Skip to contentSkip to conversation

driver core: generalize driver_override infrastructure

PATCH Superseded Danilo Krummrich 6 participants 21 messages Mar 2

Subsystems (13)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Danilo Krummrich

author maintainer
#0 lore raw

cover driver core: generalize driver_override infrastructure

Currently, there are 12 busses (including platform and PCI) that duplicate the driver_override logic for their individual devices.

All of them seem to be prone to the bug described in [1].

While this could be solved for every bus individually using a separate lock, solving this in the driver-core generically results in less (and cleaner) changes overall.

Thus, move driver_override to struct device, provide corresponding accessors for busses and handle locking with a separate lock internally.

In particular, add device_set_driver_override(), device_has_driver_override(), device_match_driver_override() and a helper, DEVICE_ATTR_DRIVER_OVERRIDE(), to declare the corresponding sysfs store() and show() callbacks.

Until all busses have migrated, keep driver_set_override() in place.

Note that we can't use the device lock for the reasons described in [2].

This patch series includes the migration of the platform bus; patches for all other affected busses still need to be extracted as a follow-up of the WIP treewide patch in [3].

[1] https://bugzilla.kernel.org/show_bug.cgi?id=220789
[2] https://lore.kernel.org/driver-core/[email protected]/
[3] https://git.kernel.org/pub/scm/linux/kernel/git/dakr/linux.git/log/?h=driver_override

Danilo Krummrich (3):
driver core: generalize driver_override in struct device
hwmon: axi-fan: don't use driver_override as IRQ name driver core: platform: use generic driver_override infrastructure

 arch/sh/drivers/platform_early.c |  6 ++-
 drivers/base/core.c              |  2 +
 drivers/base/dd.c                | 60 +++++++++++++++++++++++
 drivers/base/platform.c          | 35 ++------------
 drivers/bus/simple-pm-bus.c      |  4 +-
 drivers/clk/imx/clk-scu.c        |  3 +-
 drivers/hwmon/axi-fan-control.c  |  2 +-
 drivers/slimbus/qcom-ngd-ctrl.c  |  6 +--
 include/linux/device.h           | 81 ++++++++++++++++++++++++++++++++
 include/linux/platform_device.h  |  5 --
 sound/soc/samsung/i2s.c          |  6 +--
 11 files changed, 161 insertions(+), 49 deletions(-)

base-commit: 78437ab3b769f80526416570f60173c89858dd84

Signature · 1 line
2.53.0
author maintainer
#1 lore raw

PATCH 1/3 driver core: generalize driver_override in struct device

Currently, there are 12 busses (including platform and PCI) that duplicate the driver_override logic for their individual devices.

All of them seem to be prone to the bug described in [1].

While this could be solved for every bus individually using a separate lock, solving this in the driver-core generically results in less (and cleaner) changes overall.

Thus, move driver_override to struct device, provide corresponding accessors for busses and handle locking with a separate lock internally.

In particular, add device_set_driver_override(), device_has_driver_override(), device_match_driver_override() and a helper, DEVICE_ATTR_DRIVER_OVERRIDE(), to declare the corresponding sysfs store() and show() callbacks.

Until all busses have migrated, keep driver_set_override() in place.

Note that we can't use the device lock for the reasons described in [2].

---
 drivers/base/core.c    |  2 ++
 drivers/base/dd.c      | 60 +++++++++++++++++++++++++++++++
 include/linux/device.h | 81 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 143 insertions(+)
drivers/base/core.c +2 -0
index 791f9e444df8..a8cb90577d10 100644
@@ -2566,6 +2566,7 @@ static void device_release(struct kobject *kobj)
 	else
 		WARN(1, KERN_ERR "Device '%s' does not have a release() function, it is broken and must be fixed. See Documentation/core-api/kobject.rst.\n",
 			dev_name(dev));
+	kfree(dev->driver_override.name);
 	kfree(p);
 }
 
@@ -3159,6 +3160,7 @@ void device_initialize(struct device *dev)
 	kobject_init(&dev->kobj, &device_ktype);
 	INIT_LIST_HEAD(&dev->dma_pools);
 	mutex_init(&dev->mutex);
+	spin_lock_init(&dev->driver_override.lock);
 	lockdep_set_novalidate_class(&dev->mutex);
 	spin_lock_init(&dev->devres_lock);
 	INIT_LIST_HEAD(&dev->devres_head);
drivers/base/dd.c +60 -0

66 lines not on this page · raw message

include/linux/device.h +81 -0

105 lines not on this page · raw message

Signature · 1 line
2.53.0
#2 lore raw

Re: [PATCH 1/3] driver core: generalize driver_override in struct device

Quoted 18 lines · Danilo Krummrich

Currently, there are 12 busses (including platform and PCI) that duplicate the driver_override logic for their individual devices.

All of them seem to be prone to the bug described in [1].

While this could be solved for every bus individually using a separate lock, solving this in the driver-core generically results in less (and cleaner) changes overall.

Thus, move driver_override to struct device, provide corresponding accessors for busses and handle locking with a separate lock internally.

In particular, add device_set_driver_override(), device_has_driver_override(), device_match_driver_override() and a helper, DEVICE_ATTR_DRIVER_OVERRIDE(), to declare the corresponding sysfs store() and show() callbacks.

Quoted 7 lines

Until all busses have migrated, keep driver_set_override() in place.

Note that we can't use the device lock for the reasons described in [2].

Hi Danilo,

I wanted to test if this fixes the issue using PoCs, but I'm hitting a KASAN splat right during boot. The issue disappears if I revert this patch.

KASAN report:

[    7.266874] ==================================================================
[    7.267707] BUG: KASAN: slab-use-after-free in device_release+0x1f4/0x240
[    7.267707] Read of size 8 at addr ffff888003f4a370 by task kworker/1:0/24
[    7.267707]
[    7.267707] CPU: 1 UID: 0 PID: 24 Comm: kworker/1:0 Not tainted
7.0.0-rc2-00001-gc1a10dc76109 #4 PREEMP
[    7.267707] Hardware name: QEMU Ubuntu 24.04 PC v2 (i440FX + PIIX,
arch_caps fix, 1996), BIOS 1.16.3-de4
[    7.267707] Workqueue: events_long serio_handle_event
[    7.267707] Call Trace:
[    7.267707]  <TASK>
[    7.267707]  dump_stack_lvl+0x66/0xa0
[    7.267707]  print_report+0xce/0x660
[    7.267707]  ? device_release+0x1f4/0x240
[    7.267707]  ? __virt_addr_valid+0x208/0x410
[    7.267707]  ? device_release+0x1f4/0x240
[    7.267707]  kasan_report+0xe0/0x110
[    7.267707]  ? device_release+0x1f4/0x240
[    7.267707]  device_release+0x1f4/0x240
[    7.267707]  kobject_put+0x1c8/0x450
[    7.267707]  atkbd_connect+0x615/0x9e0
[    7.267707]  ? __pfx_atkbd_connect+0x10/0x10
[    7.267707]  ? kernfs_create_link+0x169/0x230
[    7.267707]  ? do_raw_spin_unlock+0x53/0x220
[    7.267707]  serio_driver_probe+0x72/0xb0
[    7.267707]  really_probe+0x254/0x910
[    7.267707]  __driver_probe_device+0x20b/0x3d0
[    7.267707]  driver_probe_device+0x45/0x130
[    7.267707]  __driver_attach+0x1f6/0x550
[    7.267707]  ? __pfx___driver_attach+0x10/0x10
[    7.267707]  bus_for_each_dev+0x103/0x180
[    7.267707]  ? __pfx_bus_for_each_dev+0x10/0x10
[    7.267707]  ? _raw_spin_unlock_irqrestore+0x3f/0x50
[    7.267707]  ? lockdep_hardirqs_on_prepare+0xea/0x1a0
[    7.267707]  serio_handle_event+0x1ce/0x840
[    7.267707]  process_one_work+0x7fc/0x1760
[    7.267707]  ? __pfx_process_one_work+0x10/0x10
[    7.267707]  ? lock_is_held_type+0x8f/0x100
[    7.267707]  ? __pfx_serio_handle_event+0x10/0x10
[    7.267707]  worker_thread+0x593/0xfb0
[    7.267707]  ? __pfx_worker_thread+0x10/0x10
[    7.267707]  kthread+0x319/0x400
[    7.267707]  ? __pfx_kthread+0x10/0x10
[    7.267707]  ret_from_fork+0x590/0x830
[    7.267707]  ? __pfx_ret_from_fork+0x10/0x10
[    7.267707]  ? __switch_to+0x860/0xe50
[    7.267707]  ? __switch_to_asm+0x39/0x70
[    7.267707]  ? __switch_to_asm+0x33/0x70
[    7.267707]  ? __pfx_kthread+0x10/0x10
[    7.267707]  ret_from_fork_asm+0x1a/0x30
[    7.267707]  </TASK>
[    7.267707]
[    7.267707] Allocated by task 24:
[    7.267707]  kasan_save_stack+0x33/0x60
[    7.267707]  kasan_save_track+0x14/0x30
[    7.267707]  __kasan_kmalloc+0x8f/0xa0
[    7.267707]  input_allocate_device+0x3f/0x330
[    7.267707]  atkbd_connect+0x97/0x9e0
[    7.267707]  serio_driver_probe+0x72/0xb0
[    7.267707]  really_probe+0x254/0x910
[    7.267707]  __driver_probe_device+0x20b/0x3d0
[    7.267707]  driver_probe_device+0x45/0x130
[    7.267707]  __driver_attach+0x1f6/0x550
[    7.267707]  bus_for_each_dev+0x103/0x180
[    7.267707]  serio_handle_event+0x1ce/0x840
[    7.267707]  process_one_work+0x7fc/0x1760
[    7.267707]  worker_thread+0x593/0xfb0
[    7.267707]  kthread+0x319/0x400
[    7.267707]  ret_from_fork+0x590/0x830
[    7.267707]  ret_from_fork_asm+0x1a/0x30
[    7.267707]
[    7.267707] Freed by task 24:
[    7.267707]  kasan_save_stack+0x33/0x60
[    7.267707]  kasan_save_track+0x14/0x30
[    7.267707]  kasan_save_free_info+0x3b/0x60
[    7.267707]  __kasan_slab_free+0x43/0x70
[    7.267707]  kfree+0x193/0x4f0
[    7.267707]  input_dev_release+0xa6/0xd0
[    7.267707]  device_release+0x9a/0x240
[    7.267707]  kobject_put+0x1c8/0x450
[    7.267707]  atkbd_connect+0x615/0x9e0
[    7.267707]  serio_driver_probe+0x72/0xb0
[    7.267707]  really_probe+0x254/0x910
[    7.267707]  __driver_probe_device+0x20b/0x3d0
[    7.267707]  driver_probe_device+0x45/0x130
[    7.267707]  __driver_attach+0x1f6/0x550
[    7.267707]  bus_for_each_dev+0x103/0x180
[    7.267707]  serio_handle_event+0x1ce/0x840
[    7.267707]  process_one_work+0x7fc/0x1760
[    7.267707]  worker_thread+0x593/0xfb0
[    7.267707]  kthread+0x319/0x400
[    7.267707]  ret_from_fork+0x590/0x830
[    7.267707]  ret_from_fork_asm+0x1a/0x30
[    7.267707]
[    7.267707] The buggy address belongs to the object at ffff888003f4a000
[    7.267707]  which belongs to the cache kmalloc-2k of size 2048
[    7.267707] The buggy address is located 880 bytes inside of
[    7.267707]  freed 2048-byte region [ffff888003f4a000, ffff888003f4a800)
[    7.267707]
[    7.267707] The buggy address belongs to the physical page:
[    7.267707] page: refcount:0 mapcount:0 mapping:0000000000000000
index:0xffff888003f4b800 pfn:0x3f48
[    7.267707] head: order:3 mapcount:0 entire_mapcount:0
nr_pages_mapped:0 pincount:0
[    7.267707] flags: 0x100000000000240(workingset|head|node=0|zone=1)
[    7.267707] page_type: f5(slab)
[    7.267707] raw: 0100000000000240 ffff888001043240 ffff888001041088
ffff888001041088
[    7.267707] raw: ffff888003f4b800 0000000000050002 00000000f5000000
0000000000000000
[    7.267707] head: 0100000000000240 ffff888001043240
ffff888001041088 ffff888001041088
[    7.267707] head: ffff888003f4b800 0000000000050002
00000000f5000000 0000000000000000
[    7.267707] head: 0100000000000003 ffffea00000fd201
00000000ffffffff 00000000ffffffff
[    7.267707] head: 0000000000000000 0000000000000000
00000000ffffffff 0000000000000000
[    7.267707] page dumped because: kasan: bad access detected
[    7.267707]
[    7.267707] Memory state around the buggy address:
[    7.267707]  ffff888003f4a200: fb fb fb fb fb fb fb fb fb fb fb fb
fb fb fb fb
[    7.267707]  ffff888003f4a280: fb fb fb fb fb fb fb fb fb fb fb fb
fb fb fb fb
[    7.267707] >ffff888003f4a300: fb fb fb fb fb fb fb fb fb fb fb fb
fb fb fb fb
[    7.267707]                                                              ^
[    7.267707]  ffff888003f4a380: fb fb fb fb fb fb fb fb fb fb fb fb
fb fb fb fb
[    7.267707]  ffff888003f4a400: fb fb fb fb fb fb fb fb fb fb fb fb
fb fb fb fb
[    7.267707] ==================================================================
[    7.293685] Disabling lock debugging due to kernel taint

This is on a basic QEMU x86_64 VM. Note that I did not apply the "WIP:
treewide: make callsites use generic driver_override" patch.

I'm currently looking into the root cause.

Thanks.

Quoted 213 lines
---
 drivers/base/core.c    |  2 ++
 drivers/base/dd.c      | 60 +++++++++++++++++++++++++++++++
 include/linux/device.h | 81 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 143 insertions(+)
drivers/base/core.c +2 -0
index 791f9e444df8..a8cb90577d10 100644
@@ -2566,6 +2566,7 @@ static void device_release(struct kobject *kobj)
        else
                WARN(1, KERN_ERR "Device '%s' does not have a release() function, it is broken and must be fixed. See Documentation/core-api/kobject.rst.\n",
                        dev_name(dev));
+       kfree(dev->driver_override.name);
        kfree(p);
 }
 
@@ -3159,6 +3160,7 @@ void device_initialize(struct device *dev)
        kobject_init(&dev->kobj, &device_ktype);
        INIT_LIST_HEAD(&dev->dma_pools);
        mutex_init(&dev->mutex);
+       spin_lock_init(&dev->driver_override.lock);
        lockdep_set_novalidate_class(&dev->mutex);
        spin_lock_init(&dev->devres_lock);
        INIT_LIST_HEAD(&dev->devres_head);
drivers/base/dd.c +60 -0
index 0354f209529c..697e36e63cab 100644
@@ -381,6 +381,66 @@ static void __exit deferred_probe_exit(void)
 }
 __exitcall(deferred_probe_exit);
 
+int __device_set_driver_override(struct device *dev, const char *s, size_t len)
+{
+       const char *new, *old;
+       char *cp;
+
+       if (!s)
+               return -EINVAL;
+
+       /*
+        * The stored value will be used in sysfs show callback (sysfs_emit()),
+        * which has a length limit of PAGE_SIZE and adds a trailing newline.
+        * Thus we can store one character less to avoid truncation during sysfs
+        * show.
+        */
+       if (len >= (PAGE_SIZE - 1))
+               return -EINVAL;
+
+       /*
+        * Compute the real length of the string in case userspace sends us a
+        * bunch of \0 characters like python likes to do.
+        */
+       len = strlen(s);
+
+       if (!len) {
+               /* Empty string passed - clear override */
+               spin_lock(&dev->driver_override.lock);
+               old = dev->driver_override.name;
+               dev->driver_override.name = NULL;
+               spin_unlock(&dev->driver_override.lock);
+               kfree(old);
+
+               return 0;
+       }
+
+       cp = strnchr(s, len, '\n');
+       if (cp)
+               len = cp - s;
+
+       new = kstrndup(s, len, GFP_KERNEL);
+       if (!new)
+               return -ENOMEM;
+
+       spin_lock(&dev->driver_override.lock);
+       old = dev->driver_override.name;
+       if (cp != s) {
+               dev->driver_override.name = new;
+               spin_unlock(&dev->driver_override.lock);
+       } else {
+               /* "\n" passed - clear override */
+               dev->driver_override.name = NULL;
+               spin_unlock(&dev->driver_override.lock);
+
+               kfree(new);
+       }
+       kfree(old);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(__device_set_driver_override);
+
 /**
  * device_is_bound() - Check if device is bound to a driver
  * @dev: device to check
include/linux/device.h +81 -1
index 0be95294b6e6..4599156d5cbd 100644
@@ -266,6 +266,33 @@ ssize_t device_show_string(struct device *dev, struct device_attribute *attr,
        struct dev_ext_attribute dev_attr_##_name = \
                { __ATTR(_name, (_mode) & ~0222, device_show_string, NULL), (_var) }
 
+/**
+ * DEVICE_ATTR_DRIVER_OVERRIDE - Define sysfs driver_override attribute callbacks
+ *
+ * Generates the standard driver_override_show() and driver_override_store()
+ * sysfs callbacks and the static DEVICE_ATTR_RW(driver_override) declaration.
+ */
+#define DEVICE_ATTR_DRIVER_OVERRIDE()                                          \
+static ssize_t driver_override_store(struct device *dev,                       \
+                                    struct device_attribute *attr,             \
+                                    const char *buf, size_t count)             \
+{                                                                              \
+       int ret;                                                                \
+                                                                               \
+       ret = __device_set_driver_override(dev, buf, count);                    \
+       if (ret)                                                                \
+               return ret;                                                     \
+                                                                               \
+       return count;                                                           \
+}                                                                              \
+static ssize_t driver_override_show(struct device *dev,                                \
+                                   struct device_attribute *attr, char *buf)   \
+{                                                                              \
+       guard(spinlock)(&dev->driver_override.lock);                            \
+       return sysfs_emit(buf, "%s\n", dev->driver_override.name);              \
+}                                                                              \
+static DEVICE_ATTR_RW(driver_override)
+
 #define DEVICE_ATTR_IGNORE_LOCKDEP(_name, _mode, _show, _store) \
        struct device_attribute dev_attr_##_name =              \
                __ATTR_IGNORE_LOCKDEP(_name, _mode, _show, _store)
@@ -483,6 +510,8 @@ struct device_physical_location {
  *             on.  This shrinks the "Board Support Packages" (BSPs) and
  *             minimizes board-specific #ifdefs in drivers.
  * @driver_data: Private pointer for driver specific info.
+ * @driver_override: Driver name to force a match.  Do not touch directly; use
+ *                  device_set_driver_override() instead.
  * @links:     Links to suppliers and consumers of this device.
  * @power:     For device power management.
  *             See Documentation/driver-api/pm/devices.rst for details.
@@ -576,6 +605,10 @@ struct device {
                                           core doesn't touch it */
        void            *driver_data;   /* Driver data, set and get with
                                           dev_set_drvdata/dev_get_drvdata */
+       struct {
+               const char      *name;
+               spinlock_t      lock;
+       } driver_override;
        struct mutex            mutex;  /* mutex to synchronize calls to
                                         * its driver.
                                         */
@@ -701,6 +734,54 @@ struct device_link {
 
 #define kobj_to_dev(__kobj)    container_of_const(__kobj, struct device, kobj)
 
+int __device_set_driver_override(struct device *dev, const char *s, size_t len);
+
+/**
+ * device_set_driver_override() - Helper to set or clear driver override.
+ * @dev: Device to change
+ * @s: NUL-terminated string, new driver name to force a match, pass empty
+ *     string to clear it ("" or "\n", where the latter is only for sysfs
+ *     interface).
+ *
+ * Helper to set or clear driver override of a device.
+ *
+ * Returns: 0 on success or a negative error code on failure.
+ */
+static inline int device_set_driver_override(struct device *dev, const char *s)
+{
+       return __device_set_driver_override(dev, s, strlen(s));
+}
+
+/**
+ * device_has_driver_override() - Check if a driver override has been set.
+ * @dev: device to check
+ *
+ * Returns true if a driver override has been set for this device.
+ */
+static inline bool device_has_driver_override(struct device *dev)
+{
+       guard(spinlock)(&dev->driver_override.lock);
+       return !!dev->driver_override.name;
+}
+
+/**
+ * device_match_driver_override() - Match a driver against the device's driver_override.
+ * @dev: device to check
+ * @drv: driver to match against
+ *
+ * Returns > 0 if a driver override is set and matches the given driver, 0 if a
+ * driver override is set but does not match, or < 0 if a driver override is not
+ * set at all.
+ */
+static inline int device_match_driver_override(struct device *dev,
+                                              const struct device_driver *drv)
+{
+       guard(spinlock)(&dev->driver_override.lock);
+       if (dev->driver_override.name)
+               return !strcmp(dev->driver_override.name, drv->name);
+       return -1;
+}
+
 /**
  * device_iommu_mapped - Returns true when the device DMA is translated
  *                      by an IOMMU
--

2.53.0

Tested-by Reviewed-by on v1 1/3
#3 lore raw

Re: [PATCH 1/3] driver core: generalize driver_override in struct device

In reply to Danilo Krummrich
Quoted 18 lines · Danilo Krummrich

Currently, there are 12 busses (including platform and PCI) that duplicate the driver_override logic for their individual devices.

All of them seem to be prone to the bug described in [1].

While this could be solved for every bus individually using a separate lock, solving this in the driver-core generically results in less (and cleaner) changes overall.

Thus, move driver_override to struct device, provide corresponding accessors for busses and handle locking with a separate lock internally.

In particular, add device_set_driver_override(), device_has_driver_override(), device_match_driver_override() and a helper, DEVICE_ATTR_DRIVER_OVERRIDE(), to declare the corresponding sysfs store() and show() callbacks.

Quoted 7 lines

Until all busses have migrated, keep driver_set_override() in place.

Note that we can't use the device lock for the reasons described in [2].

Applying the following diff fixes the KASAN issue. It was just a minor bug.

drivers/base/core.c +1 -0
index a8cb90577d10..09b98f02f559 100644
@@ -2556,6 +2556,7 @@ static void device_release(struct kobject *kobj)
        devres_release_all(dev);
 
        kfree(dev->dma_range_map);
+       kfree(dev->driver_override.name);
 
        if (dev->release)
                dev->release(dev);
@@ -2566,7 +2567,6 @@ static void device_release(struct kobject *kobj)
        else
                WARN(1, KERN_ERR "Device '%s' does not have a

release() function, it is broken and must be fixed. See Documentation/core-api/kobject.rst.\n",

                        dev_name(dev));
-       kfree(dev->driver_override.name);
        kfree(p);

}

With this applied, along with the PCI driver diff from the WIP patch, the issue is resolved. I tested this on PCI and both PoCs no longer trigger KASAN. I also ran with other debug options enabled (lockdep, sleep inside atomic, etc.) and hit no warnings.

I was working on a similar patch recently, but your version is better. Not returning the string directly provides better encapsulation and makes the API much harder to misuse.

Quoted 213 lines
---
 drivers/base/core.c    |  2 ++
 drivers/base/dd.c      | 60 +++++++++++++++++++++++++++++++
 include/linux/device.h | 81 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 143 insertions(+)
drivers/base/core.c +2 -0
index 791f9e444df8..a8cb90577d10 100644
@@ -2566,6 +2566,7 @@ static void device_release(struct kobject *kobj)
        else
                WARN(1, KERN_ERR "Device '%s' does not have a release() function, it is broken and must be fixed. See Documentation/core-api/kobject.rst.\n",
                        dev_name(dev));
+       kfree(dev->driver_override.name);
        kfree(p);
 }
 
@@ -3159,6 +3160,7 @@ void device_initialize(struct device *dev)
        kobject_init(&dev->kobj, &device_ktype);
        INIT_LIST_HEAD(&dev->dma_pools);
        mutex_init(&dev->mutex);
+       spin_lock_init(&dev->driver_override.lock);
        lockdep_set_novalidate_class(&dev->mutex);
        spin_lock_init(&dev->devres_lock);
        INIT_LIST_HEAD(&dev->devres_head);
drivers/base/dd.c +60 -0
index 0354f209529c..697e36e63cab 100644
@@ -381,6 +381,66 @@ static void __exit deferred_probe_exit(void)
 }
 __exitcall(deferred_probe_exit);
 
+int __device_set_driver_override(struct device *dev, const char *s, size_t len)
+{
+       const char *new, *old;
+       char *cp;
+
+       if (!s)
+               return -EINVAL;
+
+       /*
+        * The stored value will be used in sysfs show callback (sysfs_emit()),
+        * which has a length limit of PAGE_SIZE and adds a trailing newline.
+        * Thus we can store one character less to avoid truncation during sysfs
+        * show.
+        */
+       if (len >= (PAGE_SIZE - 1))
+               return -EINVAL;
+
+       /*
+        * Compute the real length of the string in case userspace sends us a
+        * bunch of \0 characters like python likes to do.
+        */
+       len = strlen(s);
+
+       if (!len) {
+               /* Empty string passed - clear override */
+               spin_lock(&dev->driver_override.lock);
+               old = dev->driver_override.name;
+               dev->driver_override.name = NULL;
+               spin_unlock(&dev->driver_override.lock);
+               kfree(old);
+
+               return 0;
+       }
+
+       cp = strnchr(s, len, '\n');
+       if (cp)
+               len = cp - s;
+
+       new = kstrndup(s, len, GFP_KERNEL);
+       if (!new)
+               return -ENOMEM;
+
+       spin_lock(&dev->driver_override.lock);
+       old = dev->driver_override.name;
+       if (cp != s) {
+               dev->driver_override.name = new;
+               spin_unlock(&dev->driver_override.lock);
+       } else {
+               /* "\n" passed - clear override */
+               dev->driver_override.name = NULL;
+               spin_unlock(&dev->driver_override.lock);
+
+               kfree(new);
+       }
+       kfree(old);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(__device_set_driver_override);
+
 /**
  * device_is_bound() - Check if device is bound to a driver
  * @dev: device to check
include/linux/device.h +81 -1
index 0be95294b6e6..4599156d5cbd 100644
@@ -266,6 +266,33 @@ ssize_t device_show_string(struct device *dev, struct device_attribute *attr,
        struct dev_ext_attribute dev_attr_##_name = \
                { __ATTR(_name, (_mode) & ~0222, device_show_string, NULL), (_var) }
 
+/**
+ * DEVICE_ATTR_DRIVER_OVERRIDE - Define sysfs driver_override attribute callbacks
+ *
+ * Generates the standard driver_override_show() and driver_override_store()
+ * sysfs callbacks and the static DEVICE_ATTR_RW(driver_override) declaration.
+ */
+#define DEVICE_ATTR_DRIVER_OVERRIDE()                                          \
+static ssize_t driver_override_store(struct device *dev,                       \
+                                    struct device_attribute *attr,             \
+                                    const char *buf, size_t count)             \
+{                                                                              \
+       int ret;                                                                \
+                                                                               \
+       ret = __device_set_driver_override(dev, buf, count);                    \
+       if (ret)                                                                \
+               return ret;                                                     \
+                                                                               \
+       return count;                                                           \
+}                                                                              \
+static ssize_t driver_override_show(struct device *dev,                                \
+                                   struct device_attribute *attr, char *buf)   \
+{                                                                              \
+       guard(spinlock)(&dev->driver_override.lock);                            \
+       return sysfs_emit(buf, "%s\n", dev->driver_override.name);              \
+}                                                                              \
+static DEVICE_ATTR_RW(driver_override)
+
 #define DEVICE_ATTR_IGNORE_LOCKDEP(_name, _mode, _show, _store) \
        struct device_attribute dev_attr_##_name =              \
                __ATTR_IGNORE_LOCKDEP(_name, _mode, _show, _store)
@@ -483,6 +510,8 @@ struct device_physical_location {
  *             on.  This shrinks the "Board Support Packages" (BSPs) and
  *             minimizes board-specific #ifdefs in drivers.
  * @driver_data: Private pointer for driver specific info.
+ * @driver_override: Driver name to force a match.  Do not touch directly; use
+ *                  device_set_driver_override() instead.
  * @links:     Links to suppliers and consumers of this device.
  * @power:     For device power management.
  *             See Documentation/driver-api/pm/devices.rst for details.
@@ -576,6 +605,10 @@ struct device {
                                           core doesn't touch it */
        void            *driver_data;   /* Driver data, set and get with
                                           dev_set_drvdata/dev_get_drvdata */
+       struct {
+               const char      *name;
+               spinlock_t      lock;
+       } driver_override;
        struct mutex            mutex;  /* mutex to synchronize calls to
                                         * its driver.
                                         */
@@ -701,6 +734,54 @@ struct device_link {
 
 #define kobj_to_dev(__kobj)    container_of_const(__kobj, struct device, kobj)
 
+int __device_set_driver_override(struct device *dev, const char *s, size_t len);
+
+/**
+ * device_set_driver_override() - Helper to set or clear driver override.
+ * @dev: Device to change
+ * @s: NUL-terminated string, new driver name to force a match, pass empty
+ *     string to clear it ("" or "\n", where the latter is only for sysfs
+ *     interface).
+ *
+ * Helper to set or clear driver override of a device.
+ *
+ * Returns: 0 on success or a negative error code on failure.
+ */
+static inline int device_set_driver_override(struct device *dev, const char *s)
+{
+       return __device_set_driver_override(dev, s, strlen(s));
+}
+
+/**
+ * device_has_driver_override() - Check if a driver override has been set.
+ * @dev: device to check
+ *
+ * Returns true if a driver override has been set for this device.
+ */
+static inline bool device_has_driver_override(struct device *dev)
+{
+       guard(spinlock)(&dev->driver_override.lock);
+       return !!dev->driver_override.name;
+}
+
+/**
+ * device_match_driver_override() - Match a driver against the device's driver_override.
+ * @dev: device to check
+ * @drv: driver to match against
+ *
+ * Returns > 0 if a driver override is set and matches the given driver, 0 if a
+ * driver override is set but does not match, or < 0 if a driver override is not
+ * set at all.
+ */
+static inline int device_match_driver_override(struct device *dev,
+                                              const struct device_driver *drv)
+{
+       guard(spinlock)(&dev->driver_override.lock);
+       if (dev->driver_override.name)
+               return !strcmp(dev->driver_override.name, drv->name);
+       return -1;
+}
+
 /**
  * device_iommu_mapped - Returns true when the device DMA is translated
  *                      by an IOMMU
--

2.53.0

author maintainer
#4 lore raw

Re: [PATCH 1/3] driver core: generalize driver_override in struct device

Quoted 19 lines · Gui-Dong Han

Applying the following diff fixes the KASAN issue. It was just a minor bug.

drivers/base/core.c +1 -0
index a8cb90577d10..09b98f02f559 100644
@@ -2556,6 +2556,7 @@ static void device_release(struct kobject *kobj)
        devres_release_all(dev);
 
        kfree(dev->dma_range_map);
+       kfree(dev->driver_override.name);
 
        if (dev->release)
                dev->release(dev);
@@ -2566,7 +2567,6 @@ static void device_release(struct kobject *kobj)
        else
                WARN(1, KERN_ERR "Device '%s' does not have a

release() function, it is broken and must be fixed. See Documentation/core-api/kobject.rst.\n",

Quoted 4 lines
                        dev_name(dev));
-       kfree(dev->driver_override.name);
        kfree(p);

}

Yes, we must not access dev after the release callbacks has been called; no idea how this kfree() ended up below. Thanks for catching!

Quoted 8 lines

With this applied, along with the PCI driver diff from the WIP patch, the issue is resolved. I tested this on PCI and both PoCs no longer trigger KASAN. I also ran with other debug options enabled (lockdep, sleep inside atomic, etc.) and hit no warnings.

I was working on a similar patch recently, but your version is better. Not returning the string directly provides better encapsulation and makes the API much harder to misuse.

Ah, right, I remember you mentioned that! If you want I can add your
Co-developed-by: to this patch to account for your work.

Quoted 2 lines
#5 lore raw

Re: [PATCH 1/3] driver core: generalize driver_override in struct device

Quoted 34 lines · Danilo Krummrich
Quoted 19 lines · Gui-Dong Han

Applying the following diff fixes the KASAN issue. It was just a minor bug.

drivers/base/core.c +1 -0
index a8cb90577d10..09b98f02f559 100644
@@ -2556,6 +2556,7 @@ static void device_release(struct kobject *kobj)
        devres_release_all(dev);
 
        kfree(dev->dma_range_map);
+       kfree(dev->driver_override.name);
 
        if (dev->release)
                dev->release(dev);
@@ -2566,7 +2567,6 @@ static void device_release(struct kobject *kobj)
        else
                WARN(1, KERN_ERR "Device '%s' does not have a

release() function, it is broken and must be fixed. See Documentation/core-api/kobject.rst.\n",

Quoted 4 lines
                        dev_name(dev));
-       kfree(dev->driver_override.name);
        kfree(p);

}

Yes, we must not access dev after the release callbacks has been called; no idea how this kfree() ended up below. Thanks for catching!

Quoted 5 lines

With this applied, along with the PCI driver diff from the WIP patch, the issue is resolved. I tested this on PCI and both PoCs no longer trigger KASAN. I also ran with other debug options enabled (lockdep, sleep inside atomic, etc.) and hit no warnings.

Quoted 6 lines
Quoted 3 lines

I was working on a similar patch recently, but your version is better. Not returning the string directly provides better encapsulation and makes the API much harder to misuse.

Ah, right, I remember you mentioned that! If you want I can add your
Co-developed-by: to this patch to account for your work.

That would be great, yes please. Thank you!

And thanks for all your hard work on this patch series, it's a really solid improvement.

#6 lore raw

Re: [PATCH 1/3] driver core: generalize driver_override in struct device

In reply to Danilo Krummrich

Hi Danilo,

Quoted 17 lines · Danilo Krummrich

Currently, there are 12 busses (including platform and PCI) that duplicate the driver_override logic for their individual devices.

All of them seem to be prone to the bug described in [1].

While this could be solved for every bus individually using a separate lock, solving this in the driver-core generically results in less (and cleaner) changes overall.

Thus, move driver_override to struct device, provide corresponding accessors for busses and handle locking with a separate lock internally.

In particular, add device_set_driver_override(), device_has_driver_override(), device_match_driver_override() and a helper, DEVICE_ATTR_DRIVER_OVERRIDE(), to declare the corresponding sysfs store() and show() callbacks.

Quoted 7 lines

Until all busses have migrated, keep driver_set_override() in place.

Note that we can't use the device lock for the reasons described in [2].

Thanks for your patch!

Quoted 19 lines
drivers/base/dd.c +13 -0
@@ -381,6 +381,66 @@ static void __exit deferred_probe_exit(void)
 }
 __exitcall(deferred_probe_exit);
 
+int __device_set_driver_override(struct device *dev, const char *s, size_t len)
+{
+       const char *new, *old;
+       char *cp;
+
+       if (!s)
+               return -EINVAL;
+
+       /*
+        * The stored value will be used in sysfs show callback (sysfs_emit()),
+        * which has a length limit of PAGE_SIZE and adds a trailing newline.
+        * Thus we can store one character less to avoid truncation during sysfs
+        * show.

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 406

Quoted 10 lines
drivers/base/dd.c +10 -0
+        */
+       if (len >= (PAGE_SIZE - 1))
+               return -EINVAL;
+
+       /*
+        * Compute the real length of the string in case userspace sends us a
+        * bunch of \0 characters like python likes to do.
+        */
+       len = strlen(s);
+

The newline case below is is basically the same case as the empty string. Hence if you would move the newline check here:

    if (len) {
            cp = strnchr(s, len, '\n');
            if (cp)
                    len = cp - s;
    }

then the "cp != s" check below is no longer needed.

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 416

Quoted 10 lines
drivers/base/dd.c +10 -0
+       if (!len) {
+               /* Empty string passed - clear override */
+               spin_lock(&dev->driver_override.lock);
+               old = dev->driver_override.name;
+               dev->driver_override.name = NULL;
+               spin_unlock(&dev->driver_override.lock);
+               kfree(old);
+
+               return 0;
+       }

Also, this block can be eliminated completely...

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 424

Quoted 8 lines
drivers/base/dd.c +8 -0
+
+       cp = strnchr(s, len, '\n');
+       if (cp)
+               len = cp - s;
+
+       new = kstrndup(s, len, GFP_KERNEL);
+       if (!new)
+               return -ENOMEM;

... by pre-initializing new to NULL, and making the allocation of new conditional on len being non-zero.

Quoted 12 lines
drivers/base/dd.c +12 -0
+
+       spin_lock(&dev->driver_override.lock);
+       old = dev->driver_override.name;
+       if (cp != s) {
+               dev->driver_override.name = new;
+               spin_unlock(&dev->driver_override.lock);
+       } else {
+               /* "\n" passed - clear override */
+               dev->driver_override.name = NULL;
+               spin_unlock(&dev->driver_override.lock);
+
+               kfree(new);

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 446

Quoted 10 lines
drivers/base/dd.c +7 -0
+       }
+       kfree(old);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(__device_set_driver_override);
+
 /**
  * device_is_bound() - Check if device is bound to a driver
  * @dev: device to check

Gr{oetje,eeting}s,

Geert

Signature · 5 lines
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds
author maintainer
#7 lore raw

Re: [PATCH 1/3] driver core: generalize driver_override in struct device

Quoted 76 lines · Geert Uytterhoeven
Quoted 19 lines
drivers/base/dd.c +13 -0
@@ -381,6 +381,66 @@ static void __exit deferred_probe_exit(void)
 }
 __exitcall(deferred_probe_exit);
 
+int __device_set_driver_override(struct device *dev, const char *s, size_t len)
+{
+       const char *new, *old;
+       char *cp;
+
+       if (!s)
+               return -EINVAL;
+
+       /*
+        * The stored value will be used in sysfs show callback (sysfs_emit()),
+        * which has a length limit of PAGE_SIZE and adds a trailing newline.
+        * Thus we can store one character less to avoid truncation during sysfs
+        * show.

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 406

Quoted 10 lines
drivers/base/dd.c +10 -0
+        */
+       if (len >= (PAGE_SIZE - 1))
+               return -EINVAL;
+
+       /*
+        * Compute the real length of the string in case userspace sends us a
+        * bunch of \0 characters like python likes to do.
+        */
+       len = strlen(s);
+

The newline case below is is basically the same case as the empty string. Hence if you would move the newline check here:

    if (len) {
            cp = strnchr(s, len, '\n');
            if (cp)
                    len = cp - s;
    }

then the "cp != s" check below is no longer needed.

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 416

Quoted 10 lines
drivers/base/dd.c +10 -0
+       if (!len) {
+               /* Empty string passed - clear override */
+               spin_lock(&dev->driver_override.lock);
+               old = dev->driver_override.name;
+               dev->driver_override.name = NULL;
+               spin_unlock(&dev->driver_override.lock);
+               kfree(old);
+
+               return 0;
+       }

Also, this block can be eliminated completely...

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 424

Quoted 8 lines
drivers/base/dd.c +8 -0
+
+       cp = strnchr(s, len, '\n');
+       if (cp)
+               len = cp - s;
+
+       new = kstrndup(s, len, GFP_KERNEL);
+       if (!new)
+               return -ENOMEM;

... by pre-initializing new to NULL, and making the allocation of new conditional on len being non-zero.

Quoted 10 lines
drivers/base/dd.c +10 -0
+
+       spin_lock(&dev->driver_override.lock);
+       old = dev->driver_override.name;
+       if (cp != s) {
+               dev->driver_override.name = new;
+               spin_unlock(&dev->driver_override.lock);
+       } else {
+               /* "\n" passed - clear override */
+               dev->driver_override.name = NULL;
+               spin_unlock(&dev->driver_override.lock);
Quoted 8 lines

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 442

Quoted 8 lines
drivers/base/dd.c +8 -0
+
+               kfree(new);
+       }
+       kfree(old);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(__device_set_driver_override);

This is essentially a copy of driver_set_override(). Except for the required minor changes I intentionally kept it "as is" as it will go through -fixes and we know it works properly.

Do you mind sending a follow-up patch with your suggested improvements?

#8 lore raw

Re: [PATCH 1/3] driver core: generalize driver_override in struct device

Hi Danilo,

Quoted 81 lines · Danilo Krummrich
Quoted 80 lines · Geert Uytterhoeven
Quoted 19 lines
drivers/base/dd.c +13 -0
@@ -381,6 +381,66 @@ static void __exit deferred_probe_exit(void)
 }
 __exitcall(deferred_probe_exit);
 
+int __device_set_driver_override(struct device *dev, const char *s, size_t len)
+{
+       const char *new, *old;
+       char *cp;
+
+       if (!s)
+               return -EINVAL;
+
+       /*
+        * The stored value will be used in sysfs show callback (sysfs_emit()),
+        * which has a length limit of PAGE_SIZE and adds a trailing newline.
+        * Thus we can store one character less to avoid truncation during sysfs
+        * show.

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 406

Quoted 10 lines
drivers/base/dd.c +10 -0
+        */
+       if (len >= (PAGE_SIZE - 1))
+               return -EINVAL;
+
+       /*
+        * Compute the real length of the string in case userspace sends us a
+        * bunch of \0 characters like python likes to do.
+        */
+       len = strlen(s);
+

The newline case below is is basically the same case as the empty string. Hence if you would move the newline check here:

    if (len) {
            cp = strnchr(s, len, '\n');
            if (cp)
                    len = cp - s;
    }

then the "cp != s" check below is no longer needed.

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 416

Quoted 10 lines
drivers/base/dd.c +10 -0
+       if (!len) {
+               /* Empty string passed - clear override */
+               spin_lock(&dev->driver_override.lock);
+               old = dev->driver_override.name;
+               dev->driver_override.name = NULL;
+               spin_unlock(&dev->driver_override.lock);
+               kfree(old);
+
+               return 0;
+       }

Also, this block can be eliminated completely...

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 424

Quoted 8 lines
drivers/base/dd.c +8 -0
+
+       cp = strnchr(s, len, '\n');
+       if (cp)
+               len = cp - s;
+
+       new = kstrndup(s, len, GFP_KERNEL);
+       if (!new)
+               return -ENOMEM;

... by pre-initializing new to NULL, and making the allocation of new conditional on len being non-zero.

Quoted 14 lines
drivers/base/dd.c +14 -0
+
+       spin_lock(&dev->driver_override.lock);
+       old = dev->driver_override.name;
+       if (cp != s) {
+               dev->driver_override.name = new;
+               spin_unlock(&dev->driver_override.lock);
+       } else {
+               /* "\n" passed - clear override */
+               dev->driver_override.name = NULL;
+               spin_unlock(&dev->driver_override.lock);
+
+               kfree(new);
+       }
+       kfree(old);
Quoted 8 lines
Quoted 4 lines

drivers/base/dd.c static void __exit deferred_probe_exit(void) · line 442

Quoted 4 lines
drivers/base/dd.c +4 -0
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(__device_set_driver_override);

This is essentially a copy of driver_set_override(). Except for the required minor changes I intentionally kept it "as is" as it will go through -fixes and we know it works properly.

So I will have two to fix? ;-)

Quoted 1 line

Do you mind sending a follow-up patch with your suggested improvements?

Adding it to my TODO list, if this patch makes it as-is.

Gr{oetje,eeting}s,

Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- [email protected]

In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds

author maintainer
#9 lore raw

Re: [PATCH 1/3] driver core: generalize driver_override in struct device

Quoted 1 line · Geert Uytterhoeven

So I will have two to fix? ;-)

No, once all busses have been migrated driver_set_override() will be removed. :)

#10 lore raw

Re: [PATCH 1/3] driver core: generalize driver_override in struct device

In reply to Danilo Krummrich

Am 02.03.26 um 01:25 schrieb Danilo Krummrich:

Quoted 9 lines

Currently, there are 12 busses (including platform and PCI) that duplicate the driver_override logic for their individual devices.

All of them seem to be prone to the bug described in [1].

While this could be solved for every bus individually using a separate lock, solving this in the driver-core generically results in less (and cleaner) changes overall.

Quoted 7 lines

Thus, move driver_override to struct device, provide corresponding accessors for busses and handle locking with a separate lock internally.

In particular, add device_set_driver_override(), device_has_driver_override(), device_match_driver_override() and a helper, DEVICE_ATTR_DRIVER_OVERRIDE(), to declare the corresponding sysfs store() and show() callbacks.

Nice patch series, centralizing the handling of driver_override in the device core removes a lot of duplicated code.

I wonder if we can also move DEVICE_ATTR_DRIVER_OVERRIDE() into the device core by registering the associated sysfs attribute inside bus_add_device().

Bus types that honor driver_override could for example set a boolean flag inside struct bus_type to tell the device core that driver_override is supported.

However this is just a suggestion, the current patch series seems fine to me.

Thanks,
Armin Wolf

Quoted 217 lines

Until all busses have migrated, keep driver_set_override() in place.

Note that we can't use the device lock for the reasons described in [2].

---
  drivers/base/core.c    |  2 ++
  drivers/base/dd.c      | 60 +++++++++++++++++++++++++++++++
  include/linux/device.h | 81 ++++++++++++++++++++++++++++++++++++++++++
  3 files changed, 143 insertions(+)
drivers/base/core.c +2 -0
index 791f9e444df8..a8cb90577d10 100644
@@ -2566,6 +2566,7 @@ static void device_release(struct kobject *kobj)
  	else
  		WARN(1, KERN_ERR "Device '%s' does not have a release() function, it is broken and must be fixed. See Documentation/core-api/kobject.rst.\n",
  			dev_name(dev));
+	kfree(dev->driver_override.name);
  	kfree(p);
  }
 
@@ -3159,6 +3160,7 @@ void device_initialize(struct device *dev)
  	kobject_init(&dev->kobj, &device_ktype);
  	INIT_LIST_HEAD(&dev->dma_pools);
  	mutex_init(&dev->mutex);
+	spin_lock_init(&dev->driver_override.lock);
  	lockdep_set_novalidate_class(&dev->mutex);
  	spin_lock_init(&dev->devres_lock);
  	INIT_LIST_HEAD(&dev->devres_head);
drivers/base/dd.c +60 -0
index 0354f209529c..697e36e63cab 100644
@@ -381,6 +381,66 @@ static void __exit deferred_probe_exit(void)
  }
  __exitcall(deferred_probe_exit);
 
+int __device_set_driver_override(struct device *dev, const char *s, size_t len)
+{
+	const char *new, *old;
+	char *cp;
+
+	if (!s)
+		return -EINVAL;
+
+	/*
+	 * The stored value will be used in sysfs show callback (sysfs_emit()),
+	 * which has a length limit of PAGE_SIZE and adds a trailing newline.
+	 * Thus we can store one character less to avoid truncation during sysfs
+	 * show.
+	 */
+	if (len >= (PAGE_SIZE - 1))
+		return -EINVAL;
+
+	/*
+	 * Compute the real length of the string in case userspace sends us a
+	 * bunch of \0 characters like python likes to do.
+	 */
+	len = strlen(s);
+
+	if (!len) {
+		/* Empty string passed - clear override */
+		spin_lock(&dev->driver_override.lock);
+		old = dev->driver_override.name;
+		dev->driver_override.name = NULL;
+		spin_unlock(&dev->driver_override.lock);
+		kfree(old);
+
+		return 0;
+	}
+
+	cp = strnchr(s, len, '\n');
+	if (cp)
+		len = cp - s;
+
+	new = kstrndup(s, len, GFP_KERNEL);
+	if (!new)
+		return -ENOMEM;
+
+	spin_lock(&dev->driver_override.lock);
+	old = dev->driver_override.name;
+	if (cp != s) {
+		dev->driver_override.name = new;
+		spin_unlock(&dev->driver_override.lock);
+	} else {
+		/* "\n" passed - clear override */
+		dev->driver_override.name = NULL;
+		spin_unlock(&dev->driver_override.lock);
+
+		kfree(new);
+	}
+	kfree(old);
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(__device_set_driver_override);
+
  /**
   * device_is_bound() - Check if device is bound to a driver
   * @dev: device to check
include/linux/device.h +81 -0
index 0be95294b6e6..4599156d5cbd 100644
@@ -266,6 +266,33 @@ ssize_t device_show_string(struct device *dev, struct device_attribute *attr,
  	struct dev_ext_attribute dev_attr_##_name = \
  		{ __ATTR(_name, (_mode) & ~0222, device_show_string, NULL), (_var) }
 
+/**
+ * DEVICE_ATTR_DRIVER_OVERRIDE - Define sysfs driver_override attribute callbacks
+ *
+ * Generates the standard driver_override_show() and driver_override_store()
+ * sysfs callbacks and the static DEVICE_ATTR_RW(driver_override) declaration.
+ */
+#define DEVICE_ATTR_DRIVER_OVERRIDE()						\
+static ssize_t driver_override_store(struct device *dev,			\
+				     struct device_attribute *attr,		\
+				     const char *buf, size_t count)		\
+{										\
+	int ret;								\
+										\
+	ret = __device_set_driver_override(dev, buf, count);			\
+	if (ret)								\
+		return ret;							\
+										\
+	return count;								\
+}										\
+static ssize_t driver_override_show(struct device *dev,				\
+				    struct device_attribute *attr, char *buf)	\
+{										\
+	guard(spinlock)(&dev->driver_override.lock);				\
+	return sysfs_emit(buf, "%s\n", dev->driver_override.name);		\
+}										\
+static DEVICE_ATTR_RW(driver_override)
+
  #define DEVICE_ATTR_IGNORE_LOCKDEP(_name, _mode, _show, _store) \
  	struct device_attribute dev_attr_##_name =		\
  		__ATTR_IGNORE_LOCKDEP(_name, _mode, _show, _store)
@@ -483,6 +510,8 @@ struct device_physical_location {
   * 		on.  This shrinks the "Board Support Packages" (BSPs) and
   * 		minimizes board-specific #ifdefs in drivers.
   * @driver_data: Private pointer for driver specific info.
+ * @driver_override: Driver name to force a match.  Do not touch directly; use
+ *		     device_set_driver_override() instead.
   * @links:	Links to suppliers and consumers of this device.
   * @power:	For device power management.
   *		See Documentation/driver-api/pm/devices.rst for details.
@@ -576,6 +605,10 @@ struct device {
  					   core doesn't touch it */
  	void		*driver_data;	/* Driver data, set and get with
  					   dev_set_drvdata/dev_get_drvdata */
+	struct {
+		const char	*name;
+		spinlock_t	lock;
+	} driver_override;
  	struct mutex		mutex;	/* mutex to synchronize calls to
  					 * its driver.
  					 */
@@ -701,6 +734,54 @@ struct device_link {
 
  #define kobj_to_dev(__kobj)	container_of_const(__kobj, struct device, kobj)
 
+int __device_set_driver_override(struct device *dev, const char *s, size_t len);
+
+/**
+ * device_set_driver_override() - Helper to set or clear driver override.
+ * @dev: Device to change
+ * @s: NUL-terminated string, new driver name to force a match, pass empty
+ *     string to clear it ("" or "\n", where the latter is only for sysfs
+ *     interface).
+ *
+ * Helper to set or clear driver override of a device.
+ *
+ * Returns: 0 on success or a negative error code on failure.
+ */
+static inline int device_set_driver_override(struct device *dev, const char *s)
+{
+	return __device_set_driver_override(dev, s, strlen(s));
+}
+
+/**
+ * device_has_driver_override() - Check if a driver override has been set.
+ * @dev: device to check
+ *
+ * Returns true if a driver override has been set for this device.
+ */
+static inline bool device_has_driver_override(struct device *dev)
+{
+	guard(spinlock)(&dev->driver_override.lock);
+	return !!dev->driver_override.name;
+}
+
+/**
+ * device_match_driver_override() - Match a driver against the device's driver_override.
+ * @dev: device to check
+ * @drv: driver to match against
+ *
+ * Returns > 0 if a driver override is set and matches the given driver, 0 if a
+ * driver override is set but does not match, or < 0 if a driver override is not
+ * set at all.
+ */
+static inline int device_match_driver_override(struct device *dev,
+					       const struct device_driver *drv)
+{
+	guard(spinlock)(&dev->driver_override.lock);
+	if (dev->driver_override.name)
+		return !strcmp(dev->driver_override.name, drv->name);
+	return -1;
+}
+
  /**
   * device_iommu_mapped - Returns true when the device DMA is translated
   *			 by an IOMMU
author maintainer
#11 lore raw

Re: [PATCH 1/3] driver core: generalize driver_override in struct device

Quoted 11 lines · Armin Wolf

Nice patch series, centralizing the handling of driver_override in the device core removes a lot of duplicated code.

I wonder if we can also move DEVICE_ATTR_DRIVER_OVERRIDE() into the device core by registering the associated sysfs attribute inside bus_add_device().

Bus types that honor driver_override could for example set a boolean flag inside struct bus_type to tell the device core that driver_override is supported.

However this is just a suggestion, the current patch series seems fine to me.

That's a very good suggestion, thanks for pointing it out. In fact, I already thought of this at some point, but then forgot to implement it. :)

author maintainer
#12 lore raw

PATCH 2/3 hwmon: axi-fan: don't use driver_override as IRQ name

In reply to Danilo Krummrich

Do not use driver_override as IRQ name, as it is not guaranteed to point to a valid string; use dev_name() instead.

Fixes 8412b410fa5e ("hwmon: Support ADI Fan Control IP")
---
 drivers/hwmon/axi-fan-control.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hwmon/axi-fan-control.c +1 -1
index b7bb325c3ad9..ec4bbb104449 100644
@@ -507,7 +507,7 @@ static int axi_fan_control_probe(struct platform_device *pdev)
 	ret = devm_request_threaded_irq(&pdev->dev, ctl->irq, NULL,
 					axi_fan_control_irq_handler,
 					IRQF_ONESHOT | IRQF_TRIGGER_HIGH,
-					pdev->driver_override, ctl);
+					dev_name(&pdev->dev), ctl);
 	if (ret)
 		return dev_err_probe(&pdev->dev, ret,
 				     "failed to request an irq\n");
Signature · 1 line
2.53.0
maintainer
#13 lore raw

Re: [PATCH 2/3] hwmon: axi-fan: don't use driver_override as IRQ name

Quoted 14 lines · Danilo Krummrich

Do not use driver_override as IRQ name, as it is not guaranteed to point to a valid string; use dev_name() instead.

Fixes 8412b410fa5e ("hwmon: Support ADI Fan Control IP")
---
  drivers/hwmon/axi-fan-control.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hwmon/axi-fan-control.c +0 -0
index b7bb325c3ad9..ec4bbb104449 100644

drivers/hwmon/axi-fan-control.c static int axi_fan_control_probe(struct platform_device *pdev) · line 510

Quoted 5 lines
drivers/hwmon/axi-fan-control.c +1 -1
@@ -507,7 +507,7 @@ static int axi_fan_control_probe(struct platform_device *pdev)
  	ret = devm_request_threaded_irq(&pdev->dev, ctl->irq, NULL,
  					axi_fan_control_irq_handler,
  					IRQF_ONESHOT | IRQF_TRIGGER_HIGH,
-					pdev->driver_override, ctl);
+					dev_name(&pdev->dev), ctl);

The devm_request_threaded_irq() API documentation says:

@devname: An ascii name for the claiming device, dev_name(dev) if NULL

So NULL should be sufficient.

Nuno, was there a special reason to use driver_override ?

Thanks,
Guenter

author maintainer
#14 lore raw

Re: [PATCH 2/3] hwmon: axi-fan: don't use driver_override as IRQ name

Quoted 23 lines · Guenter Roeck
Quoted 14 lines · Danilo Krummrich

Do not use driver_override as IRQ name, as it is not guaranteed to point to a valid string; use dev_name() instead.

Fixes 8412b410fa5e ("hwmon: Support ADI Fan Control IP")
---
  drivers/hwmon/axi-fan-control.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hwmon/axi-fan-control.c +0 -0
index b7bb325c3ad9..ec4bbb104449 100644

drivers/hwmon/axi-fan-control.c static int axi_fan_control_probe(struct platform_device *pdev) · line 510

Quoted 5 lines
drivers/hwmon/axi-fan-control.c +1 -1
@@ -507,7 +507,7 @@ static int axi_fan_control_probe(struct platform_device *pdev)
  	ret = devm_request_threaded_irq(&pdev->dev, ctl->irq, NULL,
  					axi_fan_control_irq_handler,
  					IRQF_ONESHOT | IRQF_TRIGGER_HIGH,
-					pdev->driver_override, ctl);
+					dev_name(&pdev->dev), ctl);

The devm_request_threaded_irq() API documentation says:

Quoted 3 lines

@devname: An ascii name for the claiming device, dev_name(dev) if NULL

So NULL should be sufficient.

I usually prefer to be explicit, but I can change it to NULL.

Thanks,
Danilo

#15 lore raw

Re: [PATCH 2/3] hwmon: axi-fan: don't use driver_override as IRQ name

In reply to Guenter Roeck
Quoted 23 lines · Guenter Roeck
Quoted 14 lines · Danilo Krummrich

Do not use driver_override as IRQ name, as it is not guaranteed to point to a valid string; use dev_name() instead.

Fixes 8412b410fa5e ("hwmon: Support ADI Fan Control IP")

---
  drivers/hwmon/axi-fan-control.c | 2 +-

  1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hwmon/axi-fan-control.c +0 -0
index b7bb325c3ad9..ec4bbb104449 100644

drivers/hwmon/axi-fan-control.c static int axi_fan_control_probe(struct platform_device *pdev) · line 510

Quoted 5 lines
drivers/hwmon/axi-fan-control.c +1 -1
@@ -507,7 +507,7 @@ static int axi_fan_control_probe(struct platform_device *pdev)
   	ret = devm_request_threaded_irq(&pdev->dev, ctl->irq, NULL,
   					axi_fan_control_irq_handler,
   					IRQF_ONESHOT | IRQF_TRIGGER_HIGH,
-					pdev->driver_override, ctl);
+					dev_name(&pdev->dev), ctl);

The devm_request_threaded_irq() API documentation says:

Quoted 7 lines

@devname:    An ascii name for the claiming device, dev_name(dev) if NULL

So NULL should be sufficient.

Nuno, was there a special reason to use driver_override ?

Not really. That driver was one of my first patches so most likely it was just some copy pasting. NULL should be fine.

- Nuno Sá

author maintainer
#16 lore raw

PATCH 3/3 driver core: platform: use generic driver_override infrastructure

In reply to Danilo Krummrich

When a driver is probed through __driver_attach(), the bus' match() callback is called without the device lock held, thus accessing the driver_override field without a lock, which can cause a UAF.

Fix this by using the driver-core driver_override infrastructure taking care of proper locking internally.

Note that calling match() from __driver_attach() without the device lock held is intentional. [1]

Fixes 3d713e0e382e ("driver core: platform: add device binding path 'driver_override'")
---
 arch/sh/drivers/platform_early.c |  6 ++++--
 drivers/base/platform.c          | 35 +++++---------------------------
 drivers/bus/simple-pm-bus.c      |  4 ++--
 drivers/clk/imx/clk-scu.c        |  3 +--
 drivers/slimbus/qcom-ngd-ctrl.c  |  6 ++----
 include/linux/platform_device.h  |  5 -----
 sound/soc/samsung/i2s.c          |  6 +++---
 7 files changed, 17 insertions(+), 48 deletions(-)
arch/sh/drivers/platform_early.c +4 -2
index 143747c45206..3cd17bb0be67 100644
@@ -25,10 +25,12 @@ static int platform_match(struct device *dev, struct device_driver *drv)
 {
 	struct platform_device *pdev = to_platform_device(dev);
 	struct platform_driver *pdrv = to_platform_driver(drv);
+	int ret;
 
 	/* When driver_override is set, only bind to the matching driver */
-	if (pdev->driver_override)
-		return !strcmp(pdev->driver_override, drv->name);
+	ret = device_match_driver_override(dev, drv);
+	if (ret >= 0)
+		return ret;
 
 	/* Then try to match against the id table */
 	if (pdrv->id_table)
drivers/base/platform.c +5 -30
index b45d41b018ca..22ae87921a7a 100644
@@ -603,7 +603,6 @@ static void platform_device_release(struct device *dev)
 	kfree(pa->pdev.dev.platform_data);
 	kfree(pa->pdev.mfd_cell);
 	kfree(pa->pdev.resource);
-	kfree(pa->pdev.driver_override);
 	kfree(pa);
 }
 
@@ -1306,33 +1305,7 @@ static ssize_t numa_node_show(struct device *dev,
 }
 static DEVICE_ATTR_RO(numa_node);
 
-static ssize_t driver_override_show(struct device *dev,
-				    struct device_attribute *attr, char *buf)
-{
-	struct platform_device *pdev = to_platform_device(dev);
-	ssize_t len;
-
-	device_lock(dev);
-	len = sysfs_emit(buf, "%s\n", pdev->driver_override);
-	device_unlock(dev);
-
-	return len;
-}
-
-static ssize_t driver_override_store(struct device *dev,
-				     struct device_attribute *attr,
-				     const char *buf, size_t count)
-{
-	struct platform_device *pdev = to_platform_device(dev);
-	int ret;
-
-	ret = driver_set_override(dev, &pdev->driver_override, buf, count);
-	if (ret)
-		return ret;
-
-	return count;
-}
-static DEVICE_ATTR_RW(driver_override);
+DEVICE_ATTR_DRIVER_OVERRIDE();
 
 static struct attribute *platform_dev_attrs[] = {
 	&dev_attr_modalias.attr,
@@ -1377,10 +1350,12 @@ static int platform_match(struct device *dev, const struct device_driver *drv)
 {
 	struct platform_device *pdev = to_platform_device(dev);
 	struct platform_driver *pdrv = to_platform_driver(drv);
+	int ret;
 
 	/* When driver_override is set, only bind to the matching driver */
-	if (pdev->driver_override)
-		return !strcmp(pdev->driver_override, drv->name);
+	ret = device_match_driver_override(dev, drv);
+	if (ret >= 0)
+		return ret;
 
 	/* Attempt an OF style match first */
 	if (of_driver_match_device(dev, drv))
drivers/bus/simple-pm-bus.c +2 -2
index 3f00d953fb9a..c920bd6fbaaf 100644
@@ -36,7 +36,7 @@ static int simple_pm_bus_probe(struct platform_device *pdev)
 	 * that's not listed in simple_pm_bus_of_match. We don't want to do any
 	 * of the simple-pm-bus tasks for these devices, so return early.
 	 */
-	if (pdev->driver_override)
+	if (device_has_driver_override(&pdev->dev))
 		return 0;
 
 	match = of_match_device(dev->driver->of_match_table, dev);
@@ -78,7 +78,7 @@ static void simple_pm_bus_remove(struct platform_device *pdev)
 {
 	const void *data = of_device_get_match_data(&pdev->dev);
 
-	if (pdev->driver_override || data)
+	if (device_has_driver_override(&pdev->dev) || data)
 		return;
 
 	dev_dbg(&pdev->dev, "%s\n", __func__);
drivers/clk/imx/clk-scu.c +1 -2
index a85ec48a798b..9b33df9967ec 100644
@@ -706,8 +706,7 @@ struct clk_hw *imx_clk_scu_alloc_dev(const char *name,
 	if (ret)
 		goto put_device;
 
-	ret = driver_set_override(&pdev->dev, &pdev->driver_override,
-				  "imx-scu-clk", strlen("imx-scu-clk"));
+	ret = device_set_driver_override(&pdev->dev, "imx-scu-clk");
 	if (ret)
 		goto put_device;
 
drivers/slimbus/qcom-ngd-ctrl.c +2 -4
index 9aa7218b4e8d..1ed6be6e85d2 100644
@@ -1535,10 +1535,8 @@ static int of_qcom_slim_ngd_register(struct device *parent,
 		ngd->id = id;
 		ngd->pdev->dev.parent = parent;
 
-		ret = driver_set_override(&ngd->pdev->dev,
-					  &ngd->pdev->driver_override,
-					  QCOM_SLIM_NGD_DRV_NAME,
-					  strlen(QCOM_SLIM_NGD_DRV_NAME));
+		ret = device_set_driver_override(&ngd->pdev->dev,
+						 QCOM_SLIM_NGD_DRV_NAME);
 		if (ret) {
 			platform_device_put(ngd->pdev);
 			kfree(ngd);
include/linux/platform_device.h +0 -5
index 813da101b5bf..ed1d50d1c3c1 100644
@@ -31,11 +31,6 @@ struct platform_device {
 	struct resource	*resource;
 
 	const struct platform_device_id	*id_entry;
-	/*
-	 * Driver name to force a match.  Do not set directly, because core
-	 * frees it.  Use driver_set_override() to set or clear it.
-	 */
-	const char *driver_override;
 
 	/* MFD cell pointer */
 	struct mfd_cell *mfd_cell;
sound/soc/samsung/i2s.c +3 -3
index e9964f0e010a..140907a41a70 100644
@@ -1360,10 +1360,10 @@ static int i2s_create_secondary_device(struct samsung_i2s_priv *priv)
 	if (!pdev_sec)
 		return -ENOMEM;
 
-	pdev_sec->driver_override = kstrdup("samsung-i2s", GFP_KERNEL);
-	if (!pdev_sec->driver_override) {
+	ret = device_set_driver_override(&pdev_sec->dev, "samsung-i2s");
+	if (ret) {
 		platform_device_put(pdev_sec);
-		return -ENOMEM;
+		return ret;
 	}
 
 	ret = platform_device_add(pdev_sec);
Signature · 1 line
2.53.0
Reviewed-by on v1 3/3
#17 lore raw

Re: [PATCH 3/3] driver core: platform: use generic driver_override infrastructure

Quoted 8 lines · Danilo Krummrich

When a driver is probed through __driver_attach(), the bus' match() callback is called without the device lock held, thus accessing the driver_override field without a lock, which can cause a UAF.

Fix this by using the driver-core driver_override infrastructure taking care of proper locking internally.

Quoted 8 lines

Note that calling match() from __driver_attach() without the device lock held is intentional. [1]

Fixes 3d713e0e382e ("driver core: platform: add device binding path 'driver_override'")

Looks good to me.

One minor note for the commit message: this UAF can also be triggered from the bind_store path, not just __driver_attach(). It would be good to mention this as well.

Quoted 181 lines
---
 arch/sh/drivers/platform_early.c |  6 ++++--
 drivers/base/platform.c          | 35 +++++---------------------------
 drivers/bus/simple-pm-bus.c      |  4 ++--
 drivers/clk/imx/clk-scu.c        |  3 +--
 drivers/slimbus/qcom-ngd-ctrl.c  |  6 ++----
 include/linux/platform_device.h  |  5 -----
 sound/soc/samsung/i2s.c          |  6 +++---
 7 files changed, 17 insertions(+), 48 deletions(-)
arch/sh/drivers/platform_early.c +4 -2
index 143747c45206..3cd17bb0be67 100644
@@ -25,10 +25,12 @@ static int platform_match(struct device *dev, struct device_driver *drv)
 {
        struct platform_device *pdev = to_platform_device(dev);
        struct platform_driver *pdrv = to_platform_driver(drv);
+       int ret;
 
        /* When driver_override is set, only bind to the matching driver */
-       if (pdev->driver_override)
-               return !strcmp(pdev->driver_override, drv->name);
+       ret = device_match_driver_override(dev, drv);
+       if (ret >= 0)
+               return ret;
 
        /* Then try to match against the id table */
        if (pdrv->id_table)
drivers/base/platform.c +5 -30
index b45d41b018ca..22ae87921a7a 100644
@@ -603,7 +603,6 @@ static void platform_device_release(struct device *dev)
        kfree(pa->pdev.dev.platform_data);
        kfree(pa->pdev.mfd_cell);
        kfree(pa->pdev.resource);
-       kfree(pa->pdev.driver_override);
        kfree(pa);
 }
 
@@ -1306,33 +1305,7 @@ static ssize_t numa_node_show(struct device *dev,
 }
 static DEVICE_ATTR_RO(numa_node);
 
-static ssize_t driver_override_show(struct device *dev,
-                                   struct device_attribute *attr, char *buf)
-{
-       struct platform_device *pdev = to_platform_device(dev);
-       ssize_t len;
-
-       device_lock(dev);
-       len = sysfs_emit(buf, "%s\n", pdev->driver_override);
-       device_unlock(dev);
-
-       return len;
-}
-
-static ssize_t driver_override_store(struct device *dev,
-                                    struct device_attribute *attr,
-                                    const char *buf, size_t count)
-{
-       struct platform_device *pdev = to_platform_device(dev);
-       int ret;
-
-       ret = driver_set_override(dev, &pdev->driver_override, buf, count);
-       if (ret)
-               return ret;
-
-       return count;
-}
-static DEVICE_ATTR_RW(driver_override);
+DEVICE_ATTR_DRIVER_OVERRIDE();
 
 static struct attribute *platform_dev_attrs[] = {
        &dev_attr_modalias.attr,
@@ -1377,10 +1350,12 @@ static int platform_match(struct device *dev, const struct device_driver *drv)
 {
        struct platform_device *pdev = to_platform_device(dev);
        struct platform_driver *pdrv = to_platform_driver(drv);
+       int ret;
 
        /* When driver_override is set, only bind to the matching driver */
-       if (pdev->driver_override)
-               return !strcmp(pdev->driver_override, drv->name);
+       ret = device_match_driver_override(dev, drv);
+       if (ret >= 0)
+               return ret;
 
        /* Attempt an OF style match first */
        if (of_driver_match_device(dev, drv))
drivers/bus/simple-pm-bus.c +2 -2
index 3f00d953fb9a..c920bd6fbaaf 100644
@@ -36,7 +36,7 @@ static int simple_pm_bus_probe(struct platform_device *pdev)
         * that's not listed in simple_pm_bus_of_match. We don't want to do any
         * of the simple-pm-bus tasks for these devices, so return early.
         */
-       if (pdev->driver_override)
+       if (device_has_driver_override(&pdev->dev))
                return 0;
 
        match = of_match_device(dev->driver->of_match_table, dev);
@@ -78,7 +78,7 @@ static void simple_pm_bus_remove(struct platform_device *pdev)
 {
        const void *data = of_device_get_match_data(&pdev->dev);
 
-       if (pdev->driver_override || data)
+       if (device_has_driver_override(&pdev->dev) || data)
                return;
 
        dev_dbg(&pdev->dev, "%s\n", __func__);
drivers/clk/imx/clk-scu.c +1 -2
index a85ec48a798b..9b33df9967ec 100644
@@ -706,8 +706,7 @@ struct clk_hw *imx_clk_scu_alloc_dev(const char *name,
        if (ret)
                goto put_device;
 
-       ret = driver_set_override(&pdev->dev, &pdev->driver_override,
-                                 "imx-scu-clk", strlen("imx-scu-clk"));
+       ret = device_set_driver_override(&pdev->dev, "imx-scu-clk");
        if (ret)
                goto put_device;
 
drivers/slimbus/qcom-ngd-ctrl.c +2 -4
index 9aa7218b4e8d..1ed6be6e85d2 100644
@@ -1535,10 +1535,8 @@ static int of_qcom_slim_ngd_register(struct device *parent,
                ngd->id = id;
                ngd->pdev->dev.parent = parent;
 
-               ret = driver_set_override(&ngd->pdev->dev,
-                                         &ngd->pdev->driver_override,
-                                         QCOM_SLIM_NGD_DRV_NAME,
-                                         strlen(QCOM_SLIM_NGD_DRV_NAME));
+               ret = device_set_driver_override(&ngd->pdev->dev,
+                                                QCOM_SLIM_NGD_DRV_NAME);
                if (ret) {
                        platform_device_put(ngd->pdev);
                        kfree(ngd);
include/linux/platform_device.h +0 -5
index 813da101b5bf..ed1d50d1c3c1 100644
@@ -31,11 +31,6 @@ struct platform_device {
        struct resource *resource;
 
        const struct platform_device_id *id_entry;
-       /*
-        * Driver name to force a match.  Do not set directly, because core
-        * frees it.  Use driver_set_override() to set or clear it.
-        */
-       const char *driver_override;
 
        /* MFD cell pointer */
        struct mfd_cell *mfd_cell;
sound/soc/samsung/i2s.c +3 -4
index e9964f0e010a..140907a41a70 100644
@@ -1360,10 +1360,10 @@ static int i2s_create_secondary_device(struct samsung_i2s_priv *priv)
        if (!pdev_sec)
                return -ENOMEM;
 
-       pdev_sec->driver_override = kstrdup("samsung-i2s", GFP_KERNEL);
-       if (!pdev_sec->driver_override) {
+       ret = device_set_driver_override(&pdev_sec->dev, "samsung-i2s");
+       if (ret) {
                platform_device_put(pdev_sec);
-               return -ENOMEM;
+               return ret;
        }
 
        ret = platform_device_add(pdev_sec);
--

2.53.0

#18 lore raw
In reply to Danilo Krummrich
Quoted 25 lines · Danilo Krummrich

Currently, there are 12 busses (including platform and PCI) that duplicate the driver_override logic for their individual devices.

All of them seem to be prone to the bug described in [1].

While this could be solved for every bus individually using a separate lock, solving this in the driver-core generically results in less (and cleaner) changes overall.

Thus, move driver_override to struct device, provide corresponding accessors for busses and handle locking with a separate lock internally.

In particular, add device_set_driver_override(), device_has_driver_override(), device_match_driver_override() and a helper, DEVICE_ATTR_DRIVER_OVERRIDE(), to declare the corresponding sysfs store() and show() callbacks.

Until all busses have migrated, keep driver_set_override() in place.

Note that we can't use the device lock for the reasons described in [2].

This patch series includes the migration of the platform bus; patches for all other affected busses still need to be extracted as a follow-up of the WIP treewide patch in [3].

Quoted 8 lines

[1] https://bugzilla.kernel.org/show_bug.cgi?id=220789
[2] https://lore.kernel.org/driver-core/[email protected]/
[3] https://git.kernel.org/pub/scm/linux/kernel/git/dakr/linux.git/log/?h=driver_override

Danilo Krummrich (3):
driver core: generalize driver_override in struct device
hwmon: axi-fan: don't use driver_override as IRQ name driver core: platform: use generic driver_override infrastructure

Hi Danilo,

It looks like some usages of platform_device->driver_override were missed. I found them here:
- drivers/bus/simple-pm-bus.c
- drivers/clk/imx/clk-scu.c
- drivers/slimbus/qcom-ngd-ctrl.c
- sound/soc/samsung/i2s.c

The good news is these can be easily updated to use the new APIs. This is required to avoid breaking the build, since the field is removed from struct platform_device. The previous build likely passed because these weren't enabled. I will use allyesconfig for testing going forward.

I scanned for similar cases and most fit the new APIs perfectly. One exception is drivers/xen/xen-pciback/pci_stub.c. It does strcmp(dev->driver_override, PCISTUB_DRIVER_NAME) instead of using drv->name. We might want to change device_match_driver_override() to take a const char * instead to handle this.

Besides axi-fan, I didn't find any other drivers that need to read driver_override. This is great, as it means we hopefully won't need to expose a read API at all.

Thanks.

Quoted 19 lines
 arch/sh/drivers/platform_early.c |  6 ++-
 drivers/base/core.c              |  2 +
 drivers/base/dd.c                | 60 +++++++++++++++++++++++
 drivers/base/platform.c          | 35 ++------------
 drivers/bus/simple-pm-bus.c      |  4 +-
 drivers/clk/imx/clk-scu.c        |  3 +-
 drivers/hwmon/axi-fan-control.c  |  2 +-
 drivers/slimbus/qcom-ngd-ctrl.c  |  6 +--
 include/linux/device.h           | 81 ++++++++++++++++++++++++++++++++
 include/linux/platform_device.h  |  5 --
 sound/soc/samsung/i2s.c          |  6 +--
 11 files changed, 161 insertions(+), 49 deletions(-)

base-commit: 78437ab3b769f80526416570f60173c89858dd84 --
2.53.0

author maintainer
#19 lore raw
Quoted 5 lines · Gui-Dong Han
Quoted 4 lines

Danilo Krummrich (3):
driver core: generalize driver_override in struct device
hwmon: axi-fan: don't use driver_override as IRQ name driver core: platform: use generic driver_override infrastructure

Quoted 8 lines

Hi Danilo,

It looks like some usages of platform_device->driver_override were missed. I found them here:
- drivers/bus/simple-pm-bus.c
- drivers/clk/imx/clk-scu.c
- drivers/slimbus/qcom-ngd-ctrl.c
- sound/soc/samsung/i2s.c

They should all be covered by patch 3, no?

Quoted 11 lines

The good news is these can be easily updated to use the new APIs. This is required to avoid breaking the build, since the field is removed from struct platform_device. The previous build likely passed because these weren't enabled. I will use allyesconfig for testing going forward.

I scanned for similar cases and most fit the new APIs perfectly. One exception is drivers/xen/xen-pciback/pci_stub.c. It does strcmp(dev->driver_override, PCISTUB_DRIVER_NAME) instead of using drv->name. We might want to change device_match_driver_override() to take a const char * instead to handle this.

xen_pcibk_pci_driver should use the exact same define, so we can just convert this to:

drivers/xen/xen-pciback/pci_stub.c +3 -3
index e4b27aecbf0591..301207b4a30dac 100644
@@ -609,9 +609,9 @@ static int pcistub_probe(struct pci_dev *dev, const struct pci_device_id *id)
 
 	match = pcistub_match(dev);
 
-	if ((dev->driver_override &&
-	     !strcmp(dev->driver_override, PCISTUB_DRIVER_NAME)) ||
-	    match) {
+	if (device_match_driver_override(&dev->dev,
+					 &xen_pcibk_pci_driver.driver) > 0 ||
+					 match) {
 
 		if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL
 		    && dev->hdr_type != PCI_HEADER_TYPE_BRIDGE) {

I.e. no separate API needed.

Quoted 3 lines

Besides axi-fan, I didn't find any other drivers that need to read driver_override. This is great, as it means we hopefully won't need to expose a read API at all.

Great, thanks for checking.

#20 lore raw
Quoted 9 lines · Danilo Krummrich
Quoted 7 lines · Gui-Dong Han
Quoted 4 lines

Danilo Krummrich (3):
driver core: generalize driver_override in struct device
hwmon: axi-fan: don't use driver_override as IRQ name driver core: platform: use generic driver_override infrastructure

Hi Danilo,

Quoted 8 lines
Quoted 6 lines

It looks like some usages of platform_device->driver_override were missed. I found them here:
- drivers/bus/simple-pm-bus.c
- drivers/clk/imx/clk-scu.c
- drivers/slimbus/qcom-ngd-ctrl.c
- sound/soc/samsung/i2s.c

They should all be covered by patch 3, no?

My apologies, I got a bit confused here.

Quoted 34 lines
Quoted 7 lines

The good news is these can be easily updated to use the new APIs. This is required to avoid breaking the build, since the field is removed from struct platform_device. The previous build likely passed because these weren't enabled. I will use allyesconfig for testing going forward.

Quoted 5 lines

I scanned for similar cases and most fit the new APIs perfectly. One exception is drivers/xen/xen-pciback/pci_stub.c. It does strcmp(dev->driver_override, PCISTUB_DRIVER_NAME) instead of using drv->name. We might want to change device_match_driver_override() to take a const char * instead to handle this.

xen_pcibk_pci_driver should use the exact same define, so we can just convert this to:

drivers/xen/xen-pciback/pci_stub.c +3 -3
index e4b27aecbf0591..301207b4a30dac 100644
@@ -609,9 +609,9 @@ static int pcistub_probe(struct pci_dev *dev, const struct pci_device_id *id)
 
        match = pcistub_match(dev);
 
-       if ((dev->driver_override &&
-            !strcmp(dev->driver_override, PCISTUB_DRIVER_NAME)) ||
-           match) {
+       if (device_match_driver_override(&dev->dev,
+                                        &xen_pcibk_pci_driver.driver) > 0 ||
+                                        match) {
 
                if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL
                    && dev->hdr_type != PCI_HEADER_TYPE_BRIDGE) {
Quoted 1 line

I.e. no separate API needed.

Thanks for the explanation, you are absolutely right.

Quoted 6 lines
Quoted 4 lines

Besides axi-fan, I didn't find any other drivers that need to read driver_override. This is great, as it means we hopefully won't need to expose a read API at all.

Great, thanks for checking.