Skip to contentSkip to conversation

fix aic7xxx probing

PATCH In review James Bottomley 2 participants 7 messages Jun 24, 2004 – Jun 25, 2004

Subsystems (2)
SCSI AIC7XXX / AIC79XX SCSI driver
lore
Share & tools
lore

James Bottomley

author
#0 lore raw

PATCH 1/1 fix aic7xxx probing

It looks like there's still a problem with the probing:

ahc_linux_pci_init() returns the number of boards attached ahc_linux_eisa_init() returns 0 on success or error

The only reason the current code actually works on PCI seems to be that it's already attached the devices by the time it returns what the ahc_linux_detect() assumes to be a failure code.

James

===== drivers/scsi/aic7xxx/aic7xxx_osm.c 1.54 vs edited =====

drivers/scsi/aic7xxx/aic7xxx_osm.c +5 -11
@@ -842,9 +842,7 @@
 static int
 ahc_linux_detect(Scsi_Host_Template *template)
 {
-	struct	ahc_softc *ahc;
 	int     found = 0;
-	int	eisa_err, pci_err;
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 	/*

@@ -892,13 +890,11 @@
 	 */
 	ahc_list_lockinit();
 
-	pci_err = ahc_linux_pci_init();
-	eisa_err = ahc_linux_eisa_init();
-
-	if(pci_err && eisa_err)
-		goto out;
-
-
+	found = ahc_linux_pci_init();
+	if (!ahc_linux_eisa_init())
+		found++;
+	
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 	/*
 	 * Register with the SCSI layer all
 	 * controllers we've found.

@@ -908,9 +904,7 @@
 		if (ahc_linux_register_host(ahc, template) == 0)
 			found++;
 	}
-out:
 
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 	spin_lock_irq(&io_request_lock);
 #endif
 	aic7xxx_detect_complete++;

===== drivers/scsi/aic7xxx/aic7xxx_osm.h 1.54 vs edited =====

drivers/scsi/aic7xxx/aic7xxx_osm.h +1 -1
@@ -940,7 +940,7 @@
 }
 #else
 static inline int ahc_linux_pci_init(void) {
-	return -ENODEV;
+	return 0;
 }
 static inline void ahc_linux_pci_exit(void) {
 }
#1 lore raw
Quoted 8 lines · James Bottomley

It looks like there's still a problem with the probing:

ahc_linux_pci_init() returns the number of boards attached ahc_linux_eisa_init() returns 0 on success or error

The only reason the current code actually works on PCI seems to be that it's already attached the devices by the time it returns what the ahc_linux_detect() assumes to be a failure code.

This patch applied against 2.6.7-mm2 doesn't detect my PCI HBA whereas 2.6.7-mm2 currently does.

author
#2 lore raw
Quoted 2 lines · Zwane Mwaikambo

This patch applied against 2.6.7-mm2 doesn't detect my PCI HBA whereas 2.6.7-mm2 currently does.

Thanks, had another look at the code and it seems that the meaning of the return value for ahc_linux_pci_init() is different depending on kernel version, sigh. For < 2.4.0 it means number of boards found, otherwise it's zero or error.

We really need to scrub this 2.2 code out of the driver, but in the meantime, can you test the attached? (backing out the original). It just makes the return of ahc_linux_pci_init() consistent.

Thanks,

James

===== aic7xxx_osm.c 1.54 vs edited =====

aic7xxx_osm.c renamed from drivers/scsi/aic7xxx/aic7xxx_osm.c +5 -11
@@ -842,9 +842,7 @@
 static int
 ahc_linux_detect(Scsi_Host_Template *template)
 {
-	struct	ahc_softc *ahc;
 	int     found = 0;
-	int	eisa_err, pci_err;
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 	/*

@@ -892,13 +890,11 @@
 	 */
 	ahc_list_lockinit();
 
-	pci_err = ahc_linux_pci_init();
-	eisa_err = ahc_linux_eisa_init();
-
-	if(pci_err && eisa_err)
-		goto out;
-
-
+	found = ahc_linux_pci_init();
+	if (!ahc_linux_eisa_init())
+		found++;
+	
+#if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 	/*
 	 * Register with the SCSI layer all
 	 * controllers we've found.

@@ -908,9 +904,7 @@
 		if (ahc_linux_register_host(ahc, template) == 0)
 			found++;
 	}
-out:
 
-#if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 	spin_lock_irq(&io_request_lock);
 #endif
 	aic7xxx_detect_complete++;

===== aic7xxx_osm.h 1.54 vs edited =====

aic7xxx_osm.h renamed from drivers/scsi/aic7xxx/aic7xxx_osm.h +1 -1
@@ -940,7 +940,7 @@
 }
 #else
 static inline int ahc_linux_pci_init(void) {
-	return -ENODEV;
+	return 0;
 }
 static inline void ahc_linux_pci_exit(void) {
 }

===== aic7xxx_osm_pci.c 1.12 vs edited =====

aic7xxx_osm_pci.c renamed from drivers/scsi/aic7xxx/aic7xxx_osm_pci.c +2 -1
@@ -198,7 +198,8 @@
 ahc_linux_pci_init(void)
 {
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
-	return (pci_module_init(&aic7xxx_pci_driver));
+	/* Translate error or zero return into zero or one */
+	return pci_module_init(&aic7xxx_pci_driver) ? 0 : 1;
 #else
 	struct pci_dev *pdev;
 	u_int class;
#3 lore raw
Quoted 4 lines · James Bottomley
Quoted 2 lines · Zwane Mwaikambo

This patch applied against 2.6.7-mm2 doesn't detect my PCI HBA whereas 2.6.7-mm2 currently does.

Quoted 8 lines

Thanks, had another look at the code and it seems that the meaning of the return value for ahc_linux_pci_init() is different depending on kernel version, sigh. For < 2.4.0 it means number of boards found, otherwise it's zero or error.

We really need to scrub this 2.2 code out of the driver, but in the meantime, can you test the attached? (backing out the original). It just makes the return of ahc_linux_pci_init() consistent.

Doesn't detect the HBA with that patch either.

author
#4 lore raw
Quoted 1 line · Zwane Mwaikambo

Doesn't detect the HBA with that patch either.

Hmm, probably doesn't do direct host addition in certain cases. How about third time lucky?

James

===== drivers/scsi/aic7xxx/aic7xxx_osm.c 1.54 vs edited =====

drivers/scsi/aic7xxx/aic7xxx_osm.c +4 -9
@@ -844,7 +844,6 @@
 {
 	struct	ahc_softc *ahc;
 	int     found = 0;
-	int	eisa_err, pci_err;
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 	/*

@@ -892,13 +891,10 @@
 	 */
 	ahc_list_lockinit();
 
-	pci_err = ahc_linux_pci_init();
-	eisa_err = ahc_linux_eisa_init();
-
-	if(pci_err && eisa_err)
-		goto out;
-
-
+	found = ahc_linux_pci_init();
+	if (!ahc_linux_eisa_init())
+		found++;
+	
 	/*
 	 * Register with the SCSI layer all
 	 * controllers we've found.

@@ -908,7 +904,6 @@
 		if (ahc_linux_register_host(ahc, template) == 0)
 			found++;
 	}
-out:
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,5,0)
 	spin_lock_irq(&io_request_lock);

===== drivers/scsi/aic7xxx/aic7xxx_osm.h 1.54 vs edited =====

drivers/scsi/aic7xxx/aic7xxx_osm.h +1 -1
@@ -940,7 +940,7 @@
 }
 #else
 static inline int ahc_linux_pci_init(void) {
-	return -ENODEV;
+	return 0;
 }
 static inline void ahc_linux_pci_exit(void) {
 }

===== drivers/scsi/aic7xxx/aic7xxx_osm_pci.c 1.12 vs edited =====

drivers/scsi/aic7xxx/aic7xxx_osm_pci.c +2 -1
@@ -198,7 +198,8 @@
 ahc_linux_pci_init(void)
 {
 #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,4,0)
-	return (pci_module_init(&aic7xxx_pci_driver));
+	/* Translate error or zero return into zero or one */
+	return pci_module_init(&aic7xxx_pci_driver) ? 0 : 1;
 #else
 	struct pci_dev *pdev;
 	u_int class;
#5 lore raw
Quoted 5 lines · James Bottomley
Quoted 1 line · Zwane Mwaikambo

Doesn't detect the HBA with that patch either.

Hmm, probably doesn't do direct host addition in certain cases. How about third time lucky?

Indeed, third time is the charm ;)

Thanks,
Zwane

author
#6 lore raw
Quoted 1 line · Zwane Mwaikambo

Indeed, third time is the charm ;)

Thanks, for testing this; sorry I have so much difficulty working out how aic7xxx should be operating...

I'll apply this one...it should get my eagle back working again.

James