[PATCH] PCI: spacemit: Ignore PERST# bit for controllers

[PATCH] PCI: spacemit: Ignore PERST# bit for controllers

[Ali Tariq]

The k1_pcie_init() does not set PCIE_IGNORE_PERSTN bit of the
control logic register, which causes controller to reset itself once
PERST# signal is asserted even when acting as host.
On Orange Pi RV2 (SpacemiT K1), this resets the controller and wipes
the IDs. k1_pcie_deinit() leaves PERST# asserted, so when k1_pcie_init()
runs again, its first DBI access hangs the CPU.

Set PCIE_IGNORE_PERSTN bit of the control logic register of APMU
in k1_pcie_init() before enabling resources so controllers
dont end up resetting themselves when PERST# signal is asserted.

Fixes: ff64e078e45f ("PCI: spacemit: Add SpacemiT PCIe host driver")
Signed-off-by: Ali Tariq <alitariq45892@gmail.com>
---
Hang log: https://kernelci.cloud-v.co/scheduler/job/420
Fixed Log: https://kernelci.cloud-v.co/scheduler/job/430
---
 drivers/pci/controller/dwc/pcie-spacemit-k1.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/drivers/pci/controller/dwc/pcie-spacemit-k1.c b/drivers/pci/controller/dwc/pcie-spacemit-k1.c
index 92a71ea90ac3..af6028e8825a 100644
--- a/drivers/pci/controller/dwc/pcie-spacemit-k1.c
+++ b/drivers/pci/controller/dwc/pcie-spacemit-k1.c
@@ -47,6 +47,7 @@
 
 #define PCIE_CONTROL_LOGIC			0x0004
 #define PCIE_SOFT_RESET			BIT(0)
+#define PCIE_IGNORE_PERSTN		BIT(2)	/* 1: ignore PERST# */
 
 struct k1_pcie {
 	struct dw_pcie pci;
@@ -131,11 +132,19 @@ static int k1_pcie_init(struct dw_pcie_rp *pp)
 	struct dw_pcie *pci = to_dw_pcie_from_pp(pp);
 	struct k1_pcie *k1 = to_k1_pcie(pci);
 	u32 reset_ctrl;
+	u32 ctrl_logic;
 	u32 val;
 	int ret;
 
 	k1_pcie_toggle_soft_reset(k1);
 
+	/*
+	 * Ignore the PERST# signal so the controller does not reset
+	 * itself when it pulls PERST# low.
+	 */
+	ctrl_logic = k1->pmu_off + PCIE_CONTROL_LOGIC;
+	regmap_set_bits(k1->pmu, ctrl_logic, PCIE_IGNORE_PERSTN);
+
 	ret = k1_pcie_enable_resources(k1);
 	if (ret)
 		return ret;
-- 
2.34.1

Re: [PATCH] PCI: spacemit: Ignore PERST# bit for controllers

[Troy Mitchell]

[-- Attachment #1: Type: text/plain, Size: 1622 bytes --]

On Wed Nov 4, 2026 at 2:07 PM GMT, Ali Tariq wrote:
> The k1_pcie_init() does not set PCIE_IGNORE_PERSTN bit of the
> control logic register, which causes controller to reset itself once
> PERST# signal is asserted even when acting as host.
> On Orange Pi RV2 (SpacemiT K1), this resets the controller and wipes
> the IDs. k1_pcie_deinit() leaves PERST# asserted, so when k1_pcie_init()
> runs again, its first DBI access hangs the CPU.
>
> Set PCIE_IGNORE_PERSTN bit of the control logic register of APMU
> in k1_pcie_init() before enabling resources so controllers
> dont end up resetting themselves when PERST# signal is asserted.
>
> Fixes: ff64e078e45f ("PCI: spacemit: Add SpacemiT PCIe host driver")
> Signed-off-by: Ali Tariq <alitariq45892@gmail.com>
> ---
> Hang log: https://kernelci.cloud-v.co/scheduler/job/420
> Fixed Log: https://kernelci.cloud-v.co/scheduler/job/430
> ---
>  drivers/pci/controller/dwc/pcie-spacemit-k1.c | 9 +++++++++
>  1 file changed, 9 insertions(+)
>
> diff --git a/drivers/pci/controller/dwc/pcie-spacemit-k1.c b/drivers/pci/controller/dwc/pcie-spacemit-k1.c
> index 92a71ea90ac3..af6028e8825a 100644
> --- a/drivers/pci/controller/dwc/pcie-spacemit-k1.c
> +++ b/drivers/pci/controller/dwc/pcie-spacemit-k1.c
> @@ -47,6 +47,7 @@
>  
>  #define PCIE_CONTROL_LOGIC			0x0004
>  #define PCIE_SOFT_RESET			BIT(0)
> +#define PCIE_IGNORE_PERSTN		BIT(2)	/* 1: ignore PERST# */
A similar but earlier patch can be found here [1]

Link: https://lore.kernel.org/all/20260924-spacemit-k1-pcie-fix-v1-1-002b5c1ba9fc@linux.spacemit.com/ [1]

-- 
Troy Mitchell


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 273 bytes --]

Re: [PATCH] PCI: spacemit: Ignore PERST# bit for controllers

[Ali Tariq]

On 11/4/26 7:14 PM, Troy Mitchell wrote:
> On Wed Nov 4, 2026 at 2:07 PM GMT, Ali Tariq wrote:
>> The k1_pcie_init() does not set PCIE_IGNORE_PERSTN bit of the
>> control logic register, which causes controller to reset itself once
>> PERST# signal is asserted even when acting as host.
>> On Orange Pi RV2 (SpacemiT K1), this resets the controller and wipes
>> the IDs. k1_pcie_deinit() leaves PERST# asserted, so when k1_pcie_init()
>> runs again, its first DBI access hangs the CPU.
>>
>> Set PCIE_IGNORE_PERSTN bit of the control logic register of APMU
>> in k1_pcie_init() before enabling resources so controllers
>> dont end up resetting themselves when PERST# signal is asserted.
>>
>> Fixes: ff64e078e45f ("PCI: spacemit: Add SpacemiT PCIe host driver")
>> Signed-off-by: Ali Tariq <alitariq45892@gmail.com>
>> ---
>> Hang log: https://kernelci.cloud-v.co/scheduler/job/420
>> Fixed Log: https://kernelci.cloud-v.co/scheduler/job/430
>> ---
>>   drivers/pci/controller/dwc/pcie-spacemit-k1.c | 9 +++++++++
>>   1 file changed, 9 insertions(+)
>>
>> diff --git a/drivers/pci/controller/dwc/pcie-spacemit-k1.c b/drivers/pci/controller/dwc/pcie-spacemit-k1.c
>> index 92a71ea90ac3..af6028e8825a 100644
>> --- a/drivers/pci/controller/dwc/pcie-spacemit-k1.c
>> +++ b/drivers/pci/controller/dwc/pcie-spacemit-k1.c
>> @@ -47,6 +47,7 @@
>>   
>>   #define PCIE_CONTROL_LOGIC			0x0004
>>   #define PCIE_SOFT_RESET			BIT(0)
>> +#define PCIE_IGNORE_PERSTN		BIT(2)	/* 1: ignore PERST# */
> A similar but earlier patch can be found here [1]
> 
> Link: https://lore.kernel.org/all/20260924-spacemit-k1-pcie-fix-v1-1-002b5c1ba9fc@linux.spacemit.com/ [1]
> 

Thank you for pointing.

I missed that. I will check that patch.

You can drop my patch.

Regards,
Ali

Re: [PATCH] PCI: spacemit: Ignore PERST# bit for controllers

[Encrow Thorne]

On Wed Nov 4, 2026 at 10:07 PM CST, Ali Tariq wrote:

Hi Ali,
> The k1_pcie_init() does not set PCIE_IGNORE_PERSTN bit of the
> control logic register, which causes controller to reset itself once
> PERST# signal is asserted even when acting as host.
> On Orange Pi RV2 (SpacemiT K1), this resets the controller and wipes
> the IDs. k1_pcie_deinit() leaves PERST# asserted, so when k1_pcie_init()
> runs again, its first DBI access hangs the CPU.
>
> Set PCIE_IGNORE_PERSTN bit of the control logic register of APMU
> in k1_pcie_init() before enabling resources so controllers
> dont end up resetting themselves when PERST# signal is asserted.
>
> Fixes: ff64e078e45f ("PCI: spacemit: Add SpacemiT PCIe host driver")
> Signed-off-by: Ali Tariq <alitariq45892@gmail.com>
> ---
> Hang log: https://kernelci.cloud-v.co/scheduler/job/420
> Fixed Log: https://kernelci.cloud-v.co/scheduler/job/430
> ---
>  drivers/pci/controller/dwc/pcie-spacemit-k1.c | 9 +++++++++
>  1 file changed, 9 insertions(+)
>
> diff --git a/drivers/pci/controller/dwc/pcie-spacemit-k1.c b/drivers/pci/controller/dwc/pcie-spacemit-k1.c
> index 92a71ea90ac3..af6028e8825a 100644
> --- a/drivers/pci/controller/dwc/pcie-spacemit-k1.c
> +++ b/drivers/pci/controller/dwc/pcie-spacemit-k1.c
> @@ -47,6 +47,7 @@
>  
>  #define PCIE_CONTROL_LOGIC			0x0004
>  #define PCIE_SOFT_RESET			BIT(0)
> +#define PCIE_IGNORE_PERSTN		BIT(2)	/* 1: ignore PERST# */
>  
>  struct k1_pcie {
>  	struct dw_pcie pci;
> @@ -131,11 +132,19 @@ static int k1_pcie_init(struct dw_pcie_rp *pp)
>  	struct dw_pcie *pci = to_dw_pcie_from_pp(pp);
>  	struct k1_pcie *k1 = to_k1_pcie(pci);
>  	u32 reset_ctrl;
> +	u32 ctrl_logic;
>  	u32 val;
>  	int ret;
>  
>  	k1_pcie_toggle_soft_reset(k1);
>  
> +	/*
> +	 * Ignore the PERST# signal so the controller does not reset
> +	 * itself when it pulls PERST# low.
> +	 */
> +	ctrl_logic = k1->pmu_off + PCIE_CONTROL_LOGIC;
> +	regmap_set_bits(k1->pmu, ctrl_logic, PCIE_IGNORE_PERSTN);
> +
>  	ret = k1_pcie_enable_resources(k1);
>  	if (ret)
>  		return ret;

Thanks for investigating this and submitting a fix.

Our testing also confirms that setting PCIE_IGNORE_PERSTN prevents
PERST# from resetting the local RC and resolves the DBI hang.

I will send a v2 of my earlier patch with an updated explanation
and the test results.

Regards,
Encrow Thorne