From: Logan Gunthorpe <logang@deltatee.com>
To: dmaengine@vger.kernel.org, linux-kernel@vger.kernel.org,
Vinod Koul <vkoul@kernel.org>
Cc: "Frank Li" <Frank.li@nxp.com>,
"Christoph Hellwig" <hch@infradead.org>,
"Christophe Jaillet" <christophe.jaillet@wanadoo.fr>,
"Dave Jiang" <dave.jiang@intel.com>,
"Thomas Weißschuh" <linux@weissschuh.net>,
"Kelvin Cao" <kelvin.cao@microchip.com>,
"Logan Gunthorpe" <logang@deltatee.com>
Subject: [PATCH v3 1/4] dmaengine: add per-channel sysfs attribute groups via chan_groups
Date: Tue, 3 Nov 2026 09:40:18 -0600 [thread overview]
Message-ID: <20261103154022.459620-2-logang@deltatee.com> (raw)
In-Reply-To: <20261103154022.459620-1-logang@deltatee.com>
Each channel already gets its own struct device (dma_chan_dev),
registered with device_register()/device_unregister() and torn down
correctly by chan_dev_release(). Let drivers add their own sysfs
attributes to that device directly, via a new dma_device.chan_groups
field.
This avoids drivers needing a second, separately-managed kobject to
expose driver-specific attributes. That's a use-after-free issue:
kobject_put() can invoke the ktype's release() and free the structure
embedding dma_chan while the core (or the driver) still expects to
reference it afterward, so avoiding it requires careful ordering and
bookkeeping that's easy to get wrong. The channel's struct device
already has the correct lifetime, and sysfs_create_group() supports
the same named-subdirectory layout (attribute_group.name) that a bare
kobject would provide.
To let show()/store() callbacks safely recover the struct dma_chan from
the struct device they're attached to without exposing dma_list_mutex
itself to drivers, add a small dmaengine_chan_from_dev_lock() /
dmaengine_chan_from_dev_unlock() pair that take / release the lock, and
wrap them in a dmaengine_chan_from_dev CLASS so callers can write:
CLASS(dmaengine_chan_from_dev, c)(dev);
if (!c)
return -ENODEV;
Signed-off-by: Logan Gunthorpe <logang@deltatee.com>
---
drivers/dma/dmaengine.c | 27 +++++++++++++++++++++++++++
drivers/dma/dmaengine.h | 19 +++++++++++++++++++
include/linux/dmaengine.h | 4 ++++
3 files changed, 50 insertions(+)
diff --git a/drivers/dma/dmaengine.c b/drivers/dma/dmaengine.c
index 34be53487a21..38f601bf41e1 100644
--- a/drivers/dma/dmaengine.c
+++ b/drivers/dma/dmaengine.c
@@ -161,6 +161,32 @@ static struct dma_chan *dev_to_dma_chan(struct device *dev)
return chan_dev->chan;
}
+/**
+ * dmaengine_chan_from_dev_lock - take dma_list_mutex and convert a channel's
+ * struct device to its dma_chan
+ * @dev: the channel's struct device, embedded in struct dma_chan_dev
+ *
+ * Returns NULL if the channel has already been unregistered. Pairs with
+ * dmaengine_chan_from_dev_unlock(); see the dmaengine_chan_from_dev CLASS in
+ * drivers/dma/dmaengine.h.
+ */
+struct dma_chan *dmaengine_chan_from_dev_lock(struct device *dev)
+{
+ mutex_lock(&dma_list_mutex);
+ return dev_to_dma_chan(dev);
+}
+EXPORT_SYMBOL_GPL(dmaengine_chan_from_dev_lock);
+
+/**
+ * dmaengine_chan_from_dev_unlock - release the channel lookup lock
+ * @chan: unused; matches the value produced by dmaengine_chan_from_dev_lock()
+ */
+void dmaengine_chan_from_dev_unlock(struct dma_chan *chan)
+{
+ mutex_unlock(&dma_list_mutex);
+}
+EXPORT_SYMBOL_GPL(dmaengine_chan_from_dev_unlock);
+
static ssize_t memcpy_count_show(struct device *dev,
struct device_attribute *attr, char *buf)
{
@@ -1117,6 +1143,7 @@ static int __dma_async_device_channel_register(struct dma_device *device,
dmaengine_chan_dev(chan)->class = &dma_devclass;
dmaengine_chan_dev(chan)->parent = device->dev;
+ dmaengine_chan_dev(chan)->groups = device->chan_groups;
chan->chan_dev->chan = chan;
chan->chan_dev->dev_id = device->dev_id;
spin_lock_init(&chan->lock);
diff --git a/drivers/dma/dmaengine.h b/drivers/dma/dmaengine.h
index 42a3cfa7e597..f03248a091df 100644
--- a/drivers/dma/dmaengine.h
+++ b/drivers/dma/dmaengine.h
@@ -214,6 +214,25 @@ int dma_async_device_channel_register(struct dma_device *device,
void dma_async_device_channel_unregister(struct dma_device *device,
struct dma_chan *chan);
+/*
+ * dmaengine_chan_from_dev_lock() / dmaengine_chan_from_dev_unlock() bracket
+ * a critical section across which a channel's struct device can be safely
+ * converted back to its struct dma_chan: dmaengine_chan_from_dev_lock()
+ * returns NULL if the channel has already been unregistered, and the lock
+ * it takes must be held for as long as the returned channel (or anything
+ * derived from it) is accessed.
+ *
+ * Use these through the dmaengine_chan_from_dev CLASS below rather than calling
+ * them directly.
+ */
+struct dma_chan *dmaengine_chan_from_dev_lock(struct device *dev);
+void dmaengine_chan_from_dev_unlock(struct dma_chan *chan);
+
+DEFINE_CLASS(dmaengine_chan_from_dev, struct dma_chan *,
+ dmaengine_chan_from_dev_unlock(_T),
+ dmaengine_chan_from_dev_lock(dev),
+ struct device *dev)
+
#ifdef CONFIG_DEBUG_FS
#include <linux/debugfs.h>
diff --git a/include/linux/dmaengine.h b/include/linux/dmaengine.h
index 805668fbe437..944c8b0f5848 100644
--- a/include/linux/dmaengine.h
+++ b/include/linux/dmaengine.h
@@ -839,6 +839,9 @@ struct dma_filter {
* @dev: struct device reference for dma mapping api
* @owner: owner module (automatically set based on the provided dev)
* @chan_ida: unique channel ID
+ * @chan_groups: optional NULL-terminated array of sysfs attribute groups
+ * added to each channel's struct device, for driver-specific per-channel
+ * attributes
* @src_addr_widths: bit mask of src addr widths the device supports
* Width is specified in bytes, e.g. for a device supporting
* a width of 4 the mask should have BIT(4) set.
@@ -923,6 +926,7 @@ struct dma_device {
struct device *dev;
struct module *owner;
struct ida chan_ida;
+ const struct attribute_group **chan_groups;
u32 src_addr_widths;
u32 dst_addr_widths;
--
2.47.3
next prev parent reply other threads:[~2026-11-03 15:40 UTC|newest]
Thread overview: 6+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-11-03 15:40 [PATCH v3 0/4] Add sysfs interface to switchtec-dma Logan Gunthorpe
2026-11-03 15:40 ` Logan Gunthorpe [this message]
2026-11-03 15:40 ` [PATCH v3 2/4] dmaengine: ioatdma: convert per-channel sysfs to chan_groups Logan Gunthorpe
2026-11-03 16:00 ` Dave Jiang
2026-11-03 15:40 ` [PATCH v3 3/4] dmaengine: switchtec-dma: add config sysfs attributes Logan Gunthorpe
2026-11-03 15:40 ` [PATCH v3 4/4] dmaengine: switchtec-dma: add pmon " Logan Gunthorpe
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20261103154022.459620-2-logang@deltatee.com \
--to=logang@deltatee.com \
--cc=Frank.li@nxp.com \
--cc=christophe.jaillet@wanadoo.fr \
--cc=dave.jiang@intel.com \
--cc=dmaengine@vger.kernel.org \
--cc=hch@infradead.org \
--cc=kelvin.cao@microchip.com \
--cc=linux-kernel@vger.kernel.org \
--cc=linux@weissschuh.net \
--cc=vkoul@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®