Hello Kyle,
In ccio-dma.c, I noticed that eventhought the prototype of the function ccio_clear_io_tlb() is:
static CCIO_INLINE void
ccio_clear_io_tlb(struct ioc *ioc, dma_addr_t iovp, size_t byte_cnt)
i.e. the second arg being iovp (afaik the io virtual page number) deduced from the iova (io virtual address)
it's so useless to:
iovp &= IOVP_MASK; /* clear offset bits, just want pagenum */
and the only place where it's called is:
ccio_clear_io_tlb(ioc, CCIO_IOVP(iova), saved_byte_cnt);
it's well an iovp given the macro definition:
#define CCIO_IOVP(iova) ((iova) & IOVP_MASK)
May I so suggest following patch:
--- a/drivers/parisc/ccio-dma.c 2008-07-19 21:17:37.000000000 +0000
+++ b/drivers/parisc/ccio-dma.c 2008-07-19 21:32:54.000000000 +0000
@@ -643,7 +643,6 @@
{
u32 chain_size = 1 << ioc->chainid_shift;
- iovp &= IOVP_MASK; /* clear offset bits, just want pagenum */
byte_cnt += chain_size;
while(byte_cnt > chain_size) {
=== <> ===
Tia,
J.
On Sat, Jul 19, 2008 at 09:34:08PM +0000, Joel Soete wrote:
> --- a/drivers/parisc/ccio-dma.c 2008-07-19 21:17:37.000000000 +0000
> +++ b/drivers/parisc/ccio-dma.c 2008-07-19 21:32:54.000000000 +0000
> @@ -643,7 +643,6 @@
> {
> u32 chain_size = 1 << ioc->chainid_shift;
>
> - iovp &= IOVP_MASK; /* clear offset bits, just want pagenum */
> byte_cnt += chain_size;
>
> while(byte_cnt > chain_size) {
applied.