dmaengine.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
* [PATCH v4] dmaengine: idxd: fix wq cleanup of WQCFG registers
@ 2021-04-12 16:02 Dave Jiang
  2021-04-12 16:38 ` Vinod Koul
  0 siblings, 1 reply; 2+ messages in thread
From: Dave Jiang @ 2021-04-12 16:02 UTC (permalink / raw)
  To: vkoul; +Cc: Shreenivaas Devarajan, dmaengine

A pre-release silicon erratum workaround where wq reset does not clear
WQCFG registers was leaked into upstream code. Use wq reset command
instead of blasting the MMIO region. This also address an issue where
we clobber registers in future devices.

Fixes: da32b28c95a7 ("dmaengine: idxd: cleanup workqueue config after disabling")
Reported-by: Shreenivaas Devarajan <shreenivaas.devarajan@intel.com>
Signed-off-by: Dave Jiang <dave.jiang@intel.com>
---
v4:
- Remove noisy debug output (Vinod)
v3:
- Remove unused vars
v2:
- Set IDXD_WQ_DISABLED for internal state after reset.

 drivers/dma/idxd/device.c |   35 ++++++++++++++++++++++++-----------
 drivers/dma/idxd/idxd.h   |    1 +
 drivers/dma/idxd/sysfs.c  |    9 ++-------
 3 files changed, 27 insertions(+), 18 deletions(-)

diff --git a/drivers/dma/idxd/device.c b/drivers/dma/idxd/device.c
index 84a6ea60ecf0..0375d9459e74 100644
--- a/drivers/dma/idxd/device.c
+++ b/drivers/dma/idxd/device.c
@@ -282,6 +282,22 @@ void idxd_wq_drain(struct idxd_wq *wq)
 	idxd_cmd_exec(idxd, IDXD_CMD_DRAIN_WQ, operand, NULL);
 }
 
+void idxd_wq_reset(struct idxd_wq *wq)
+{
+	struct idxd_device *idxd = wq->idxd;
+	struct device *dev = &idxd->pdev->dev;
+	u32 operand;
+
+	if (wq->state != IDXD_WQ_ENABLED) {
+		dev_dbg(dev, "WQ %d in wrong state: %d\n", wq->id, wq->state);
+		return;
+	}
+
+	operand = BIT(wq->id % 16) | ((wq->id / 16) << 16);
+	idxd_cmd_exec(idxd, IDXD_CMD_RESET_WQ, operand, NULL);
+	wq->state = IDXD_WQ_DISABLED;
+}
+
 int idxd_wq_map_portal(struct idxd_wq *wq)
 {
 	struct idxd_device *idxd = wq->idxd;
@@ -363,8 +379,6 @@ int idxd_wq_disable_pasid(struct idxd_wq *wq)
 void idxd_wq_disable_cleanup(struct idxd_wq *wq)
 {
 	struct idxd_device *idxd = wq->idxd;
-	struct device *dev = &idxd->pdev->dev;
-	int i, wq_offset;
 
 	lockdep_assert_held(&idxd->dev_lock);
 	memset(wq->wqcfg, 0, idxd->wqcfg_size);
@@ -376,14 +390,6 @@ void idxd_wq_disable_cleanup(struct idxd_wq *wq)
 	wq->ats_dis = 0;
 	clear_bit(WQ_FLAG_DEDICATED, &wq->flags);
 	memset(wq->name, 0, WQ_NAME_SIZE);
-
-	for (i = 0; i < WQCFG_STRIDES(idxd); i++) {
-		wq_offset = WQCFG_OFFSET(idxd, wq->id, i);
-		iowrite32(0, idxd->reg_base + wq_offset);
-		dev_dbg(dev, "WQ[%d][%d][%#x]: %#x\n",
-			wq->id, i, wq_offset,
-			ioread32(idxd->reg_base + wq_offset));
-	}
 }
 
 /* Device control bits */
@@ -642,7 +648,14 @@ static int idxd_wq_config_write(struct idxd_wq *wq)
 	if (!wq->group)
 		return 0;
 
-	memset(wq->wqcfg, 0, idxd->wqcfg_size);
+	/*
+	 * Instead of memset the entire shadow copy of WQCFG, copy from the hardware after
+	 * wq reset. This will copy back the sticky values that are present on some devices.
+	 */
+	for (i = 0; i < WQCFG_STRIDES(idxd); i++) {
+		wq_offset = WQCFG_OFFSET(idxd, wq->id, i);
+		wq->wqcfg->bits[i] = ioread32(idxd->reg_base + wq_offset);
+	}
 
 	/* byte 0-3 */
 	wq->wqcfg->wq_size = wq->size;
diff --git a/drivers/dma/idxd/idxd.h b/drivers/dma/idxd/idxd.h
index 81a0e65fd316..730745d331e3 100644
--- a/drivers/dma/idxd/idxd.h
+++ b/drivers/dma/idxd/idxd.h
@@ -341,6 +341,7 @@ void idxd_wq_free_resources(struct idxd_wq *wq);
 int idxd_wq_enable(struct idxd_wq *wq);
 int idxd_wq_disable(struct idxd_wq *wq);
 void idxd_wq_drain(struct idxd_wq *wq);
+void idxd_wq_reset(struct idxd_wq *wq);
 int idxd_wq_map_portal(struct idxd_wq *wq);
 void idxd_wq_unmap_portal(struct idxd_wq *wq);
 void idxd_wq_disable_cleanup(struct idxd_wq *wq);
diff --git a/drivers/dma/idxd/sysfs.c b/drivers/dma/idxd/sysfs.c
index 5f7bc4b1621a..18bf4d148989 100644
--- a/drivers/dma/idxd/sysfs.c
+++ b/drivers/dma/idxd/sysfs.c
@@ -275,7 +275,6 @@ static void disable_wq(struct idxd_wq *wq)
 {
 	struct idxd_device *idxd = wq->idxd;
 	struct device *dev = &idxd->pdev->dev;
-	int rc;
 
 	mutex_lock(&wq->wq_lock);
 	dev_dbg(dev, "%s removing WQ %s\n", __func__, dev_name(&wq->conf_dev));
@@ -296,17 +295,13 @@ static void disable_wq(struct idxd_wq *wq)
 	idxd_wq_unmap_portal(wq);
 
 	idxd_wq_drain(wq);
-	rc = idxd_wq_disable(wq);
+	idxd_wq_reset(wq);
 
 	idxd_wq_free_resources(wq);
 	wq->client_count = 0;
 	mutex_unlock(&wq->wq_lock);
 
-	if (rc < 0)
-		dev_warn(dev, "Failed to disable %s: %d\n",
-			 dev_name(&wq->conf_dev), rc);
-	else
-		dev_info(dev, "wq %s disabled\n", dev_name(&wq->conf_dev));
+	dev_info(dev, "wq %s disabled\n", dev_name(&wq->conf_dev));
 }
 
 static int idxd_config_bus_remove(struct device *dev)



^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH v4] dmaengine: idxd: fix wq cleanup of WQCFG registers
  2021-04-12 16:02 [PATCH v4] dmaengine: idxd: fix wq cleanup of WQCFG registers Dave Jiang
@ 2021-04-12 16:38 ` Vinod Koul
  0 siblings, 0 replies; 2+ messages in thread
From: Vinod Koul @ 2021-04-12 16:38 UTC (permalink / raw)
  To: Dave Jiang; +Cc: Shreenivaas Devarajan, dmaengine

On 12-04-21, 09:02, Dave Jiang wrote:
> A pre-release silicon erratum workaround where wq reset does not clear
> WQCFG registers was leaked into upstream code. Use wq reset command
> instead of blasting the MMIO region. This also address an issue where
> we clobber registers in future devices.

Applied, thanks

-- 
~Vinod

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2021-04-12 16:48 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-04-12 16:02 [PATCH v4] dmaengine: idxd: fix wq cleanup of WQCFG registers Dave Jiang
2021-04-12 16:38 ` Vinod Koul

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).