Re: [PATCH] STAGING: Comedi: Build only on arches providing PAGE_KERNEL_NOCACHE

On 23/06/11 13:02, Ralf Baechle wrote:

On Thu, Jun 23, 2011 at 12:53:36PM +0100, Martyn Welch wrote:

quoted

On 23/06/11 12:45, Ralf Baechle wrote:

quoted

On architectures that don't define PAGE_KERNEL_NOCACHE, the Comedi driver turns
into tragedy:

  CC [M]  drivers/staging/comedi/drivers.o
drivers/staging/comedi/drivers.c: In function ‘comedi_buf_alloc’:
drivers/staging/comedi/drivers.c:505:41: error: ‘PAGE_KERNEL_NOCACHE’ undeclared (first use in this function)
drivers/staging/comedi/drivers.c:505:41: note: each undeclared identifier is rep orted only once for each function it appears in
make[3]: *** [drivers/staging/comedi/drivers.o] Error 1

Restrict the driver to only those architectures that define PAGE_KERNEL_NOCACHE.

PAGE_KERNEL_NOCACHE is a kludge - some system architectures such as SGI IP27
are even uable to offer uncached operation - at least in the way an unwitting
driver might assume.  I haven't looked in details how the driver is using
the area vmaped with PAGE_KERNEL_NOCACHE but maybe doing it XFS-style using
cached memory and the flush_kernel_vmap_range / invalidate_kernel_vmap_range
APIs in conjunction with the DMA API is a practical alternative.

Signed-off-by: Ralf Baechle <redacted>

 drivers/staging/comedi/Kconfig |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/drivers/staging/comedi/Kconfig b/drivers/staging/comedi/Kconfig
index 1502d80..bccdc12 100644
--- a/drivers/staging/comedi/Kconfig
+++ b/drivers/staging/comedi/Kconfig

@@ -2,6 +2,7 @@ config COMEDI
 	tristate "Data acquisition support (comedi)"
 	default N
 	depends on m
+	depends on BROKEN || FRV || M32R || MN10300 || SUPERH || TILE || X86

I'm sure I got comedi to compile on a 32-bit PPC board not that long ago. Has
something changed, or is this just not an exhaustive list?

(Adding the PPC folks to cc.)

A "git grep -w PAGE_KERNEL_NOCACHE arch/powerpc/" doesn't find anything so
I don't think the driver will build there.  I don't have a PPC toolchain
to verify that.

Ah, just found the tree looks like I had a little patch to make it seemingly
work, though it will make it "not work" for most other archs at a guess (it
may not even be right on PPC):

diff --git a/drivers/staging/comedi/drivers.c b/drivers/staging/comedi/drivers.c
index 44d6b62..263ad3d 100644
--- a/drivers/staging/comedi/drivers.c
+++ b/drivers/staging/comedi/drivers.c

@@ -505,7 +505,7 @@ int comedi_buf_alloc(struct comedi_device *dev, struct

comedi_subdevice *s,
                }
                if (i == n_pages) {
                        async->prealloc_buf =
-                           vmap(pages, n_pages, VM_MAP, PAGE_KERNEL_NOCACHE);
+                           vmap(pages, n_pages, VM_MAP, PAGE_KERNEL_NC);
                }
                vfree(pages);

(Sorry for the wrapping)

-- 
Martyn Welch (Principal Software Engineer) | Registered in England and
GE Intelligent Platforms                   | Wales (3828642) at 100
T +44(0)127322748                          | Barbirolli Square, Manchester,
E martyn.welch@ge.com                      | M2 3AB  VAT:GB 927559189