Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Todd Poynor <toddpoynor@gmail.com>
To:     Rob Springer <rspringer@google.com>,
        John Joseph <jnjoseph@google.com>,
        Ben Chan <benchan@chromium.org>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc:     devel@driverdev.osuosl.org, linux-kernel@vger.kernel.org,
        Todd Poynor <toddpoynor@google.com>
Subject: [PATCH 10/16] staging: gasket: page table: remove extraneous memory barriers
Date:   Thu,  9 Aug 2018 20:21:05 -0700
Message-Id: <20180810032111.197743-11-toddpoynor@gmail.com>
In-Reply-To: <20180810032111.197743-1-toddpoynor@gmail.com>
References: <20180810032111.197743-1-toddpoynor@gmail.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

From: Todd Poynor <toddpoynor@google.com>

Some explicit memory barriers in the page table code are not necessary,
either because:

(a) The barrier follows a non-relaxed MMIO access that already performs
a read or write memory barrier.

(b) The barrier follows DMA API calls for which the device-visible
effects of IOMMU programming are guaranteed to be flushed to the IOMMU
prior to the call returning, and doesn't need to sync with normal memory
access.

Signed-off-by: Todd Poynor <toddpoynor@google.com>
---
 drivers/staging/gasket/gasket_page_table.c | 15 +++++----------
 1 file changed, 5 insertions(+), 10 deletions(-)

diff --git a/drivers/staging/gasket/gasket_page_table.c b/drivers/staging/gasket/gasket_page_table.c
index 4d2499269499b..53492f4fad6aa 100644
--- a/drivers/staging/gasket/gasket_page_table.c
+++ b/drivers/staging/gasket/gasket_page_table.c
@@ -317,8 +317,6 @@ static void gasket_free_extended_subtable(struct gasket_page_table *pg_tbl,
 
 	/* Release the page table from the device */
 	writeq(0, slot);
-	/* Force sync around the address release. */
-	mb();
 
 	if (pte->dma_addr)
 		dma_unmap_page(pg_tbl->device, pte->dma_addr, PAGE_SIZE,
@@ -504,8 +502,6 @@ static int gasket_perform_mapping(struct gasket_page_table *pg_tbl,
 					(void *)page_to_phys(page));
 				return -1;
 			}
-			/* Wait until the page is mapped. */
-			mb();
 		}
 
 		/* Make the DMA-space address available to the device. */
@@ -604,12 +600,13 @@ static void gasket_perform_unmapping(struct gasket_page_table *pg_tbl,
 	 */
 	for (i = 0; i < num_pages; i++) {
 		/* release the address from the device, */
-		if (is_simple_mapping || ptes[i].status == PTE_INUSE)
+		if (is_simple_mapping || ptes[i].status == PTE_INUSE) {
 			writeq(0, &slots[i]);
-		else
+		} else {
 			((u64 __force *)slots)[i] = 0;
-		/* Force sync around the address release. */
-		mb();
+			/* sync above PTE update before updating mappings */
+			wmb();
+		}
 
 		/* release the address from the driver, */
 		if (ptes[i].status == PTE_INUSE) {
@@ -898,8 +895,6 @@ static int gasket_alloc_extended_subtable(struct gasket_page_table *pg_tbl,
 	/* Map the page into DMA space. */
 	pte->dma_addr = dma_map_page(pg_tbl->device, pte->page, 0, PAGE_SIZE,
 				     DMA_BIDIRECTIONAL);
-	/* Wait until the page is mapped. */
-	mb();
 
 	/* make the addresses available to the device */
 	dma_addr = (pte->dma_addr + pte->offset) | GASKET_VALID_SLOT_FLAG;
-- 
2.18.0.597.ga71716f1ad-goog