Re: [PATCH V6 2/5] PCI: Adapt pci_register_io_range() for indirect-IO and PCI I/O translation

From: John Garry
Date: Tue Jan 31 2017 - 08:41:54 EST

On 31/01/2017 00:10, Bjorn Helgaas wrote:
On Tue, Jan 24, 2017 at 03:05:22PM +0800, zhichang.yuan wrote:
After indirect-IO is introduced, system must can assigned indirect-IO devices
with logical I/O ranges which are different from those for PCI I/O devices.
Otherwise, I/O accessors can't identify whether the I/O port is for memory
mapped I/O or indirect-IO.


We must assign logical I/O port space for indirect I/O such that the
I/O accessors can tell whether a logical I/O port refers to memory-
mapped I/O space or indirect I/O space.

It's better

As current helper, pci_register_io_range(), is used for PCI I/O ranges
registration and translation, indirect-IO devices should also apply these
helpers to manage the I/O ranges. It will be easy to ensure the assigned
logical I/O ranges unique.
But for indirect-IO devices, there is no cpu address. The current
pci_register_io_range() can not work for this case.

This patch makes some changes on the pci_register_io_range() to support the
I/O range registration with device's fwnode also. After this, the indirect-IO
devices can register the device-local I/O range to system logical I/O and
easily perform the translation between device-local I/O range and sytem
logical I/O range.

-int __weak pci_register_io_range(phys_addr_t addr, resource_size_t size)
+int __weak pci_register_io_range(struct fwnode_handle *node, phys_addr_t addr,
+ resource_size_t size, unsigned long *port)

Why is this __weak? It looks like it's been __weak since its
introduction by 41f8bba7f555 ("of/pci: Add pci_register_io_range() and
pci_pio_to_address()"), but I don't see any other implementations of

Can you add a patch that does nothing but make this non-weak?


+ /*
+ * powerpc and microblaze have their own registration,
+ * just look up the value here

Can you include a pointer to the powerpc and microblaze registration
code here? It's conceivable that somebody could generalize this
enough to support powerpc and microblaze as well.

It should be no problem

--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -34,6 +34,9 @@

#include <linux/pci_ids.h>

+/* the macro below flags an invalid cpu address
+ * and is used by IO special hosts */



Use conventional multi-line comment style:

* IO_RANGE_IOEXT flags an invalid CPU address ...

+#define IO_RANGE_IOEXT (resource_size_t)(-1ull)

And put this close to related things, e.g., pci_register_io_range(),
instead of just dropping it in at the top of the file.


* The PCI interface treats multi-function devices as independent
* devices. The slot/function address of each device is encoded
@@ -1197,8 +1200,8 @@ int __must_check pci_bus_alloc_resource(struct pci_bus *bus,
void *alignf_data);

-int pci_register_io_range(phys_addr_t addr, resource_size_t size);
+int pci_register_io_range(struct fwnode_handle *node, phys_addr_t addr,
+ resource_size_t size, unsigned long *port);
unsigned long pci_address_to_pio(phys_addr_t addr);
phys_addr_t pci_pio_to_address(unsigned long pio);
int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr);