Re: [patch 2/4] PNP: replace pnp_resource_table with dynamically allocated resources

[Bjorn Helgaas <bjorn.helgaas@hp.com>]

On Monday 19 May 2008 05:14:07 pm Rene Herman wrote:
> On 16-05-08 00:07, Bjorn Helgaas wrote:
> 
> > PNP used to have a fixed-size pnp_resource_table for tracking the
> > resources used by a device.  This table often overflowed, so we've
> > had to increase the table size, which wastes memory because most
> > devices have very few resources.
> > 
> > This patch replaces the table with a linked list of resources where
> > the entries are allocated on demand.
> 
> [ ... ]
> 
> >  static void pnp_clean_resource_table(struct pnp_dev *dev)
> >  {
> > -	struct resource *res;
> > -	int idx;
> > +	struct pnp_resource *pnp_res, *tmp;
> >  
> > -	for (idx = 0; idx < PNP_MAX_IRQ; idx++) {
> > -		res = &dev->res->irq[idx].res;
> > -		if (res->flags & IORESOURCE_AUTO) {
> > -			res->flags = IORESOURCE_IRQ;
> > -			pnp_init_resource(res);
> > -		}
> > -	}
> > -	for (idx = 0; idx < PNP_MAX_DMA; idx++) {
> > -		res = &dev->res->dma[idx].res;
> > -		if (res->flags & IORESOURCE_AUTO) {
> > -			res->flags = IORESOURCE_DMA;
> > -			pnp_init_resource(res);
> > -		}
> > -	}
> > -	for (idx = 0; idx < PNP_MAX_PORT; idx++) {
> > -		res = &dev->res->port[idx].res;
> > -		if (res->flags & IORESOURCE_AUTO) {
> > -			res->flags = IORESOURCE_IO;
> > -			pnp_init_resource(res);
> > -		}
> > -	}
> > -	for (idx = 0; idx < PNP_MAX_MEM; idx++) {
> > -		res = &dev->res->mem[idx].res;
> > -		if (res->flags & IORESOURCE_AUTO) {
> > -			res->flags = IORESOURCE_MEM;
> > -			pnp_init_resource(res);
> > -		}
> > +	list_for_each_entry_safe(pnp_res, tmp, &dev->resources, list) {
> > +		if (pnp_res->res.flags & IORESOURCE_AUTO)
> > +			pnp_free_resource(pnp_res);
> >  	}
> >  }
> 
> 
> Do correct me if I'm wrong but I don't believe this will do. The index 
> for the resources is preserved simply due to the list position after the 
> DISABLED thing but here the list is reshuffled.
> 
> So say I have an ISAPnP device with 2 port resources the second of which 
> I force to manual setting through sysfs (ie, AUTO is cleared). This API 
> would then delete the first port resource after which the second port 
> resource is the first entry in the list which would make for example 
> isapnp_set_resource() program it into hardware index 1, no?

My reasoning was that all AUTO entries should be at the end of the
list, so deleting them should not change the order of other entries.

I don't think there's a way to create a non-AUTO entry that follows
an AUTO one.  We add non-AUTO resources in:
  - isapnp_get_resources()
  - pnpbios_parse_allocated_resource_data()
  - pnpacpi_parse_allocated_resource()
  - pnp_set_current_resources() (sysfs interface)
and in all those cases, we use pnp_init_resources() first, which
clears out all existing resources.

We add AUTO resources in pnp_assign_{port,mem,etc}(), and those bail
out early if there's already a resource at the specified index.

All this is certainly not obvious from the code and it took me a
long time to convince myself of it, so I could easily be mistaken.

Bjorn