Skip to contentSkip to conversation

restore libata build on frv

PATCH NAK Al Viro 7 participants 24 messages Sep 24, 2006 – Sep 27, 2006

Subsystems (2)
lore
Share & tools
lore

Al Viro

author
#0 lore raw

PATCH 1/1 restore libata build on frv

---
 include/asm-frv/libata-portmap.h |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
include/asm-frv/libata-portmap.h new file +1 -0
new file mode 100644
index 0000000..75484ef
@@ -0,0 +1 @@
+#include <asm-generic/libata-portmap.h>
Signature · 1 line
1.4.2.GIT
#1 lore raw

include/asm-frv/libata-portmap.h line 1

Quoted 3 lines · Al Viro

+++ b/include/asm-frv/libata-portmap.h

include/asm-frv/libata-portmap.h new file +1 -0
@@ -0,0 +1 @@
+#include <asm-generic/libata-portmap.h>

NAK... These settings are totally meaningless on FRV.

David

#2 lore raw

Ar Llu, 2006-09-25 am 11:44 +0100, ysgrifennodd David Howells:

Quoted 7 lines

include/asm-frv/libata-portmap.h line 1

Quoted 3 lines · Al Viro

+++ b/include/asm-frv/libata-portmap.h

include/asm-frv/libata-portmap.h new file +1 -0
@@ -0,0 +1 @@
+#include <asm-generic/libata-portmap.h>

NAK... These settings are totally meaningless on FRV.

You have no PCI bus ?

#3 lore raw
Quoted 10 lines · Alan Cox

Ar Llu, 2006-09-25 am 11:44 +0100, ysgrifennodd David Howells:

Quoted 7 lines

include/asm-frv/libata-portmap.h line 1

Quoted 3 lines · Al Viro

+++ b/include/asm-frv/libata-portmap.h

include/asm-frv/libata-portmap.h new file +1 -0
@@ -0,0 +1 @@
+#include <asm-generic/libata-portmap.h>

NAK... These settings are totally meaningless on FRV.

You have no PCI bus ?

Note that if you don't provide an asm/libata-portmap.h file, you can't build libata at the moment - linux/libata.h requires this file to be present.

Signature · 3 lines
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core
#4 lore raw
Quoted 3 lines · Russell King

Note that if you don't provide an asm/libata-portmap.h file, you can't build libata at the moment - linux/libata.h requires this file to be present.

Yes, but it should be empty as there is no ISA bus.

David

#5 lore raw
In reply to Alan Cox
Quoted 3 lines · Alan Cox
Quoted 1 line

NAK... These settings are totally meaningless on FRV.

You have no PCI bus ?

I do, but I don't have an ISA bus, and the stuff in asm-generic is worse than useless. #if'ing out all the legacy-handling code that uses this stuff seems to work fine.

These are all legacy ISA settings, and not applicable to FRV:

	#define ATA_PRIMARY_CMD		0x1F0
	#define ATA_PRIMARY_CTL		0x3F6
	#define ATA_PRIMARY_IRQ		14

	#define ATA_SECONDARY_CMD	0x170
	#define ATA_SECONDARY_CTL	0x376
	#define ATA_SECONDARY_IRQ	15

Note that the ata_pci_init_legacy_port() explicitly states the IRQ numbers as 14 and 15 without reference to the macros and so is bad, eg:

		probe_ent->irq = 14;

The attached patch makes it possible to dispense with the settings completely.

David

---
[PATCH] FRV: Make libata build on FRV

Make FRV build with libata enabled. This is done by making the legacy ISA interface support conditional on the definition of the legacy ISA port settings. If there's no ISA bus, we shouldn't even attempt to pretend that there is.

---

 drivers/ata/libata-core.c        |    2 ++
 drivers/ata/libata-sff.c         |   10 ++++++++++
 include/asm-frv/libata-portmap.h |   17 ++++++++++++++++-
 3 files changed, 28 insertions(+), 1 deletions(-)
drivers/ata/libata-core.c +2 -0
index 753b015..2afa510 100644
@@ -5714,6 +5714,7 @@ void ata_host_remove(struct ata_host *ho
 
 		ata_scsi_release(ap->scsi_host);
 
+#ifdef ATA_PRIMARY_CMD
 		if ((ap->flags & ATA_FLAG_NO_LEGACY) == 0) {
 			struct ata_ioports *ioaddr = &ap->ioaddr;
 
@@ -5723,6 +5724,7 @@ void ata_host_remove(struct ata_host *ho
 			else if (ioaddr->cmd_addr == ATA_SECONDARY_CMD)
 				release_region(ATA_SECONDARY_CMD, 8);
 		}
+#endif
 
 		scsi_host_put(ap->scsi_host);
 	}
drivers/ata/libata-sff.c +10 -0
index 688bb55..63fdf73 100644
@@ -867,6 +867,7 @@ ata_pci_init_native_mode(struct pci_dev 
 }
 
 
+#ifdef ATA_PRIMARY_CMD
 static struct ata_probe_ent *ata_pci_init_legacy_port(struct pci_dev *pdev,
 				struct ata_port_info **port, int port_mask)
 {
@@ -914,6 +915,7 @@ static struct ata_probe_ent *ata_pci_ini
 
 	return probe_ent;
 }
+#endif
 
 
 /**
@@ -993,6 +995,7 @@ int ata_pci_init_one (struct pci_dev *pd
 		goto err_out;
 	}
 
+#ifdef ATA_PRIMARY_CMD
 	if (legacy_mode) {
 		if (!request_region(ATA_PRIMARY_CMD, 8, "libata")) {
 			struct resource *conflict, res;
@@ -1032,6 +1035,7 @@ int ata_pci_init_one (struct pci_dev *pd
 		} else
 			legacy_mode |= ATA_PORT_SECONDARY;
 	}
+#endif
 
 	/* we have legacy mode, but all ports are unavailable */
 	if (legacy_mode == (1 << 3)) {
@@ -1047,14 +1051,18 @@ int ata_pci_init_one (struct pci_dev *pd
 	if (rc)
 		goto err_out_regions;
 
+#ifdef ATA_PRIMARY_CMD
 	if (legacy_mode) {
 		probe_ent = ata_pci_init_legacy_port(pdev, port, legacy_mode);
 	} else {
+#endif
 		if (n_ports == 2)
 			probe_ent = ata_pci_init_native_mode(pdev, port, ATA_PORT_PRIMARY | ATA_PORT_SECONDARY);
 		else
 			probe_ent = ata_pci_init_native_mode(pdev, port, ATA_PORT_PRIMARY);
+#ifdef ATA_PRIMARY_CMD
 	}
+#endif
 	if (!probe_ent) {
 		rc = -ENOMEM;
 		goto err_out_regions;
@@ -1070,10 +1078,12 @@ int ata_pci_init_one (struct pci_dev *pd
 	return 0;
 
 err_out_regions:
+#ifdef ATA_PRIMARY_CMD
 	if (legacy_mode & ATA_PORT_PRIMARY)
 		release_region(ATA_PRIMARY_CMD, 8);
 	if (legacy_mode & ATA_PORT_SECONDARY)
 		release_region(ATA_SECONDARY_CMD, 8);
+#endif
 	pci_release_regions(pdev);
 err_out:
 	if (disable_dev_on_err)
include/asm-frv/libata-portmap.h +16 -1
index 75484ef..2403f0b 100644
@@ -1 +1,16 @@
-#include <asm-generic/libata-portmap.h>
+/* ISA bus mappings for legacy ATA ports - which don't exist on FRV
+ *
+ * Copyright (C) 2006 Red Hat, Inc. All Rights Reserved.
+ * Written by David Howells ([email protected])
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#ifndef _ASM_LIBATA_PORTMAP_H
+#define _ASM_LIBATA_PORTMAP_H
+
+
+#endif /* _ASM_LIBATA_PORTMAP_H */
Nacked-by NAK
#6 lore raw

Ar Llu, 2006-09-25 am 12:27 +0100, ysgrifennodd David Howells:

Quoted 10 lines

Alan Cox <[email protected]> wrote:
These are all legacy ISA settings, and not applicable to FRV:

	#define ATA_PRIMARY_CMD		0x1F0
	#define ATA_PRIMARY_CTL		0x3F6
	#define ATA_PRIMARY_IRQ		14

	#define ATA_SECONDARY_CMD	0x170
	#define ATA_SECONDARY_CTL	0x376
	#define ATA_SECONDARY_IRQ	15

Wrong these are PCI settings. Please read the PCI specifications. In particular the handling of non-native mode IDE storage class devices on a PCI bus. For the IRQ mapping of the non-native ports consult your bridge documentation.

Quoted 4 lines

Note that the ata_pci_init_legacy_port() explicitly states the IRQ numbers as 14 and 15 without reference to the macros and so is bad, eg:

		probe_ent->irq = 14;

That is indeed a bug

Quoted 6 lines

Make FRV build with libata enabled. This is done by making the legacy ISA interface support conditional on the definition of the legacy ISA port settings. If there's no ISA bus, we shouldn't even attempt to pretend that there is.

#7 lore raw
Quoted 4 lines · Alan Cox

Wrong these are PCI settings. Please read the PCI specifications. In particular the handling of non-native mode IDE storage class devices on a PCI bus. For the IRQ mapping of the non-native ports consult your bridge documentation.

Even if that is the case, they are all invalid/incorrect, and so Al Viro's patch is _still_ NAK'd.

David

author
#8 lore raw
Quoted 9 lines · David Howells
Quoted 4 lines · Alan Cox

Wrong these are PCI settings. Please read the PCI specifications. In particular the handling of non-native mode IDE storage class devices on a PCI bus. For the IRQ mapping of the non-native ports consult your bridge documentation.

Even if that is the case, they are all invalid/incorrect, and so Al Viro's patch is _still_ NAK'd.

Fine by me. In that case we need to add

	depends on !FRV || BROKEN

to drivers/ata/Kconfig and be done with that. BTW, empty libata-portmap.h is equivalent to absent one - it still won't build.

#9 lore raw
Quoted 2 lines · Al Viro

Fine by me. In that case we need to add

	depends on !FRV || BROKEN

My point is that all the numbers are invalid or incorrect. They will cause the arch to oops, and so need completely replacing. So that patch you added is incorrect and should not include asm-generic/libata-portmap.h as nothing in there is correct on this arch. So your patch should *not* be applied, but should instead be replaced.

Quoted 2 lines

to drivers/ata/Kconfig and be done with that. BTW, empty libata-portmap.h is equivalent to absent one - it still won't build.

Why does the arch have to supply those numbers? What's wrong with my suggested patch? According to code in libata, these are _legacy_ access methods, and on FRV they aren't currently required, so why can't I dispense with them if they're not needed? Doing that not only skips legacy accesses for ISA compatibility, it also reduces the code size by not actually emitting the code for the accesses, thus making the kernel smaller...

David

#10 lore raw

Ar Llu, 2006-09-25 am 15:39 +0100, ysgrifennodd David Howells:

Quoted 3 lines

Why does the arch have to supply those numbers? What's wrong with my suggested patch? According to code in libata, these are _legacy_ access methods, and on FRV they aren't currently required, so why can't I dispense

"legacy, legacy, legacy" "wont wont wont"

The ports in question are PCI values. They come from the PCI specifications and apply to any device with PCI bus, unless it has special mappings. The same logic you are whining about is already partly handled in the generic pci quirks code, and in time will end up with the I/O port value fixups there anyway.

See quirk_ide_bases in drivers/pci/quirks.c

Go read the specifications and stop whining about legacy ports and ISA bus for things that are not.

Ack Al Viro's changes but with IRQ set to zero.

Alan

#11 lore raw
Quoted 1 line · Alan Cox

Ack Al Viro's changes but with IRQ set to zero.

The PCI bus has special mappings. You may not address it with those numbers. Any of those numbers. Al's patch is 100% incorrect. Sorry.

David

author
#12 lore raw
Quoted 6 lines · David Howells
Quoted 1 line · Alan Cox

Ack Al Viro's changes but with IRQ set to zero.

The PCI bus has special mappings. You may not address it with those numbers. Any of those numbers. Al's patch is 100% incorrect. Sorry.

Oh, for fuck sake... 2.6.18 drivers/scsi/libata-core.c:

                if ((ap->flags & ATA_FLAG_NO_LEGACY) == 0) {
                        struct ata_ioports *ioaddr = &ap->ioaddr;

                        if (ioaddr->cmd_addr == 0x1f0)
                                release_region(0x1f0, 8);
                        else if (ioaddr->cmd_addr == 0x170)
                                release_region(0x170, 8);
                }

current drivers/ata/libata-core.c:

                if ((ap->flags & ATA_FLAG_NO_LEGACY) == 0) {
                        struct ata_ioports *ioaddr = &ap->ioaddr;

                        /* FIXME: Add -ac IDE pci mods to remove these special cases */
                        if (ioaddr->cmd_addr == ATA_PRIMARY_CMD)
                                release_region(ATA_PRIMARY_CMD, 8);
                        else if (ioaddr->cmd_addr == ATA_SECONDARY_CMD)
                                release_region(ATA_SECONDARY_CMD, 8);
                }

Patch in question restores the situation prior to libata merge. That's what FRV had been doing all along if SATA had been enabled. No more, mo less.

Now, if you want to change that behaviour, more power to you. But that's a separate patch, obviously, and all issues related to that exist in vanilla 2.6.18 just as in the current tree.

#13 lore raw
In reply to Alan Cox
Quoted 7 lines · Alan Cox

Ar Llu, 2006-09-25 am 15:39 +0100, ysgrifennodd David Howells:

Quoted 3 lines

Why does the arch have to supply those numbers? What's wrong with my suggested patch? According to code in libata, these are _legacy_ access methods, and on FRV they aren't currently required, so why can't I dispense

"legacy, legacy, legacy" "wont wont wont"

Quoted 7 lines

The ports in question are PCI values. They come from the PCI specifications and apply to any device with PCI bus, unless it has special mappings. The same logic you are whining about is already partly handled in the generic pci quirks code, and in time will end up with the I/O port value fixups there anyway.

See quirk_ide_bases in drivers/pci/quirks.c

If we can do that with PCI quirks, why the need to hard-code it in the IDE driver too?

And IRQ zero isn't particularly helpful suggestion -- using an invalid IRQ number would be better. Like NO_IRQ or IDE_NO_IRQ, which should be -1.

Don't make me dig out the board where the PCI slots all get IRQ 0 :)

Signature · 1 line
dwmw2
#14 lore raw
Quoted 15 lines · David Woodhouse
Quoted 6 lines · Alan Cox

Ar Llu, 2006-09-25 am 15:39 +0100, ysgrifennodd David Howells:

Quoted 3 lines

Why does the arch have to supply those numbers? What's wrong with my suggested patch? According to code in libata, these are _legacy_ access methods, and on FRV they aren't currently required, so why can't I dispense

"legacy, legacy, legacy" "wont wont wont"

Quoted 7 lines

The ports in question are PCI values. They come from the PCI specifications and apply to any device with PCI bus, unless it has special mappings. The same logic you are whining about is already partly handled in the generic pci quirks code, and in time will end up with the I/O port value fixups there anyway.

See quirk_ide_bases in drivers/pci/quirks.c

Quoted 8 lines

If we can do that with PCI quirks, why the need to hard-code it in the IDE driver too?

And IRQ zero isn't particularly helpful suggestion -- using an invalid IRQ number would be better. Like NO_IRQ or IDE_NO_IRQ, which should be -1.

Don't make me dig out the board where the PCI slots all get IRQ 0 :)

The irq is a special case no matter how we try to prettyify it. We need two irqs, and PCI only gives us one per device.

Jeff

#15 lore raw
Quoted 2 lines · Jeff Garzik

The irq is a special case no matter how we try to prettyify it. We need two irqs, and PCI only gives us one per device.

That's fine -- but don't use zero to mean none. We have NO_IRQ for that, and zero isn't an appropriate choice.

Signature · 1 line
dwmw2
#16 lore raw

Ar Maw, 2006-09-26 am 09:56 +0100, ysgrifennodd David Woodhouse:

Quoted 6 lines · Jeff Garzik
Quoted 2 lines

The irq is a special case no matter how we try to prettyify it. We need two irqs, and PCI only gives us one per device.

That's fine -- but don't use zero to mean none. We have NO_IRQ for that, and zero isn't an appropriate choice.

Zero means "no IRQ". That's official kernel policy and true for both old and new IDE. Architectures are supposed to remap any real "irq 0".

Might as well use NO_IRQ though, as its clearer.

Alan

#17 lore raw
In reply to David Woodhouse

Ar Maw, 2006-09-26 am 09:56 +0100, ysgrifennodd David Woodhouse:

Quoted 6 lines · Jeff Garzik
Quoted 2 lines

The irq is a special case no matter how we try to prettyify it. We need two irqs, and PCI only gives us one per device.

That's fine -- but don't use zero to mean none. We have NO_IRQ for that, and zero isn't an appropriate choice.

Let me correct that - you *had* NO_IRQ. It isn't defined for most platforms any more in -mm it seems.

Alan

#18 lore raw
In reply to David Woodhouse
Quoted 7 lines · David Woodhouse
Quoted 2 lines · Jeff Garzik

The irq is a special case no matter how we try to prettyify it. We need two irqs, and PCI only gives us one per device.

That's fine -- but don't use zero to mean none. We have NO_IRQ for that, and zero isn't an appropriate choice.

Zero _is_ an appropriate choice, dammit!

That NO_IRQ thing should be zero, and any architecture that thinks that zero is a valid IRQ just needs to fix its own irq mapping so that the "cookie" doesn't work.

The thing is, it's zero. Get over it. It can't be "-1" or some other random value like people have indicated, because that thing is often read from places where "-1" simply isn't a possible value (eg it gets its default value initialized from a "unsigned char" in MMIO space on x86).

So instead of making everybody and their dog to silly things with some NO_IRQ define that they haven't historically done, the rule is simple: "0" means "no irq", so that you can test for it with obvious code like

	if (!dev->irq)
		..

and then, if your actual _hardware_ things that the bit-pattern with all bits clear is a valid irq that can be used for normal devices, then what you do is you add a irq number translation layer (WHICH WE NEED AND HAVE _ANYWAY_) and make sure that nobody sees that on a _software_ level.

Linus

#19 lore raw
Quoted 4 lines · Linus Torvalds

Zero _is_ an appropriate choice, dammit!
...
and then, if your actual _hardware_ things that the bit-pattern with all bits clear is a valid irq that can be used for normal devices,

PCI, for example. According to the spec, 0x00 is valid in the Interrupt Line register and 0xFF indicates unconnected or unset.

Quoted 3 lines

then what you do is you add a irq number translation layer (WHICH WE NEED AND HAVE _ANYWAY_) and make sure that nobody sees that on a _software_ level.

So, by your argument, if you _have_ to have an IRQ translation layer anyway, then what's the problem with having zero as a valid IRQ and using some other value to indicate an invalid IRQ? As you say, you have a translation layer anyway...

That would mean the arch maintainer could make the optimal choice for their arch - perhaps picking 255 which would be consistent with PCI.

As far as FRV goes, I'm quite happy with 0 as being NO_IRQ, since the 0 bit in the primary PIC registers is the master switch, not a per-level bit (there are no source indicators unfortunately).

However, x86, x86_64, and others *do* treat IRQ 0 as being valid, and expose it to userspace in various ways:

	warthog>cat /proc/interrupts 
		   CPU0       
	  0:  287035291    IO-APIC-edge  timer
	...

So on *that* basis, using IRQ 0 to indicate unset/invalid/etc would seem to be bad.

David

#20 lore raw
In reply to Linus Torvalds
Quoted 5 lines · Linus Torvalds

That NO_IRQ thing should be zero, and any architecture that thinks that zero is a valid IRQ just needs to fix its own irq mapping so that the "cookie" doesn't work.

The thing is, it's zero. Get over it.

kernel/irq/manage.c +1 -1
index 92be519..0cdf8ad 100644
@@ -219,7 +219,7 @@ int setup_irq(unsigned int irq, struct i
 	unsigned long flags;
 	int shared = 0;
 
-	if (irq >= NR_IRQS)
+	if (!irq || irq >= NR_IRQS)
 		return -EINVAL;
 
 	if (desc->chip == &no_irq_chip)
Signature · 1 line
dwmw2
#21 lore raw
In reply to David Howells
Quoted 2 lines · David Howells

My point is that all the numbers are invalid or incorrect. They will cause the arch to oops, and so need completely replacing.

The way we deal with code which blindly pokes at legacy I/O addresses on PowerPC is to have a check_legacy_ioport() function which will tell you if it's safe to poke at the address in question. Perhaps that should be extended to other architectures (some can just #define it to 0) and used in the IDE code.

Signature · 1 line
dwmw2
#22 lore raw
In reply to Al Viro

Ar Llu, 2006-09-25 am 15:20 +0100, ysgrifennodd Al Viro:

Quoted 6 lines

Fine by me. In that case we need to add

	depends on !FRV || BROKEN

to drivers/ata/Kconfig and be done with that. BTW, empty libata-portmap.h is equivalent to absent one - it still won't build.

From every public piece of info I can find and from looking at the FRV

tree your changes are correct for the ports Al. I can't find any info on how legacy IRQ routing is done on FRV systems but if it is not then set the IRQ values to zero and maybe Dave will stop complaining.

Alan

#23 lore raw
Quoted 9 lines · Alan Cox
Quoted 4 lines

Fine by me. In that case we need to add

	depends on !FRV || BROKEN

to drivers/ata/Kconfig and be done with that. BTW, empty libata-portmap.h is equivalent to absent one - it still won't build.

From every public piece of info I can find and from looking at the FRV tree your changes are correct for the ports Al. I can't find any info on how legacy IRQ routing is done on FRV systems but if it is not then set the IRQ values to zero and maybe Dave will stop complaining.

Sigh.

On FRV, inX() and outX() take fully qualified memory-space addresses, exactly as readX() and writeX() (in/out just wrap readX/writeX). This is because:

(1) The FRV has a limited number of static mappings, and these have to specify _all_ access windows to I/O, RAM, ROM, etc. The FRV arch uses a single mapping to handle *all* I/O (which happens to be through the region from 0xE0000000 to 0xFFFFFFFF) thus allowing it to use the remaining mappings for other purposes.

(2) inX() and outX() would have to adjust the addresses to otherwise make them appear PC compatible. Making in() and out() just pass the addresses straight through means I don't have to do any calculation on the address in order to use it.

inb(0x1F0) will, for example, oops because there's no mapping for the bottom virtual megabyte to anywhere, otherwise NULL pointer detection would not be possible.

Don't forget, also, that things like FRV systems generally _won't_ have pluggable PCI buses, and so any devices attached to it will be known in advance, and generalisations can be waived.

David