Skip to contentSkip to conversation

Revert "memory: ti-aemif: don't needlessly iterate over child nodes"

PATCH Applied Bartosz Golaszewski 3 participants 4 messages Aug 24, 2024 – Aug 25, 2024

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Revert "memory: ti-aemif: don't needlessly iterate over child nodes"

From: Bartosz Golaszewski <[email protected]>

This reverts commit 23a641d5c2bce4c723fff9118a5d865ee6b9d05a.

The first-level children of the aemif node are not the device nodes (ones containing the 'compatible' property) but the chip-select nodes which instead have their own children.

of_platform_populate() will skip such nodes so we must indeed iterate over the direct children of the aemif node. The problem here is that we never call of_platform_depopulate() as it takes the root device as argument. We only have an unpopulated chip-select nodes so we will leak these devices if any of the calls to of_platform_populate() fails.

I don't have a batter idea right now but my patch was not correct so we need to revert it. While at it: at least use the scoped variant of the OF node iterator. Down the line, we should find a better solution to fix this potential resource leak in error path.

---
 drivers/memory/ti-aemif.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)
drivers/memory/ti-aemif.c +5 -3
index 3b546eddf5fe9..d54dc3cfff73c 100644
@@ -379,9 +379,11 @@ static int aemif_probe(struct platform_device *pdev)
 	 * child will be probed after the AEMIF timing parameters are set.
 	 */
 	if (np) {
-		ret = devm_of_platform_populate(dev);
-		if (ret)
-			return ret;
+		for_each_available_child_of_node_scoped(np, child_np) {
+			ret = of_platform_populate(child_np, NULL, NULL, dev);
+			if (ret < 0)
+				return ret;
+		}
 	}
 
 	return 0;
Signature · 1 line
2.43.0
#1 lore raw
Quoted 19 lines · Bartosz Golaszewski

From: Bartosz Golaszewski <[email protected]>

This reverts commit 23a641d5c2bce4c723fff9118a5d865ee6b9d05a.

The first-level children of the aemif node are not the device nodes (ones containing the 'compatible' property) but the chip-select nodes which instead have their own children.

of_platform_populate() will skip such nodes so we must indeed iterate over the direct children of the aemif node. The problem here is that we never call of_platform_depopulate() as it takes the root device as argument. We only have an unpopulated chip-select nodes so we will leak these devices if any of the calls to of_platform_populate() fails.

I don't have a batter idea right now but my patch was not correct so we need to revert it. While at it: at least use the scoped variant of the OF node iterator. Down the line, we should find a better solution to fix this potential resource leak in error path.

Quoted 4 lines
---
 drivers/memory/ti-aemif.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

I'll drop the original commit, because my upstream (arm/soc) might question this.

Best regards,
Krzysztof

#2 lore raw
Quoted 20 lines · Krzysztof Kozlowski
Quoted 19 lines · Bartosz Golaszewski

From: Bartosz Golaszewski <[email protected]>

This reverts commit 23a641d5c2bce4c723fff9118a5d865ee6b9d05a.

The first-level children of the aemif node are not the device nodes (ones containing the 'compatible' property) but the chip-select nodes which instead have their own children.

of_platform_populate() will skip such nodes so we must indeed iterate over the direct children of the aemif node. The problem here is that we never call of_platform_depopulate() as it takes the root device as argument. We only have an unpopulated chip-select nodes so we will leak these devices if any of the calls to of_platform_populate() fails.

I don't have a batter idea right now but my patch was not correct so we need to revert it. While at it: at least use the scoped variant of the OF node iterator. Down the line, we should find a better solution to fix this potential resource leak in error path.

Quoted 8 lines
Quoted 4 lines
---
 drivers/memory/ti-aemif.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

I'll drop the original commit, because my upstream (arm/soc) might question this.

Dropping is non-trivial, so indeed revert.

Best regards,
Krzysztof

#3 lore raw
In reply to Bartosz Golaszewski
Quoted 6 lines · Bartosz Golaszewski

This reverts commit 23a641d5c2bce4c723fff9118a5d865ee6b9d05a.

The first-level children of the aemif node are not the device nodes (ones containing the 'compatible' property) but the chip-select nodes which instead have their own children.

Quoted 7 lines

of_platform_populate() will skip such nodes so we must indeed iterate over the direct children of the aemif node. The problem here is that we never call of_platform_depopulate() as it takes the root device as argument. We only have an unpopulated chip-select nodes so we will leak these devices if any of the calls to of_platform_populate() fails.

[...]

Applied, thanks!

[1/1] Revert "memory: ti-aemif: don't needlessly iterate over child nodes"

      https://git.kernel.org/krzk/linux-mem-ctrl/c/c7d2f3fbdf59b206414ddc306b0fb74cd174c0ed

Best regards,

Signature · 1 line
Krzysztof Kozlowski <[email protected]>