Re: [PATCH] irqchip/qcom-pdc: add support for v3.2 HW

[Neil Armstrong <neil.armstrong@linaro.org>]

On 21/08/2023 10:41, Maulik Shah (mkshah) wrote:
> Hi,
> 
> On 8/21/2023 1:00 PM, Neil Armstrong wrote:
>> The PDC Hw version 3.2 and later doesn't have the enable registers
>> anymore, get the HW version from registers and stop accessing those
>> registers starting at this version.
> 
> Starting PDC v3.2, enable bit is moved from IRQ_ENABLE_BANK to the IRQ_i_CFG register.
> we need to set enable bit to tell PDC HW whether IRQ is enabled / disabled in order to wake up from SoC sleep states.
> 
>>
>> Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
>> ---
>>   drivers/irqchip/qcom-pdc.c | 64 ++++++++++++++++++++++++++++++++++++++--------
>>   1 file changed, 53 insertions(+), 11 deletions(-)
>>
>> diff --git a/drivers/irqchip/qcom-pdc.c b/drivers/irqchip/qcom-pdc.c
>> index d96916cf6a41..620efb9fcc96 100644
>> --- a/drivers/irqchip/qcom-pdc.c
>> +++ b/drivers/irqchip/qcom-pdc.c
>> @@ -26,6 +26,13 @@
>>   #define IRQ_ENABLE_BANK        0x10
>>   #define IRQ_i_CFG        0x110
>> +#define PDC_VERSION        0x1000
>> +
>> +/* Notable PDC versions */
>> +enum {
>> +    PDC_VERSION_3_2    = 0x30200,
>> +};
>> +
>>   struct pdc_pin_region {
>>       u32 pin_base;
>>       u32 parent_base;
>> @@ -38,6 +45,7 @@ static DEFINE_RAW_SPINLOCK(pdc_lock);
>>   static void __iomem *pdc_base;
>>   static struct pdc_pin_region *pdc_region;
>>   static int pdc_region_cnt;
>> +static unsigned int pdc_version;
>>   static void pdc_reg_write(int reg, u32 i, u32 val)
>>   {
>> @@ -183,6 +191,25 @@ static struct irq_chip qcom_pdc_gic_chip = {
>>       .irq_set_affinity    = irq_chip_set_affinity_parent,
>>   };
>> +static struct irq_chip qcom_pdc_gic_chip_3_2 = {
>> +    .name            = "PDC",
>> +    .irq_eoi        = irq_chip_eoi_parent,
>> +    .irq_mask        = irq_chip_mask_parent,
>> +    .irq_unmask        = irq_chip_unmask_parent,
>> +    .irq_disable        = irq_chip_disable_parent,
>> +    .irq_enable        = irq_chip_enable_parent,
>> +    .irq_get_irqchip_state    = irq_chip_get_parent_state,
>> +    .irq_set_irqchip_state    = irq_chip_set_parent_state,
>> +    .irq_retrigger        = irq_chip_retrigger_hierarchy,
>> +    .irq_set_type        = qcom_pdc_gic_set_type,
>> +    .flags            = IRQCHIP_MASK_ON_SUSPEND |
>> +                  IRQCHIP_SET_TYPE_MASKED |
>> +                  IRQCHIP_SKIP_SET_WAKE |
>> +                  IRQCHIP_ENABLE_WAKEUP_ON_SUSPEND,
>> +    .irq_set_vcpu_affinity    = irq_chip_set_vcpu_affinity_parent,
>> +    .irq_set_affinity    = irq_chip_set_affinity_parent,
>> +};
>> +
> we should not need to have another irqchip defined for same as these remains same for old/new version.
> 
> Not sure if you have access to downstream change, or let me know i can post the same.
> In downstream we do like...
> 
> The 3rd bit in IRQ_i_CFG in PDC v3.2 is for enable/disable,
> The 0-2 bits are for the IRQ type in all PDC versions..
> 
>      #define IRQ_i_CFG_IRQ_ENABLE   BIT(3)
>      #define IRQ_i_CFG_TYPE_MASK    0x7
> 
> and modify pdc_enable_intr()
> 
>         if (pdc_version < PDC_VERSION_3_2) {
>                 index = pin_out / 32;
>                 mask = pin_out % 32;
>                 enable = pdc_reg_read(IRQ_ENABLE_BANK, index);
>                 __assign_bit(mask, &enable, on);
>                 pdc_reg_write(IRQ_ENABLE_BANK, index, enable);
>         } else {
>                 index = d->hwirq;
>                 enable = pdc_reg_read(IRQ_i_CFG, index);
>                 __assign_bit(IRQ_i_CFG_IRQ_ENABLE, &enable, on);
>                 pdc_reg_write(IRQ_i_CFG, index, enable);
>         }
> 
> and qcom_pdc_gic_set_type(), add line to only take 0-2 bits from old_pdc_type as they are for type.
> 
>          old_pdc_type = pdc_reg_read(IRQ_i_CFG, d->hwirq);
> +       pdc_type |= (old_pdc_type & ~IRQ_i_CFG_TYPE_MASK);
>> +        if (pdc_version < PDC_VERSION_3_2)
>> +            pdc_enable_region(n);
>>       }
> 
> you will still need to clear enable bits in version 3.2.

Ack, thx for the review, I'll fix all that for v2.

Neil

> 
> Thanks,
> Maulik
>