Re: [Patch] ARCH_HAS_DEV_MEM patch for Xwindow on IA64machine

["Jan Beulich" <jbeulich@novell.com>]

>>> Jun Kamada <kama@jp.fujitsu.com> 04.07.07 13:43 >>>
>We would like to support Xwindow on IA64 by remapping machine address to
>pseudo physical address at PCI device initialization (pci_acpi_scan_root()
>in arch/ia64/pci/pci.c) but at "mmap /dev/mem" systemcall.
>With the new scheme, we need no longer the archtecture specific memory
>operations described above.

You only talk about X - isn't it possible that there are other users of this interface?

Also,

>--- a/drivers/xen/Kconfig	Thu Jun 21 15:12:20 2007 +0900
>+++ b/drivers/xen/Kconfig	Thu Jun 21 15:13:21 2007 +0900
>@@ -273,4 +273,8 @@ config XEN_SMPBOOT
> 	default y
> 	depends on SMP
> 
>+config XEN_DEVMEM
>+	bool
>+	default y
>+
> endif

This seems pretty useless unless ia64 doesn't use drivers/xen/Kconfig (which
would be a mistake I think).

>--- a/drivers/xen/Makefile	Thu Jun 21 15:12:20 2007 +0900
>+++ b/drivers/xen/Makefile	Thu Jun 21 15:13:21 2007 +0900
>@@ -5,9 +5,9 @@ obj-y	+= xenbus/
> obj-y	+= xenbus/
> obj-y	+= gntdev/
> obj-y	+= balloon/
>-obj-y	+= char/
> 
> obj-y	+= util.o
>+obj-$(CONFIG_XEN_DEVMEM)		+= char/
> obj-$(CONFIG_XEN_BLKDEV_BACKEND)	+= blkback/
> obj-$(CONFIG_XEN_BLKDEV_TAP)		+= blktap/
> obj-$(CONFIG_XEN_NETDEV_BACKEND)	+= netback/

This is certainly wrong. Even if driver/xen/char/ currently only has a
single file, you should descend into that directory. The conditional
(if needed at all) should then be in driver/xen/char/Makefile.

Jan