Junling Ma, le mar. 04 août 2020 17:59:57 -0700, a ecrit:
> > On Aug 4, 2020, at 5:30 PM, Samuel Thibault wrote:
> >
> > Junling Ma, le mar. 04 août 2020 17:08:03 -0700, a ecrit:
> static user_intr_t *
> search_intr (struct irqdev *dev, ipc_port_t dst_port)
> {
> user_intr_t
> On Aug 4, 2020, at 5:30 PM, Samuel Thibault wrote:
>
> Junling Ma, le mar. 04 août 2020 17:08:03 -0700, a ecrit:
static user_intr_t *
search_intr (struct irqdev *dev, ipc_port_t dst_port)
{
user_intr_t *e;
- queue_iterate (dev->intr_queue, e, user_intr_t *, chain)
>>>
Junling Ma, le mar. 04 août 2020 17:11:06 -0700, a ecrit:
> In the next patch, we disallow deliver_user_intr from touching the queue, so
> we do not need to lock it.
We don't want separate patch to depend on each other. Each patch should
by its own just *work* so we can test all of them one after
Junling Ma, le mar. 04 août 2020 17:08:03 -0700, a ecrit:
> >> static user_intr_t *
> >> search_intr (struct irqdev *dev, ipc_port_t dst_port)
> >> {
> >> user_intr_t *e;
> >> - queue_iterate (dev->intr_queue, e, user_intr_t *, chain)
> >> + simple_lock(&dev->lock);
> >> + queue_iterate (&dev-
In the next patch, we disallow deliver_user_intr from touching the queue, so we
do not need to lock it. In fact, by doing that, we prevent interrupts from
pulling the rug under our feet, so there is no need to do splhigh/splx in
irq_acknowledge and intr_thread.
Junling
> On Aug 4, 2020, at 4:5
Hi Samual,
> On Aug 4, 2020, at 4:51 PM, Samuel Thibault wrote:
>
> Junling Ma, le dim. 02 août 2020 21:04:24 -0700, a ecrit:
>> -struct irqdev irqtab = {
>> - "irq", irq_eoi, &main_intr_queue, 0,
>> - {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15},
>> -};
>> +struct irqdev irqtab;
>>
Oh, btw, in queue_intr you also need to take the simple_lock which
modifying the entry values.
Samuel
Junling Ma, le dim. 02 août 2020 21:04:24 -0700, a ecrit:
> -struct irqdev irqtab = {
> - "irq", irq_eoi, &main_intr_queue, 0,
> - {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15},
> -};
> +struct irqdev irqtab;
> +
> +void irq_init()
> +{
> + irqtab.name = "irq";
> + irqtab.irqdev_ack
1. The queue manipulated by two threads, the registration server and the
intr_thread, so we need a lock to protect it.
2. To initialize the lock, we introduce a irq_init function in
i386/i386/irq.[ch] and call it in device/device_init.c
---
device/device_init.c | 3 ++-
device/intr.c|
