Skip to contentSkip to conversation

scsi/initio: section mismatches with HOTPLUG=n

PATCH In review Randy Dunlap 3 participants 7 messages Nov 20, 2006 – Nov 26, 2006 SCSI

lore
Share & tools
lore

Randy Dunlap

author
#0 lore raw

PATCH 1/1 scsi/initio: section mismatches with HOTPLUG=n

From: Randy Dunlap <[email protected]>

WARNING: drivers/scsi/initio.o - Section mismatch: reference to .init.data: from .text between 'i91u_detect' (at offset 0x26e8) and 'i91uSCBPost'
WARNING: drivers/scsi/initio.o - Section mismatch: reference to .init.data:i91u_pci_devices from .text between 'i91u_detect' (at offset 0x26ef) and 'i91uSCBPost'

Signed-off-by Randy Dunlap <[email protected]>
---
 drivers/scsi/initio.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/scsi/initio.c +1 -1
@@ -170,7 +170,7 @@ static int setup_debug = 0;
 static void i91uSCBPost(BYTE * pHcb, BYTE * pScb);
 
 /* PCI Devices supported by this driver */
-static struct pci_device_id i91u_pci_devices[] __devinitdata = {
+static struct pci_device_id i91u_pci_devices[] = {
 	{ PCI_VENDOR_ID_INIT,  I950_DEVICE_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
 	{ PCI_VENDOR_ID_INIT,  I940_DEVICE_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
 	{ PCI_VENDOR_ID_INIT,  I935_DEVICE_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},

---

#1 lore raw
Quoted 4 lines · Randy Dunlap

From: Randy Dunlap <[email protected]>

WARNING: drivers/scsi/initio.o - Section mismatch: reference to .init.data: from .text between 'i91u_detect' (at offset 0x26e8) and 'i91uSCBPost'
WARNING: drivers/scsi/initio.o - Section mismatch: reference to .init.data:i91u_pci_devices from .text between 'i91u_detect' (at offset 0x26ef) and 'i91uSCBPost'

Hmm, this apparently got into the tree unreviewed via Ben Collins:

commit 012887327875915f76a6208e81fe0d67a682ec15

Author: Ben Collins <[email protected]>
Date:   Wed Oct 18 08:40:57 2006 -0400

[initio] Add pci dev table for module auto loading.

Signed-off-by Ben Collins <[email protected]>

Ben, for future reference, would you send SCSI patches over the SCSI list, please. Secondly, do you intend to complete this change? What Randy proposes will restore the driver to its original state. However, it could also be fixed properly by adding __devinit to all the relevant PCI initialisation routines, which would be preferable.

James

#2 lore raw
Quoted 14 lines · James Bottomley
Quoted 4 lines · Randy Dunlap

From: Randy Dunlap <[email protected]>

WARNING: drivers/scsi/initio.o - Section mismatch: reference to .init.data: from .text between 'i91u_detect' (at offset 0x26e8) and 'i91uSCBPost'
WARNING: drivers/scsi/initio.o - Section mismatch: reference to .init.data:i91u_pci_devices from .text between 'i91u_detect' (at offset 0x26ef) and 'i91uSCBPost'

Hmm, this apparently got into the tree unreviewed via Ben Collins:

commit 012887327875915f76a6208e81fe0d67a682ec15

Author: Ben Collins <[email protected]>
Date:   Wed Oct 18 08:40:57 2006 -0400

[initio] Add pci dev table for module auto loading.

Quoted 7 lines
Signed-off-by Ben Collins <[email protected]>

Ben, for future reference, would you send SCSI patches over the SCSI list, please. Secondly, do you intend to complete this change? What Randy proposes will restore the driver to its original state. However, it could also be fixed properly by adding __devinit to all the relevant PCI initialisation routines, which would be preferable.

The patch went in with about half a dozen others for similar changes, so it was a group set.

Sorry about the mismatches. I've only been building with HOTPLUG=y.

I can do a patch if you want, just let me know.

#3 lore raw
Quoted 6 lines · Ben Collins

The patch went in with about half a dozen others for similar changes, so it was a group set.

Sorry about the mismatches. I've only been building with HOTPLUG=y.

I can do a patch if you want, just let me know.

Well, I was sort of hoping you weren't just randomly introducing __devinitdata but had all the necessary support patches for converting the driver to hotplug. However, if that's not the case, the safest thing to do is remove the section annotation ... I don't think anyone cares enough about this driver to convert it fully and test it.

James

#4 lore raw
Quoted 6 lines · James Bottomley
Quoted 5 lines · Ben Collins

The patch went in with about half a dozen others for similar changes, so it was a group set.

Sorry about the mismatches. I've only been building with HOTPLUG=y.

Quoted 7 lines
Quoted 1 line

I can do a patch if you want, just let me know.

Well, I was sort of hoping you weren't just randomly introducing __devinitdata but had all the necessary support patches for converting the driver to hotplug. However, if that's not the case, the safest thing to do is remove the section annotation ... I don't think anyone cares enough about this driver to convert it fully and test it.

The MODULE_DEV_TABLE was added so userspace has an easier time loading a driver for the device, without special rules. It wasn't meant to make the driver hot-pluggable. I think the idea that a module alias is just for hotplug is a little short-sighted.

#5 lore raw
Quoted 5 lines · Ben Collins

The MODULE_DEV_TABLE was added so userspace has an easier time loading a
driver for the device, without special rules. It wasn't meant to make the driver hot-pluggable. I think the idea that a module alias is just for hotplug is a little short-sighted.

The module device table isn't the problem. The problem is the fact that you marked the table __devinitdata. __devinitdata is a sectional marking that causes the table to be placed in a discardable init data section for CONFIG_HOTPLUG=n. However, the routines accessing it still occur in the standard text segment hence the mismatch. If you want to mark data __devinitdata, you have to ensure that all the routines accessing it are marked __devinit.

James

#6 lore raw
Quoted 7 lines · James Bottomley
Quoted 5 lines · Ben Collins

The MODULE_DEV_TABLE was added so userspace has an easier time loading a
driver for the device, without special rules. It wasn't meant to make the driver hot-pluggable. I think the idea that a module alias is just for hotplug is a little short-sighted.

Quoted 7 lines

The module device table isn't the problem. The problem is the fact that you marked the table __devinitdata. __devinitdata is a sectional marking that causes the table to be placed in a discardable init data section for CONFIG_HOTPLUG=n. However, the routines accessing it still occur in the standard text segment hence the mismatch. If you want to mark data __devinitdata, you have to ensure that all the routines accessing it are marked __devinit.

I know what the section is :) I wasn't aware that it was causing a problem with HOTPLUG=n. I'll get a patch to fix it up.