On Mon, Jan 26, 2015 at 7:22 PM, Bjorn Helgaas [off-list ref] wrote:

On Thu, Jan 22, 2015 at 02:33:36PM -0600, Bjorn Helgaas wrote:

quoted

On Mon, Jan 12, 2015 at 01:05:12AM +0100, Linus Walleij wrote:

quoted

On Sat, Jan 10, 2015 at 10:53 PM, Arnd Bergmann [off-list ref] wrote:

quoted

On Saturday 10 January 2015 22:40:22 Linus Walleij wrote:

quoted

quoted

 static int v3_read_config(struct pci_bus *bus, unsigned int devfn, int where,
                          int size, u32 *val)
 {
-       addr = v3_open_config_window(bus, devfn, where);
+       int ret = pci_generic_config_read(bus, devfn, where, size, val);
        v3_close_config_window();
+       return ret;
 }

 static int v3_write_config(struct pci_bus *bus, unsigned int devfn, int where,
                           int size, u32 val)
 {
+       int ret = pci_generic_config_write(bus, devfn, where, size, val);
        v3_close_config_window();
-       raw_spin_unlock_irqrestore(&v3_lock, flags);
+       return ret;
 }

 static struct pci_ops pci_v3_ops = {
+       .map_bus = v3_open_config_window,
        .read   = v3_read_config,
        .write  = v3_write_config,

So .map_bus is called before every .read/.write operation I take it.

Wouldn't it be proper to call the v3_close_config_window() from a
matching .unmap_bus() callback for symmetry?

It would be nicer for integrator but useless for anything else, so
I'd vote for leaving it the way Rob posted.

OK I buy that, throw in a comment about it in the code if there
is some time for iterating the patch.

Would you prefer something like the following instead?  It keeps the
v3_open/close symmetry, but it does break apart and duplicate some of the
logic from v3_open_config_window().

Hearing nothing, I kept Rob's initial version.  It can be tweaked later if
anybody wants to.  I just want to get all this into v3.20.

Sorry, I'm OK with either version, this latest iteration looks very nice.
Acked-by.

Yours,
Linus Walleij