[PATCH v2 0/6] spacemit: k3: Replace bad CLINT node with ACLINT nodes

Junhui Liu posted 6 patches 2 weeks, 3 days ago
There is a newer version of this series
.../thead,c900-aclint-mswi.yaml                    |  1 +
.../thead,c900-aclint-sswi.yaml                    |  5 +--
.../devicetree/bindings/timer/sifive,clint.yaml    |  5 ++-
.../bindings/timer/thead,c900-aclint-mtimer.yaml   |  1 +
arch/riscv/boot/dts/spacemit/k3.dtsi               | 37 ++++++++++++++++------
drivers/irqchip/irq-aclint-sswi.c                  |  1 +
6 files changed, 37 insertions(+), 13 deletions(-)
[PATCH v2 0/6] spacemit: k3: Replace bad CLINT node with ACLINT nodes
Posted by Junhui Liu 2 weeks, 3 days ago
When testing OpenSBI with the upstream Device Tree, the CLINT controller
failed to function properly with the irqchip.riscv_imsic_noipi boot
parameter:

  Platform IPI Device         : aclint-mswi
  Platform Timer Device       : aclint-mtimer @ 24000000Hz
  
  Domain0 Region03            : 0x00000000e081c000-0x00000000e081ffff M: (I,R,W) S/U: ()
  
  [    0.091014] smp: Bringing up secondary CPUs ...
  [   10.104917] CPU1 failed to report alive state
  [   20.118689] CPU2 failed to report alive state
  [   30.132288] CPU3 failed to report alive state
  [   40.145953] CPU4 failed to report alive state
  [   50.159604] CPU5 failed to report alive state
  [   60.173212] CPU6 failed to report alive state
  [   70.186981] CPU7 failed to report alive state
  [   70.191411] smp: Brought up 1 node, 1 CPU

Through exploring and testing, I discovered that the K3 actually
provides all three complete ACLINT components, and subsequent
confirmation with the vendor yielded the following information:

The ACLINT controllers are mapped to two physical memory regions:
secure (0xf1810000-0xf1820000) and non-secure (0xe0810000-0xe0820000),
with varying accessibility per controller and execution mode:

+---------+--------+------------------------+-----------------------+
| Offset  | Module | Secure Region          | Non-Secure Region     |
+---------+--------+------------------------+-----------------------+
| +0x0000 | MSWI   | 0xf1810000-0xf1813fff  | 0xe0810000-0xe0813fff |
|         |        | (M-mode only)          | (Inaccessible)        |
+---------+--------+------------------------+-----------------------+
| +0x4000 | MTIMER | 0xf1814000-0xf181bfff  | 0xe0814000-0xe081bfff |
|         |        | (M-mode only)          | (Inaccessible)        |
+---------+--------+------------------------+-----------------------+
| +0xc000 | SSWI   | 0xf181c000-0xf181ffff  | 0xe081c000-0xe081ffff |
|         |        | (M-mode only if efused)| (M/S-mode Accessible) |
+---------+--------+------------------------+-----------------------+

Taking these constraints into account, we choose the secure region
(0xf1810000-0xf181bfff) for the M-mode ACLINT components (MSWI and
MTIMER), and the non-secure region (0xe081c000-0xe081ffff) for the
S-mode SSWI component.

Signed-off-by: Junhui Liu <junhui.liu@pigmoral.tech>
---
Changes in v2:
- Drop the generic riscv,aclint-* fallback compatibles and use only
  SpacemiT K3-specific compatibles.
- Match spacemit,k3-aclint-sswi directly in the SSWI driver.
- Squash the ACLINT MSWI and MTIMER DT additions into the patch that
  replaces the incorrect CLINT node.
- Link to v1: https://patch.msgid.link/20260729-k3-aclint-v1-0-aef88614d3a5@pigmoral.tech

---
Junhui Liu (6):
      dt-bindings: timer: thead,c900-aclint-mtimer: Add SpacemiT K3
      dt-bindings: interrupt-controller: thead,c900-aclint-mswi: Add SpacemiT K3
      dt-bindings: interrupt-controller: thead,c900-aclint-sswi: Add SpacemiT K3
      dt-bindings: timer: sifive,clint: Deprecate spacemit,k3-clint
      irqchip/aclint-sswi: Add support for SpacemiT K3
      riscv: dts: spacemit: k3: Replace incorrect CLINT node with ACLINT nodes

 .../thead,c900-aclint-mswi.yaml                    |  1 +
 .../thead,c900-aclint-sswi.yaml                    |  5 +--
 .../devicetree/bindings/timer/sifive,clint.yaml    |  5 ++-
 .../bindings/timer/thead,c900-aclint-mtimer.yaml   |  1 +
 arch/riscv/boot/dts/spacemit/k3.dtsi               | 37 ++++++++++++++++------
 drivers/irqchip/irq-aclint-sswi.c                  |  1 +
 6 files changed, 37 insertions(+), 13 deletions(-)
---
base-commit: cee9395acd8043be0644b25c34bfa86623f2b935
change-id: 20260729-k3-aclint-94ca9d3f818f

Best regards,
--  
Junhui Liu <junhui.liu@pigmoral.tech>
Re: [PATCH v2 0/6] spacemit: k3: Replace bad CLINT node with ACLINT nodes
Posted by Bo Gan 2 weeks ago
Hi Junhui,

Thanks for your patch. This would give me more clarity while reviewing
the OpenSBI patchset from Troy, and I was wondering why Troy sent this
https://lore.kernel.org/opensbi/20260908-ipi-clear-all-v1-1-b1bd5d016eb6@linux.dev/T/#u

Now it's much clearer -- K3 has two ways to do IPI, ACLINT and IMSIC.

One thing that may help us further: Can you clarify on the exact meaning
of "M-mode only"? OpenSBI would use PMP to protect those MSWI/MTIMER
regions, so I wonder if you meant that it's enforced by the PMP settings,
or it's *actually* enforced by the device itself (or fabric)? AFAIK, to
achieve this, the device must have a way to distinguish the privilege-
level of the CPU at the time it initiated the MMIO request. Hence, the
core must be putting the privilege level as some kind of attribute in the
MMIO request. Is that the case in k3? Can you confirm with SpacemiT?

On 9/8/26 09:28, Junhui Liu wrote:
> When testing OpenSBI with the upstream Device Tree, the CLINT controller
> failed to function properly with the irqchip.riscv_imsic_noipi boot
> parameter:
> 
>    Platform IPI Device         : aclint-mswi
>    Platform Timer Device       : aclint-mtimer @ 24000000Hz
>    
>    Domain0 Region03            : 0x00000000e081c000-0x00000000e081ffff M: (I,R,W) S/U: ()
>    
>    [    0.091014] smp: Bringing up secondary CPUs ...
>    [   10.104917] CPU1 failed to report alive state
>    [   20.118689] CPU2 failed to report alive state
>    [   30.132288] CPU3 failed to report alive state
>    [   40.145953] CPU4 failed to report alive state
>    [   50.159604] CPU5 failed to report alive state
>    [   60.173212] CPU6 failed to report alive state
>    [   70.186981] CPU7 failed to report alive state
>    [   70.191411] smp: Brought up 1 node, 1 CPU
> 
> Through exploring and testing, I discovered that the K3 actually
> provides all three complete ACLINT components, and subsequent
> confirmation with the vendor yielded the following information:
> 
> The ACLINT controllers are mapped to two physical memory regions:
> secure (0xf1810000-0xf1820000) and non-secure (0xe0810000-0xe0820000),
> with varying accessibility per controller and execution mode:
> 
> +---------+--------+------------------------+-----------------------+
> | Offset  | Module | Secure Region          | Non-Secure Region     |
> +---------+--------+------------------------+-----------------------+
> | +0x0000 | MSWI   | 0xf1810000-0xf1813fff  | 0xe0810000-0xe0813fff |
> |         |        | (M-mode only)          | (Inaccessible)        |
> +---------+--------+------------------------+-----------------------+
> | +0x4000 | MTIMER | 0xf1814000-0xf181bfff  | 0xe0814000-0xe081bfff |
> |         |        | (M-mode only)          | (Inaccessible)        |
> +---------+--------+------------------------+-----------------------+
> | +0xc000 | SSWI   | 0xf181c000-0xf181ffff  | 0xe081c000-0xe081ffff |
> |         |        | (M-mode only if efused)| (M/S-mode Accessible) |
> +---------+--------+------------------------+-----------------------+
> 
> Taking these constraints into account, we choose the secure region
> (0xf1810000-0xf181bfff) for the M-mode ACLINT components (MSWI and
> MTIMER), and the non-secure region (0xe081c000-0xe081ffff) for the
> S-mode SSWI component.
> 
> Signed-off-by: Junhui Liu <junhui.liu@pigmoral.tech>
> ---
> Changes in v2:
> - Drop the generic riscv,aclint-* fallback compatibles and use only
>    SpacemiT K3-specific compatibles.
> - Match spacemit,k3-aclint-sswi directly in the SSWI driver.
> - Squash the ACLINT MSWI and MTIMER DT additions into the patch that
>    replaces the incorrect CLINT node.
> - Link to v1: https://patch.msgid.link/20260729-k3-aclint-v1-0-aef88614d3a5@pigmoral.tech
> 
> ---
> Junhui Liu (6):
>        dt-bindings: timer: thead,c900-aclint-mtimer: Add SpacemiT K3
>        dt-bindings: interrupt-controller: thead,c900-aclint-mswi: Add SpacemiT K3
>        dt-bindings: interrupt-controller: thead,c900-aclint-sswi: Add SpacemiT K3
>        dt-bindings: timer: sifive,clint: Deprecate spacemit,k3-clint
>        irqchip/aclint-sswi: Add support for SpacemiT K3
>        riscv: dts: spacemit: k3: Replace incorrect CLINT node with ACLINT nodes
> 
>   .../thead,c900-aclint-mswi.yaml                    |  1 +
>   .../thead,c900-aclint-sswi.yaml                    |  5 +--
>   .../devicetree/bindings/timer/sifive,clint.yaml    |  5 ++-
>   .../bindings/timer/thead,c900-aclint-mtimer.yaml   |  1 +
>   arch/riscv/boot/dts/spacemit/k3.dtsi               | 37 ++++++++++++++++------
>   drivers/irqchip/irq-aclint-sswi.c                  |  1 +
>   6 files changed, 37 insertions(+), 13 deletions(-)
> ---
> base-commit: cee9395acd8043be0644b25c34bfa86623f2b935
> change-id: 20260729-k3-aclint-94ca9d3f818f
> 
> Best regards,
> --
> Junhui Liu <junhui.liu@pigmoral.tech>
> 
> 
> _______________________________________________
> linux-riscv mailing list
> linux-riscv@lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv

Bo
Re: [PATCH v2 0/6] spacemit: k3: Replace bad CLINT node with ACLINT nodes
Posted by Junhui Liu 1 week, 4 days ago
Hi Bo,

On Fri Sep 11, 2026 at 4:43 PM CST, Bo Gan wrote:
> Hi Junhui,
>
> Thanks for your patch. This would give me more clarity while reviewing
> the OpenSBI patchset from Troy, and I was wondering why Troy sent this
> https://lore.kernel.org/opensbi/20260908-ipi-clear-all-v1-1-b1bd5d016eb6@linux.dev/T/#u
>
> Now it's much clearer -- K3 has two ways to do IPI, ACLINT and IMSIC.
>
> One thing that may help us further: Can you clarify on the exact meaning
> of "M-mode only"? OpenSBI would use PMP to protect those MSWI/MTIMER
> regions, so I wonder if you meant that it's enforced by the PMP settings,
> or it's *actually* enforced by the device itself (or fabric)? AFAIK, to
> achieve this, the device must have a way to distinguish the privilege-
> level of the CPU at the time it initiated the MMIO request. Hence, the
> core must be putting the privilege level as some kind of attribute in the
> MMIO request. Is that the case in k3? Can you confirm with SpacemiT?

Thanks for pointing this out. I have double-checked and found that my
earlier "M-mode only" description was incorrect. The restriction depends
on the eFuse configuration rather than CPU privilege modes. Under the
relevant eFuse configuration, the non-secure address window becomes
inaccessible, while the secure window remains available. PMP handles
access restrictions based on privilege mode.

I also verified that S-mode can read MSIP and MTIME through the secure
window once PMP allows those reads.

I will correct the explanation and use the secure address window
consistently for all ACLINT components in the next version.

>
> On 9/8/26 09:28, Junhui Liu wrote:
>> When testing OpenSBI with the upstream Device Tree, the CLINT controller
>> failed to function properly with the irqchip.riscv_imsic_noipi boot
>> parameter:
>> 
>>    Platform IPI Device         : aclint-mswi
>>    Platform Timer Device       : aclint-mtimer @ 24000000Hz
>>    
>>    Domain0 Region03            : 0x00000000e081c000-0x00000000e081ffff M: (I,R,W) S/U: ()
>>    
>>    [    0.091014] smp: Bringing up secondary CPUs ...
>>    [   10.104917] CPU1 failed to report alive state
>>    [   20.118689] CPU2 failed to report alive state
>>    [   30.132288] CPU3 failed to report alive state
>>    [   40.145953] CPU4 failed to report alive state
>>    [   50.159604] CPU5 failed to report alive state
>>    [   60.173212] CPU6 failed to report alive state
>>    [   70.186981] CPU7 failed to report alive state
>>    [   70.191411] smp: Brought up 1 node, 1 CPU
>> 
>> Through exploring and testing, I discovered that the K3 actually
>> provides all three complete ACLINT components, and subsequent
>> confirmation with the vendor yielded the following information:
>> 
>> The ACLINT controllers are mapped to two physical memory regions:
>> secure (0xf1810000-0xf1820000) and non-secure (0xe0810000-0xe0820000),
>> with varying accessibility per controller and execution mode:
>> 
>> +---------+--------+------------------------+-----------------------+
>> | Offset  | Module | Secure Region          | Non-Secure Region     |
>> +---------+--------+------------------------+-----------------------+
>> | +0x0000 | MSWI   | 0xf1810000-0xf1813fff  | 0xe0810000-0xe0813fff |
>> |         |        | (M-mode only)          | (Inaccessible)        |
>> +---------+--------+------------------------+-----------------------+
>> | +0x4000 | MTIMER | 0xf1814000-0xf181bfff  | 0xe0814000-0xe081bfff |
>> |         |        | (M-mode only)          | (Inaccessible)        |
>> +---------+--------+------------------------+-----------------------+
>> | +0xc000 | SSWI   | 0xf181c000-0xf181ffff  | 0xe081c000-0xe081ffff |
>> |         |        | (M-mode only if efused)| (M/S-mode Accessible) |
>> +---------+--------+------------------------+-----------------------+
>> 
>> Taking these constraints into account, we choose the secure region
>> (0xf1810000-0xf181bfff) for the M-mode ACLINT components (MSWI and
>> MTIMER), and the non-secure region (0xe081c000-0xe081ffff) for the
>> S-mode SSWI component.
>> 
>> Signed-off-by: Junhui Liu <junhui.liu@pigmoral.tech>
>> ---
>> Changes in v2:
>> - Drop the generic riscv,aclint-* fallback compatibles and use only
>>    SpacemiT K3-specific compatibles.
>> - Match spacemit,k3-aclint-sswi directly in the SSWI driver.
>> - Squash the ACLINT MSWI and MTIMER DT additions into the patch that
>>    replaces the incorrect CLINT node.
>> - Link to v1: https://patch.msgid.link/20260729-k3-aclint-v1-0-aef88614d3a5@pigmoral.tech
>> 
>> ---
>> Junhui Liu (6):
>>        dt-bindings: timer: thead,c900-aclint-mtimer: Add SpacemiT K3
>>        dt-bindings: interrupt-controller: thead,c900-aclint-mswi: Add SpacemiT K3
>>        dt-bindings: interrupt-controller: thead,c900-aclint-sswi: Add SpacemiT K3
>>        dt-bindings: timer: sifive,clint: Deprecate spacemit,k3-clint
>>        irqchip/aclint-sswi: Add support for SpacemiT K3
>>        riscv: dts: spacemit: k3: Replace incorrect CLINT node with ACLINT nodes
>> 
>>   .../thead,c900-aclint-mswi.yaml                    |  1 +
>>   .../thead,c900-aclint-sswi.yaml                    |  5 +--
>>   .../devicetree/bindings/timer/sifive,clint.yaml    |  5 ++-
>>   .../bindings/timer/thead,c900-aclint-mtimer.yaml   |  1 +
>>   arch/riscv/boot/dts/spacemit/k3.dtsi               | 37 ++++++++++++++++------
>>   drivers/irqchip/irq-aclint-sswi.c                  |  1 +
>>   6 files changed, 37 insertions(+), 13 deletions(-)
>> ---
>> base-commit: cee9395acd8043be0644b25c34bfa86623f2b935
>> change-id: 20260729-k3-aclint-94ca9d3f818f
>> 
>> Best regards,
>> --
>> Junhui Liu <junhui.liu@pigmoral.tech>
>> 
>> 
>> _______________________________________________
>> linux-riscv mailing list
>> linux-riscv@lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/linux-riscv
>
> Bo

-- 
Best regards,
Junhui Liu