Re: [PATCH 1/5] Make linux drivers optional

[Samuel Thibault]

Damien Zammit, le ven. 26 mars 2021 20:48:46 +1100, a ecrit:
> +#ifndef LINUX_DEV
> +struct intr_list {
> +  user_intr_t *user_intr;
> +  struct intr_list *next;
> +};
> +static struct intr_list *user_intr_handlers;
> +#endif
> +
>  static user_intr_t *
>  search_intr (struct irqdev *dev, ipc_port_t dst_port)
>  {
> @@ -147,6 +156,33 @@ out:
>    return ret;
>  }
>  
> +#ifndef LINUX_DEV
> +int
> +install_user_intr_handler (struct irqdev *dev, int id, unsigned long flags,
> +                       user_intr_t *user_intr)
> +{
> +  unsigned int irq = dev->irq[id];
> +  struct intr_list *old = user_intr_handlers;
> +
> +  assert (irq < NINTR);
> +
> +  while (old)
> +    {
> +      /* Test whether the irq handler has been set */
> +      if (old->user_intr->dst_port && old->user_intr->dst_port == 
> user_intr->dst_port)

Did you mean old->user_intr like the linux code does?

> +        {
> +          printf ("This interrupt handler has already been installed on line 
> %d", irq);
> +          return 1;
> +        }
> +      old = old->next;
> +    }
> +
> +  old = kalloc (sizeof (struct intr_list));
> +  *(old->user_intr) = *user_intr;

? did you mean
old->user_intr = user_intr;
?

Also, you need to add the new entry to the user_intr_handlers list.

> +  return deliver_user_intr(dev, irq, old->user_intr);

?? what for in install_user_intr_handler??



Also, this patch seems to be lacking the actual interrupt delivery.
The interrupt delivery also has to check for the value returned by
deliver_user_intr, so the entry can be removed whenever it is not needed
any more. See how linux_intr does it.

> diff --git a/linux/configfrag.ac b/linux/configfrag.ac
> index 78b59d7f..26c99e45 100644
> --- a/linux/configfrag.ac
> +++ b/linux/configfrag.ac

I do not understand this part without explanation of your approach.

> @@ -33,6 +33,8 @@ dnl USE OF THIS SOFTWARE.
> +    qemu)
> +      device_driver_group_qemu=selected;;

? but qemu drivers is not a group?

> +AC_Linux_DRIVER([ne],
> +  [Ethernet controller NE2000/NE1000 ISA (ne, ne1000, ne2000)],
> +  [CONFIG_NE2000],
> +  [qemu])

No, ne is part of the net group.

What is the actual intermediate problem you are trying to solve here by
making qemu a group?

Put another way, is it not possible to add a block group and use

--enable-block-group=no --enable-net-group=no --enable-scsi-group=no 
--enable-wireless-group=no --enable-pcmcia-group=no

?

Samuel