linux-kernel - [PATCH] driver core: platform: Allow using a dedicated dma_mask for platform

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Date:   Fri, 28 Jun 2019 16:15:50 +0200
From:   Maxime Chevallier <maxime.chevallier@...tlin.com>
To:     Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        "Rafael J . Wysocki" <rafael@...nel.org>,
        Rob Herring <robh+dt@...nel.org>,
        Frank Rowand <frowand.list@...il.com>,
        Russell King <linux@...linux.org.uk>,
        Christoph Hellwig <hch@...radead.org>
Cc:     Maxime Chevallier <maxime.chevallier@...tlin.com>,
        David Miller <davem@...emloft.net>, brian.brooks@...aro.org,
        linux-kernel@...r.kernel.org,
        Antoine Tenart <antoine.tenart@...tlin.com>,
        thomas.petazzoni@...tlin.com, linux-arm-kernel@...ts.infradead.org,
        nadavh@...vell.com, stefanc@...vell.com,
        Russell King <rmk+kernel@...linux.org.uk>
Subject: [PATCH] driver core: platform: Allow using a dedicated dma_mask for platform_device

This patch attempts to solve a long standing situation where
dev->dma_mask is a pointer to dev->dma_coherent_mask, meaning that any
change to the coherent mask will affect the streaming mask.

The API allows to use different values for both masks, but for now
platform_device built from DT will simply make the dma_mask point to the
coherent mask.

This is a problem on a least one driver, the PPv2 network driver, which
needs different streaming and coherent masks to overcome a HW
limitation. In this case, the issue is a performance degradation since
the streaming mask isn't as big as it ought to be, causing a lot of
buffer bounces.

There were previous attempts to fix this issue. One of them by Brian
Brooks, where the dma_mask is reallocated in the driver itself,
which wasn't considered to be the best approach :

https://lore.kernel.org/netdev/20180820024730.9147-1-brian.brooks@linaro.org/

This lead to a discussion pointing to another attempt to solve the issue,
by Christoph Hellwig :

https://lore.kernel.org/lkml/20180829062401.8701-2-hch@lst.de/

This more generic approach ended-up causing regressions on some mfd
drivers (the sm501 was one of the reports).

The current patch tries to be a bit less generic, and allows setting-up
the dma_mask for platform devices using a dedicated helper. In this case,
the dma_mask is allocated in struct platform_object, as suggested by
Russell King.

This helper is then used in platform_device creation code from the DT.

Suggested-by: Russell King <rmk+kernel@...linux.org.uk>
Signed-off-by: Maxime Chevallier <maxime.chevallier@...tlin.com>
---
Hi everyone,

This patch, if suitable, would require a lot of testing to detect
drivers that rely on the streaming mask being the same as the coherent
mask.

Thanks,

Maxime

 drivers/base/platform.c         | 17 +++++++++++++++++
 drivers/of/platform.c           |  7 +++++--
 include/linux/platform_device.h |  1 +
 3 files changed, 23 insertions(+), 2 deletions(-)

diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index 4d1729853d1a..35e7bdb8576c 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -256,9 +256,26 @@ EXPORT_SYMBOL_GPL(platform_add_devices);
 
 struct platform_object {
 	struct platform_device pdev;
+	u64 dma_mask;
 	char name[];
 };
 
+/**
+ * platform_device_setup_dma_mask - Sets the dma_mask pointer
+ * @pdev: platform device to configure the device's mask
+ *
+ * Sets the dma_mask of the underlying device to point to a dedicated region,
+ * that belongs to the platform_device.
+ */
+void platform_device_setup_dma_mask(struct platform_device *pdev)
+{
+	struct platform_object *pa = container_of(pdev, struct platform_object,
+						  pdev);
+
+	pa->pdev.dev.dma_mask = &pa->dma_mask;
+}
+EXPORT_SYMBOL_GPL(platform_device_setup_dma_mask);
+
 /**
  * platform_device_put - destroy a platform device
  * @pdev: platform device to free
diff --git a/drivers/of/platform.c b/drivers/of/platform.c
index 04ad312fd85b..4a6980e3356c 100644
--- a/drivers/of/platform.c
+++ b/drivers/of/platform.c
@@ -186,8 +186,11 @@ static struct platform_device *of_platform_device_create_pdata(
 		goto err_clear_flag;
 
 	dev->dev.coherent_dma_mask = DMA_BIT_MASK(32);
-	if (!dev->dev.dma_mask)
-		dev->dev.dma_mask = &dev->dev.coherent_dma_mask;
+	if (!dev->dev.dma_mask) {
+		platform_device_setup_dma_mask(dev);
+		*dev->dev.dma_mask = dev->dev.coherent_dma_mask;
+	}
+
 	dev->dev.bus = &platform_bus_type;
 	dev->dev.platform_data = platform_data;
 	of_msi_configure(&dev->dev, dev->dev.of_node);
diff --git a/include/linux/platform_device.h b/include/linux/platform_device.h
index cc464850b71e..a95c2d224de9 100644
--- a/include/linux/platform_device.h
+++ b/include/linux/platform_device.h
@@ -181,6 +181,7 @@ extern int platform_device_add_properties(struct platform_device *pdev,
 extern int platform_device_add(struct platform_device *pdev);
 extern void platform_device_del(struct platform_device *pdev);
 extern void platform_device_put(struct platform_device *pdev);
+extern void platform_device_setup_dma_mask(struct platform_device *pdev);
 
 struct platform_driver {
 	int (*probe)(struct platform_device *);
-- 
2.20.1