Fix libata resource conflict for legacy mode
PATCH 1/1 Fix libata resource conflict for legacy mode
Hi,
Someone reported a bug about not being able to use his box with ICH6 chipset unless booting with PnP disabled on the kernel command-line.
When failing, he had in the logs :
ata_piix 0000:00:1f.2: MAP [ P0 P2 IDE IDE ]
ACPI: PCI Interrupt 0000:00:1f.2[C] -> GSI 20 (level, low) -> IRQ 21
ata: 0x1f0 IDE port busy
After some debugging, it turns out that with PnP activated the resources are looking like this :
0100-01fe : pnp 00:09
0170-0177 : libata
01f0-01f7 : libata
Thus, with the check done in libata, it finds "pnp 00:09" instead of "libata" and fails.
The quick fix I made for this is to check wether the resource has a child or not and then check which child is conflicting.
This patch was tested on 2.6.17 but I rediffed it for 2.6.18-git as the same bug will likely occur. Of course, if preferred, I can rediff it for libata#upstream.
Also, I added some printks to print the name of the conflicting resource but this can also be removed if preferred.
Regards,
Arnaud
When the libata is trying to handle legacy ide ports (0x1f0 for instance), it doesn't take care if the resource has childs or not.
The result is that this situation :
0100-01fe : pnp 00:09
0170-0177 : libata
01f0-01f7 : libata
is seen as conflict, which is wrong. The 'pnp 00:09' part is there due to the PnP support. If PnP support is off, this doesn't appear.
The proposed fix is to detect childs and in this case, look at which child is conflicting.
---
drivers/scsi/libata-bmdma.c +6 -0
@@ -1016,10 +1016,13 @@ int ata_pci_init_one (struct pci_dev *pd res.start =0x1f0 ; res.end =0x1f0 +8 -1 ; conflict =____request_resource (&ioport_resource, &res); +if (conflict->child) + conflict =____request_resource (conflict,&res);if (!strcmp (conflict->name,"libata" )) legacy_mode |= (1 <<0 );else { disable_dev_on_err =0 ; +printk (KERN_WARNING "ata: conflict with %s\n" ,conflict->name);printk (KERN_WARNING "ata: 0x1f0 IDE port busy\n" ); } }else
@@ -1030,10 +1033,13 @@ int ata_pci_init_one (struct pci_dev *pd res.start =0x170 ; res.end =0x170 +8 -1 ; conflict =____request_resource (&ioport_resource, &res); +if (conflict->child) + conflict =____request_resource (conflict,&res);if (!strcmp (conflict->name,"libata" )) legacy_mode |= (1 <<1 );else { disable_dev_on_err =0 ; +printk (KERN_WARNING "ata: conflict with %s\n" ,conflict->name);printk (KERN_WARNING "ata: 0x170 IDE port busy\n" ); } }else
Hello,
Quoted 3 lines · Arnaud Patard
0100-01fe : pnp 00:09
0170-0177 : libata
01f0-01f7 : libata
Arghh...
[--snip--]
drivers/scsi/libata-bmdma.c int ata_pci_init_one (struct pci_dev *pd · line 1020
Quoted 10 lines
Index:
linux-2.6/drivers/scsi/libata-bmdma.c
===================================================================drivers/scsi/libata-bmdma.c +2
-0@@ -1016,10 +1016,13 @@ int ata_pci_init_one (struct pci_dev *pd res.start =0x1f0 ; res.end =0x1f0 +8 -1 ; conflict =____request_resource (&ioport_resource, &res); +if (conflict->child) + conflict =____request_resource (conflict,&res);
IMHO, having something which is slightly more generic would be better - e.g. looping till no child.
drivers/scsi/libata-bmdma.c int ata_pci_init_one (struct pci_dev *pd · line 1026
Quoted 6 lines
drivers/scsi/libata-bmdma.c +1
-0if (!strcmp (conflict->name,"libata" )) legacy_mode |= (1 <<0 );else { disable_dev_on_err =0 ; +printk (KERN_WARNING "ata: conflict with %s\n" ,conflict->name);printk (KERN_WARNING "ata: 0x1f0 IDE port busy\n" );
Please merge two printks into one. Other than that, I think printing conflict name is nice addition.
drivers/scsi/libata-bmdma.c int ata_pci_init_one (struct pci_dev *pd · line 1028
Quoted 2 lines
drivers/scsi/libata-bmdma.c +0
-0} }else
Thanks.
Signature · 1 line
tejun
Quoted 1 line · Tejun Heo
Hello,
Hi,
Quoted 5 lines
Quoted 5 lines
Index:
linux-2.6/drivers/scsi/libata-bmdma.c
===================================================================drivers/scsi/libata-bmdma.c +0
-0@@ -1016,10 +1016,13 @@ int ata_pci_init_one (struct pci_dev *pd
Quoted 8 lines
drivers/scsi/libata-bmdma.c int ata_pci_init_one (struct pci_dev *pd · line 1020
Quoted 5 lines
drivers/scsi/libata-bmdma.c +2
-0res.start =0x1f0 ; res.end =0x1f0 +8 -1 ; conflict =____request_resource (&ioport_resource, &res); +if (conflict->child) + conflict =____request_resource (conflict,&res);IMHO, having something which is slightly more generic would be better - e.g. looping till no child.
Ok. Done in the attached patch.
Quoted 9 lines
drivers/scsi/libata-bmdma.c int ata_pci_init_one (struct pci_dev *pd · line 1026
Quoted 7 lines
drivers/scsi/libata-bmdma.c +1
-0if (!strcmp (conflict->name,"libata" )) legacy_mode |= (1 <<0 );else { disable_dev_on_err =0 ; +printk (KERN_WARNING "ata: conflict with %s\n" ,conflict->name);printk (KERN_WARNING "ata: 0x1f0 IDE port busy\n" );Please merge two printks into one. Other than that, I think printing
Fixed too.
Regards,
Arnaud
When the libata is trying to handle legacy ide ports (0x1f0 for instance), it doesn't take care if the resource has childs or not.
The result is that this situation :
0100-01fe : pnp 00:09
0170-0177 : libata
01f0-01f7 : libata
is seen as conflict, which is wrong.
The proposed fix is to detect childs and in this case, look at which child is conflicting.
---
drivers/scsi/libata-bmdma.c +8 -2
@@ -1016,11 +1016,14 @@ int ata_pci_init_one (struct pci_dev *pd res.start =0x1f0 ; res.end =0x1f0 +8 -1 ; conflict =____request_resource (&ioport_resource, &res); +while (conflict->child) + conflict =____request_resource (conflict, &res);if (!strcmp (conflict->name,"libata" )) legacy_mode |= (1 <<0 );else { disable_dev_on_err =0 ; -printk (KERN_WARNING "ata: 0x1f0 IDE port busy\n" ); +printk (KERN_WARNING "ata: 0x1f0 IDE port busy\n" \ +"ata: conflict with %s\n" , conflict->name); } }else legacy_mode |= (1 <<0 );
@@ -1030,11 +1033,14 @@ int ata_pci_init_one (struct pci_dev *pd res.start =0x170 ; res.end =0x170 +8 -1 ; conflict =____request_resource (&ioport_resource, &res); +while (conflict->child) + conflict =____request_resource (conflict, &res);if (!strcmp (conflict->name,"libata" )) legacy_mode |= (1 <<1 );else { disable_dev_on_err =0 ; -printk (KERN_WARNING "ata: 0x170 IDE port busy\n" ); +printk (KERN_WARNING "ata: 0x170 IDE port busy\n" \ +"ata: conflict with %s\n" , conflict->name); } }else legacy_mode |= (1 <<1 );
Tejun Heo
Acked-byHello, again.
drivers/scsi/libata-bmdma.c int ata_pci_init_one (struct pci_dev *pd · line 1026
Quoted 3 lines · Arnaud Patard
drivers/scsi/libata-bmdma.c +2
-1-printk (KERN_WARNING "ata: 0x1f0 IDE port busy\n" ); +printk (KERN_WARNING "ata: 0x1f0 IDE port busy\n" \ +"ata: conflict with %s\n" , conflict->name);
I was thinking something like
"ata: 0x1f0 IDE port busy (conflict with %s)\n"
Other than that,
Signature · 1 line
tejun
Jeff Garzik
AppliedQuoted 11 lines · Arnaud Patard
When the libata is trying to handle legacy ide ports (
0x1f0for instance), it doesn't take care if the resource has childs or not. The result is that this situation :
0100-01fe : pnp 00:09
0170-0177 : libata
01f0-01f7 : libatais seen as conflict, which is wrong.
The proposed fix is to detect childs and in this case, look at which child is conflicting.Signed-off-by Arnaud Patard <[email protected]>
Applied. For the future, please inline both the patch description, and the patch itself. I had to apply this manually, rather than use the usual scripts.
For more info read:Documentation/SubmittingPatches
http://linux.yyz.us/patch-format.html
Jeff