Patchwork fix bug of isa_bus irq

login
register
mail settings
Submitter Wanpeng Li
Date March 11, 2012, 7:04 a.m.
Message ID <1331449479-8197-1-git-send-email-liwp@linux.vnet.ibm.com>
Download mbox | patch
Permalink /patch/145945/
State New
Headers show

Comments

Wanpeng Li - March 11, 2012, 7:04 a.m.
ISA bus only use IRQ 0~15, so don't need to give an array qemu_irq 0~23, just
array qemu_irq i8259 is ok.

Signed-off-by: Wanpeng Li <liwp@linux.vnet.ibm.com>
---
 hw/pc_piix.c |    3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)
Jan Kiszka - March 11, 2012, 7:46 a.m.
On 2012-03-11 08:04, Wanpeng Li wrote:
> ISA bus only use IRQ 0~15, so don't need to give an array qemu_irq 0~23, just
> array qemu_irq i8259 is ok.
> 
> Signed-off-by: Wanpeng Li <liwp@linux.vnet.ibm.com>
> ---
>  hw/pc_piix.c |    3 ++-
>  1 files changed, 2 insertions(+), 1 deletions(-)
> 
> diff --git a/hw/pc_piix.c b/hw/pc_piix.c
> index 63dba4e..52f7cf8 100644
> --- a/hw/pc_piix.c
> +++ b/hw/pc_piix.c
> @@ -210,7 +210,6 @@ static void pc_init1(MemoryRegion *system_memory,
>          isa_bus = isa_bus_new(NULL, system_io);
>          no_hpet = 1;
>      }
> -    isa_bus_irqs(isa_bus, gsi);
>  
>      if (kvm_irqchip_in_kernel()) {
>          i8259 = kvm_i8259_init(isa_bus);
> @@ -221,6 +220,8 @@ static void pc_init1(MemoryRegion *system_memory,
>          i8259 = i8259_init(isa_bus, cpu_irq[0]);
>      }
>  
> +    isa_bus_irqs(isa_bus, i8259);
> +
>      for (i = 0; i < ISA_NUM_IRQS; i++) {
>          gsi_state->i8259_irq[i] = i8259[i];
>      }

This is bogus. isa_bus_irqs sets the output IRQs of the ISA bus. And
those are not only delivered to the PIC on the PIIX2, but also the
IOAPIC. Thus we have to pass in the GSI input lines which dispatch to
both. Of those lines, only the first 16 will be used by the ISA bus
(there is even an assert to ensure this).

Did you see any concrete bug in the context of this logic?

Jan
Wanpeng Li - March 12, 2012, 6:08 a.m.
On Sun, Mar 11, 2012 at 08:46:38AM +0100, Jan Kiszka wrote:
>On 2012-03-11 08:04, Wanpeng Li wrote:
>> ISA bus only use IRQ 0~15, so don't need to give an array qemu_irq 0~23, just
>> array qemu_irq i8259 is ok.
>> 
>> Signed-off-by: Wanpeng Li <liwp@linux.vnet.ibm.com>
>> ---
>>  hw/pc_piix.c |    3 ++-
>>  1 files changed, 2 insertions(+), 1 deletions(-)
>> 
>> diff --git a/hw/pc_piix.c b/hw/pc_piix.c
>> index 63dba4e..52f7cf8 100644
>> --- a/hw/pc_piix.c
>> +++ b/hw/pc_piix.c
>> @@ -210,7 +210,6 @@ static void pc_init1(MemoryRegion *system_memory,
>>          isa_bus = isa_bus_new(NULL, system_io);
>>          no_hpet = 1;
>>      }
>> -    isa_bus_irqs(isa_bus, gsi);
>>  
>>      if (kvm_irqchip_in_kernel()) {
>>          i8259 = kvm_i8259_init(isa_bus);
>> @@ -221,6 +220,8 @@ static void pc_init1(MemoryRegion *system_memory,
>>          i8259 = i8259_init(isa_bus, cpu_irq[0]);
>>      }
>>  
>> +    isa_bus_irqs(isa_bus, i8259);
>> +
>>      for (i = 0; i < ISA_NUM_IRQS; i++) {
>>          gsi_state->i8259_irq[i] = i8259[i];
>>      }
>
>This is bogus. isa_bus_irqs sets the output IRQs of the ISA bus. And
>those are not only delivered to the PIC on the PIIX2, but also the
>IOAPIC. Thus we have to pass in the GSI input lines which dispatch to
>both. Of those lines, only the first 16 will be used by the ISA bus
>(there is even an assert to ensure this).
>
>Did you see any concrete bug in the context of this logic?
>
>Jan
>

Yes, but actually PIC is being used at present, whether passing qemu_irq
0~23 to isa_bus is not safe or not.

Wanpeng Li
Jan Kiszka - March 12, 2012, 6:55 a.m.
On 2012-03-12 07:08, Wanpeng Li wrote:
> On Sun, Mar 11, 2012 at 08:46:38AM +0100, Jan Kiszka wrote:
>> On 2012-03-11 08:04, Wanpeng Li wrote:
>>> ISA bus only use IRQ 0~15, so don't need to give an array qemu_irq 0~23, just
>>> array qemu_irq i8259 is ok.
>>>
>>> Signed-off-by: Wanpeng Li <liwp@linux.vnet.ibm.com>
>>> ---
>>>  hw/pc_piix.c |    3 ++-
>>>  1 files changed, 2 insertions(+), 1 deletions(-)
>>>
>>> diff --git a/hw/pc_piix.c b/hw/pc_piix.c
>>> index 63dba4e..52f7cf8 100644
>>> --- a/hw/pc_piix.c
>>> +++ b/hw/pc_piix.c
>>> @@ -210,7 +210,6 @@ static void pc_init1(MemoryRegion *system_memory,
>>>          isa_bus = isa_bus_new(NULL, system_io);
>>>          no_hpet = 1;
>>>      }
>>> -    isa_bus_irqs(isa_bus, gsi);
>>>  
>>>      if (kvm_irqchip_in_kernel()) {
>>>          i8259 = kvm_i8259_init(isa_bus);
>>> @@ -221,6 +220,8 @@ static void pc_init1(MemoryRegion *system_memory,
>>>          i8259 = i8259_init(isa_bus, cpu_irq[0]);
>>>      }
>>>  
>>> +    isa_bus_irqs(isa_bus, i8259);
>>> +
>>>      for (i = 0; i < ISA_NUM_IRQS; i++) {
>>>          gsi_state->i8259_irq[i] = i8259[i];
>>>      }
>>
>> This is bogus. isa_bus_irqs sets the output IRQs of the ISA bus. And
>> those are not only delivered to the PIC on the PIIX2, but also the
>> IOAPIC. Thus we have to pass in the GSI input lines which dispatch to
>> both. Of those lines, only the first 16 will be used by the ISA bus
>> (there is even an assert to ensure this).
>>
>> Did you see any concrete bug in the context of this logic?
>>
>> Jan
>>
> 
> Yes, but actually PIC is being used at present, whether passing qemu_irq
> 0~23 to isa_bus is not safe or not.

Sorry, IRQ routing to PIC and IOAPIC is actually not a property of the
PIIX3 but the board we emulate. And here we follow the Multiprocessor
Specification of Intel and route ISA bus IRQs to both interrupt
controllers. Thus the bus must be connected to the GSIs. And, again,
GSI[16..13] aren't referenced by the ISA bus at any time.

Jan

Patch

diff --git a/hw/pc_piix.c b/hw/pc_piix.c
index 63dba4e..52f7cf8 100644
--- a/hw/pc_piix.c
+++ b/hw/pc_piix.c
@@ -210,7 +210,6 @@  static void pc_init1(MemoryRegion *system_memory,
         isa_bus = isa_bus_new(NULL, system_io);
         no_hpet = 1;
     }
-    isa_bus_irqs(isa_bus, gsi);
 
     if (kvm_irqchip_in_kernel()) {
         i8259 = kvm_i8259_init(isa_bus);
@@ -221,6 +220,8 @@  static void pc_init1(MemoryRegion *system_memory,
         i8259 = i8259_init(isa_bus, cpu_irq[0]);
     }
 
+    isa_bus_irqs(isa_bus, i8259);
+
     for (i = 0; i < ISA_NUM_IRQS; i++) {
         gsi_state->i8259_irq[i] = i8259[i];
     }