* [PATCH] Connecting the SB600's i8259 controller rather in pasemi's pci.c than in pasemi's setup.c.
@ 2026-07-01 10:54 Christian Zigotzky
2026-07-01 11:14 ` Christian Zigotzky
0 siblings, 1 reply; 3+ messages in thread
From: Christian Zigotzky @ 2026-07-01 10:54 UTC (permalink / raw)
To: Madhavan Srinivasan, Michael Ellerman, Nicholas Piggin,
Christophe Leroy (CS GROUP), Christian Zigotzky,
Krzysztof Kozlowski,
open list:LINUX FOR POWERPC (32-BIT AND 64-BIT), open list
"pas_pci_init" was before "nemo_init_IRQ".
Now "pas_pci_init" is after "nemo_init_IRQ" in the official kernel source
code.
I think "pas_pci_init" scans (discovers) the PCI(e) devices
and after that, "nemo_init_IRQ" assigns interrupt numbers
to these devices if required.
It's not possible to assigns interrupt numbers to PCI(e) devices
which have not been discovered yet.
Signed-off-by: Christian Zigotzky <chzigotzky@xenosoft.de>
---
arch/powerpc/platforms/pasemi/pci.c | 7 +++++++
arch/powerpc/platforms/pasemi/setup.c | 7 ++++---
2 files changed, 11 insertions(+), 3 deletions(-)
diff --git a/arch/powerpc/platforms/pasemi/pci.c b/arch/powerpc/platforms/pasemi/pci.c
index 2df955274652..7208c325bfc5 100644
--- a/arch/powerpc/platforms/pasemi/pci.c
+++ b/arch/powerpc/platforms/pasemi/pci.c
@@ -25,6 +25,8 @@
#define PA_PXP_CFA(bus, devfn, off) (((bus) << 20) | ((devfn) << 12) | (off))
+extern void nemo_init_IRQ(void);
+
static inline int pa_pxp_offset_valid(u8 bus, u8 devfn, int offset)
{
/* Device 0 Function 0 is special: It's config space spans function 1 as
@@ -265,6 +267,11 @@ static int __init pas_add_bridge(struct device_node *dev)
*/
isa_bridge_find_early(hose);
+ /*
+ * ISA bridge is now active, add the i8259 cascade (if needed)
+ */
+ nemo_init_IRQ();
+
return 0;
}
diff --git a/arch/powerpc/platforms/pasemi/setup.c b/arch/powerpc/platforms/pasemi/setup.c
index d03b41336901..eec74611be46 100644
--- a/arch/powerpc/platforms/pasemi/setup.c
+++ b/arch/powerpc/platforms/pasemi/setup.c
@@ -214,10 +214,12 @@ static void sb600_8259_cascade(struct irq_desc *desc)
chip->irq_eoi(&desc->irq_data);
}
-static void __init nemo_init_IRQ(struct mpic *mpic)
+void nemo_init_IRQ(void)
{
struct device_node *np;
int gpio_virq;
+ struct mpic *mpic;
+
/* Connect the SB600's legacy i8259 controller */
np = of_find_node_by_path("/pxp@0,e0000000");
i8259_init(np, 0);
@@ -228,6 +230,7 @@ static void __init nemo_init_IRQ(struct mpic *mpic)
irq_set_chained_handler(gpio_virq, sb600_8259_cascade);
mpic_unmask_irq(irq_get_irq_data(gpio_virq));
+ mpic = irq_get_chip_data(gpio_virq);
irq_set_default_domain(mpic->irqhost);
}
@@ -298,8 +301,6 @@ static __init void pas_init_IRQ(void)
mpic_unmask_irq(irq_get_irq_data(nmi_virq));
}
- nemo_init_IRQ(mpic);
-
of_node_put(mpic_node);
of_node_put(root);
}
--
2.55.0.windows.1
^ permalink raw reply related [flat|nested] 3+ messages in thread
* Re: [PATCH] Connecting the SB600's i8259 controller rather in pasemi's pci.c than in pasemi's setup.c.
2026-07-01 10:54 Christian Zigotzky
@ 2026-07-01 11:14 ` Christian Zigotzky
0 siblings, 0 replies; 3+ messages in thread
From: Christian Zigotzky @ 2026-07-01 11:14 UTC (permalink / raw)
To: Christian Zigotzky, Krzysztof Kozlowski, Madhavan Srinivasan,
Michael Ellerman, linux-kernel, Nicholas Piggin, linuxppc-dev,
Christophe Leroy
Cc: Darren Stevens, R.T.Dickinson, hypexed
[-- Attachment #1: Type: text/plain, Size: 4024 bytes --]
The Nemo board [1] doesn’t boot without this patch. Darren explained it really well:
Originally we initialised the PCI-e ports in setup arch, this is quite early, and it seems uses some kernel functions that were not recommended (They changed a whole lot of different platforms at the same time for the same reason)
After this we added the ISA bridge, then the kernel would init the IRQ contollers.
The patch that broke booting on the X1000 moved the pas_pci_init to a node in the machine description, where it called later in the boot sequence. Unfortunately this is after we've tried to add the i8259 contoller from the pas_init_IRQ. Since our ISA bridge can't be found until we've connected the PCI-e ports the system tries to write to registers that aren't yet mapped - result a kernel panic, but before console I/O has been initialised so it appears to be a hang. We had a similar problem when they were introducing Radix support.
My patch changes our code so that it works with the new kernel code in place. Basically I moved the code that adds the i8259 cascade to after we've scanned for the ISA bridge where I know it will work.
Hopefully this makes sense, shout out if it doesn't
Regards
Darren
[1] https://en.wikipedia.org/wiki/AmigaOne_X1000
> On 01 July 2026 at 12:55 pm, Christian Zigotzky <chzigotzky@xenosoft.de> wrote :
>
> "pas_pci_init" was before "nemo_init_IRQ".
> Now "pas_pci_init" is after "nemo_init_IRQ" in the official kernel source
> code.
> I think "pas_pci_init" scans (discovers) the PCI(e) devices
> and after that, "nemo_init_IRQ" assigns interrupt numbers
> to these devices if required.
> It's not possible to assigns interrupt numbers to PCI(e) devices
> which have not been discovered yet.
>
> Signed-off-by: Christian Zigotzky <chzigotzky@xenosoft.de>
> ---
> arch/powerpc/platforms/pasemi/pci.c | 7 +++++++
> arch/powerpc/platforms/pasemi/setup.c | 7 ++++---
> 2 files changed, 11 insertions(+), 3 deletions(-)
>
> diff --git a/arch/powerpc/platforms/pasemi/pci.c b/arch/powerpc/platforms/pasemi/pci.c
> index 2df955274652..7208c325bfc5 100644
> --- a/arch/powerpc/platforms/pasemi/pci.c
> +++ b/arch/powerpc/platforms/pasemi/pci.c
> @@ -25,6 +25,8 @@
>
> #define PA_PXP_CFA(bus, devfn, off) (((bus) << 20) | ((devfn) << 12) | (off))
>
> +extern void nemo_init_IRQ(void);
> +
> static inline int pa_pxp_offset_valid(u8 bus, u8 devfn, int offset)
> {
> /* Device 0 Function 0 is special: It's config space spans function 1 as
> @@ -265,6 +267,11 @@ static int __init pas_add_bridge(struct device_node *dev)
> */
> isa_bridge_find_early(hose);
>
> + /*
> + * ISA bridge is now active, add the i8259 cascade (if needed)
> + */
> + nemo_init_IRQ();
> +
> return 0;
> }
>
> diff --git a/arch/powerpc/platforms/pasemi/setup.c b/arch/powerpc/platforms/pasemi/setup.c
> index d03b41336901..eec74611be46 100644
> --- a/arch/powerpc/platforms/pasemi/setup.c
> +++ b/arch/powerpc/platforms/pasemi/setup.c
> @@ -214,10 +214,12 @@ static void sb600_8259_cascade(struct irq_desc *desc)
> chip->irq_eoi(&desc->irq_data);
> }
>
> -static void __init nemo_init_IRQ(struct mpic *mpic)
> +void nemo_init_IRQ(void)
> {
> struct device_node *np;
> int gpio_virq;
> + struct mpic *mpic;
> +
> /* Connect the SB600's legacy i8259 controller */
> np = of_find_node_by_path("/pxp@0,e0000000");
> i8259_init(np, 0);
> @@ -228,6 +230,7 @@ static void __init nemo_init_IRQ(struct mpic *mpic)
> irq_set_chained_handler(gpio_virq, sb600_8259_cascade);
> mpic_unmask_irq(irq_get_irq_data(gpio_virq));
>
> + mpic = irq_get_chip_data(gpio_virq);
> irq_set_default_domain(mpic->irqhost);
> }
>
> @@ -298,8 +301,6 @@ static __init void pas_init_IRQ(void)
> mpic_unmask_irq(irq_get_irq_data(nmi_virq));
> }
>
> - nemo_init_IRQ(mpic);
> -
> of_node_put(mpic_node);
> of_node_put(root);
> }
> --
> 2.55.0.windows.1
>
[-- Attachment #2: Type: text/html, Size: 5950 bytes --]
^ permalink raw reply [flat|nested] 3+ messages in thread
* [PATCH] Connecting the SB600's i8259 controller rather in pasemi's pci.c than in pasemi's setup.c.
@ 2026-07-01 12:59 Christian Zigotzky
0 siblings, 0 replies; 3+ messages in thread
From: Christian Zigotzky @ 2026-07-01 12:59 UTC (permalink / raw)
To: Madhavan Srinivasan, Michael Ellerman, Nicholas Piggin,
Christophe Leroy (CS GROUP), Krzysztof Kozlowski,
Christian Zigotzky,
open list:LINUX FOR POWERPC (32-BIT AND 64-BIT), open list
Hello,
Here is the email without HTML (Sorry):
The Nemo board [1] doesn’t boot without this patch. Darren explained it really well:
Originally we initialised the PCI-e ports in setup arch, this is quite early, and it seems uses some kernel functions that were not recommended (They changed a whole lot of different platforms at the same time for the same reason)
After this we added the ISA bridge, then the kernel would init the IRQ contollers.
The patch that broke booting on the X1000 moved the pas_pci_init to a node in the machine description, where it called later in the boot sequence. Unfortunately this is after we've tried to add the i8259 contoller from the pas_init_IRQ. Since our ISA bridge can't be found until we've connected the PCI-e ports the system tries to write to registers that aren't yet mapped - result a kernel panic, but before console I/O has been initialised so it appears to be a hang. We had a similar problem when they were introducing Radix support.
My patch changes our code so that it works with the new kernel code in place. Basically I moved the code that adds the i8259 cascade to after we've scanned for the ISA bridge where I know it will work.
Hopefully this makes sense, shout out if it doesn't.
Regards
Darren
[1] https://en.wikipedia.org/wiki/AmigaOne_X1000
Signed-off-by: Christian Zigotzky <chzigotzky@xenosoft.de>
---
arch/powerpc/platforms/pasemi/pci.c | 7 +++++++
arch/powerpc/platforms/pasemi/setup.c | 7 ++++---
2 files changed, 11 insertions(+), 3 deletions(-)
diff --git a/arch/powerpc/platforms/pasemi/pci.c b/arch/powerpc/platforms/pasemi/pci.c
index 2df955274652..7208c325bfc5 100644
--- a/arch/powerpc/platforms/pasemi/pci.c
+++ b/arch/powerpc/platforms/pasemi/pci.c
@@ -25,6 +25,8 @@
#define PA_PXP_CFA(bus, devfn, off) (((bus) << 20) | ((devfn) << 12) | (off))
+extern void nemo_init_IRQ(void);
+
static inline int pa_pxp_offset_valid(u8 bus, u8 devfn, int offset)
{
/* Device 0 Function 0 is special: It's config space spans function 1 as
@@ -265,6 +267,11 @@ static int __init pas_add_bridge(struct device_node *dev)
*/
isa_bridge_find_early(hose);
+ /*
+ * ISA brigde is now active, add the i8259 cascade (if needed)
+ */
+ nemo_init_IRQ();
+
return 0;
}
diff --git a/arch/powerpc/platforms/pasemi/setup.c b/arch/powerpc/platforms/pasemi/setup.c
index d03b41336901..eec74611be46 100644
--- a/arch/powerpc/platforms/pasemi/setup.c
+++ b/arch/powerpc/platforms/pasemi/setup.c
@@ -214,10 +214,12 @@ static void sb600_8259_cascade(struct irq_desc *desc)
chip->irq_eoi(&desc->irq_data);
}
-static void __init nemo_init_IRQ(struct mpic *mpic)
+void nemo_init_IRQ(void)
{
struct device_node *np;
int gpio_virq;
+ struct mpic *mpic;
+
/* Connect the SB600's legacy i8259 controller */
np = of_find_node_by_path("/pxp@0,e0000000");
i8259_init(np, 0);
@@ -228,6 +230,7 @@ static void __init nemo_init_IRQ(struct mpic *mpic)
irq_set_chained_handler(gpio_virq, sb600_8259_cascade);
mpic_unmask_irq(irq_get_irq_data(gpio_virq));
+ mpic = irq_get_chip_data(gpio_virq);
irq_set_default_domain(mpic->irqhost);
}
@@ -298,8 +301,6 @@ static __init void pas_init_IRQ(void)
mpic_unmask_irq(irq_get_irq_data(nmi_virq));
}
- nemo_init_IRQ(mpic);
-
of_node_put(mpic_node);
of_node_put(root);
}
--
2.55.0.windows.1
^ permalink raw reply related [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-07-01 13:03 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-07-01 12:59 [PATCH] Connecting the SB600's i8259 controller rather in pasemi's pci.c than in pasemi's setup.c Christian Zigotzky
-- strict thread matches above, loose matches on Subject: below --
2026-07-01 10:54 Christian Zigotzky
2026-07-01 11:14 ` Christian Zigotzky
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.