Skip to contentSkip to conversation

x86, irq: Check if irq is remapped before freeing irte

PATCH In review Yinghai Lu 2 participants 8 messages Oct 18, 2010 – Oct 19, 2010

Subsystems (2)
lore
Share & tools
lore

Yinghai Lu

author
#0 lore raw

PATCH 1/1 x86, irq: Check if irq is remapped before freeing irte

On one system that support intr-rempping when boot with "intremap=off"

got:

[  177.824202] calling  alsa_card_azx_init+0x0/0x1b @ 1
[  177.843968] HDA Intel 0000:00:1b.0: PCI INT A -> GSI 22 (level, low) -> IRQ 22
[  177.848210] HDA Intel 0000:00:1b.0: irq 1435 for MSI/MSI-X
[  177.863797] HDA Intel 0000:00:1b.0: setting latency timer to 64
[  177.895084] hda-intel: no codecs found!
[  177.895501] BUG: unable to handle kernel NULL pointer dereference at 00000000000000f8
[  177.913316] IP: [<ffffffff8145fc18>] free_irte+0x47/0xc0
[  177.913859] PGD 0 
[  177.914037] Oops: 0000 [#1] SMP 
[  177.933240] last sysfs file: 
[  177.933501] CPU 0 
[  177.933655] Modules linked in:
[  177.933937] 
[  177.934078] Pid: 15044, comm: work_for_cpu Not tainted 2.6.36-rc8-tip-yh-01994-g95100d0-dirty #198      /Sun Fire X4800
[  177.953986] RIP: 0010:[<ffffffff8145fc18>]  [<ffffffff8145fc18>] free_irte+0x47/0xc0
...
[  178.173326] Call Trace:
[  178.173574]  [<ffffffff810515b4>] destroy_irq+0x3a/0x75
[  178.192934]  [<ffffffff81051834>] arch_teardown_msi_irq+0xe/0x10
[  178.193418]  [<ffffffff81458dc3>] arch_teardown_msi_irqs+0x56/0x7f
[  178.213021]  [<ffffffff81458e79>] free_msi_irqs+0x8d/0xeb
[  178.213490]  [<ffffffff81459673>] pci_disable_msi+0x35/0x3a
[  178.232956]  [<ffffffff81b68917>] azx_free+0x83/0x11c
[  178.233301]  [<ffffffff81cb1ec7>] azx_probe+0x7b1/0xab4
[  178.252885]  [<ffffffff810a59ef>] ? trace_hardirqs_on+0xd/0xf
[  178.253303]  [<ffffffff81442a50>] local_pci_probe+0x4d/0x96
[  178.272801]  [<ffffffff8108e72c>] ? do_work_for_cpu+0x0/0x2b
[  178.273270]  [<ffffffff8108e744>] do_work_for_cpu+0x18/0x2b
[  178.292785]  [<ffffffff8108e72c>] ? do_work_for_cpu+0x0/0x2b
[  178.293220]  [<ffffffff81094135>] kthread+0x9d/0xa5
[  178.312742]  [<ffffffff81034954>] kernel_thread_helper+0x4/0x10
[  178.313222]  [<ffffffff81cc96fc>] ? restore_args+0x0/0x30
[  178.332746]  [<ffffffff81094098>] ? kthread+0x0/0xa5
[  178.333207]  [<ffffffff81034950>] ? kernel_thread_helper+0x0/0x10

Root cause is that irq_2_iommu is embedded into irq_cfg now...

Need to check if that irq is really mapped, before free irte.

---
 drivers/pci/intr_remapping.c |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/pci/intr_remapping.c +2 -2
Index: linux-2.6/drivers/pci/intr_remapping.c
===================================================================
@@ -60,7 +60,7 @@ int get_irte(int irq, struct irte *entry
 	unsigned long flags;
 	int index;
 
-	if (!entry || !irq_iommu)
+	if (!entry || !irq_iommu || !irq_iommu->iommu)
 		return -1;
 
 	spin_lock_irqsave(&irq_2_ir_lock, flags);
@@ -268,7 +268,7 @@ int free_irte(int irq)
 	unsigned long flags;
 	int rc;
 
-	if (!irq_iommu)
+	if (!irq_iommu || !irq_iommu->iommu)
 		return -1;
 
 	spin_lock_irqsave(&irq_2_ir_lock, flags);
#1 lore raw

drivers/pci/intr_remapping.c int get_irte(int irq, struct irte *entry · line 64

Quoted 12 lines · Yinghai Lu

Index: linux-2.6/drivers/pci/intr_remapping.c
===================================================================

drivers/pci/intr_remapping.c +1 -1
@@ -60,7 +60,7 @@ int get_irte(int irq, struct irte *entry
 	unsigned long flags;
 	int index;
 
-	if (!entry || !irq_iommu)
+	if (!entry || !irq_iommu || !irq_iommu->iommu)
 		return -1;

Hmm, why do we need this? This is only called from ir_ioapic_set_affinity() and ir_msi_set_affinity().

We should never end up there when intr_remapping=off, right ?

drivers/pci/intr_remapping.c int free_irte(int irq) · line 272

Quoted 8 lines
drivers/pci/intr_remapping.c +1 -1
 	spin_lock_irqsave(&irq_2_ir_lock, flags);
@@ -268,7 +268,7 @@ int free_irte(int irq)
 	unsigned long flags;
 	int rc;
 
-	if (!irq_iommu)
+	if (!irq_iommu || !irq_iommu->iommu)
 		return -1;

That one makes sense

Thanks,

tglx

#2 lore raw
Quoted 10 lines · Thomas Gleixner
Quoted 9 lines · Yinghai Lu

Index: linux-2.6/drivers/pci/intr_remapping.c
===================================================================

drivers/pci/intr_remapping.c +0 -0
@@ -60,7 +60,7 @@ int get_irte(int irq, struct irte *entry
 	unsigned long flags;
 	int index;
 
Quoted 8 lines

drivers/pci/intr_remapping.c int get_irte(int irq, struct irte *entry · line 64

Quoted 3 lines
drivers/pci/intr_remapping.c +1 -1
-	if (!entry || !irq_iommu)
+	if (!entry || !irq_iommu || !irq_iommu->iommu)
 		return -1;

Hmm, why do we need this? This is only called from ir_ioapic_set_affinity() and ir_msi_set_affinity().

We should never end up there when intr_remapping=off, right ?

Thinking more about it, this check is actively bogus. The call sites do:

        struct irte irte;

        if (get_irte(irq, &irte))
                return -1;

So entry _CANNOT_ be NULL.

And in fact we should change get_irte() to

get_irte(struct irq_2_iommu *irq_iommu, struct irte *entry)

The call site already knows about it. No need to lookup irq_iommu based on the irq number.

Thanks,

tglx

author
#3 lore raw
Quoted 30 lines · Thomas Gleixner
Quoted 10 lines · Thomas Gleixner
Quoted 9 lines · Yinghai Lu

Index: linux-2.6/drivers/pci/intr_remapping.c
===================================================================

drivers/pci/intr_remapping.c +0 -0
@@ -60,7 +60,7 @@ int get_irte(int irq, struct irte *entry
   unsigned long flags;
   int index;
 
Quoted 8 lines

drivers/pci/intr_remapping.c int get_irte(int irq, struct irte *entry · line 64

Quoted 3 lines
drivers/pci/intr_remapping.c +1 -1
-    if (!entry || !irq_iommu)
+    if (!entry || !irq_iommu || !irq_iommu->iommu)
       return -1;

Hmm, why do we need this? This is only called from ir_ioapic_set_affinity() and ir_msi_set_affinity().

We should never end up there when intr_remapping=off, right ?

Thinking more about it, this check is actively bogus. The call sites do:

       struct irte irte;

       if (get_irte(irq, &irte))
               return -1;
Quoted 8 lines

So entry _CANNOT_ be NULL.

And in fact we should change get_irte() to

get_irte(struct irq_2_iommu *irq_iommu, struct irte *entry)

The call site already knows about it. No need to lookup irq_iommu based on the irq number.

Yes

author
#4 lore raw
In reply to Thomas Gleixner
Quoted 30 lines · Thomas Gleixner
Quoted 10 lines · Thomas Gleixner
Quoted 9 lines · Yinghai Lu

Index: linux-2.6/drivers/pci/intr_remapping.c
===================================================================

drivers/pci/intr_remapping.c +0 -0
@@ -60,7 +60,7 @@ int get_irte(int irq, struct irte *entry
 	unsigned long flags;
 	int index;
 
Quoted 8 lines

drivers/pci/intr_remapping.c int get_irte(int irq, struct irte *entry · line 64

Quoted 3 lines
drivers/pci/intr_remapping.c +1 -1
-	if (!entry || !irq_iommu)
+	if (!entry || !irq_iommu || !irq_iommu->iommu)
 		return -1;

Hmm, why do we need this? This is only called from ir_ioapic_set_affinity() and ir_msi_set_affinity().

We should never end up there when intr_remapping=off, right ?

Thinking more about it, this check is actively bogus. The call sites do:

        struct irte irte;

        if (get_irte(irq, &irte))
                return -1;
Quoted 8 lines

So entry _CANNOT_ be NULL.

And in fact we should change get_irte() to

get_irte(struct irq_2_iommu *irq_iommu, struct irte *entry)

The call site already knows about it. No need to lookup irq_iommu based on the irq number.

looks like all irq-irte related API could replace "int irq" to "struct irq_2_iommu *irq_iommu"

extern int get_irte(int irq, struct irte *entry);
extern int modify_irte(int irq, struct irte *irte_modified); extern int alloc_irte(struct intel_iommu *iommu, int irq, u16 count); extern int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index,

                        u16 sub_handle);

extern int map_irq_to_irte_handle(int irq, u16 *sub_handle); extern int free_irte(int irq);

Thanks

Yinghai

#5 lore raw
Quoted 15 lines · Yinghai Lu
Quoted 14 lines · Thomas Gleixner
Quoted 10 lines · Thomas Gleixner
Quoted 9 lines · Yinghai Lu

Index: linux-2.6/drivers/pci/intr_remapping.c
===================================================================

drivers/pci/intr_remapping.c +0 -0
@@ -60,7 +60,7 @@ int get_irte(int irq, struct irte *entry
 	unsigned long flags;
 	int index;
 
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

drivers/pci/intr_remapping.c int get_irte(int irq, struct irte *entry · line 64

Quoted 3 lines
drivers/pci/intr_remapping.c +1 -1
-	if (!entry || !irq_iommu)
+	if (!entry || !irq_iommu || !irq_iommu->iommu)
 		return -1;

Hmm, why do we need this? This is only called from ir_ioapic_set_affinity() and ir_msi_set_affinity().

That does not answer that question !

Quoted 20 lines
Quoted 9 lines
Quoted 1 line

We should never end up there when intr_remapping=off, right ?

Thinking more about it, this check is actively bogus. The call sites do:

        struct irte irte;

        if (get_irte(irq, &irte))
                return -1;
Quoted 8 lines

So entry _CANNOT_ be NULL.

And in fact we should change get_irte() to

get_irte(struct irq_2_iommu *irq_iommu, struct irte *entry)

The call site already knows about it. No need to lookup irq_iommu based on the irq number.

looks like all irq-irte related API could replace "int irq" to "struct irq_2_iommu *irq_iommu"

Quoted 7 lines

extern int get_irte(int irq, struct irte *entry);
extern int modify_irte(int irq, struct irte *irte_modified); extern int alloc_irte(struct intel_iommu *iommu, int irq, u16 count); extern int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index,

                        u16 sub_handle);

extern int map_irq_to_irte_handle(int irq, u16 *sub_handle); extern int free_irte(int irq);

Probably, but we need to figure out which functions need which checks instead of having either redundant or superflous ones there.

Thanks,

tglx

author
#6 lore raw
Quoted 17 lines · Thomas Gleixner
Quoted 15 lines · Yinghai Lu
Quoted 14 lines · Thomas Gleixner
Quoted 10 lines · Thomas Gleixner
Quoted 9 lines · Yinghai Lu

Index: linux-2.6/drivers/pci/intr_remapping.c
===================================================================

drivers/pci/intr_remapping.c +0 -0
@@ -60,7 +60,7 @@ int get_irte(int irq, struct irte *entry
 	unsigned long flags;
 	int index;
 
Quoted 8 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines

drivers/pci/intr_remapping.c int get_irte(int irq, struct irte *entry · line 64

Quoted 3 lines
drivers/pci/intr_remapping.c +1 -1
-	if (!entry || !irq_iommu)
+	if (!entry || !irq_iommu || !irq_iommu->iommu)
 		return -1;

Hmm, why do we need this? This is only called from ir_ioapic_set_affinity() and ir_msi_set_affinity().

That does not answer that question !

we don't need that checking there.

Quoted 29 lines
Quoted 21 lines
Quoted 10 lines
Quoted 2 lines

We should never end up there when intr_remapping=off, right ?

Thinking more about it, this check is actively bogus. The call sites do:

        struct irte irte;

        if (get_irte(irq, &irte))
                return -1;
Quoted 8 lines

So entry _CANNOT_ be NULL.

And in fact we should change get_irte() to

get_irte(struct irq_2_iommu *irq_iommu, struct irte *entry)

The call site already knows about it. No need to lookup irq_iommu based on the irq number.

looks like all irq-irte related API could replace "int irq" to "struct irq_2_iommu *irq_iommu"

Quoted 7 lines

extern int get_irte(int irq, struct irte *entry);
extern int modify_irte(int irq, struct irte *irte_modified); extern int alloc_irte(struct intel_iommu *iommu, int irq, u16 count); extern int set_irte_irq(int irq, struct intel_iommu *iommu, u16 index,

                        u16 sub_handle);

extern int map_irq_to_irte_handle(int irq, u16 *sub_handle); extern int free_irte(int irq);

Quoted 3 lines

Probably, but we need to figure out which functions need which checks instead of having either redundant or superflous ones there.

only free_irte().

because other have have protection from intr_remapping_enabled or irq_remapped(get_irq_chip_data(irq)) or with ir related chip.

this one works too.

---
 arch/x86/kernel/apic/io_apic.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
arch/x86/kernel/apic/io_apic.c +2 -1
Index: linux-2.6/arch/x86/kernel/apic/io_apic.c
===================================================================
@@ -3109,7 +3109,8 @@ void destroy_irq(unsigned int irq)
 
 	irq_set_status_flags(irq, IRQ_NOREQUEST|IRQ_NOPROBE);
 
-	free_irte(irq);
+	if (intr_remapping_enabled)
+		free_irte(irq);
 	raw_spin_lock_irqsave(&vector_lock, flags);
 	__clear_irq_vector(irq, cfg);
 	raw_spin_unlock_irqrestore(&vector_lock, flags);
#7 lore raw

[tip:irq/core] x86: ioapic: Call free_irte only if interrupt remapping enabled

In reply to Yinghai Lu
Commit-ID:  9717967c4b704ce344c954afb5bb160aa9c01c34
Gitweb:     http://git.kernel.org/tip/9717967c4b704ce344c954afb5bb160aa9c01c34
Author:     Yinghai Lu <[email protected]>
AuthorDate: Mon, 18 Oct 2010 13:47:48 -0700
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Tue, 19 Oct 2010 09:25:33 +0200

x86: ioapic: Call free_irte only if interrupt remapping enabled

On a system that support intr-rempping when booting with "intremap=off"

[  177.895501] BUG: unable to handle kernel NULL pointer dereference at 00000000000000f8
[  177.913316] IP: [<ffffffff8145fc18>] free_irte+0x47/0xc0
...
[  178.173326] Call Trace:
[  178.173574]  [<ffffffff810515b4>] destroy_irq+0x3a/0x75
[  178.192934]  [<ffffffff81051834>] arch_teardown_msi_irq+0xe/0x10
[  178.193418]  [<ffffffff81458dc3>] arch_teardown_msi_irqs+0x56/0x7f
[  178.213021]  [<ffffffff81458e79>] free_msi_irqs+0x8d/0xeb

Call free_irte only when interrupt remapping is enabled.

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/apic/io_apic.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
arch/x86/kernel/apic/io_apic.c +2 -1
index 20e47e0..8ae808d 100644
@@ -3109,7 +3109,8 @@ void destroy_irq(unsigned int irq)
 
 	irq_set_status_flags(irq, IRQ_NOREQUEST|IRQ_NOPROBE);
 
-	free_irte(irq);
+	if (intr_remapping_enabled)
+		free_irte(irq);
 	raw_spin_lock_irqsave(&vector_lock, flags);
 	__clear_irq_vector(irq, cfg);
 	raw_spin_unlock_irqrestore(&vector_lock, flags);