Messages in this thread Patch in this message | | | From | Allen Hubbe <> | Subject | [PATCH v2 14/17] NTB: Improve performance with write combining | Date | Thu, 21 May 2015 04:20:58 -0400 |
| |
From: Dave Jiang <dave.jiang@intel.com>
Changing the memory window BAR mappings to write combining significantly boosts the performance. We will also use memcpy that utilizies non-temporal store which showed performance improement when doing non-cached memcpys.
Signed-off-by: Dave Jiang <dave.jiang@intel.com> --- drivers/ntb/ntb_transport.c | 9 +++++++-- 1 file changed, 7 insertions(+), 2 deletions(-)
diff --git a/drivers/ntb/ntb_transport.c b/drivers/ntb/ntb_transport.c index 733c0ec..95252ea 100644 --- a/drivers/ntb/ntb_transport.c +++ b/drivers/ntb/ntb_transport.c @@ -58,6 +58,7 @@ #include <linux/pci.h> #include <linux/slab.h> #include <linux/types.h> +#include <linux/uaccess.h> #include "linux/ntb.h" #include "linux/ntb_transport.h" @@ -990,7 +991,7 @@ static int ntb_transport_probe(struct ntb_client *self, struct ntb_dev *ndev) if (rc) goto err1; - mw->vbase = ioremap(mw->phys_addr, mw->phys_size); + mw->vbase = ioremap_wc(mw->phys_addr, mw->phys_size); if (!mw->vbase) { rc = -ENOMEM; goto err1; @@ -1361,7 +1362,11 @@ static void ntb_tx_copy_callback(void *data) static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset) { - memcpy_toio(offset, entry->buf, entry->len); + /* + * 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); /* Ensure that the data is fully copied out before setting the flags */ wmb(); -- 2.4.0.rc0.43.gcf8a8c6
| |