Skip to contentSkip to conversation

Add proper module ID tables to Adaptec aic7[9x]xx drivers

PATCH In review Andi Kleen 4 participants 10 messages Jun 21, 2004

Subsystems (2)
lore
Share & tools
lore

Andi Kleen

author
#0 lore raw

PATCH 1/1 Add proper module ID tables to Adaptec aic7[9x]xx drivers

[I didn't find an aic7xxx maintainer in MAINTAINERS, so just sending to the list]

This is needed for 2.6 hotplug where the driver is autoloaded. When you have multiple conflicting entries the hotplug module loader usually loads the first one listed, which may be correct or may be not.

With these changes the drivers announce the correct PCI IDs.

The patches are quite big (i had to move the ID tables into an include), but mostly mechanical.

I only tested it on a single 78xx, so it's possible that i made a typo somewhere. Some testing would be appreciated, especially on 79xx

There is still an conflicting entry for aic7xxx_old unfortunately. This is hard to handle in hotplug. I think the right solution would be to somehow tag drivers as obsolete so that they aren't loaded by default. This will need an extension in the module pci table mechanism. I'm thinking about a a new MODULE_OBSOLETE macro for this and a obsolete_modules table generated at modules_install time. But that's for another patch. Better would be maybe to just drop the aic7xxx_old driver. Is it really still needed?

-Andi

drivers/scsi/aic7xxx/aic79xx_pci.c renamed from drivers/scsi/aic7xxx/aic79xx_pci.c-o +1 -40
diff -u linux/drivers/scsi/aic7xxx/aic79xx_pci.c-o linux/drivers/scsi/aic7xxx/aic79xx_pci.c
@@ -51,46 +51,7 @@
 #include <dev/aic7xxx/aic79xx_inline.h>
 #endif
 
-static __inline uint64_t
-ahd_compose_id(u_int device, u_int vendor, u_int subdevice, u_int subvendor)
-{
-	uint64_t id;
-
-	id = subvendor
-	   | (subdevice << 16)
-	   | ((uint64_t)vendor << 32)
-	   | ((uint64_t)device << 48);
-
-	return (id);
-}
-
-#define ID_ALL_MASK			0xFFFFFFFFFFFFFFFFull
-#define ID_ALL_IROC_MASK		0xFFFFFF7FFFFFFFFFull
-#define ID_DEV_VENDOR_MASK		0xFFFFFFFF00000000ull
-#define ID_9005_GENERIC_MASK		0xFFF0FFFF00000000ull
-#define ID_9005_GENERIC_IROC_MASK	0xFFF0FF7F00000000ull
-
-#define ID_AIC7901			0x800F9005FFFF9005ull
-#define ID_AHA_29320A			0x8000900500609005ull
-#define ID_AHA_29320ALP			0x8017900500449005ull
-
-#define ID_AIC7901A			0x801E9005FFFF9005ull
-#define ID_AHA_29320			0x8012900500429005ull
-#define ID_AHA_29320B			0x8013900500439005ull
-#define ID_AHA_29320LP			0x8014900500449005ull
-
-#define ID_AIC7902			0x801F9005FFFF9005ull
-#define ID_AIC7902_B			0x801D9005FFFF9005ull
-#define ID_AHA_39320			0x8010900500409005ull
-#define ID_AHA_39320_B			0x8015900500409005ull
-#define ID_AHA_39320A			0x8016900500409005ull
-#define ID_AHA_39320D			0x8011900500419005ull
-#define ID_AHA_39320D_B			0x801C900500419005ull
-#define ID_AHA_39320D_HP		0x8011900500AC0E11ull
-#define ID_AHA_39320D_B_HP		0x801C900500AC0E11ull
-#define ID_AIC7902_PCI_REV_A4		0x3
-#define ID_AIC7902_PCI_REV_B0		0x10
-#define SUBID_HP			0x0E11
+#include "aic79xx_pci.h"
 
 #define DEVID_9005_TYPE(id) ((id) & 0xF)
 #define		DEVID_9005_TYPE_HBA		0x0	/* Standard Card */
drivers/scsi/aic7xxx/aic79xx_osm_pci.c renamed from drivers/scsi/aic7xxx/aic79xx_osm_pci.c-o +24 -5
diff -u linux/drivers/scsi/aic7xxx/aic79xx_osm_pci.c-o linux/drivers/scsi/aic7xxx/aic79xx_osm_pci.c
@@ -41,6 +41,7 @@
 
 #include "aic79xx_osm.h"
 #include "aic79xx_inline.h"
+#include "aic79xx_pci.h"
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,4,0)
 struct pci_device_id
@@ -58,12 +59,30 @@
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
 static void	ahd_linux_pci_dev_remove(struct pci_dev *pdev);
 
-/* We do our own ID filtering.  So, grab all SCSI storage class devices. */
 static struct pci_device_id ahd_linux_pci_id_table[] = {
-	{
-		0x9005, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID,
-		PCI_CLASS_STORAGE_SCSI << 8, 0xFFFF00, 0
-	},
+#define LINUXID(x,s) (unsigned)((((x) >> s) & 0xffff) ?: PCI_ANY_ID)
+#define ID(x) \
+	{ \
+		LINUXID(x,32), LINUXID(x,48), LINUXID(x,0), LINUXID(x,16), \
+		PCI_CLASS_STORAGE_SCSI << 8, 0xFFFF00, 0 \
+	}
+	ID(ID_AHA_29320A),
+	ID(ID_AHA_29320ALP),
+	ID(ID_AHA_29320),
+	ID(ID_AHA_29320B),
+	ID(ID_AHA_29320LP),
+	ID(ID_AHA_39320),
+	ID(ID_AHA_39320_B),
+	ID(ID_AHA_39320A),
+	ID(ID_AHA_39320D),
+	ID(ID_AHA_39320D_HP),
+	ID(ID_AHA_39320D_B),
+	ID(ID_AHA_39320D_B_HP),
+	ID(ID_AHA_29320),
+	ID(ID_AHA_29320B),
+	ID(ID_AIC7901 & ID_DEV_VENDOR_MASK),
+	ID(ID_AIC7901A & ID_DEV_VENDOR_MASK),
+	ID(ID_AIC7902 & ID_9005_GENERIC_MASK),
 	{ 0 }
 };
 
drivers/scsi/aic7xxx/aic79xx_pci.h new file +41 -0

41 lines not on this page · raw message

drivers/scsi/aic7xxx/aic7xxx_pci.h new file +92 -0

92 lines not on this page · raw message

drivers/scsi/aic7xxx/aic7xxx_pci.c renamed from drivers/scsi/aic7xxx/aic7xxx_pci.c-o +2 -92

103 lines not on this page · raw message

drivers/scsi/aic7xxx/aic7xxx_osm_pci.c renamed from drivers/scsi/aic7xxx/aic7xxx_osm_pci.c-o +67 -9

89 lines not on this page · raw message

#1 lore raw
Quoted 5 lines · On Mon, 21 Jun 2004 16:14:41 +0200 Andi Kleen wrote:

This is needed for 2.6 hotplug where the driver is autoloaded. When you have multiple conflicting entries the hotplug module loader usually loads the first one listed, which may be correct or may be not.

With these changes the drivers announce the correct PCI IDs.

Unfortunately, the patch does not seem to be correct :(

struct pci_device_id does not have the mask field, therefore the ID_9005_GENERIC_MASK restriction cannot be specified other than by listing all 16 possible IDs as separate entries. Your patch adds only one entry, thus losing 15 other possible IDs.

author
#2 lore raw
Quoted 6 lines · On Mon, 21 Jun 2004 17:44:08 +0400 Sergey Vlasov <[email protected]> wrote:
Quoted 4 lines · On Mon, 21 Jun 2004 16:14:41 +0200 Andi Kleen wrote:

This is needed for 2.6 hotplug where the driver is autoloaded. When you have multiple conflicting entries the hotplug module loader usually loads the first one listed, which may be correct or may be not.

Quoted 8 lines
Quoted 1 line

With these changes the drivers announce the correct PCI IDs.

Unfortunately, the patch does not seem to be correct :(

struct pci_device_id does not have the mask field, therefore the ID_9005_GENERIC_MASK restriction cannot be specified other than by listing all 16 possible IDs as separate entries. Your patch adds only one entry, thus losing 15 other possible IDs.

Hmm, good point. Thanks for catching this.

Here is a new patch. Does this one look better?

-Andi

drivers/scsi/aic7xxx/aic79xx_pci.c renamed from drivers/scsi/aic7xxx/aic79xx_pci.c-o +1 -40
diff -u linux/drivers/scsi/aic7xxx/aic79xx_pci.c-o linux/drivers/scsi/aic7xxx/aic79xx_pci.c
@@ -51,46 +51,7 @@
 #include <dev/aic7xxx/aic79xx_inline.h>
 #endif
 
-static __inline uint64_t
-ahd_compose_id(u_int device, u_int vendor, u_int subdevice, u_int subvendor)
-{
-	uint64_t id;
-
-	id = subvendor
-	   | (subdevice << 16)
-	   | ((uint64_t)vendor << 32)
-	   | ((uint64_t)device << 48);
-
-	return (id);
-}
-
-#define ID_ALL_MASK			0xFFFFFFFFFFFFFFFFull
-#define ID_ALL_IROC_MASK		0xFFFFFF7FFFFFFFFFull
-#define ID_DEV_VENDOR_MASK		0xFFFFFFFF00000000ull
-#define ID_9005_GENERIC_MASK		0xFFF0FFFF00000000ull
-#define ID_9005_GENERIC_IROC_MASK	0xFFF0FF7F00000000ull
-
-#define ID_AIC7901			0x800F9005FFFF9005ull
-#define ID_AHA_29320A			0x8000900500609005ull
-#define ID_AHA_29320ALP			0x8017900500449005ull
-
-#define ID_AIC7901A			0x801E9005FFFF9005ull
-#define ID_AHA_29320			0x8012900500429005ull
-#define ID_AHA_29320B			0x8013900500439005ull
-#define ID_AHA_29320LP			0x8014900500449005ull
-
-#define ID_AIC7902			0x801F9005FFFF9005ull
-#define ID_AIC7902_B			0x801D9005FFFF9005ull
-#define ID_AHA_39320			0x8010900500409005ull
-#define ID_AHA_39320_B			0x8015900500409005ull
-#define ID_AHA_39320A			0x8016900500409005ull
-#define ID_AHA_39320D			0x8011900500419005ull
-#define ID_AHA_39320D_B			0x801C900500419005ull
-#define ID_AHA_39320D_HP		0x8011900500AC0E11ull
-#define ID_AHA_39320D_B_HP		0x801C900500AC0E11ull
-#define ID_AIC7902_PCI_REV_A4		0x3
-#define ID_AIC7902_PCI_REV_B0		0x10
-#define SUBID_HP			0x0E11
+#include "aic79xx_pci.h"
 
 #define DEVID_9005_TYPE(id) ((id) & 0xF)
 #define		DEVID_9005_TYPE_HBA		0x0	/* Standard Card */
drivers/scsi/aic7xxx/aic79xx_osm_pci.c renamed from drivers/scsi/aic7xxx/aic79xx_osm_pci.c-o +28 -5
diff -u linux/drivers/scsi/aic7xxx/aic79xx_osm_pci.c-o linux/drivers/scsi/aic7xxx/aic79xx_osm_pci.c
@@ -41,6 +41,7 @@
 
 #include "aic79xx_osm.h"
 #include "aic79xx_inline.h"
+#include "aic79xx_pci.h"
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,4,0)
 struct pci_device_id
@@ -58,12 +59,34 @@
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
 static void	ahd_linux_pci_dev_remove(struct pci_dev *pdev);
 
-/* We do our own ID filtering.  So, grab all SCSI storage class devices. */
 static struct pci_device_id ahd_linux_pci_id_table[] = {
-	{
-		0x9005, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID,
-		PCI_CLASS_STORAGE_SCSI << 8, 0xFFFF00, 0
-	},
+#define LINUXID(x,s) (unsigned)((((x) >> s) & 0xffff) ?: PCI_ANY_ID)
+#define ID(x) \
+	{ \
+		LINUXID(x,32), LINUXID(x,48), LINUXID(x,0), LINUXID(x,16), \
+		PCI_CLASS_STORAGE_SCSI << 8, 0xFFFF00, 0 \
+	}
+#define ID4(x,y) \
+	ID(x | ((y+0)<<48)), ID(x | ((y+1)<<48)), ID(x | ((y+2)<<48)), \
+	ID(x | ((y+3)<<48))
+#define ID16(x) ID4(x,0ULL), ID4(x,4ULL), ID4(x,8ULL), ID4(x,12ULL)
+	ID(ID_AHA_29320A),
+	ID(ID_AHA_29320ALP),
+	ID(ID_AHA_29320),
+	ID(ID_AHA_29320B),
+	ID(ID_AHA_29320LP),
+	ID(ID_AHA_39320),
+	ID(ID_AHA_39320_B),
+	ID(ID_AHA_39320A),
+	ID(ID_AHA_39320D),
+	ID(ID_AHA_39320D_HP),
+	ID(ID_AHA_39320D_B),
+	ID(ID_AHA_39320D_B_HP),
+	ID(ID_AHA_29320),
+	ID(ID_AHA_29320B),
+	ID(ID_AIC7901 & ID_DEV_VENDOR_MASK),
+	ID(ID_AIC7901A & ID_DEV_VENDOR_MASK),
+	ID16(ID_AIC7902),
 	{ 0 }
 };
 
drivers/scsi/aic7xxx/aic79xx_pci.h new file +41 -0

41 lines not on this page · raw message

drivers/scsi/aic7xxx/aic7xxx_pci.h new file +92 -0

92 lines not on this page · raw message

drivers/scsi/aic7xxx/aic7xxx_pci.c renamed from drivers/scsi/aic7xxx/aic7xxx_pci.c-o +2 -92

103 lines not on this page · raw message

drivers/scsi/aic7xxx/aic7xxx_osm_pci.c renamed from drivers/scsi/aic7xxx/aic7xxx_osm_pci.c-o +71 -9

93 lines not on this page · raw message

#4 lore raw
In reply to Andi Kleen

Thank you Andi for the patches. aic7xxx drivers are due for update to use the new timeout infrastructure. I'll incorporate them in, and do some testing.

Thanks,

Signature · 1 line
Luben
Quoted 26 lines · Andi Kleen

[I didn't find an aic7xxx maintainer in MAINTAINERS, so just sending to the list]

This is needed for 2.6 hotplug where the driver is autoloaded. When you have multiple conflicting entries the hotplug module loader usually loads the first one listed, which may be correct or may be not.

With these changes the drivers announce the correct PCI IDs.

The patches are quite big (i had to move the ID tables into an include), but mostly mechanical.

I only tested it on a single 78xx, so it's possible that i made a typo somewhere. Some testing would be appreciated, especially on 79xx

There is still an conflicting entry for aic7xxx_old unfortunately. This is hard to handle in hotplug. I think the right solution would be to somehow tag drivers as obsolete so that they aren't loaded by default. This will need an extension in the module pci table mechanism. I'm thinking about a a new MODULE_OBSOLETE macro for this and a obsolete_modules table generated at modules_install time. But that's for another patch. Better would be maybe to just drop the aic7xxx_old driver. Is it really still needed?

-Andi

#5 lore raw
Quoted 3 lines · Luben Tuikov

Thank you Andi for the patches. aic7xxx drivers are due for update to use the new timeout infrastructure. I'll incorporate them in, and do some testing.

cool.
Would you please consider the update to be done in small incremental changes/steps ? Eg each separate change/bugfix a separate patch ? Would make life a lot easier for me (distro kernel guy) and for James to be able to diagnose regressions a lot better by binary searching changesets etc etc.

Greetings,
Arjan van de Ven

author
#8 lore raw
In reply to Luben Tuikov
Quoted 3 lines · On Mon, 21 Jun 2004 11:30:04 -0400 Luben Tuikov <[email protected]>…

Thank you Andi for the patches. aic7xxx drivers are due for update to use the new timeout infrastructure. I'll incorporate them in, and do some testing.

Thanks. Please use the second patch i posted, not the first.

-Andi

#6 lore raw

Re: [PATCH] Add proper module ID tables to Adaptec aic7[9x]xxdrivers

In reply to Arjan van de Ven
Quoted 7 lines · Arjan van de Ven
Quoted 3 lines · Luben Tuikov

Thank you Andi for the patches. aic7xxx drivers are due for update to use the new timeout infrastructure. I'll incorporate them in, and do some testing.

Quoted 6 lines

cool.
Would you please consider the update to be done in small incremental changes/steps ? Eg each separate change/bugfix a separate patch ? Would make life a lot easier for me (distro kernel guy) and for James to be able to diagnose regressions a lot better by binary searching changesets etc etc.

Yep, no problem.

Current kernel version of, say, aic79xxx is 1.3.11, and I have 2.0.12, they are 226 patches away. Each patch is a submission into perforce. I can extract each and every one of those 226 and can post them somewhere on a website (hosting anyone? ;-) ), so that anyone can review them.

I can also provide a single patch from 1.3.11 to 2.0.11 so that a "to<-->from" picture can be seen as some patches in between may "cancel" each other out.

And I'll also provide a bk-send and a tar.gz of the whole thing as it looks in my tree.

Let me know if anything else is needed.

Thanks,

Signature · 1 line
Luben
#9 lore raw
In reply to Andi Kleen
Quoted 8 lines · Andi Kleen
Quoted 3 lines · On Mon, 21 Jun 2004 11:30:04 -0400 Luben Tuikov <[email protected]>…

Thank you Andi for the patches. aic7xxx drivers are due for update to use the new timeout infrastructure. I'll incorporate them in, and do some testing.

Thanks. Please use the second patch i posted, not the first.

No problem.

Signature · 1 line
Luben
#3 lore raw
In reply to Andi Kleen
Quoted 13 lines · Andi Kleen
Quoted 10 lines · On Mon, 21 Jun 2004 17:44:08 +0400 Sergey Vlasov <[email protected]> wrote:
Quoted 5 lines · On Mon, 21 Jun 2004 16:14:41 +0200 Andi Kleen wrote:

This is needed for 2.6 hotplug where the driver is autoloaded. When you have multiple conflicting entries the hotplug module loader usually loads the first one listed, which may be correct or may be not.

With these changes the drivers announce the correct PCI IDs.

Unfortunately, the patch does not seem to be correct :(

Quoted 8 lines
Quoted 4 lines

struct pci_device_id does not have the mask field, therefore the ID_9005_GENERIC_MASK restriction cannot be specified other than by listing all 16 possible IDs as separate entries. Your patch adds only one entry, thus losing 15 other possible IDs.

Hmm, good point. Thanks for catching this.

Here is a new patch. Does this one look better?

It fixes the above problem, but it's hard to say that the patch is correct without carefully checking all the tables. I'm trying to hack up something to autogenerate the pci_device_id table from the aic7xxx internal table.

BTW, ID_AIC7810 and ID_AIC7815 probably should not be in the PCI ID table at all - ahc_raid_setup() just prints "RAID functionality unsupported" for them. And some more IDs generated by ID16 are really rejected by the driver due to the (ID_9005_SISL_ID, ID_9005_SISL_MASK) exclusion entry.

#7 lore raw

Re: [PATCH] Add proper module ID tables to Adaptec aic7[9x]xxdrivers

In reply to Luben Tuikov
Quoted 15 lines · Luben Tuikov
Quoted 7 lines · Arjan van de Ven
Quoted 3 lines · Luben Tuikov

Thank you Andi for the patches. aic7xxx drivers are due for update to use the new timeout infrastructure. I'll incorporate them in, and do some testing.

Quoted 6 lines

cool.
Would you please consider the update to be done in small incremental changes/steps ? Eg each separate change/bugfix a separate patch ? Would make life a lot easier for me (distro kernel guy) and for James to be able to diagnose regressions a lot better by binary searching changesets etc etc.

Quoted 4 lines

Yep, no problem.

Current kernel version of, say, aic79xxx is 1.3.11, and I have 2.0.12, they are 226 patches away. Each patch is a submission

I was actually hoping you would have started from the kernel.org driver not the adaptec weird fork justin has been maintaining ;(

Quoted 3 lines

into perforce. I can extract each and every one of those 226 and can post them somewhere on a website (hosting anyone? ;-) ), so that anyone can review them.

it's not just about review, it's also about merging it in parts so that you don't need to binary chop through 226 patches if something fails, but only through like 30 or so.
And then a week later the next 30 go in etc etc