On 14/02/17 00:44, Shanker Donthineni wrote:
> Hi Marc,
>
>
> On 01/17/2017 04:20 AM, Marc Zyngier wrote:
>> When we don't have the DirectLPI feature, we must work around the
>> architecture shortcomings to be able to perform the required
>> invalidation.
>>
>> For this, we create a fake device whose sole purpose is to
>> provide a way to issue a map/inv/unmap sequence (and the corresponding
>> sync operations). That's 6 commands and a full serialization point
>> to be able to do this.
>>
>> You just have hope the hypervisor won't do that too often...
>>
>> Signed-off-by: Marc Zyngier <[email protected]>
>> ---
>> drivers/irqchip/irq-gic-v3-its.c | 59
>> ++++++++++++++++++++++++++++++++++++++--
>> 1 file changed, 57 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/irqchip/irq-gic-v3-its.c
>> b/drivers/irqchip/irq-gic-v3-its.c
>> index 008fb71..3787579 100644
>> --- a/drivers/irqchip/irq-gic-v3-its.c
>> +++ b/drivers/irqchip/irq-gic-v3-its.c
>> @@ -133,6 +133,9 @@ struct its_device {
>> u32 device_id;
>> };
>>
>> +static struct its_device *vpe_proxy_dev;
>> +static DEFINE_RAW_SPINLOCK(vpe_proxy_dev_lock);
>> +
>> static LIST_HEAD(its_nodes);
>> static DEFINE_SPINLOCK(its_lock);
>> static struct rdists *gic_rdists;
>> @@ -993,8 +996,35 @@ static void lpi_update_config(struct irq_data *d, u8
>> clr, u8 set)
>> struct its_vpe *vpe = irq_data_get_irq_chip_data(d);
>> void __iomem *rdbase;
>>
>> - rdbase = per_cpu_ptr(gic_rdists->rdist,
>> vpe->col_idx)->rd_base;
>> - writeq_relaxed(d->hwirq, rdbase + GICR_INVLPIR);
>> + if (gic_rdists->has_direct_lpi) {
>> + rdbase = per_cpu_ptr(gic_rdists->rdist,
>> vpe->col_idx)->rd_base;
>> + writeq_relaxed(d->hwirq, rdbase + GICR_INVLPIR);
>> + } else {
>> + /*
>> + * This is insane.
>> + *
>> + * If a GICv4 doesn't implement Direct LPIs,
>> + * the only way to perform an invalidate is to
>> + * use a fake device to issue a MAP/INV/UNMAP
>> + * sequence. Since each of these commands has
>> + * a sync operation, this is really fast. Not.
>> + *
>> + * We always use event 0, and this serialize
>> + * all VPE invalidations in the system.
>> + *
>> + * Broken by design(tm).
>> + */
>> + unsigned long flags;
>> +
>> + raw_spin_lock_irqsave(&vpe_proxy_dev_lock, flags);
>> +
>> + vpe_proxy_dev->event_map.col_map[0] =
>> vpe->col_idx;
>> + its_send_mapvi(vpe_proxy_dev, vpe->vpe_db_lpi, 0);
>> + its_send_inv(vpe_proxy_dev, 0);
>> + its_send_discard(vpe_proxy_dev, 0);
>> +
>> + raw_spin_unlock_irqrestore(&vpe_proxy_dev_lock,
>> flags);
>> + }
>> }
>> }
>>
>> @@ -2481,6 +2511,31 @@ static struct irq_domain *its_init_vpe_domain(void)
>> struct fwnode_handle *handle;
>> struct irq_domain *domain;
>>
>> + if (gic_rdists->has_direct_lpi) {
>> + pr_info("ITS: Using DirectLPI for VPE invalidation\n");
>> + } else {
>> + struct its_node *its;
>> +
>> + list_for_each_entry(its, &its_nodes, entry) {
>> + u32 devid;
>> +
>> + if (!its->is_v4)
>> + continue;
>> +
>> + /* Use the last possible DevID */
>> + devid = GENMASK(its->device_ids - 1, 0);
> How do we know this 'devid' is not being used by real hardware devices?
You can't know it. Or rather, you find out once it is too late.
> I think we need some kind check in its_msi_prepare() to skip this device
> or WARN.
Yup. I've now added some code to that effect. I may add a command-line
option to specify a "safe" DevID in the future, but only if we hit this
in real life.
> Unfortunately Qualcomm doesn't support Direct LPI feature.
I feel sorry for you :-(. I don't understand why this wasn't made
mandatory with GICv4, because the amount of pain you have to go through
to invalidate a doorbell is unbelievable.
M.
--
Jazz is not dead. It just smells funny...
