@@ -11,6 +11,7 @@
#include <linux/swiotlb.h>
#include <xen/xen.h>
+#include <xen/interface/grant_table.h>
#include <xen/interface/memory.h>
#include <xen/swiotlb-xen.h>
@@ -44,6 +45,8 @@ static inline void *kmap_high_get(struct page *page)
static inline void kunmap_high(struct page *page) {}
#endif
+static bool hypercall_flush = false;
+
/* functions called by SWIOTLB */
@@ -60,17 +63,35 @@ static void dma_cache_maint(dma_addr_t handle, unsigned long offset,
do {
size_t len = left;
void *vaddr;
+
+ /* buffers in highmem or foreign pages cannot cross page
+ * boundaries */
+ if (len + offset > PAGE_SIZE)
+ len = PAGE_SIZE - offset;
if (!pfn_valid(pfn))
{
- /* TODO: cache flush */
+ struct gnttab_cache_flush cflush;
+
+ cflush.op = 0;
+ cflush.a.dev_bus_addr = pfn << PAGE_SHIFT;
+ cflush.offset = offset;
+ cflush.length = len;
+
+ if (op == dmac_unmap_area && dir != DMA_TO_DEVICE)
+ cflush.op = GNTTAB_CACHE_INVAL;
+ if (op == dmac_map_area) {
+ if (dir == DMA_FROM_DEVICE)
+ cflush.op = GNTTAB_CACHE_INVAL;
+ else
+ cflush.op = GNTTAB_CACHE_CLEAN;
+ }
+ if (cflush.op)
+ HYPERVISOR_grant_table_op(GNTTABOP_cache_flush, &cflush, 1);
} else {
struct page *page = pfn_to_page(pfn);
if (PageHighMem(page)) {
- if (len + offset > PAGE_SIZE)
- len = PAGE_SIZE - offset;
-
if (cache_is_vipt_nonaliasing()) {
vaddr = kmap_atomic(page);
op(vaddr + offset, len, dir);
@@ -143,7 +164,7 @@ bool xen_arch_need_swiotlb(struct device *dev,
unsigned long pfn,
unsigned long mfn)
{
- return ((pfn != mfn) && !is_dma_coherent(dev));
+ return (!hypercall_flush && (pfn != mfn) && !is_dma_coherent(dev));
}
int xen_create_contiguous_region(phys_addr_t pstart, unsigned int order,
@@ -186,10 +207,18 @@ static struct dma_map_ops xen_swiotlb_dma_ops = {
int __init xen_mm_init(void)
{
+ struct gnttab_cache_flush cflush;
if (!xen_initial_domain())
return 0;
xen_swiotlb_init(1, false);
xen_dma_ops = &xen_swiotlb_dma_ops;
+
+ cflush.op = 0;
+ cflush.a.dev_bus_addr = 0;
+ cflush.offset = 0;
+ cflush.length = 0;
+ if (HYPERVISOR_grant_table_op(GNTTABOP_cache_flush, &cflush, 1) != -ENOSYS)
+ hypercall_flush = true;
return 0;
}
-arch_initcall(xen_mm_init);
+arch_initcall(xen_mm_init)
@@ -479,6 +479,25 @@ struct gnttab_get_version {
DEFINE_GUEST_HANDLE_STRUCT(gnttab_get_version);
/*
+ * Issue one or more cache maintenance operations on a portion of a
+ * page granted to the calling domain by a foreign domain.
+ */
+#define GNTTABOP_cache_flush 12
+struct gnttab_cache_flush {
+ union {
+ uint64_t dev_bus_addr;
+ grant_ref_t ref;
+ } a;
+ uint16_t offset; /* offset from start of grant */
+ uint16_t length; /* size within the grant */
+#define GNTTAB_CACHE_CLEAN (1<<0)
+#define GNTTAB_CACHE_INVAL (1<<1)
+#define GNTTAB_CACHE_SOURCE_GREF (1<<31)
+ uint32_t op;
+};
+DEFINE_GUEST_HANDLE_STRUCT(gnttab_cache_flush);
+
+/*
* Bitfield values for update_pin_status.flags.
*/
/* Map the grant entry for access by I/O devices. */
Introduce support for new hypercall GNTTABOP_cache_flush. Use it to perform cache flashing on pages used for dma when necessary. If GNTTABOP_cache_flush is supported by the hypervisor, we don't need to bounce dma map operations that involve foreign grants and non-coherent devices. Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com> --- Changes in v4: - add comment; - avoid bouncing dma map operations that involve foreign grants and non-coherent devices if GNTTABOP_cache_flush is provided by Xen. Changes in v3: - fix the cache maintenance op call to match what Linux does natively; - update the hypercall interface to match Xen. Changes in v2: - update the hypercall interface to match Xen; - call the interface on a single page at a time. --- arch/arm/xen/mm.c | 41 ++++++++++++++++++++++++++++++----- include/xen/interface/grant_table.h | 19 ++++++++++++++++ 2 files changed, 54 insertions(+), 6 deletions(-)