[PATCH 0/5 v2] x86: boot interrupt reroute fix

[PATCH 0/5 v2] x86: boot interrupt reroute fix

[Stefan Assmann]

During the 2.6.29 merge parts of e1d3a90846b40ad3160bf4b648d36c6badad39ac got
lost because of other ACPI changes. This patchset gets the reroute code working
again with the current ACPI code. As suggested by Ingo Molnar the missing boot
interrupt reroute parts are re-added in separate files.
Therefore some smaller changes to the ACPI and PCI code are necessary to allow
the boot interrupt reroute code to access required structures and functionality.
In detail, the reroute code requires access to:
- struct acpi_prt_entry
- pci_find_upstream_pcie_bridge()

Additionally to not change any information in the global acpi_prt_list it is
desirable to have acpi_pci_irq_find_prt_entry() return a pointer to a copy of
a struct acpi_prt_entry instead of pointer into the global acpi_prt_list. The
callers of acpi_pci_irq_find_prt_entry() have been updated to free the
allocated structure as well.

By request of Bjorn Helgaas I've created a bugzilla entry at
https://bugzilla.kernel.org/show_bug.cgi?id=21882
to gather information about the boot interrupt reroute problem.

Jesse Barnes suggested to integrate the changes through the ACPI tree.

v2:
- split pci_find_upstream_pcie_bridge() changes into separate patches

[PATCH 1/5 v2] ACPI: move struct acpi_prt_entry to actbl1.h

[Stefan Assmann]

From: Stefan Assmann <sassmann@redhat.com>

Moving struct acpi_prt_entry from drivers/acpi/pci_irq.c to
include/acpi/actbl1.h to make it available outside of pci_irq.c. This will
allow the boot interrupt reroute code, and others, to operate with
acpi_prt_entrys.

Signed-off-by: Stefan Assmann <sassmann@redhat.com>
---
 drivers/acpi/pci_irq.c |    8 --------
 include/acpi/actbl1.h  |   10 ++++++++++
 2 files changed, 10 insertions(+), 8 deletions(-)

diff --git a/drivers/acpi/pci_irq.c b/drivers/acpi/pci_irq.c
index f907cfb..3bc2164 100644
--- a/drivers/acpi/pci_irq.c
+++ b/drivers/acpi/pci_irq.c
@@ -45,14 +45,6 @@
 #define _COMPONENT		ACPI_PCI_COMPONENT
 ACPI_MODULE_NAME("pci_irq");
 
-struct acpi_prt_entry {
-	struct list_head	list;
-	struct acpi_pci_id	id;
-	u8			pin;
-	acpi_handle		link;
-	u32			index;		/* GSI, or link _CRS index */
-};
-
 static LIST_HEAD(acpi_prt_list);
 static DEFINE_SPINLOCK(acpi_prt_lock);
 
diff --git a/include/acpi/actbl1.h b/include/acpi/actbl1.h
index c637b75..bbc46ed 100644
--- a/include/acpi/actbl1.h
+++ b/include/acpi/actbl1.h
@@ -752,6 +752,16 @@ struct acpi_madt_local_x2apic_nmi {
 	u8 reserved[3];
 };
 
+/* _PRT entry */
+
+struct acpi_prt_entry {
+	struct list_head	list;
+	struct acpi_pci_id	id;
+	u8			pin;
+	acpi_handle		link;
+	u32			index;		/* GSI, or link _CRS index */
+};
+
 /*
  * Common flags fields for MADT subtables
  */
-- 
1.7.3.2

[PATCH 2/5 v2] ACPI: make acpi_pci_irq_find_prt_entry return a copy of a prt_entry

[Stefan Assmann]

From: Stefan Assmann <sassmann@redhat.com>

By returning a pointer to a copy of struct acpi_prt_entry in
acpi_pci_irq_find_prt_entry() the calling code may modify the returned
structure without altering entries in the global acpi_prt_list. Thus the
changes to acpi_prt_entry by the boot interrupt reroute code won't change the
original PRT entries.
Currently there are only two calls for acpi_pci_irq_find_prt_entry() and both
are in acpi_pci_irq_lookup(). acpi_pci_irq_lookup() itself is used only by
acpi_pci_irq_enable() and acpi_pci_irq_disable() so both are also updated to
free the copy of struct acpi_prt_entry appropriately.

Signed-off-by: Stefan Assmann <sassmann@redhat.com>
---
 drivers/acpi/pci_irq.c |   11 +++++++++--
 1 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/acpi/pci_irq.c b/drivers/acpi/pci_irq.c
index 3bc2164..e1632b2 100644
--- a/drivers/acpi/pci_irq.c
+++ b/drivers/acpi/pci_irq.c
@@ -60,7 +60,7 @@ static inline char pin_name(int pin)
 static struct acpi_prt_entry *acpi_pci_irq_find_prt_entry(struct pci_dev *dev,
 							  int pin)
 {
-	struct acpi_prt_entry *entry;
+	struct acpi_prt_entry *entry, *entry_copy;
 	int segment = pci_domain_nr(dev->bus);
 	int bus = dev->bus->number;
 	int device = PCI_SLOT(dev->devfn);
@@ -72,7 +72,11 @@ static struct acpi_prt_entry *acpi_pci_irq_find_prt_entry(struct pci_dev *dev,
 		    && (device == entry->id.device)
 		    && (pin == entry->pin)) {
 			spin_unlock(&acpi_prt_lock);
-			return entry;
+			entry_copy = kzalloc(sizeof(struct acpi_prt_entry), GFP_KERNEL);
+			if (!entry_copy)
+				return NULL;
+			memcpy(entry_copy, entry, sizeof(struct acpi_prt_entry));
+			return entry_copy;
 		}
 	}
 	spin_unlock(&acpi_prt_lock);
@@ -384,6 +388,7 @@ int acpi_pci_irq_enable(struct pci_dev *dev)
 							 &link);
 		else
 			gsi = entry->index;
+		kfree(entry);
 	} else
 		gsi = -1;
 
@@ -453,6 +458,8 @@ void acpi_pci_irq_disable(struct pci_dev *dev)
 	else
 		gsi = entry->index;
 
+	if (entry)
+		kfree(entry);
 	/*
 	 * TBD: It might be worth clearing dev->irq by magic constant
 	 * (e.g. PCI_UNDEFINED_IRQ).
-- 
1.7.3.2

[PATCH 3/5 v2] PCI: Add locking to pci_find_upstream_pcie_bridge()

[Stefan Assmann]

From: Stefan Assmann <sassmann@redhat.com>

pci_find_upstream_pcie_bridge() should take the pci_bus_sem semaphore while
reading global PCI structures.

Signed-off-by: Stefan Assmann <sassmann@redhat.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
---
 drivers/pci/search.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/drivers/pci/search.c b/drivers/pci/search.c
index 9d75dc8..c8159b0 100644
--- a/drivers/pci/search.c
+++ b/drivers/pci/search.c
@@ -29,6 +29,8 @@ pci_find_upstream_pcie_bridge(struct pci_dev *pdev)
 
 	if (pci_is_pcie(pdev))
 		return NULL;
+
+	down_read(&pci_bus_sem);
 	while (1) {
 		if (pci_is_root_bus(pdev->bus))
 			break;
@@ -38,6 +40,7 @@ pci_find_upstream_pcie_bridge(struct pci_dev *pdev)
 			tmp = pdev;
 			continue;
 		}
+		up_read(&pci_bus_sem);
 		/* PCI device should connect to a PCIe bridge */
 		if (pdev->pcie_type != PCI_EXP_TYPE_PCI_BRIDGE) {
 			/* Busted hardware? */
@@ -46,6 +49,7 @@ pci_find_upstream_pcie_bridge(struct pci_dev *pdev)
 		}
 		return pdev;
 	}
+	up_read(&pci_bus_sem);
 
 	return tmp;
 }
-- 
1.7.3.2

[PATCH 4/5 v2] PCI: Add header declaration for pci_find_upstream_pcie_bridge()

[Stefan Assmann]

From: Stefan Assmann <sassmann@redhat.com>

Add header declaration for pci_find_upstream_pcie_bridge to pci.h. This will
allow the boot interrupt reroute code, and others, to make use of the
implementation.

Signed-off-by: Stefan Assmann <sassmann@redhat.com>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
---
 include/linux/pci.h |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)

diff --git a/include/linux/pci.h b/include/linux/pci.h
index 7454408..7acb6b2 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -698,6 +698,7 @@ int pci_bus_find_ext_capability(struct pci_bus *bus, unsigned int devfn,
 int pci_find_ht_capability(struct pci_dev *dev, int ht_cap);
 int pci_find_next_ht_capability(struct pci_dev *dev, int pos, int ht_cap);
 struct pci_bus *pci_find_next_bus(const struct pci_bus *from);
+struct pci_dev *pci_find_upstream_pcie_bridge(struct pci_dev *pdev);
 
 struct pci_dev *pci_get_device(unsigned int vendor, unsigned int device,
 				struct pci_dev *from);
@@ -1208,6 +1209,8 @@ static inline void pci_unblock_user_cfg_access(struct pci_dev *dev)
 
 static inline struct pci_bus *pci_find_next_bus(const struct pci_bus *from)
 { return NULL; }
+static inline struct pci_dev *pci_find_upstream_pcie_bridge(struct pci_dev *pdev)
+{ return NULL; }
 
 static inline struct pci_dev *pci_get_slot(struct pci_bus *bus,
 						unsigned int devfn)
-- 
1.7.3.2

[PATCH 5/5 v2] PCI, ACPI: re-add boot interrupt reroute code in separate files

[Stefan Assmann]

From: Stefan Assmann <sassmann@redhat.com>

During the 2.6.29 merge major parts of e1d3a90846b40ad3160bf4b648d36c6badad39ac
got lost because of other ACPI changes. As suggested by Ingo Molnar the missing
boot interrupt reroute parts are re-added in separate files. The code has
been updated to be less ifdeffy and play along with the current ACPI code. The
printks have been converted to dev_info(). Also the implementation of
bridge_has_boot_interrupt_variant() has been dropped in favor to the existing
pci_find_upstream_pcie_bridge().

By request of Bjorn Helgaas I've created a bugzilla entry at
https://bugzilla.kernel.org/show_bug.cgi?id=21882
to gather information about the boot interrupt reroute problem.

Signed-off-by: Stefan Assmann <sassmann@redhat.com>
---
 arch/x86/include/asm/boot_irq.h |   28 +++++++++++++++++
 arch/x86/kernel/apic/Makefile   |    3 ++
 arch/x86/kernel/apic/boot_irq.c |   63 +++++++++++++++++++++++++++++++++++++++
 drivers/acpi/pci_irq.c          |    4 ++
 4 files changed, 98 insertions(+), 0 deletions(-)
 create mode 100644 arch/x86/include/asm/boot_irq.h
 create mode 100644 arch/x86/kernel/apic/boot_irq.c

diff --git a/arch/x86/include/asm/boot_irq.h b/arch/x86/include/asm/boot_irq.h
new file mode 100644
index 0000000..0cc35e1
--- /dev/null
+++ b/arch/x86/include/asm/boot_irq.h
@@ -0,0 +1,28 @@
+#ifndef _ASM_X86_BOOT_IRQ_H
+#define _ASM_X86_BOOT_IRQ_H
+
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Workaround for boot interrupts that cannot be disabled in hardware: makes
+ * affected devices always use the boot interrupt, the original interrupt line
+ * for the device remains disabled.
+ */
+
+#include <linux/pci.h>
+#ifdef CONFIG_ACPI
+#include <linux/acpi.h>
+#endif /* CONFIG_ACPI */
+
+#ifdef CONFIG_X86_IO_APIC
+#ifdef CONFIG_ACPI
+extern struct acpi_prt_entry *boot_irq_reroute (struct pci_dev *dev, struct acpi_prt_entry *entry);
+#endif /* CONFIG_ACPI */
+#else
+static inline struct acpi_prt_entry *boot_irq_reroute (struct pci_dev *dev, struct acpi_prt_entry *entry)
+	{ return entry; }
+#endif /* CONFIG_X86_IO_APIC */
+
+#endif /* _ASM_X86_BOOT_IRQ_H */
diff --git a/arch/x86/kernel/apic/Makefile b/arch/x86/kernel/apic/Makefile
index 910f20b..4603b66 100644
--- a/arch/x86/kernel/apic/Makefile
+++ b/arch/x86/kernel/apic/Makefile
@@ -9,6 +9,9 @@ endif
 obj-$(CONFIG_HARDLOCKUP_DETECTOR)	+= hw_nmi.o
 
 obj-$(CONFIG_X86_IO_APIC)	+= io_apic.o
+ifeq ($(CONFIG_ACPI),y)
+obj-$(CONFIG_X86_IO_APIC)	+= boot_irq.o
+endif
 obj-$(CONFIG_SMP)		+= ipi.o
 
 ifeq ($(CONFIG_X86_64),y)
diff --git a/arch/x86/kernel/apic/boot_irq.c b/arch/x86/kernel/apic/boot_irq.c
new file mode 100644
index 0000000..297d590
--- /dev/null
+++ b/arch/x86/kernel/apic/boot_irq.c
@@ -0,0 +1,63 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * Workaround for boot interrupts that cannot be disabled in hardware: makes
+ * affected devices always use the boot interrupt, the original interrupt line
+ * for the device remains disabled.
+ */
+
+#include <linux/types.h>
+#include <asm/boot_irq.h>
+
+/*
+ * On it's way to the CPU a PCI IRQ might pass multiple PCI bridges therefore
+ * we need to inspect all of them.
+ */
+static int boot_irq_check_broken_bridge(struct pci_dev *pdev)
+{
+	struct pci_dev *tmp_pdev;
+
+	tmp_pdev = pci_find_upstream_pcie_bridge(pdev);
+
+	if (tmp_pdev)
+		return tmp_pdev->irq_reroute_variant;
+	else
+		return 0;
+}
+
+/*
+ * Some chipsets (e.g. intel 6700PXH) generate a legacy INTx when the IRQ entry
+ * in the chipset's IO-APIC is masked (as, e.g. the RT kernel does during
+ * interrupt handling). When this INTx generation cannot be disabled, we
+ * reroute these interrupts to their legacy equivalent to get rid of spurious
+ * interrupts.
+ */
+struct acpi_prt_entry *boot_irq_reroute(struct pci_dev *dev, struct acpi_prt_entry *entry)
+{
+	u32 orig_index;
+
+	switch (boot_irq_check_broken_bridge(dev)) {
+	case 0:
+		break; /* no rerouting necessary */
+
+	case INTEL_IRQ_REROUTE_VARIANT:
+		/*
+		 * Remap according to INTx routing table in 6700PXH specs,
+		 * intel order number 302628-002, section 2.15.2. Other
+		 * chipsets (80332, ...) have the same mapping and are handled
+		 * here as well.
+		 */
+		orig_index = entry->index;
+		entry->index = (entry->index % 4) + 16;
+		dev_info(&dev->dev, "PCI IRQ %d -> rerouted to legacy boot IRQ %d\n", orig_index, entry->index);
+		break;
+
+	default:
+		dev_info(&dev->dev, "Not rerouting PCI IRQ %d: unknown mapping\n", entry->index);
+		break;
+	}
+
+	return entry;
+}
diff --git a/drivers/acpi/pci_irq.c b/drivers/acpi/pci_irq.c
index e1632b2..83eb5f6 100644
--- a/drivers/acpi/pci_irq.c
+++ b/drivers/acpi/pci_irq.c
@@ -39,6 +39,7 @@
 #include <linux/slab.h>
 #include <acpi/acpi_bus.h>
 #include <acpi/acpi_drivers.h>
+#include <asm/boot_irq.h>
 
 #define PREFIX "ACPI: "
 
@@ -309,6 +310,9 @@ static struct acpi_prt_entry *acpi_pci_irq_lookup(struct pci_dev *dev, int pin)
 	if (entry) {
 		ACPI_DEBUG_PRINT((ACPI_DB_INFO, "Found %s[%c] _PRT entry\n",
 				  pci_name(dev), pin_name(pin)));
+		if (!noioapicquirk)
+			entry = boot_irq_reroute(dev, entry);
+
 		return entry;
 	}
 
-- 
1.7.3.2

Re: [PATCH 0/5 v2] x86: boot interrupt reroute fix

[Stefan Assmann]

On 02.12.2010 16:33, Stefan Assmann wrote:
> During the 2.6.29 merge parts of e1d3a90846b40ad3160bf4b648d36c6badad39ac got
> lost because of other ACPI changes. This patchset gets the reroute code working
> again with the current ACPI code. As suggested by Ingo Molnar the missing boot
> interrupt reroute parts are re-added in separate files.
> Therefore some smaller changes to the ACPI and PCI code are necessary to allow
> the boot interrupt reroute code to access required structures and functionality.
> In detail, the reroute code requires access to:
> - struct acpi_prt_entry
> - pci_find_upstream_pcie_bridge()
> 
> Additionally to not change any information in the global acpi_prt_list it is
> desirable to have acpi_pci_irq_find_prt_entry() return a pointer to a copy of
> a struct acpi_prt_entry instead of pointer into the global acpi_prt_list. The
> callers of acpi_pci_irq_find_prt_entry() have been updated to free the
> allocated structure as well.
> 
> By request of Bjorn Helgaas I've created a bugzilla entry at
> https://bugzilla.kernel.org/show_bug.cgi?id=21882
> to gather information about the boot interrupt reroute problem.
> 
> Jesse Barnes suggested to integrate the changes through the ACPI tree.
> 
> v2:
> - split pci_find_upstream_pcie_bridge() changes into separate patches

ping?

  Stefan
--
Stefan Assmann         | Red Hat GmbH
Software Engineer      | Otto-Hahn-Strasse 20, 85609 Dornach
                       | HR: Amtsgericht Muenchen HRB 153243
                       | GF: Brendan Lane, Charlie Peters,
sassmann at redhat.com |     Michael Cunningham, Charles Cachera

Re: [PATCH 0/5 v2] x86: boot interrupt reroute fix

[Stefan Assmann]

On 02.12.2010 16:33, Stefan Assmann wrote:
> During the 2.6.29 merge parts of e1d3a90846b40ad3160bf4b648d36c6badad39ac got
> lost because of other ACPI changes. This patchset gets the reroute code working
> again with the current ACPI code. As suggested by Ingo Molnar the missing boot
> interrupt reroute parts are re-added in separate files.
> Therefore some smaller changes to the ACPI and PCI code are necessary to allow
> the boot interrupt reroute code to access required structures and functionality.
> In detail, the reroute code requires access to:
> - struct acpi_prt_entry
> - pci_find_upstream_pcie_bridge()
> 
> Additionally to not change any information in the global acpi_prt_list it is
> desirable to have acpi_pci_irq_find_prt_entry() return a pointer to a copy of
> a struct acpi_prt_entry instead of pointer into the global acpi_prt_list. The
> callers of acpi_pci_irq_find_prt_entry() have been updated to free the
> allocated structure as well.
> 
> By request of Bjorn Helgaas I've created a bugzilla entry at
> https://bugzilla.kernel.org/show_bug.cgi?id=21882
> to gather information about the boot interrupt reroute problem.
> 
> Jesse Barnes suggested to integrate the changes through the ACPI tree.
> 
> v2:
> - split pci_find_upstream_pcie_bridge() changes into separate patches

No response from the ACPI side.

Jesse,
could you take the PCI changes (patches 3 and 4) through your tree?

  Stefan
--
Stefan Assmann         | Red Hat GmbH
Software Engineer      | Otto-Hahn-Strasse 20, 85609 Dornach
                       | HR: Amtsgericht Muenchen HRB 153243
                       | GF: Brendan Lane, Charlie Peters,
sassmann at redhat.com |     Michael Cunningham, Charles Cachera