Skip to contentSkip to conversation

ide controller quirk to correct bad

PATCH Prarit Bhargava 2 participants 4 messages Feb 13, 2007 – Feb 15, 2007

lore
Share & tools
lore
#0 lore raw
Quoted 7 lines · Yanmin

If ide controllers are at legacy mode, only the 4th BAR is needed, so some BIOS initiate other BAR with incorrect value. ata/ata_piix calls pci_enable_device on the ide controller, which will check BAR resources. If the BAR resource values are incorrect, pci_enable_device will fail, and ata/ata_piix couldn't attach the ide controller.

Quoted 5 lines

Below patch against 2.6.20 creates a quirk to correct the bad BAR resources for a special ide controller which is popular on tiger-4.

If I understand the use of quirks, it is to fix hardware issues that cannot be resolved by bios fixes, etc.. ie) real HW problems. At least that's been my feeble understanding. If I'm wrong on that please correct me.

Putting this sort of fix in opens up the kernel to resolving many vendors' bios issues within the kernel.

I do understand that this is a special case -- it is unlikely a new bios will ship for this box. The way I see it, a future user of this platform will have to build kernels that use the old ide-cd/piix driver and/or patch the specific OS they are using with this patch.

Bottom line: I'm not convinced this belongs in the upstream kernel, but is a good candidate for Fedora.

P.

Quoted 71 lines

---

arch/ia64/kernel/quirks.c +47 -0
@@ -0,0 +1,47 @@
+/*
+ * This file contains work-arounds for ia64 platform bugs.
+ */
+#include <linux/pci.h>
+
+/*
+ * quirk_intel_ide_controller: If an ide/ata controller is
+ * at legacy mode, BIOS might initiates BAR(bar 0~3 and 5)
+ * with incorrect value. This quirk will reset the incorrect
+ * value to 0.
+ */
+static void __devinit quirk_intel_ide_controller(struct pci_dev *dev)
+{
+	unsigned int pos;
+	struct resource *res;
+	int fixed = 0;
+	u8 tmp8;
+	u8 mask;
+
+	if ((dev->class >> 8) != PCI_CLASS_STORAGE_IDE)
+		return;
+
+	/* TODO: What if one channel is in native mode ... */
+	pci_read_config_byte(dev, PCI_CLASS_PROG, &tmp8);
+	mask = (1 << 2) | (1 << 0);
+	if ((tmp8 & mask) = mask)
+		return;
+
+	for( pos = 0; pos < 6; pos ++ ) {
+		res = &dev->resource[pos];
+		if (!(res->flags & (IORESOURCE_IO | IORESOURCE_MEM)))
+			continue;
+
+		if (!res->start && res->end) {
+			res->start = res->end = 0;
+			res->flags = 0;
+			fixed = 1;
+		}
+	}
+	if (fixed)
+		printk(KERN_WARNING
+			"PCI device %s: BIOS resource configuration fixed.\n",
+			pci_name(dev));
+}
+
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82801DB_11, quirk_intel_ide_controller);
+
arch/ia64/kernel/Makefile +1 -1
@@ -33,6 +33,7 @@ obj-$(CONFIG_CRASH_DUMP)	+= crash_dump.o
 obj-$(CONFIG_IA64_UNCACHED_ALLOCATOR)	+= uncached.o
 obj-$(CONFIG_AUDIT)		+= audit.o
 obj-$(CONFIG_PCI_MSI)		+= msi_ia64.o
+obj-$(CONFIG_PCI)		+= quirks.o
 mca_recovery-y			+= mca_drv.o mca_drv_asm.o
 
 obj-$(CONFIG_IA64_ESI)		+= esi.o
 
--

Fedora-ia64-list mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/fedora-ia64-list

#1 lore raw

RE: [Fedora-ia64-list] [PATCH] ide controller quirk to correct

Quoted 5 lines · Tony

Review

Quoted 3 lines
(quoted diff) +3 -0
+	pci_read_config_byte(dev, PCI_CLASS_PROG, &tmp8);
+	mask = (1 << 2) | (1 << 0);
+	if ((tmp8 & mask) = mask)

Are there no symbolic defines for these bits?

No.

Quoted 5 lines

If there are, then
use them. If not, then does it really help readability to assemble the bits by hand? Why not just use:

	if ((tmp8 & 5) = 5)

Ok, I will change it.

Quoted 7 lines

-Tony

P.S. The BIOS per se isn't to blame here. If you dump pci space
from the EFI prompt (use "pci 0 1f 1") you'll see that the BIOS really hasn't initialized the BARs at all.Perhaps some stuff in ACPI sets them up (wrongly)?

I checked the BARs under EFI prompt. BAR 5 is 0 which looks like not initialized by BIOS. But my opinion is BIOS really initiates BAR 5 as 0, or it should change it to non-zero but it doesn't.
Function pci_read_bases reads all 6 bars from the device configuraton space and initiates dev->resource[]. When a BAR=0, if the sz (size) is not zero, the resource[].start will be set to 0 and resource[].end will be set to sz. resource[].flag is IORESOURCE_MEM. Such data is realy the same thing we could see under /sys/devices/_bus_/_deviceid_/resource after kernel boot.

That behavior is also consistent with what Alan replied on linux-ide maillist.

Below is the new patch. Thank Tony.

---

If ide controllers are at legacy mode, only the 4th BAR is needed, so some BIOS initiate other BAR with incorrect value. ata/ata_piix calls pci_enable_device on the ide controller, which will check BAR resources. If the BAR resource values are incorrect, pci_enable_device will fail, and ata/ata_piix couldn't attach the ide controller.

Below patch against 2.6.20 creates a quirk to correct the bad BAR resources for a special ide controller which is popular on tiger-4.

---

arch/ia64/kernel/quirks.c +45 -0
@@ -0,0 +1,45 @@
+/*
+ * This file contains work-arounds for ia64 platform bugs.
+ */
+#include <linux/pci.h>
+
+/*
+ * quirk_intel_ide_controller: If an ide/ata controller is
+ * at legacy mode, BIOS might initiates BAR(bar 0~3 and 5)
+ * with incorrect value. This quirk will reset the incorrect
+ * value to 0.
+ */
+static void __devinit quirk_intel_ide_controller(struct pci_dev *dev)
+{
+	unsigned int pos;
+	struct resource *res;
+	int fixed = 0;
+	u8 tmp8;
+
+	if ((dev->class >> 8) != PCI_CLASS_STORAGE_IDE)
+		return;
+
+	/* TODO: What if one channel is in native mode ... */
+	pci_read_config_byte(dev, PCI_CLASS_PROG, &tmp8);
+	if ((tmp8 & 5) = 5)
+		return;
+
+	for( pos = 0; pos < 6; pos ++ ) {
+		res = &dev->resource[pos];
+		if (!(res->flags & (IORESOURCE_IO | IORESOURCE_MEM)))
+			continue;
+
+		if (!res->start && res->end) {
+			res->start = res->end = 0;
+			res->flags = 0;
+			fixed = 1;
+		}
+	}
+	if (fixed)
+		printk(KERN_WARNING
+			"PCI device %s: BIOS resource configuration fixed.\n",
+			pci_name(dev));
+}
+
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82801DB_11, quirk_intel_ide_controller);
+
arch/ia64/kernel/Makefile +1 -0
@@ -33,6 +33,7 @@ obj-$(CONFIG_CRASH_DUMP)	+= crash_dump.o
 obj-$(CONFIG_IA64_UNCACHED_ALLOCATOR)	+= uncached.o
 obj-$(CONFIG_AUDIT)		+= audit.o
 obj-$(CONFIG_PCI_MSI)		+= msi_ia64.o
+obj-$(CONFIG_PCI)		+= quirks.o
 mca_recovery-y			+= mca_drv.o mca_drv_asm.o
 
 obj-$(CONFIG_IA64_ESI)		+= esi.o
#2 lore raw
In reply to Prarit Bhargava
Quoted 15 lines · Prarit Bhargava
Quoted 7 lines · Yanmin

If ide controllers are at legacy mode, only the 4th BAR is needed, so some BIOS initiate other BAR with incorrect value. ata/ata_piix calls pci_enable_device on the ide controller, which will check BAR resources. If the BAR resource values are incorrect, pci_enable_device will fail, and ata/ata_piix couldn't attach the ide controller.

Quoted 5 lines

Below patch against 2.6.20 creates a quirk to correct the bad BAR resources for a special ide controller which is popular on tiger-4.

Quoted 4 lines

If I understand the use of quirks, it is to fix hardware issues that cannot be resolved by bios fixes, etc.. ie) real HW problems. At least that's been my feeble understanding. If I'm wrong on that please correct me.

Correct.

Curent issue of ide controller also could be considered as ide hardware issue. If the ide controller is at legacy mode, bar 4 is enough, but my ide controller provides bar 5 as well as 4. If the controller hardwires bar 5 as 0, there will be no such issue.

Quoted 8 lines

Putting this sort of fix in opens up the kernel to resolving many vendors' bios issues within the kernel.

I do understand that this is a special case -- it is unlikely a new bios will ship for this box. The way I see it, a future user of this platform will have to build kernels that use the old ide-cd/piix driver and/or patch the specific OS they are using with this patch.

From Alan's reply on linux-ide maillist, we could know another platform PowerPC has the similiar issue. It assumes the BAR is not used if it is equal to 0.

From the comments of function ide_pci_enable of the old ide driver, we could see that such issue is not rare.

Is it better to put the patch into upstream kernel?

Yanmin

#3 lore raw

Re: [Fedora-ia64-list] [PATCH] ide controller quirk to correct badBAR

In reply to Prarit Bhargava
Quoted 11 lines · Yanmin
Quoted 6 lines · Tony

Review

Quoted 4 lines
(quoted diff) +3 -0
+	pci_read_config_byte(dev, PCI_CLASS_PROG, &tmp8);
+	mask = (1 << 2) | (1 << 0);
+	if ((tmp8 & mask) = mask)

Are there no symbolic defines for these bits?

No.

Quoted 16 lines
Quoted 6 lines

If there are, then
use them. If not, then does it really help readability to assemble the bits by hand? Why not just use:

	if ((tmp8 & 5) = 5)

Ok, I will change it.

Quoted 7 lines

-Tony

P.S. The BIOS per se isn't to blame here. If you dump pci space
from the EFI prompt (use "pci 0 1f 1") you'll see that the BIOS really hasn't initialized the BARs at all.Perhaps some stuff in ACPI sets them up (wrongly)?

Oops. I have a bad habit of usering BIOS/ & ACPI interchangeably .. sorry for the confusion.

Quoted 9 lines

I checked the BARs under EFI prompt. BAR 5 is 0 which looks like not initialized by BIOS. But my opinion is BIOS really initiates BAR 5 as 0, or it should change it to non-zero but it doesn't.
Function pci_read_bases reads all 6 bars from the device configuraton space and initiates dev->resource[]. When a BAR=0, if the sz (size) is not zero, the resource[].start will be set to 0 and resource[].end will be set to sz. resource[].flag is IORESOURCE_MEM. Such data is realy the same thing we could see under /sys/devices/_bus_/_deviceid_/resource after kernel boot.

Quoted 3 lines

That behavior is also consistent with what Alan replied on linux-ide maillist.

IIRC, the case that Alan was talking about was a case where the powerpc arch had devices that required start =0, not this case where the ACPI table info is inocorrect.

P.

Quoted 84 lines

Below is the new patch. Thank Tony.

---

If ide controllers are at legacy mode, only the 4th BAR is needed, so some BIOS initiate other BAR with incorrect value. ata/ata_piix calls pci_enable_device on the ide controller, which will check BAR resources. If the BAR resource values are incorrect, pci_enable_device will fail, and ata/ata_piix couldn't attach the ide controller.

Below patch against 2.6.20 creates a quirk to correct the bad BAR resources for a special ide controller which is popular on tiger-4.

---

arch/ia64/kernel/quirks.c +45 -0
@@ -0,0 +1,45 @@
+/*
+ * This file contains work-arounds for ia64 platform bugs.
+ */
+#include <linux/pci.h>
+
+/*
+ * quirk_intel_ide_controller: If an ide/ata controller is
+ * at legacy mode, BIOS might initiates BAR(bar 0~3 and 5)
+ * with incorrect value. This quirk will reset the incorrect
+ * value to 0.
+ */
+static void __devinit quirk_intel_ide_controller(struct pci_dev *dev)
+{
+	unsigned int pos;
+	struct resource *res;
+	int fixed = 0;
+	u8 tmp8;
+
+	if ((dev->class >> 8) != PCI_CLASS_STORAGE_IDE)
+		return;
+
+	/* TODO: What if one channel is in native mode ... */
+	pci_read_config_byte(dev, PCI_CLASS_PROG, &tmp8);
+	if ((tmp8 & 5) = 5)
+		return;
+
+	for( pos = 0; pos < 6; pos ++ ) {
+		res = &dev->resource[pos];
+		if (!(res->flags & (IORESOURCE_IO | IORESOURCE_MEM)))
+			continue;
+
+		if (!res->start && res->end) {
+			res->start = res->end = 0;
+			res->flags = 0;
+			fixed = 1;
+		}
+	}
+	if (fixed)
+		printk(KERN_WARNING
+			"PCI device %s: BIOS resource configuration fixed.\n",
+			pci_name(dev));
+}
+
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_82801DB_11, quirk_intel_ide_controller);
+
arch/ia64/kernel/Makefile +1 -1
@@ -33,6 +33,7 @@ obj-$(CONFIG_CRASH_DUMP)	+= crash_dump.o
 obj-$(CONFIG_IA64_UNCACHED_ALLOCATOR)	+= uncached.o
 obj-$(CONFIG_AUDIT)		+= audit.o
 obj-$(CONFIG_PCI_MSI)		+= msi_ia64.o
+obj-$(CONFIG_PCI)		+= quirks.o
 mca_recovery-y			+= mca_drv.o mca_drv_asm.o
 
 obj-$(CONFIG_IA64_ESI)		+= esi.o
 
--

Fedora-ia64-list mailing list
[email protected]
https://www.redhat.com/mailman/listinfo/fedora-ia64-list