Skip to contentSkip to conversation

x86: Reserve FIRST_DEVICE_VECTOR in used_vectors bitmap.

PATCH In review Stefan Bader 2 participants 5 messages Sep 29, 2008 – Oct 3, 2008 x86

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86: Reserve FIRST_DEVICE_VECTOR in used_vectors bitmap.

From a272a28f520c58a84853e098b4ee565486ca5062 Mon Sep 17 00:00:00 2001
From: Stefan Bader <[email protected]>
Date: Sat, 27 Sep 2008 11:07:30 -0400
Subject: [PATCH] x86: Reserve FIRST_DEVICE_VECTOR in used_vectors bitmap.

Someone from the community found out, that after repeatedly unloading and loading a device driver that uses MSI IRQs, the system eventually assigned the vector initially reserved for IRQ0 to the device driver.

The reason for this is, that although IRQ0 is tied to the FIRST_DEVICE_VECTOR when declaring the irq_vector table, the corresponding bit in the used_vectors map is not set. So, if vectors are released and assigned often enough, the vector will get assigned to another interrupt. This happens more often with MSI interrupts as those are exclusively using a vector.

Fix this by setting the bit for the FIRST_DEVICE_VECTOR in the bitmap.

---
 arch/x86/kernel/io_apic_32.c |    3 +++
 1 files changed, 3 insertions(+), 0 deletions(-)
arch/x86/kernel/io_apic_32.c +3 -0
index 09cddb5..bfd9fc5 100644
@@ -2314,6 +2314,9 @@ void __init setup_IO_APIC(void)
 	for (i = first_system_vector; i < NR_VECTORS; i++)
 		set_bit(i, used_vectors);
 
+	/* Mark FIRST_DEVICE_VECTOR which is assigned to IRQ0 as used. */
+	set_bit(FIRST_DEVICE_VECTOR, used_vectors);
+
 	enable_IO_APIC();
 
 	io_apic_irqs = ~PIC_IRQS;
Signature · 1 line
1.5.4.3
#1 lore raw
Quoted 5 lines

Someone from the community found out, that after repeatedly unloading and loading a device driver that uses MSI IRQs, the system eventually assigned the vector initially reserved for IRQ0 to the device driver.

Quoted 8 lines

The reason for this is, that although IRQ0 is tied to the FIRST_DEVICE_VECTOR when declaring the irq_vector table, the corresponding bit in the used_vectors map is not set. So, if vectors are released and assigned often enough, the vector will get assigned to another interrupt. This happens more often with MSI interrupts as those are exclusively using a vector.

Fix this by setting the bit for the FIRST_DEVICE_VECTOR in the bitmap.

hi, the IO-APIC code got unified in the x86 tree for v2.6.28.

So could you please send the fix against tip/master:

http://people.redhat.com/mingo/tip.git/README

Thanks,

Ingo

#2 lore raw
Quoted 15 lines · Ingo Molnar
Quoted 4 lines

Someone from the community found out, that after repeatedly unloading and loading a device driver that uses MSI IRQs, the system eventually assigned the vector initially reserved for IRQ0 to the device driver.

Quoted 8 lines

The reason for this is, that although IRQ0 is tied to the FIRST_DEVICE_VECTOR when declaring the irq_vector table, the corresponding bit in the used_vectors map is not set. So, if vectors are released and assigned often enough, the vector will get assigned to another interrupt. This happens more often with MSI interrupts as those are exclusively using a vector.

Fix this by setting the bit for the FIRST_DEVICE_VECTOR in the bitmap.

hi, the IO-APIC code got unified in the x86 tree for v2.6.28.

Quoted 7 lines

So could you please send the fix against tip/master:

http://people.redhat.com/mingo/tip.git/README

Thanks,

Ingo

I can't see this as required for tip. FIRST_DEVICE_VECTOR is there defined distinct to the ISA IRQs (IRQ15_VECTOR+2) while it is defined to be equal to IRQ0_VECTOR on 2.6.27 and before (at least down to 2.6.24).

Regards,
Stefan

Signature · 1 line
When all other means of communication fail, try words!
#3 lore raw

2008/9/30 Stefan Bader <[email protected]>:

Quoted 20 lines
Quoted 19 lines · Ingo Molnar
Quoted 4 lines

Someone from the community found out, that after repeatedly unloading and loading a device driver that uses MSI IRQs, the system eventually assigned the vector initially reserved for IRQ0 to the device driver.

Quoted 8 lines

The reason for this is, that although IRQ0 is tied to the FIRST_DEVICE_VECTOR when declaring the irq_vector table, the corresponding bit in the used_vectors map is not set. So, if vectors are released and assigned often enough, the vector will get assigned to another interrupt. This happens more often with MSI interrupts as those are exclusively using a vector.

Fix this by setting the bit for the FIRST_DEVICE_VECTOR in the bitmap.

hi, the IO-APIC code got unified in the x86 tree for v2.6.28.

So could you please send the fix against tip/master:

http://people.redhat.com/mingo/tip.git/README

Quoted 8 lines
Quoted 3 lines

Thanks,

Ingo

I can't see this as required for tip. FIRST_DEVICE_VECTOR is there defined distinct to the ISA IRQs (IRQ15_VECTOR+2) while it is defined to be equal to IRQ0_VECTOR on 2.6.27 and before (at least down to 2.6.24).

Hi Ingo,

I didn't see any follow up, so for clarification: there is nothing to fix in tip but the bug is there in kernels between 2.6.24 and 2.6.27. Could you push it to 2.6.27 and possibly to the stable trees?

Stefan

#4 lore raw
Quoted 5 lines · Stefan Bader

Hi Ingo,

I didn't see any follow up, so for clarification: there is nothing to fix in tip but the bug is there in kernels between 2.6.24 and 2.6.27. Could you push it to 2.6.27 and possibly to the stable trees?

it's too late for v2.6.27 - this bug permeates so many kernel versions that it's clearly not in the '2.6.27 regression' category. But it can be asked to be included in -stable after v2.6.27 is released.

Ingo