NTB: Improve performance with write combining

Changing the memory window BAR mappings to write combining significantly
boosts the performance.  We will also use memcpy that uses non-temporal
store, which showed performance improvement when doing non-cached
memcpys.

Signed-off-by: Dave Jiang <dave.jiang@intel.com>
Signed-off-by: Jon Mason <jdmason@kudzu.us>
This commit is contained in:
Dave Jiang 2015-05-19 16:45:46 -04:00 committed by Jon Mason
parent 0e041fb536
commit 06917f7535
1 changed files with 10 additions and 1 deletions

View File

@ -58,6 +58,7 @@
#include <linux/pci.h> #include <linux/pci.h>
#include <linux/slab.h> #include <linux/slab.h>
#include <linux/types.h> #include <linux/types.h>
#include <linux/uaccess.h>
#include "linux/ntb.h" #include "linux/ntb.h"
#include "linux/ntb_transport.h" #include "linux/ntb_transport.h"
@ -993,7 +994,7 @@ static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev)
if (rc) if (rc)
goto err1; goto err1;
mw->vbase = ioremap(mw->phys_addr, mw->phys_size); mw->vbase = ioremap_wc(mw->phys_addr, mw->phys_size);
if (!mw->vbase) { if (!mw->vbase) {
rc = -ENOMEM; rc = -ENOMEM;
goto err1; goto err1;
@ -1375,7 +1376,15 @@ static void ntb_tx_copy_callback(void *data)
static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset) static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset)
{ {
#ifdef ARCH_HAS_NOCACHE_UACCESS
/*
* Using non-temporal mov to improve performance on non-cached
* writes, even though we aren't actually copying from user space.
*/
__copy_from_user_inatomic_nocache(offset, entry->buf, entry->len);
#else
memcpy_toio(offset, entry->buf, entry->len); memcpy_toio(offset, entry->buf, entry->len);
#endif
/* Ensure that the data is fully copied out before setting the flags */ /* Ensure that the data is fully copied out before setting the flags */
wmb(); wmb();