
> AFAIK).
> 
> > Out of curiousity, what are the mmio_tra/ioport_tra fields in the PCDP PCI
> > interface structure intended for if not for this purpose?
> 
> You *can* use the _tra fields in the PCDP -- that's what they're
> there for.  But the PCDP should be *optional*, and I think it's
> not for your current setup.
> 
> If we don't build pcdp.o on an HP box, the console still works
> just fine; it just doesn't work until the 8250 driver discovers
> the port, which is fairly late.
> 
> So I think all you need to do is add some SN-specific code
> somewhere that discovers the VGA device.  It sounds like the
> only way to do that right now is to get the information from
> the PCDP, so I guess that's what you need to do.  But having
> this duplicated in the SN code means that you won't need the
> change to pcdp.c.
> 

Finally getting back to this ...

So we will dup some of the PCDP scan code in sn/kernel/setup.c for the time
being.  Does anyone have a problem with moving drivers/firmware/pcdp.h under
include/asm-ia64 so that I don't also have to dup the structure definitions?

thanks
Mark

Re: [PATCH] set altix preferred console

[Bjorn Helgaas]

On Tuesday 11 October 2005 2:00 pm, Mark Maule wrote:
> So we will dup some of the PCDP scan code in sn/kernel/setup.c for the time
> being.  Does anyone have a problem with moving drivers/firmware/pcdp.h under
> include/asm-ia64 so that I don't also have to dup the structure definitions?

I don't want to move pcdp.h under asm-ia64.  EFI (and, I hope,
PCDP) is coming to x86 soon.  There's nothing ia64-specific
in pcdp.h or pcdp.c, so I think it would be a mistake to
move it to an arch-specific directory.

Re: [PATCH] set altix preferred console

[Mark Maule]

On Tue, Oct 11, 2005 at 03:19:39PM -0600, Bjorn Helgaas wrote:
> On Tuesday 11 October 2005 2:00 pm, Mark Maule wrote:
> > So we will dup some of the PCDP scan code in sn/kernel/setup.c for the time
> > being.  Does anyone have a problem with moving drivers/firmware/pcdp.h under
> > include/asm-ia64 so that I don't also have to dup the structure definitions?
> 
> I don't want to move pcdp.h under asm-ia64.  EFI (and, I hope,
> PCDP) is coming to x86 soon.  There's nothing ia64-specific
> in pcdp.h or pcdp.c, so I think it would be a mistake to
> move it to an arch-specific directory.

Ok, thanks Bjorn.  I was under the impression that stuff under dig64.org
was ia64 specific ...

Would moving it to include/linux be an option?  If not I'll just dup the
structs that I need for this.

Mark

Re: [PATCH] set altix preferred console

[Bjorn Helgaas]

On Tuesday 11 October 2005 3:31 pm, Mark Maule wrote:
> On Tue, Oct 11, 2005 at 03:19:39PM -0600, Bjorn Helgaas wrote:
> > On Tuesday 11 October 2005 2:00 pm, Mark Maule wrote:
> > > So we will dup some of the PCDP scan code in sn/kernel/setup.c for the time
> > > being.  Does anyone have a problem with moving drivers/firmware/pcdp.h under
> > > include/asm-ia64 so that I don't also have to dup the structure definitions?
> > 
> > I don't want to move pcdp.h under asm-ia64.  EFI (and, I hope,
> > PCDP) is coming to x86 soon.  There's nothing ia64-specific
> > in pcdp.h or pcdp.c, so I think it would be a mistake to
> > move it to an arch-specific directory.
> 
> Ok, thanks Bjorn.  I was under the impression that stuff under dig64.org
> was ia64 specific ...
> 
> Would moving it to include/linux be an option?

I sort of hate to put it in include/linux, too, but I guess
it's already a dumping ground for all sorts of miscellaneous
stuff, so one more wouldn't make it much worse.

And I can't think of any better options.

Re: [PATCH] set altix preferred console

[Mark Maule]

New version of Greg's original patch per review comments from Bjorn:


Fix default VGA console on SN platforms.  Since SN firmware does not pass
enough ACPI information to identify VGA cards and the associated legacy IO/MEM
addresses, we rely on the EFI PCDP table.  Since the linux pcdp driver is
optional (and overridden if console= directives are used) SN duplicates a
portion of the pcdp scan code to identify if there is a usable console VGA
adapter.

To avoid also dup'ing PCDP structures, drivers/firmware/pcdp.h is moved to
include/linux/pcdp.h.

Signed-off-by: Mark Maule <maule@sgi.com>

Index: vga/arch/ia64/sn/kernel/setup.c
=================================--- vga.orig/arch/ia64/sn/kernel/setup.c	2005-10-12 10:36:40.914797412 -0500
+++ vga/arch/ia64/sn/kernel/setup.c	2005-10-12 11:50:44.543433497 -0500
@@ -30,6 +30,8 @@
 #include <linux/root_dev.h>
 #include <linux/nodemask.h>
 #include <linux/pm.h>
+#include <linux/efi.h>
+#include <linux/pcdp.h>
 
 #include <asm/io.h>
 #include <asm/sal.h>
@@ -242,6 +244,48 @@
 	}
 }
 
+static void
+sn_scan_pcdp(void)
+{
+	u8 *bp;
+	struct pcdp *pcdp;
+	struct pcdp_device device;
+	struct pcdp_if_pci if_pci;
+	extern struct efi efi;
+
+	pcdp = efi.hcdp;
+	if (! pcdp)
+		return;		/* no hcdp/pcdp table */
+
+	if (pcdp->rev < 3)
+		return;		/* only support PCDP (rev >= 3) */
+
+	for (bp = (u8 *)&pcdp->uart[pcdp->num_uarts];
+	     bp < (u8 *)pcdp + pcdp->length;
+	     bp += device.length) {
+		memcpy(&device, bp, sizeof(device));
+		if (! (device.flags & PCDP_PRIMARY_CONSOLE))
+			continue;	/* not primary console */
+
+		if (device.type != PCDP_CONSOLE_VGA)
+			continue;	/* not VGA descriptor */
+
+		memcpy(&if_pci, bp+sizeof(device), sizeof(if_pci));
+		if (if_pci.interconnect != PCDP_IF_PCI)
+			continue;	/* not PCI interconnect */
+
+		if (if_pci.trans & PCDP_PCI_TRANS_IOPORT)
+			vga_console_iobase +				if_pci.ioport_tra | __IA64_UNCACHED_OFFSET;
+
+		if (if_pci.trans & PCDP_PCI_TRANS_MMIO)
+			vga_console_membase +				if_pci.mmio_tra | __IA64_UNCACHED_OFFSET;
+
+		break; /* once we find the primary, we're done */
+	}
+}
+
 /**
  * sn_setup - SN platform setup routine
  * @cmdline_p: kernel command line
@@ -263,16 +307,35 @@
 
 #if defined(CONFIG_VT) && defined(CONFIG_VGA_CONSOLE)
 	/*
-	 * If there was a primary vga adapter identified through the
-	 * EFI PCDP table, make it the preferred console.  Otherwise
-	 * zero out conswitchp.
+	 * Handle SN vga console.
+	 *
+	 * SN systems do not have enough ACPI table information
+	 * being passed from prom to identify VGA adapters and the legacy
+	 * addresses to access them.  Until that is done, SN systems rely
+	 * on the PCDP table to identify the primary VGA console if one
+	 * exists.
+	 *
+	 * However, kernel PCDP support is optional, and even if it is built
+	 * into the kernel, it will not be used if the boot cmdline contains
+	 * console= directives.
+	 *
+	 * So, to work around this mess, we duplicate some of the PCDP code
+	 * here so that the primary VGA console (as defined by PCDP) will
+	 * work on SN systems even if a different console (e.g. serial) is
+	 * selected on the boot line (or CONFIG_EFI_PCDP is off).
 	 */
 
+	if (! vga_console_membase)
+		sn_scan_pcdp();
+
 	if (vga_console_membase) {
 		/* usable vga ... make tty0 the preferred default console */
-		add_preferred_console("tty", 0, NULL);
+		if (!strstr(*cmdline_p, "console="))
+			add_preferred_console("tty", 0, NULL);
 	} else {
 		printk(KERN_DEBUG "SGI: Disabling VGA console\n");
+		if (!strstr(*cmdline_p, "console="))
+			add_preferred_console("ttySG", 0, NULL);
 #ifdef CONFIG_DUMMY_CONSOLE
 		conswitchp = &dummy_con;
 #else
Index: vga/drivers/firmware/pcdp.c
=================================--- vga.orig/drivers/firmware/pcdp.c	2005-10-12 10:36:40.915773877 -0500
+++ vga/drivers/firmware/pcdp.c	2005-10-12 10:43:38.156192381 -0500
@@ -16,8 +16,8 @@
 #include <linux/console.h>
 #include <linux/efi.h>
 #include <linux/serial.h>
+#include <linux/pcdp.h>
 #include <asm/vga.h>
-#include "pcdp.h"
 
 static int __init
 setup_serial_console(struct pcdp_uart *uart)
Index: vga/drivers/firmware/pcdp.h
=================================--- vga.orig/drivers/firmware/pcdp.h	2005-10-12 10:36:40.914797412 -0500
+++ /dev/null	1970-01-01 00:00:00.000000000 +0000
@@ -1,111 +0,0 @@
-/*
- * Definitions for PCDP-defined console devices
- *
- * v1.0a: http://www.dig64.org/specifications/DIG64_HCDPv10a_01.pdf
- * v2.0:  http://www.dig64.org/specifications/DIG64_PCDPv20.pdf
- *
- * (c) Copyright 2002, 2004 Hewlett-Packard Development Company, L.P.
- *	Khalid Aziz <khalid.aziz@hp.com>
- *	Bjorn Helgaas <bjorn.helgaas@hp.com>
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License version 2 as
- * published by the Free Software Foundation.
- */
-
-#define PCDP_CONSOLE			0
-#define PCDP_DEBUG			1
-#define PCDP_CONSOLE_OUTPUT		2
-#define PCDP_CONSOLE_INPUT		3
-
-#define PCDP_UART			(0 << 3)
-#define PCDP_VGA			(1 << 3)
-#define PCDP_USB			(2 << 3)
-
-/* pcdp_uart.type and pcdp_device.type */
-#define PCDP_CONSOLE_UART		(PCDP_UART | PCDP_CONSOLE)
-#define PCDP_DEBUG_UART			(PCDP_UART | PCDP_DEBUG)
-#define PCDP_CONSOLE_VGA		(PCDP_VGA  | PCDP_CONSOLE_OUTPUT)
-#define PCDP_CONSOLE_USB		(PCDP_USB  | PCDP_CONSOLE_INPUT)
-
-/* pcdp_uart.flags */
-#define PCDP_UART_EDGE_SENSITIVE	(1 << 0)
-#define PCDP_UART_ACTIVE_LOW		(1 << 1)
-#define PCDP_UART_PRIMARY_CONSOLE	(1 << 2)
-#define PCDP_UART_IRQ			(1 << 6) /* in pci_func for rev < 3 */
-#define PCDP_UART_PCI			(1 << 7) /* in pci_func for rev < 3 */
-
-struct pcdp_uart {
-	u8				type;
-	u8				bits;
-	u8				parity;
-	u8				stop_bits;
-	u8				pci_seg;
-	u8				pci_bus;
-	u8				pci_dev;
-	u8				pci_func;
-	u64				baud;
-	struct acpi_generic_address	addr;
-	u16				pci_dev_id;
-	u16				pci_vendor_id;
-	u32				gsi;
-	u32				clock_rate;
-	u8				pci_prog_intfc;
-	u8				flags;
-	u16				conout_index;
-	u32				reserved;
-} __attribute__((packed));
-
-#define PCDP_IF_PCI	1
-
-/* pcdp_if_pci.trans */
-#define PCDP_PCI_TRANS_IOPORT	0x02
-#define PCDP_PCI_TRANS_MMIO	0x01
-
-struct pcdp_if_pci {
-	u8			interconnect;
-	u8			reserved;
-	u16			length;
-	u8			segment;
-	u8			bus;
-	u8			dev;
-	u8			fun;
-	u16			dev_id;
-	u16			vendor_id;
-	u32			acpi_interrupt;
-	u64			mmio_tra;
-	u64			ioport_tra;
-	u8			flags;
-	u8			trans;
-} __attribute__((packed));
-
-struct pcdp_vga {
-	u8			count;		/* address space descriptors */
-} __attribute__((packed));
-
-/* pcdp_device.flags */
-#define PCDP_PRIMARY_CONSOLE	1
-
-struct pcdp_device {
-	u8			type;
-	u8			flags;
-	u16			length;
-	u16			efi_index;
-	/* next data is pcdp_if_pci or pcdp_if_acpi (not yet supported) */
-	/* next data is device specific type (currently only pcdp_vga) */
-} __attribute__((packed));
-
-struct pcdp {
-	u8			signature[4];
-	u32			length;
-	u8			rev;		/* PCDP v2.0 is rev 3 */
-	u8			chksum;
-	u8			oemid[6];
-	u8			oem_tabid[8];
-	u32			oem_rev;
-	u8			creator_id[4];
-	u32			creator_rev;
-	u32			num_uarts;
-	struct pcdp_uart	uart[0];	/* actual size is num_uarts */
-	/* remainder of table is pcdp_device structures */
-} __attribute__((packed));
Index: vga/include/linux/pcdp.h
=================================--- /dev/null	1970-01-01 00:00:00.000000000 +0000
+++ vga/include/linux/pcdp.h	2005-10-12 10:43:38.172792280 -0500
@@ -0,0 +1,111 @@
+/*
+ * Definitions for PCDP-defined console devices
+ *
+ * v1.0a: http://www.dig64.org/specifications/DIG64_HCDPv10a_01.pdf
+ * v2.0:  http://www.dig64.org/specifications/DIG64_PCDPv20.pdf
+ *
+ * (c) Copyright 2002, 2004 Hewlett-Packard Development Company, L.P.
+ *	Khalid Aziz <khalid.aziz@hp.com>
+ *	Bjorn Helgaas <bjorn.helgaas@hp.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#define PCDP_CONSOLE			0
+#define PCDP_DEBUG			1
+#define PCDP_CONSOLE_OUTPUT		2
+#define PCDP_CONSOLE_INPUT		3
+
+#define PCDP_UART			(0 << 3)
+#define PCDP_VGA			(1 << 3)
+#define PCDP_USB			(2 << 3)
+
+/* pcdp_uart.type and pcdp_device.type */
+#define PCDP_CONSOLE_UART		(PCDP_UART | PCDP_CONSOLE)
+#define PCDP_DEBUG_UART			(PCDP_UART | PCDP_DEBUG)
+#define PCDP_CONSOLE_VGA		(PCDP_VGA  | PCDP_CONSOLE_OUTPUT)
+#define PCDP_CONSOLE_USB		(PCDP_USB  | PCDP_CONSOLE_INPUT)
+
+/* pcdp_uart.flags */
+#define PCDP_UART_EDGE_SENSITIVE	(1 << 0)
+#define PCDP_UART_ACTIVE_LOW		(1 << 1)
+#define PCDP_UART_PRIMARY_CONSOLE	(1 << 2)
+#define PCDP_UART_IRQ			(1 << 6) /* in pci_func for rev < 3 */
+#define PCDP_UART_PCI			(1 << 7) /* in pci_func for rev < 3 */
+
+struct pcdp_uart {
+	u8				type;
+	u8				bits;
+	u8				parity;
+	u8				stop_bits;
+	u8				pci_seg;
+	u8				pci_bus;
+	u8				pci_dev;
+	u8				pci_func;
+	u64				baud;
+	struct acpi_generic_address	addr;
+	u16				pci_dev_id;
+	u16				pci_vendor_id;
+	u32				gsi;
+	u32				clock_rate;
+	u8				pci_prog_intfc;
+	u8				flags;
+	u16				conout_index;
+	u32				reserved;
+} __attribute__((packed));
+
+#define PCDP_IF_PCI	1
+
+/* pcdp_if_pci.trans */
+#define PCDP_PCI_TRANS_IOPORT	0x02
+#define PCDP_PCI_TRANS_MMIO	0x01
+
+struct pcdp_if_pci {
+	u8			interconnect;
+	u8			reserved;
+	u16			length;
+	u8			segment;
+	u8			bus;
+	u8			dev;
+	u8			fun;
+	u16			dev_id;
+	u16			vendor_id;
+	u32			acpi_interrupt;
+	u64			mmio_tra;
+	u64			ioport_tra;
+	u8			flags;
+	u8			trans;
+} __attribute__((packed));
+
+struct pcdp_vga {
+	u8			count;		/* address space descriptors */
+} __attribute__((packed));
+
+/* pcdp_device.flags */
+#define PCDP_PRIMARY_CONSOLE	1
+
+struct pcdp_device {
+	u8			type;
+	u8			flags;
+	u16			length;
+	u16			efi_index;
+	/* next data is pcdp_if_pci or pcdp_if_acpi (not yet supported) */
+	/* next data is device specific type (currently only pcdp_vga) */
+} __attribute__((packed));
+
+struct pcdp {
+	u8			signature[4];
+	u32			length;
+	u8			rev;		/* PCDP v2.0 is rev 3 */
+	u8			chksum;
+	u8			oemid[6];
+	u8			oem_tabid[8];
+	u32			oem_rev;
+	u8			creator_id[4];
+	u32			creator_rev;
+	u32			num_uarts;
+	struct pcdp_uart	uart[0];	/* actual size is num_uarts */
+	/* remainder of table is pcdp_device structures */
+} __attribute__((packed));

Re: [PATCH] set altix preferred console

[Greg Edwards]

On Wed, Oct 12, 2005 at 01:55:02PM -0500, Mark Maule wrote:
| New version of Greg's original patch per review comments from Bjorn:
| 
| 
| Fix default VGA console on SN platforms.  Since SN firmware does not pass
| enough ACPI information to identify VGA cards and the associated legacy IO/MEM
| addresses, we rely on the EFI PCDP table.  Since the linux pcdp driver is
| optional (and overridden if console= directives are used) SN duplicates a
| portion of the pcdp scan code to identify if there is a usable console VGA
| adapter.
| 
| To avoid also dup'ing PCDP structures, drivers/firmware/pcdp.h is moved to
| include/linux/pcdp.h.
| 
| Signed-off-by: Mark Maule <maule@sgi.com>

Acked-by: Greg Edwards <edwardsg@sgi.com>

This does fix the problem I was experiencing earlier while booting with
console= boot options.

RE: [PATCH] set altix preferred console

[Luck, Tony]

>> > I don't want to move pcdp.h under asm-ia64.  EFI (and, I hope,
>> > PCDP) is coming to x86 soon.  There's nothing ia64-specific
>> > in pcdp.h or pcdp.c, so I think it would be a mistake to
>> > move it to an arch-specific directory.
>> 
>> Ok, thanks Bjorn.  I was under the impression that stuff under dig64.org
>> was ia64 specific ...
>> 
>> Would moving it to include/linux be an option?
>
>I sort of hate to put it in include/linux, too, but I guess
>it's already a dumping ground for all sorts of miscellaneous
>stuff, so one more wouldn't make it much worse.
>
>And I can't think of any better options.

The comments at the head of pcdp.h certainly make it look
very ia64 specific.  Are there any newer EFI documents that
it could point to instead of the dig64.org URLs?

Linus is just as fond of EFI as he is of ACPI (i.e. not at all),
so I don't think he'll be too happy with include/linux/pcdp.h
If it really is the only place for this to move, then this discussion
needs to happen on linux-kernel, not on linux-ia64.

Since PCDP is closely linked to ACPI, would include/acpi/pcdp.h
be at all plausible?  At least this keeps all the bits Linus
doesn't like together and he can more easily avert his eyes.

Or, merge the contents into include/linux/efi.h?

-Tony

Re: [PATCH] set altix preferred console

[Bjorn Helgaas]

On Tuesday 18 October 2005 11:06 am, Luck, Tony wrote:
> >> > I don't want to move pcdp.h under asm-ia64.  EFI (and, I hope,
> >> > PCDP) is coming to x86 soon.  There's nothing ia64-specific
> >> > in pcdp.h or pcdp.c, so I think it would be a mistake to
> >> > move it to an arch-specific directory.
> >> 
> >> Ok, thanks Bjorn.  I was under the impression that stuff under dig64.org
> >> was ia64 specific ...
> >> 
> >> Would moving it to include/linux be an option?
> >
> >I sort of hate to put it in include/linux, too, but I guess
> >it's already a dumping ground for all sorts of miscellaneous
> >stuff, so one more wouldn't make it much worse.
> >
> >And I can't think of any better options.
> 
> The comments at the head of pcdp.h certainly make it look
> very ia64 specific.  Are there any newer EFI documents that
> it could point to instead of the dig64.org URLs?

The point of the PCDP is to tell the OS what console firmware
was using, so you don't need "console=" arguments.  It's
defined to be very easy to parse, so the OS can use it before
ACPI or PCI device enumeration.

So other than the URL that contains "dig64", there's nothing
ia64-specific.  Matt Tolentino (Intel) was hoping that the
PCDP or something similar would eventually be supported on
x86 as well.  But that world still has its blinders on and
assumes all the world is VGA, and seems to think that if you
want something other than VGA, by George, you ought to *have*
to use "console=".

> Linus is just as fond of EFI as he is of ACPI (i.e. not at all),
> so I don't think he'll be too happy with include/linux/pcdp.h
> If it really is the only place for this to move, then this discussion
> needs to happen on linux-kernel, not on linux-ia64.
> 
> Since PCDP is closely linked to ACPI, would include/acpi/pcdp.h
> be at all plausible?

PCDP is not linked with ACPI at all, except that it happens to
use an address structure that matches the ACPI "generic address".
We should probably just duplicate that structure definition and
kill the linux/acpi.h include.

> Or, merge the contents into include/linux/efi.h?

This would be a possibility.  But other than the fact that we
get the address of the PCDP from EFI, there's no EFI dependency
either.

The only problem with the current location is that SGI firmware
doesn't have quite enough device enumeration support for their
VGA devices, so they have to depend on the PCDP as a crutch.  So
maybe we should just leave pcdp.h alone and make SGI duplicate
what they need.

Re: [PATCH] set altix preferred console

[Mark Maule]

On Tue, Oct 18, 2005 at 12:33:06PM -0600, Bjorn Helgaas wrote:
> On Tuesday 18 October 2005 11:06 am, Luck, Tony wrote:
> > >> > I don't want to move pcdp.h under asm-ia64.  EFI (and, I hope,
> > >> > PCDP) is coming to x86 soon.  There's nothing ia64-specific
> > >> > in pcdp.h or pcdp.c, so I think it would be a mistake to
> > >> > move it to an arch-specific directory.
> > >> 
> > >> Ok, thanks Bjorn.  I was under the impression that stuff under dig64.org
> > >> was ia64 specific ...
> > >> 
> > >> Would moving it to include/linux be an option?
> > >
> > >I sort of hate to put it in include/linux, too, but I guess
> > >it's already a dumping ground for all sorts of miscellaneous
> > >stuff, so one more wouldn't make it much worse.
> > >
> > >And I can't think of any better options.
> > 
> > The comments at the head of pcdp.h certainly make it look
> > very ia64 specific.  Are there any newer EFI documents that
> > it could point to instead of the dig64.org URLs?
> 
> The point of the PCDP is to tell the OS what console firmware
> was using, so you don't need "console=" arguments.  It's
> defined to be very easy to parse, so the OS can use it before
> ACPI or PCI device enumeration.
> 
> So other than the URL that contains "dig64", there's nothing
> ia64-specific.  Matt Tolentino (Intel) was hoping that the
> PCDP or something similar would eventually be supported on
> x86 as well.  But that world still has its blinders on and
> assumes all the world is VGA, and seems to think that if you
> want something other than VGA, by George, you ought to *have*
> to use "console=".
> 
> > Linus is just as fond of EFI as he is of ACPI (i.e. not at all),
> > so I don't think he'll be too happy with include/linux/pcdp.h
> > If it really is the only place for this to move, then this discussion
> > needs to happen on linux-kernel, not on linux-ia64.
> > 
> > Since PCDP is closely linked to ACPI, would include/acpi/pcdp.h
> > be at all plausible?
> 
> PCDP is not linked with ACPI at all, except that it happens to
> use an address structure that matches the ACPI "generic address".
> We should probably just duplicate that structure definition and
> kill the linux/acpi.h include.
> 
> > Or, merge the contents into include/linux/efi.h?
> 
> This would be a possibility.  But other than the fact that we
> get the address of the PCDP from EFI, there's no EFI dependency
> either.
> 
> The only problem with the current location is that SGI firmware
> doesn't have quite enough device enumeration support for their
> VGA devices, so they have to depend on the PCDP as a crutch.  So
> maybe we should just leave pcdp.h alone and make SGI duplicate
> what they need.

I'm ok with that too.  Just thought it was a little neater to not dup the
structs.  This is probably better in the long run anyway as we can more
easily rip the crutches out as we move forward with ACPI.

Mark

[PATCH] set altix preferred console

[Mark Maule]

Fix default VGA console on SN platforms.  Since SN firmware does not pass
enough ACPI information to identify VGA cards and the associated legacy IO/MEM
addresses, we rely on the EFI PCDP table.  Since the linux pcdp driver is
optional (and overridden if console= directives are used) SN duplicates a
portion of the pcdp scan code to identify if there is a usable console VGA
adapter.  Additionally, dup necessary pcdp related structs to avoid dragging
drivers/pcdp.h into a more public location.

Signed-off-by: Mark Maule <maule@sgi.com>

Index: vga/arch/ia64/sn/kernel/setup.c
=================================--- vga.orig/arch/ia64/sn/kernel/setup.c	2005-10-12 10:36:40.914797412 -0500
+++ vga/arch/ia64/sn/kernel/setup.c	2005-11-09 11:28:04.181164653 -0600
@@ -30,6 +30,7 @@
 #include <linux/root_dev.h>
 #include <linux/nodemask.h>
 #include <linux/pm.h>
+#include <linux/efi.h>
 
 #include <asm/io.h>
 #include <asm/sal.h>
@@ -242,6 +243,135 @@
 	}
 }
 
+/*
+ * Scan the EFI PCDP table (if it exists) for an acceptable VGA console
+ * output device.  If one exists, pick it and set sn_legacy_{io,mem} to
+ * reflect the bus offsets needed to address it.
+ *
+ * Since pcdp support in SN is not supported in the 2.4 kernel (or at least
+ * the one lbs is based on) just declare the needed structs here.
+ *
+ * Reference spec http://www.dig64.org/specifications/DIG64_PCDPv20.pdf
+ *
+ * Returns 0 if no acceptable vga is found, !0 otherwise.
+ *
+ * Note:  This stuff is duped here because Altix requires the PCDP to
+ * locate a usable VGA device due to lack of proper ACPI support.  Structures
+ * could be used from drivers/firmware/pcdp.h, but it was decided that moving
+ * this file to a more public location just for Altix use was undesireable.
+ */
+
+struct hcdp_uart_desc {
+	u8	pad[45];
+};
+
+struct pcdp {
+	u8	signature[4];	/* should be 'HCDP' */
+	u32	length;
+	u8	rev;		/* should be >=3 for pcdp, <3 for hcdp */
+	u8	sum;
+	u8	oem_id[6];
+	u64	oem_tableid;
+	u32	oem_rev;
+	u32	creator_id;
+	u32	creator_rev;
+	u32	num_type0;
+	struct hcdp_uart_desc uart[0];	/* num_type0 of these */
+	/* pcdp descriptors follow */
+}  __attribute__((packed));
+
+struct pcdp_device_desc {
+	u8	type;
+	u8	primary;
+	u16	length;
+	u16	index;
+	/* interconnect specific structure follows */
+	/* device specific structure follows that */
+}  __attribute__((packed));
+
+struct pcdp_interface_pci {
+	u8	type;		/* 1 = pci */
+	u8	reserved;
+	u16	length;
+	u8	segment;
+	u8	bus;
+	u8 	dev;
+	u8	fun;
+	u16	devid;
+	u16	vendid;
+	u32	acpi_interrupt;
+	u64	mmio_tra;
+	u64	ioport_tra;
+	u8	flags;
+	u8	translation;
+}  __attribute__((packed));
+
+struct pcdp_vga_device {
+	u8	num_eas_desc;
+	/* ACPI Extended Address Space Desc follows */
+}  __attribute__((packed));
+
+/* from pcdp_device_desc.primary */
+#define PCDP_PRIMARY_CONSOLE	0x01
+
+/* from pcdp_device_desc.type */
+#define PCDP_CONSOLE_INOUT	0x0
+#define PCDP_CONSOLE_DEBUG	0x1
+#define PCDP_CONSOLE_OUT	0x2
+#define PCDP_CONSOLE_IN		0x3
+#define PCDP_CONSOLE_TYPE_VGA	0x8
+
+#define PCDP_CONSOLE_VGA	(PCDP_CONSOLE_TYPE_VGA | PCDP_CONSOLE_OUT)
+
+/* from pcdp_interface_pci.type */
+#define PCDP_IF_PCI		1
+
+/* from pcdp_interface_pci.translation */
+#define PCDP_PCI_TRANS_IOPORT	0x02
+#define PCDP_PCI_TRANS_MMIO	0x01
+
+static void
+sn_scan_pcdp(void)
+{
+	u8 *bp;
+	struct pcdp *pcdp;
+	struct pcdp_device_desc device;
+	struct pcdp_interface_pci if_pci;
+	extern struct efi efi;
+
+	pcdp = efi.hcdp;
+	if (! pcdp)
+		return;		/* no hcdp/pcdp table */
+
+	if (pcdp->rev < 3)
+		return;		/* only support PCDP (rev >= 3) */
+
+	for (bp = (u8 *)&pcdp->uart[pcdp->num_type0];
+	     bp < (u8 *)pcdp + pcdp->length;
+	     bp += device.length) {
+		memcpy(&device, bp, sizeof(device));
+		if (! (device.primary & PCDP_PRIMARY_CONSOLE))
+			continue;	/* not primary console */
+
+		if (device.type != PCDP_CONSOLE_VGA)
+			continue;	/* not VGA descriptor */
+
+		memcpy(&if_pci, bp+sizeof(device), sizeof(if_pci));
+		if (if_pci.type != PCDP_IF_PCI)
+			continue;	/* not PCI interconnect */
+
+		if (if_pci.translation & PCDP_PCI_TRANS_IOPORT)
+			vga_console_iobase +				if_pci.ioport_tra | __IA64_UNCACHED_OFFSET;
+
+		if (if_pci.translation & PCDP_PCI_TRANS_MMIO)
+			vga_console_membase +				if_pci.mmio_tra | __IA64_UNCACHED_OFFSET;
+
+		break; /* once we find the primary, we're done */
+	}
+}
+
 /**
  * sn_setup - SN platform setup routine
  * @cmdline_p: kernel command line
@@ -263,16 +393,35 @@
 
 #if defined(CONFIG_VT) && defined(CONFIG_VGA_CONSOLE)
 	/*
-	 * If there was a primary vga adapter identified through the
-	 * EFI PCDP table, make it the preferred console.  Otherwise
-	 * zero out conswitchp.
+	 * Handle SN vga console.
+	 *
+	 * SN systems do not have enough ACPI table information
+	 * being passed from prom to identify VGA adapters and the legacy
+	 * addresses to access them.  Until that is done, SN systems rely
+	 * on the PCDP table to identify the primary VGA console if one
+	 * exists.
+	 *
+	 * However, kernel PCDP support is optional, and even if it is built
+	 * into the kernel, it will not be used if the boot cmdline contains
+	 * console= directives.
+	 *
+	 * So, to work around this mess, we duplicate some of the PCDP code
+	 * here so that the primary VGA console (as defined by PCDP) will
+	 * work on SN systems even if a different console (e.g. serial) is
+	 * selected on the boot line (or CONFIG_EFI_PCDP is off).
 	 */
 
+	if (! vga_console_membase)
+		sn_scan_pcdp();
+
 	if (vga_console_membase) {
 		/* usable vga ... make tty0 the preferred default console */
-		add_preferred_console("tty", 0, NULL);
+		if (!strstr(*cmdline_p, "console="))
+			add_preferred_console("tty", 0, NULL);
 	} else {
 		printk(KERN_DEBUG "SGI: Disabling VGA console\n");
+		if (!strstr(*cmdline_p, "console="))
+			add_preferred_console("ttySG", 0, NULL);
 #ifdef CONFIG_DUMMY_CONSOLE
 		conswitchp = &dummy_con;
 #else