Re: [PATCH v1] irqchip/irq-sifive-plic: Add syscore callbacks for hibernation

[Mason Huo]

On 2023/2/5 18:51, Marc Zyngier wrote:
> On Fri, 13 Jan 2023 09:42:16 +0000,
> Mason Huo <mason.huo@xxxxxxxxxxxxxxxx> wrote:
>> 
>> The priority and enable registers of plic will be reset
>> during hibernation power cycle in poweroff mode,
>> add the syscore callbacks to save/restore those registers.
>> 
>> Signed-off-by: Mason Huo <mason.huo@xxxxxxxxxxxxxxxx>
>> Reviewed-by: Ley Foon Tan <leyfoon.tan@xxxxxxxxxxxxxxxx>
>> Reviewed-by: Sia Jee Heng <jeeheng.sia@xxxxxxxxxxxxxxxx>
>> ---
>>  drivers/irqchip/irq-sifive-plic.c | 93 ++++++++++++++++++++++++++++++-
>>  1 file changed, 91 insertions(+), 2 deletions(-)
>> 
>> diff --git a/drivers/irqchip/irq-sifive-plic.c b/drivers/irqchip/irq-sifive-plic.c
>> index ff47bd0dec45..80306de45d2b 100644
>> --- a/drivers/irqchip/irq-sifive-plic.c
>> +++ b/drivers/irqchip/irq-sifive-plic.c
>> @@ -17,6 +17,7 @@
>>  #include <linux/of_irq.h>
>>  #include <linux/platform_device.h>
>>  #include <linux/spinlock.h>
>> +#include <linux/syscore_ops.h>
>>  #include <asm/smp.h>
>>  
>>  /*
>> @@ -67,6 +68,8 @@ struct plic_priv {
>>  	struct irq_domain *irqdomain;
>>  	void __iomem *regs;
>>  	unsigned long plic_quirks;
>> +	unsigned int nr_irqs;
>> +	u32 *priority_reg;
>>  };
>>  
>>  struct plic_handler {
>> @@ -79,10 +82,13 @@ struct plic_handler {
>>  	raw_spinlock_t		enable_lock;
>>  	void __iomem		*enable_base;
>>  	struct plic_priv	*priv;
>> +	/* To record interrupts that are enabled before suspend. */
>> +	u32 enable_reg[MAX_DEVICES / 32];
> 
> What does MAX_DEVICES represent here? How is it related to the number
> of interrupts you're trying to save? It seems to be related to the
> number of CPUs, so it hardly makes any sense so far.
> 
The comment of this macro describes that "The largest number supported
by devices marked as 'sifive,plic-1.0.0', is 1024, of which
device 0 is defined as non-existent by the RISC-V Privileged Spec."
As far as I understand, the *device* here means HW IRQ source,
and the HW IRQ 0 is non-existent.

>>  };
>>  static int plic_parent_irq __ro_after_init;
>>  static bool plic_cpuhp_setup_done __ro_after_init;
>>  static DEFINE_PER_CPU(struct plic_handler, plic_handlers);
>> +static struct plic_priv *priv_data;
>>  
>>  static int plic_irq_set_type(struct irq_data *d, unsigned int type);
>>  
>> @@ -229,6 +235,78 @@ static int plic_irq_set_type(struct irq_data *d, unsigned int type)
>>  	return IRQ_SET_MASK_OK;
>>  }
>>  
>> +static void plic_irq_resume(void)
>> +{
>> +	unsigned int i, cpu;
>> +	u32 __iomem *reg;
>> +
>> +	for (i = 0; i < priv_data->nr_irqs; i++)
>> +		writel(priv_data->priority_reg[i],
>> +				priv_data->regs + PRIORITY_BASE + i * PRIORITY_PER_ID);
> 
> From what I can tell, this driver uses exactly 2 priorities: 0 and 1.
> And yet you use a full 32bit to encode those. Does it seem like a good
> idea?
> 
Yes, currently this driver uses oly 2 priorities.
But, according to the sifive spec, the priority register is a 32bit register,
and it supports 7 levels of priority. 

>> +
>> +	for_each_cpu(cpu, cpu_present_mask) {
>> +		struct plic_handler *handler = per_cpu_ptr(&plic_handlers, cpu);
>> +
>> +		if (!handler->present)
>> +			continue;
>> +
>> +		for (i = 0; i < DIV_ROUND_UP(priv_data->nr_irqs, 32); i++) {
>> +			reg = handler->enable_base + i * sizeof(u32);
>> +			raw_spin_lock(&handler->enable_lock);
>> +			writel(handler->enable_reg[i], reg);
>> +			raw_spin_unlock(&handler->enable_lock);
> 
> Why do you need to take/release the lock around *each* register
> access? Isn't that lock constant for a given CPU?
> 
OK, will fix it in the next version.

>> +		}
>> +	}
>> +}
>> +
>> +static int plic_irq_suspend(void)
>> +{
>> +	unsigned int i, cpu;
>> +	u32 __iomem *reg;
>> +
>> +	for (i = 0; i < priv_data->nr_irqs; i++)
>> +		priv_data->priority_reg[i] =
>> +			readl(priv_data->regs + PRIORITY_BASE + i * PRIORITY_PER_ID);
>> +
>> +	for_each_cpu(cpu, cpu_present_mask) {
>> +		struct plic_handler *handler = per_cpu_ptr(&plic_handlers, cpu);
>> +
>> +		if (!handler->present)
>> +			continue;
>> +
>> +		for (i = 0; i < DIV_ROUND_UP(priv_data->nr_irqs, 32); i++) {
>> +			reg = handler->enable_base + i * sizeof(u32);
>> +			raw_spin_lock(&handler->enable_lock);
>> +			handler->enable_reg[i] = readl(reg);
>> +			raw_spin_unlock(&handler->enable_lock);
> 
> Same remarks.
> 
> 	M.
>