Re: [PATCH v4 08/20] irqchip/gic-v4.1: Plumb get/set_irqchip_state SGI callbacks

[Zenghui Yu]

Hi Marc,

On 2020/2/18 23:31, Marc Zyngier wrote:
> diff --git a/drivers/irqchip/irq-gic-v3-its.c 
> b/drivers/irqchip/irq-gic-v3-its.c
> index 7656b353a95f..0ed286dba827 100644
> --- a/drivers/irqchip/irq-gic-v3-its.c
> +++ b/drivers/irqchip/irq-gic-v3-its.c
> @@ -144,7 +144,7 @@ struct event_lpi_map {
>  ÂÂÂÂ u16ÂÂÂÂÂÂÂÂÂÂÂ *col_map;
>  ÂÂÂÂ irq_hw_number_tÂÂÂÂÂÂÂ lpi_base;
>  ÂÂÂÂ intÂÂÂÂÂÂÂÂÂÂÂ nr_lpis;
> -ÂÂÂ raw_spinlock_tÂÂÂÂÂÂÂ vlpi_lock;
> +ÂÂÂ raw_spinlock_tÂÂÂÂÂÂÂ map_lock;

So we use map_lock to protect both LPI's and VLPI's mapping affinity of
a device, and use vpe_lock to protect vPE's affinity, OK.

>  ÂÂÂÂ struct its_vmÂÂÂÂÂÂÂ *vm;
>  ÂÂÂÂ struct its_vlpi_mapÂÂÂ *vlpi_maps;
>  ÂÂÂÂ intÂÂÂÂÂÂÂÂÂÂÂ nr_vlpis;
> @@ -240,15 +240,33 @@ static struct its_vlpi_map *get_vlpi_map(struct 
> irq_data *d)
>  ÂÂÂÂ return NULL;
>  Â}
>
> -static int irq_to_cpuid(struct irq_data *d)
> +static int irq_to_cpuid_lock(struct irq_data *d, unsigned long *flags)
>  Â{
> -ÂÂÂ struct its_device *its_dev = irq_data_get_irq_chip_data(d);
>  ÂÂÂÂ struct its_vlpi_map *map = get_vlpi_map(d);
> +ÂÂÂ int cpu;
>
> -ÂÂÂ if (map)
> -ÂÂÂÂÂÂÂ return map->vpe->col_idx;
> +ÂÂÂ if (map) {
> +ÂÂÂÂÂÂÂ raw_spin_lock_irqsave(&map->vpe->vpe_lock, *flags);
> +ÂÂÂÂÂÂÂ cpu = map->vpe->col_idx;
> +ÂÂÂ } else {
> +ÂÂÂÂÂÂÂ struct its_device *its_dev = irq_data_get_irq_chip_data(d);
> +ÂÂÂÂÂÂÂ raw_spin_lock_irqsave(&its_dev->event_map.map_lock, *flags);
> +ÂÂÂÂÂÂÂ cpu = its_dev->event_map.col_map[its_get_event_id(d)];
> +ÂÂÂ }
>
> -ÂÂÂ return its_dev->event_map.col_map[its_get_event_id(d)];
> +ÂÂÂ return cpu;
> +}

This helper is correct for normal LPIs and VLPIs, but wrong for per-vPE
IRQ (doorbell) and vSGIs. irq_data_get_irq_chip_data() gets confused by
both of them.

> +
> +static void irq_to_cpuid_unlock(struct irq_data *d, unsigned long flags)
> +{
> +ÂÂÂ struct its_vlpi_map *map = get_vlpi_map(d);
> +
> +ÂÂÂ if (map) {
> +ÂÂÂÂÂÂÂ raw_spin_unlock_irqrestore(&map->vpe->vpe_lock, flags);
> +ÂÂÂ } else {
> +ÂÂÂÂÂÂÂ struct its_device *its_dev = irq_data_get_irq_chip_data(d);
> +ÂÂÂÂÂÂÂ raw_spin_unlock_irqrestore(&its_dev->event_map.map_lock, flags);
> +ÂÂÂ }
>  Â}

The same problem for this helper.

>  Âstatic struct its_collection *valid_col(struct its_collection *col)
> @@ -1384,6 +1402,8 @@ static void direct_lpi_inv(struct irq_data *d)
>  Â{
>  ÂÂÂÂ struct its_vlpi_map *map = get_vlpi_map(d);
>  ÂÂÂÂ void __iomem *rdbase;
> +ÂÂÂ unsigned long flags;
> +ÂÂÂ int cpu;
>  ÂÂÂÂ u64 val;
>
>  ÂÂÂÂ if (map) {
> @@ -1399,10 +1419,12 @@ static void direct_lpi_inv(struct irq_data *d)
>  ÂÂÂÂ }
>
>  ÂÂÂÂ /* Target the redistributor this LPI is currently routed to */
> -ÂÂÂ rdbase = per_cpu_ptr(gic_rdists->rdist, irq_to_cpuid(d))->rd_base;
> +ÂÂÂ cpu = irq_to_cpuid_lock(d, &flags);
> +ÂÂÂ rdbase = per_cpu_ptr(gic_rdists->rdist, cpu)->rd_base;
>  ÂÂÂÂ gic_write_lpir(val, rdbase + GICR_INVLPIR);
>
>  ÂÂÂÂ wait_for_syncr(rdbase);
> +ÂÂÂ irq_to_cpuid_unlock(d, flags);
>  Â}
>
>  Âstatic void lpi_update_config(struct irq_data *d, u8 clr, u8 set)
> @@ -1471,11 +1493,11 @@ static void its_unmask_irq(struct irq_data *d)
>  Âstatic int its_set_affinity(struct irq_data *d, const struct cpumask 
> *mask_val,
>  ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ bool force)
>  Â{
> -ÂÂÂ unsigned int cpu;
>  ÂÂÂÂ const struct cpumask *cpu_mask = cpu_online_mask;
>  ÂÂÂÂ struct its_device *its_dev = irq_data_get_irq_chip_data(d);
>  ÂÂÂÂ struct its_collection *target_col;
> -ÂÂÂ u32 id = its_get_event_id(d);
> +ÂÂÂ unsigned int from, cpu;
> +ÂÂÂ unsigned long flags;
>
>  ÂÂÂÂ /* A forwarded interrupt should use irq_set_vcpu_affinity */
>  ÂÂÂÂ if (irqd_is_forwarded_to_vcpu(d))
> @@ -1496,12 +1518,16 @@ static int its_set_affinity(struct irq_data *d, 
> const struct cpumask *mask_val,
>  ÂÂÂÂÂÂÂÂ return -EINVAL;
>
>  ÂÂÂÂ /* don't set the affinity when the target cpu is same as current 
> one */
> -ÂÂÂ if (cpu != its_dev->event_map.col_map[id]) {
> +ÂÂÂ from = irq_to_cpuid_lock(d, &flags);
> +ÂÂÂ if (cpu != from) {
> +ÂÂÂÂÂÂÂ u32 id = its_get_event_id(d);
> +
>  ÂÂÂÂÂÂÂÂ target_col = &its_dev->its->collections[cpu];
>  ÂÂÂÂÂÂÂÂ its_send_movi(its_dev, target_col, id);
>  ÂÂÂÂÂÂÂÂ its_dev->event_map.col_map[id] = cpu;
>  ÂÂÂÂÂÂÂÂ irq_data_update_effective_affinity(d, cpumask_of(cpu));
>  ÂÂÂÂ }
> +ÂÂÂ irq_to_cpuid_unlock(d, flags);
>
>  ÂÂÂÂ return IRQ_SET_MASK_OK_DONE;
>  Â}
> @@ -1636,7 +1662,7 @@ static int its_vlpi_map(struct irq_data *d, struct 
> its_cmd_info *info)
>  ÂÂÂÂ if (!info->map)
>  ÂÂÂÂÂÂÂÂ return -EINVAL;
>
> -ÂÂÂ raw_spin_lock(&its_dev->event_map.vlpi_lock);
> +ÂÂÂ raw_spin_lock(&its_dev->event_map.map_lock);
>
>  ÂÂÂÂ if (!its_dev->event_map.vm) {
>  ÂÂÂÂÂÂÂÂ struct its_vlpi_map *maps;
> @@ -1685,7 +1711,7 @@ static int its_vlpi_map(struct irq_data *d, struct 
> its_cmd_info *info)
>  ÂÂÂÂ }
>
>  Âout:
> -ÂÂÂ raw_spin_unlock(&its_dev->event_map.vlpi_lock);
> +ÂÂÂ raw_spin_unlock(&its_dev->event_map.map_lock);
>  ÂÂÂÂ return ret;
>  Â}
>
> @@ -1695,7 +1721,7 @@ static int its_vlpi_get(struct irq_data *d, struct 
> its_cmd_info *info)
>  ÂÂÂÂ struct its_vlpi_map *map;
>  ÂÂÂÂ int ret = 0;
>
> -ÂÂÂ raw_spin_lock(&its_dev->event_map.vlpi_lock);
> +ÂÂÂ raw_spin_lock(&its_dev->event_map.map_lock);
>
>  ÂÂÂÂ map = get_vlpi_map(d);
>
> @@ -1708,7 +1734,7 @@ static int its_vlpi_get(struct irq_data *d, struct 
> its_cmd_info *info)
>  ÂÂÂÂ *info->map = *map;
>
>  Âout:
> -ÂÂÂ raw_spin_unlock(&its_dev->event_map.vlpi_lock);
> +ÂÂÂ raw_spin_unlock(&its_dev->event_map.map_lock);
>  ÂÂÂÂ return ret;
>  Â}
>
> @@ -1718,7 +1744,7 @@ static int its_vlpi_unmap(struct irq_data *d)
>  ÂÂÂÂ u32 event = its_get_event_id(d);
>  ÂÂÂÂ int ret = 0;
>
> -ÂÂÂ raw_spin_lock(&its_dev->event_map.vlpi_lock);
> +ÂÂÂ raw_spin_lock(&its_dev->event_map.map_lock);
>
>  ÂÂÂÂ if (!its_dev->event_map.vm || !irqd_is_forwarded_to_vcpu(d)) {
>  ÂÂÂÂÂÂÂÂ ret = -EINVAL;
> @@ -1748,7 +1774,7 @@ static int its_vlpi_unmap(struct irq_data *d)
>  ÂÂÂÂ }
>
>  Âout:
> -ÂÂÂ raw_spin_unlock(&its_dev->event_map.vlpi_lock);
> +ÂÂÂ raw_spin_unlock(&its_dev->event_map.map_lock);
>  ÂÂÂÂ return ret;
>  Â}
>
> @@ -3193,7 +3219,7 @@ static struct its_device *its_create_device(struct 
> its_node *its, u32 dev_id,
>  ÂÂÂÂ dev->event_map.col_map = col_map;
>  ÂÂÂÂ dev->event_map.lpi_base = lpi_base;
>  ÂÂÂÂ dev->event_map.nr_lpis = nr_lpis;
> -ÂÂÂ raw_spin_lock_init(&dev->event_map.vlpi_lock);
> +ÂÂÂ raw_spin_lock_init(&dev->event_map.map_lock);
>  ÂÂÂÂ dev->device_id = dev_id;
>  ÂÂÂÂ INIT_LIST_HEAD(&dev->entry);
>
> @@ -3560,6 +3586,7 @@ static int its_vpe_set_affinity(struct irq_data *d,
>  Â{
>  ÂÂÂÂ struct its_vpe *vpe = irq_data_get_irq_chip_data(d);
>  ÂÂÂÂ int from, cpu = cpumask_first(mask_val);
> +ÂÂÂ unsigned long flags;
>
>  ÂÂÂÂ /*
>  ÂÂÂÂÂ * Changing affinity is mega expensive, so let's be as lazy as
> @@ -3567,6 +3594,7 @@ static int its_vpe_set_affinity(struct irq_data *d,
>  ÂÂÂÂÂ * into the proxy device, we need to move the doorbell
>  ÂÂÂÂÂ * interrupt to its new location.
>  ÂÂÂÂÂ */
> +ÂÂÂ raw_spin_lock_irqsave(&vpe->vpe_lock, flags);
>  ÂÂÂÂ if (vpe->col_idx == cpu)
>  ÂÂÂÂÂÂÂÂ goto out;
>
> @@ -3586,6 +3614,7 @@ static int its_vpe_set_affinity(struct irq_data *d,
>
>  Âout:
>  ÂÂÂÂ irq_data_update_effective_affinity(d, cpumask_of(cpu));
> +ÂÂÂ raw_spin_unlock_irqrestore(&vpe->vpe_lock, flags);
>
>  ÂÂÂÂ return IRQ_SET_MASK_OK_DONE;
>  Â}
> @@ -3695,11 +3724,15 @@ static void its_vpe_send_inv(struct irq_data *d)
>
>  ÂÂÂÂ if (gic_rdists->has_direct_lpi) {
>  ÂÂÂÂÂÂÂÂ void __iomem *rdbase;
> +ÂÂÂÂÂÂÂ unsigned long flags;
> +ÂÂÂÂÂÂÂ int cpu;
>
>  ÂÂÂÂÂÂÂÂ /* Target the redistributor this VPE is currently known on */
> -ÂÂÂÂÂÂÂ rdbase = per_cpu_ptr(gic_rdists->rdist, vpe->col_idx)->rd_base;
> +ÂÂÂÂÂÂÂ cpu = irq_to_cpuid_lock(d, &flags);
> +ÂÂÂÂÂÂÂ rdbase = per_cpu_ptr(gic_rdists->rdist, cpu)->rd_base;
>  ÂÂÂÂÂÂÂÂ gic_write_lpir(d->parent_data->hwirq, rdbase + GICR_INVLPIR);
>  ÂÂÂÂÂÂÂÂ wait_for_syncr(rdbase);
> +ÂÂÂÂÂÂÂ irq_to_cpuid_unlock(d, flags);
>  ÂÂÂÂ } else {
>  ÂÂÂÂÂÂÂÂ its_vpe_send_cmd(vpe, its_send_inv);
>  ÂÂÂÂ }

Do we really need to grab the vpe_lock for those which are belong to
the same irqchip with its_vpe_set_affinity()? The IRQ core code should
already ensure the mutual exclusion among them, wrong?

> @@ -3735,14 +3768,18 @@ static int its_vpe_set_irqchip_state(struct 
> irq_data *d,
>
>  ÂÂÂÂ if (gic_rdists->has_direct_lpi) {
>  ÂÂÂÂÂÂÂÂ void __iomem *rdbase;
> +ÂÂÂÂÂÂÂ unsigned long flags;
> +ÂÂÂÂÂÂÂ int cpu;
>
> -ÂÂÂÂÂÂÂ rdbase = per_cpu_ptr(gic_rdists->rdist, vpe->col_idx)->rd_base;
> +ÂÂÂÂÂÂÂ cpu = irq_to_cpuid_lock(d, &flags);
> +ÂÂÂÂÂÂÂ rdbase = per_cpu_ptr(gic_rdists->rdist, cpu)->rd_base;
>  ÂÂÂÂÂÂÂÂ if (state) {
>  ÂÂÂÂÂÂÂÂÂÂÂÂ gic_write_lpir(vpe->vpe_db_lpi, rdbase + GICR_SETLPIR);
>  ÂÂÂÂÂÂÂÂ } else {
>  ÂÂÂÂÂÂÂÂÂÂÂÂ gic_write_lpir(vpe->vpe_db_lpi, rdbase + GICR_CLRLPIR);
>  ÂÂÂÂÂÂÂÂÂÂÂÂ wait_for_syncr(rdbase);
>  ÂÂÂÂÂÂÂÂ }
> +ÂÂÂÂÂÂÂ irq_to_cpuid_unlock(d, flags);
>  ÂÂÂÂ } else {
>  ÂÂÂÂÂÂÂÂ if (state)
>  ÂÂÂÂÂÂÂÂÂÂÂÂ its_vpe_send_cmd(vpe, its_send_int);
> @@ -3854,14 +3891,17 @@ static void its_vpe_4_1_deschedule(struct 
> its_vpe *vpe,
>  Âstatic void its_vpe_4_1_invall(struct its_vpe *vpe)
>  Â{
>  ÂÂÂÂ void __iomem *rdbase;
> +ÂÂÂ unsigned long flags;
>  ÂÂÂÂ u64 val;
>
>   val = GICR_INVALLR_V;
>  ÂÂÂÂ val |= FIELD_PREP(GICR_INVALLR_VPEID, vpe->vpe_id);
>
>  ÂÂÂÂ /* Target the redistributor this vPE is currently known on */
> +ÂÂÂ raw_spin_lock_irqsave(&vpe->vpe_lock, flags);
>  ÂÂÂÂ rdbase = per_cpu_ptr(gic_rdists->rdist, vpe->col_idx)->rd_base;
>  ÂÂÂÂ gic_write_lpir(val, rdbase + GICR_INVALLR);
> +ÂÂÂ raw_spin_unlock_irqrestore(&vpe->vpe_lock, flags);
>  Â}
>
>  Âstatic int its_vpe_4_1_set_vcpu_affinity(struct irq_data *d, void 
> *vcpu_info)
> @@ -3960,13 +4000,17 @@ static int its_sgi_get_irqchip_state(struct 
> irq_data *d,
>  ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ enum irqchip_irq_state which, bool *val)
>  Â{
>  ÂÂÂÂ struct its_vpe *vpe = irq_data_get_irq_chip_data(d);
> -ÂÂÂ void __iomem *base = gic_data_rdist_cpu(vpe->col_idx)->rd_base + 
> SZ_128K;
> +ÂÂÂ void __iomem *base;
> +ÂÂÂ unsigned long flags;
>  ÂÂÂÂ u32 count = 1000000;ÂÂÂ /* 1s! */
>  ÂÂÂÂ u32 status;
> +ÂÂÂ int cpu;
>
>  ÂÂÂÂ if (which != IRQCHIP_STATE_PENDING)
>  ÂÂÂÂÂÂÂÂ return -EINVAL;
>
> +ÂÂÂ cpu = irq_to_cpuid_lock(d, &flags);
> +ÂÂÂ base = gic_data_rdist_cpu(cpu)->rd_base + SZ_128K;
>  ÂÂÂÂ writel_relaxed(vpe->vpe_id, base + GICR_VSGIR);
>  ÂÂÂÂ do {
>  ÂÂÂÂÂÂÂÂ status = readl_relaxed(base + GICR_VSGIPENDR);
> @@ -3983,6 +4027,7 @@ static int its_sgi_get_irqchip_state(struct 
> irq_data *d,
>  ÂÂÂÂ } while(count);
>
>  Âout:
> +ÂÂÂ irq_to_cpuid_unlock(d, flags);
>  ÂÂÂÂ *val = !!(status & (1 << d->hwirq));
>
>  ÂÂÂÂ return 0;
> @@ -4102,6 +4147,7 @@ static int its_vpe_init(struct its_vpe *vpe)
>  ÂÂÂÂÂÂÂÂ return -ENOMEM;
>  ÂÂÂÂ }
>
> +ÂÂÂ raw_spin_lock_init(&vpe->vpe_lock);
>  ÂÂÂÂ vpe->vpe_id = vpe_id;
>  ÂÂÂÂ vpe->vpt_page = vpt_page;
>  ÂÂÂÂ if (gic_rdists->has_rvpeid)
> diff --git a/include/linux/irqchip/arm-gic-v4.h 
> b/include/linux/irqchip/arm-gic-v4.h
> index 46c167a6349f..fc43a63875a3 100644
> --- a/include/linux/irqchip/arm-gic-v4.h
> +++ b/include/linux/irqchip/arm-gic-v4.h
> @@ -60,6 +60,7 @@ struct its_vpe {
>  ÂÂÂÂÂÂÂÂ };
>  ÂÂÂÂ };
>
> +ÂÂÂ raw_spinlock_tÂÂÂÂÂÂÂ vpe_lock;
>  ÂÂÂÂ /*
>  ÂÂÂÂÂ * This collection ID is used to indirect the target
>  ÂÂÂÂÂ * redistributor for this VPE. The ID itself isn't involved in

I'm not sure if it's good enough, it may gets much clearer after
splitting.


Thanks,
Zenghui