[PATCH] irqchip/bcm-6345-l1: fix SMP support

[PATCH] irqchip/bcm-6345-l1: fix SMP support

[Álvaro Fernández Rojas]

Some BCM6358 devices start with Core #1 instead of Core #0.
Apart from that, SMP is restricted to 1 CPU since BCM6358 has a shared TLB,
which makes it impossible for the current SMP support to start both CPUs.

The problem is that smp_processor_id() returns 0 and then cpu_logical_map()
converts that to 1, which accesses an uninitialized position of intc->cpus[],
resulting in a kernel panic.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
---
 drivers/irqchip/irq-bcm6345-l1.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/irqchip/irq-bcm6345-l1.c b/drivers/irqchip/irq-bcm6345-l1.c
index e3483789f4df..b2173ce4743d 100644
--- a/drivers/irqchip/irq-bcm6345-l1.c
+++ b/drivers/irqchip/irq-bcm6345-l1.c
@@ -121,7 +121,7 @@ static void bcm6345_l1_irq_handle(struct irq_desc *desc)
 	unsigned int idx;
 
 #ifdef CONFIG_SMP
-	cpu = intc->cpus[cpu_logical_map(smp_processor_id())];
+	cpu = intc->cpus[smp_processor_id()];
 #else
 	cpu = intc->cpus[0];
 #endif
-- 
2.20.1

Re: [PATCH] irqchip/bcm-6345-l1: fix SMP support

[Florian Fainelli]

On 2/22/2021 12:13 PM, Álvaro Fernández Rojas wrote:
> Some BCM6358 devices start with Core #1 instead of Core #0.
> Apart from that, SMP is restricted to 1 CPU since BCM6358 has a shared TLB,
> which makes it impossible for the current SMP support to start both CPUs.
> 
> The problem is that smp_processor_id() returns 0 and then cpu_logical_map()
> converts that to 1, which accesses an uninitialized position of intc->cpus[],
> resulting in a kernel panic.

Sounds like you nee to correct the way the cpu_logical_map[] is
populated then?
-- 
Florian

Re: [PATCH] irqchip/bcm-6345-l1: fix SMP support

[Álvaro Fernández Rojas]

Hi Florian,

Yes, you’re right, there seems to be a bug, because cpu_number_map() and cpu_logical_map() are returning the same when BCM6358 starts with CMT 1:
[    4.662896] bcm6345_l1_irq_handle: smp_id=0 cpu_logical=1 cpu_number=1

Let’s see if I can fix it…

Best regards,
Álvaro.

> El 22 feb 2021, a las 22:00, Florian Fainelli <f.fainelli@gmail.com> escribió:
> 
> 
> 
> On 2/22/2021 12:13 PM, Álvaro Fernández Rojas wrote:
>> Some BCM6358 devices start with Core #1 instead of Core #0.
>> Apart from that, SMP is restricted to 1 CPU since BCM6358 has a shared TLB,
>> which makes it impossible for the current SMP support to start both CPUs.
>> 
>> The problem is that smp_processor_id() returns 0 and then cpu_logical_map()
>> converts that to 1, which accesses an uninitialized position of intc->cpus[],
>> resulting in a kernel panic.
> 
> Sounds like you nee to correct the way the cpu_logical_map[] is
> populated then?
> -- 
> Florian