mail archive of the barebox mailing list
 help / color / mirror / Atom feed
* [PATCH master] dma: correctly honour dma-noncoherent device tree property
@ 2024-03-22 14:24 Ahmad Fatoum
  2024-03-25  9:59 ` Sascha Hauer
  2024-03-26 10:33 ` Sascha Hauer
  0 siblings, 2 replies; 4+ messages in thread
From: Ahmad Fatoum @ 2024-03-22 14:24 UTC (permalink / raw)
  To: barebox; +Cc: Ahmad Fatoum

Most barebox platforms are either completely cache-coherent with respect to
DMA or completely non-coherent. To avoid having to walk the device tree for
non-existent dma-coherent and dma-noncohrent properties, barebox thus
only does this when CONFIG_OF_DMA_COHERENCY is selected.

CONFIG_OF_DMA_COHERENCY is to be selected by platforms that have mixed
coherency for DMA masters. Forgetting to select this option can be
annoying to debug, which is why devinfo will inform the user of default
DMA coherency assignments:

  DMA Coherent: false (default)

In order to allow devinfo to differentiate implicit default and explicit
device tree DMA coherency configuration, the struct device::dma_coherent
member is not of boolean type, but Instead an enumeration that besides
coherent and non-coherent also encodes a DEV_DMA_COHERENCE_DEFAULT state.

In practice though, a boolean was saved in this dma_coherent member,
leading to any dma-noncoherent property being ignored and the default
coherency setting being taken for the node.

By fixing the type confusion, we restore working Ethernet on the
StarFive JH7100, which is the only SoC we currently support that
requires /soc/dma-noncohrent to be set for proper operation of its DMA
masters.

Fixes: fbdea8fd54fe ("of: populate new device_d::dma_coherent attribute")
Reported-by: Antony Pavlov <antonynpavlov@gmail.com>
Signed-off-by: Ahmad Fatoum <a.fatoum@pengutronix.de>
---
 drivers/of/platform.c | 34 +++++++++++++++++++++++-----------
 1 file changed, 23 insertions(+), 11 deletions(-)

diff --git a/drivers/of/platform.c b/drivers/of/platform.c
index ec1482b27797..e2c252b1ffee 100644
--- a/drivers/of/platform.c
+++ b/drivers/of/platform.c
@@ -90,6 +90,21 @@ static struct device_node *of_get_next_dma_parent(const struct device_node *np)
 	return args.np;
 }
 
+static enum dev_dma_coherence of_dma_get_coherence(struct device_node *node)
+{
+	if (IS_ENABLED(CONFIG_OF_DMA_COHERENCY)) {
+		while (node) {
+			if (of_property_read_bool(node, "dma-coherent"))
+				return DEV_DMA_COHERENT;
+			if (of_property_read_bool(node, "dma-noncoherent"))
+				return DEV_DMA_NON_COHERENT;
+			node = of_get_next_dma_parent(node);
+		}
+	}
+
+	return DEV_DMA_COHERENCE_DEFAULT;
+}
+
 /**
  * of_dma_is_coherent - Check if device is coherent
  * @np:	device node
@@ -101,17 +116,14 @@ static struct device_node *of_get_next_dma_parent(const struct device_node *np)
  */
 bool of_dma_is_coherent(struct device_node *node)
 {
-	if (IS_ENABLED(CONFIG_OF_DMA_COHERENCY)) {
-		while (node) {
-			if (of_property_read_bool(node, "dma-coherent"))
-				return true;
-			if (of_property_read_bool(node, "dma-noncoherent"))
-				return false;
-			node = of_get_next_dma_parent(node);
-		}
+	switch (of_dma_get_coherence(node)) {
+	case DEV_DMA_COHERENT:
+		return true;
+	case DEV_DMA_NON_COHERENT:
+		return false;
+	case DEV_DMA_COHERENCE_DEFAULT:
+		return IS_ENABLED(CONFIG_ARCH_DMA_DEFAULT_COHERENT);
 	}
-
-	return IS_ENABLED(CONFIG_ARCH_DMA_DEFAULT_COHERENT);
 }
 EXPORT_SYMBOL_GPL(of_dma_is_coherent);
 
@@ -129,7 +141,7 @@ static void of_dma_configure(struct device *dev, struct device_node *np)
 	}
 
 	dev->dma_offset = offset;
-	dev->dma_coherent = of_dma_is_coherent(np);
+	dev->dma_coherent = of_dma_get_coherence(np);
 }
 
 /**
-- 
2.39.2




^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2024-03-26 17:33 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-03-22 14:24 [PATCH master] dma: correctly honour dma-noncoherent device tree property Ahmad Fatoum
2024-03-25  9:59 ` Sascha Hauer
2024-03-26 10:33 ` Sascha Hauer
2024-03-26 17:32   ` Ahmad Fatoum

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox