After acpi_init_device_object(), the lifetime of struct acpi_device is
managed by the driver core through reference counting.
Both acpi_add_power_resource() and acpi_add_single_object() call
acpi_init_device_object() and then invoke acpi_device_add(). If that
fails, their error paths call the release callback directly instead of
dropping the device reference through acpi_dev_put().
This bypasses the normal device lifetime rules and frees the object
without releasing the reference acquired by device_initialize(), which
may lead to a refcount leak.
The issue was identified by a static analysis tool I developed and
confirmed by manual review.
Fix both error paths by using acpi_dev_put() and let the release
callback handle the final cleanup.
Fixes: 781d737c74 ("ACPI: Drop power resources driver")
Fixes: 718fb0de8f ("ACPI: fix NULL bug for HID/UID string")
Cc: All applicable <stable@vger.kernel.org>
Signed-off-by: Guangshuo Li <lgs201920130244@gmail.com>
Link: https://patch.msgid.link/20260413135343.2884481-1-lgs201920130244@gmail.com
Signed-off-by: Rafael J. Wysocki <rjw@rjwysocki.net>
Several core ACPI device drivers set acpi_device_class() for the given
struct acpi_device to whatever they like, but that value is never used
unless the driver itself uses it and, sadly, they neglect to clear it on
remove. Since the only one of them still using acpi_device_class()
after previous changes is the button driver, update the others to stop
setting it in vain. Also drop the related device class sybmols that
become redundant.
Since the ACPI button driver continues to use acpi_device_class(), make
it clear the struct field represented by acpi_device_class() in its
remove callback.
No intentional functional impact.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://patch.msgid.link/3706295.iIbC2pHGDl@rafael.j.wysocki
ACPI drivers usually set acpi_device_name() for the given struct
acpi_device to whatever they like, but that value is never used unless
the driver itself uses it and, quite unfortunately, drivers neglect to
clear it on remove. Some drivers use it for printing messages or
initializing the names of subordinate devices, but it is better to use
string literals for that, especially if the given one is used just once.
To eliminate unnecessary overhead related to acpi_device_name()
handling, rework multiple core ACPI device drivers to stop setting
acpi_device_name() for struct acpi_device objects manipulated
by them and use a string literal instead of it where applicable.
No intentional functional impact.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Link: https://patch.msgid.link/10840483.nUPlyArG6x@rafael.j.wysocki
This was done entirely with mindless brute force, using
git grep -l '\<k[vmz]*alloc_objs*(.*, GFP_KERNEL)' |
xargs sed -i 's/\(alloc_objs*(.*\), GFP_KERNEL)/\1)/'
to convert the new alloc_obj() users that had a simple GFP_KERNEL
argument to just drop that argument.
Note that due to the extreme simplicity of the scripting, any slightly
more complex cases spread over multiple lines would not be triggered:
they definitely exist, but this covers the vast bulk of the cases, and
the resulting diff is also then easier to check automatically.
For the same reason the 'flex' versions will be done as a separate
conversion.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is the result of running the Coccinelle script from
scripts/coccinelle/api/kmalloc_objs.cocci. The script is designed to
avoid scalar types (which need careful case-by-case checking), and
instead replace kmalloc-family calls that allocate struct or union
object instances:
Single allocations: kmalloc(sizeof(TYPE), ...)
are replaced with: kmalloc_obj(TYPE, ...)
Array allocations: kmalloc_array(COUNT, sizeof(TYPE), ...)
are replaced with: kmalloc_objs(TYPE, COUNT, ...)
Flex array allocations: kmalloc(struct_size(PTR, FAM, COUNT), ...)
are replaced with: kmalloc_flex(*PTR, FAM, COUNT, ...)
(where TYPE may also be *VAR)
The resulting allocations no longer return "void *", instead returning
"TYPE *".
Signed-off-by: Kees Cook <kees@kernel.org>
On the THUNDEROBOT ZERO laptop, the second NVMe slot and the discrete
NVIDIA GPU are both controlled by power-resource PXP. Due to the SSDT table
bug (lack of reference), PXP will be shut dow as an "unused" power resource
during initialization, making the NVMe slot #2 + NVIDIA both inaccessible.
This issue was introduced by commit a1224f34d7 ("ACPI: PM: Check
states of power resources during initialization"). Here are test
results on the three consecutive commits:
(bad again!) a1224f34d7 ACPI: PM: Check states of power resources during initialization
(good) bc28368596 ACPI: PM: Do not turn off power resources in unknown state
(bad) 519d81956e Linux 5.15-rc6
On commit bc28368596 ("ACPI: PM: Do not turn off power resources in
unknown state") this was not an issue because the power resource state
left UNKNOWN thus being ignored.
See also commit 9b04d99788 ("ACPI: PM: Do not turn of unused power
resources on the Toshiba Click Mini") which is another almost identical
case to this one.
Fixes: a1224f34d7 ("ACPI: PM: Check states of power resources during initialization")
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=221087
Signed-off-by: Zhai Can <bczhc0@126.com>
Link: https://patch.msgid.link/20260214161452.2849346-1-bczhc0@126.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This laptop (and possibly similar models too) has power resource called
"GP12.PXP_" for its Intel XMM7360 WWAN modem.
For this power resource to turn ON power for the modem it needs certain
internal flag called "ONEN" to be set:
Method (_ON, 0, NotSerialized) // _ON_: Power On
{
If (^^^LPCB.EC0.ECRG)
{
If ((ONEN == Zero))
{
Return (Zero)
}
(..)
}
}
This flag only gets set from this power resource _OFF method, while
the actual modem power gets turned off during suspend by "GP12.PTS"
method called from the global _PTS (Prepare To Sleep) method.
In fact, this power resource _OFF method implementation just sets the
aforementioned flag:
Method (_OFF, 0, NotSerialized) // _OFF: Power Off
{
OFEN = Zero
ONEN = One
}
Upon hibernation finish, the kernel tries to set this power resource
back ON since its _STA method returns 0 and the resource is still
considered in use as it is declared as required for D0 for both the
modem ACPI device (GP12.PWAN) and its parent PCIe port ACPI device
object (GP12). But the _ON method won't do anything since that "ONEN"
flag is not set.
Overall, this means the modem is dead after hibernation finish until the
laptop is rebooted since the modem power has been cut by _PTS and its
PCI configuration was lost and not able to be restored.
The easiest way to workaround this issue is to call this power resource
_OFF method before calling the _ON method to make sure the "ONEN" flag
gets properly set.
This makes the modem alive once again after hibernation finish - with
properly restored PCI configuration space.
Since this platform does *not* support S3 the fact that
acpi_resume_power_resources() is also called during resume from S3 is
not a problem there.
Do the DMI based quirk matching and quirk flag initialization just
once - in acpi_power_resources_init() function.
This way the whole resume path overhead of this change on other systems
amounts to simple hp_eb_gp12pxp_quirk flag comparison.
Signed-off-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name>
Link: https://patch.msgid.link/c6ff6931c5d27592052f30339de1b9cc298c43f0.1754243159.git.mail@maciej.szmigiero.name
[ rjw: Changelog edits ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This way DMI based quirk matching and quirk flag initialization can be done
just once - in the newly introduced acpi_power_resources_init() function,
which is similar to existing acpi_*_init() functions.
Convert the single already existing DMI match-based quirk in this ACPI
power resource handler ("leave unused power resources on" quirk) to such
one-time initialization in acpi_power_resources_init() function instead of
re-running that DMI match each time acpi_turn_off_unused_power_resources()
gets called.
Signed-off-by: Maciej S. Szmigiero <mail@maciej.szmigiero.name>
Link: https://patch.msgid.link/b173a6987f0b35597fd82400cb28f289786e03d0.1754243159.git.mail@maciej.szmigiero.name
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
While debugging a boot-time issue, it's hard to know what ACPI power
resource the kernel is initializing.
It's very helpful to print the full name path of the power resource
being added so that it is not necessary to guess which one it is,
especially on a system with 2 ore more power resources where the
last name path segment is the same.
Before:
ACPI: PM: Power Resource [RTD3]
ACPI: PM: Power Resource [RTD3]
ACPI: PM: Power Resource [PR00]
ACPI: PM: Power Resource [PR01]
After:
ACPI: \_SB_.PCI0.RP01.RTD3: New power resource
ACPI: \_SB_.PCI0.RP08.RTD3: New power resource
ACPI: \_SB_.PCI0.I2C3.H016.PR00: New power resource
ACPI: \_SB_.PCI0.SPI1.CRFP.PR01: New power resource
Signed-off-by: Kane Chen <kane.chen@intel.com>
[ rjw: Subject and changelog edits ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Instead of having acpi_device_add() defined as a wrapper around
__acpi_device_add(), export acpi_tie_acpi_dev() so it can be called
directly by acpi_add_power_resource(), fold acpi_device_add() into the
latter and rename __acpi_device_add() to acpi_device_add().
No intentional functional impact.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Punit Agrawal <punit.agrawal@bytedance.com>
The initialization of ACPI device objects is split between
acpi_init_device_object() and __acpi_device_add() that initializes
the dev field in struct acpi_device. The "release" function pointer
is passed to __acpi_device_add() for this reason.
However, that split is artificial and all of the initialization can
be carried out by acpi_init_device_object(), so rearrange the code
to that end. In particular, make acpi_init_device_object() take the
"release" pointer as an argument, along with the "type" which is
related to it, instead of __acpi_device_add().
No intentional functional impact.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Punit Agrawal <punit.agrawal@bytedance.com>
Merge ACPI device enumeration updates, ACPI power management updates
and PCI host bridge ACPI driver updates for 5.17-rc1:
- Introduce acpi_fetch_acpi_dev() as a replacement for
acpi_bus_get_device() and use it in the ACPI subsystem (Rafael
Wysocki).
- Avoid using _CID for device enumaration if _HID is missing or
invalid (Rafael Wysocki).
- Rework quirk handling during ACPI device enumeration and add some
new quirks for known broken platforms (Hans de Goede).
- Avoid unnecessary or redundant CPU cache flushing during system
PM transitions (Kirill A. Shutemov).
- Add PM debug messages related to power resources (Rafael Wysocki).
- Fix kernel-doc comment in the PCI host bridge ACPI driver (Yang Li).
* acpi-scan:
serdev: Do not instantiate serdevs on boards with known bogus DSDT entries
i2c: acpi: Do not instantiate I2C-clients on boards with known bogus DSDT entries
ACPI / x86: Add acpi_quirk_skip_[i2c_client|serdev]_enumeration() helpers
ACPI: scan: Create platform device for BCM4752 and LNV4752 ACPI nodes
ACPI: Use acpi_fetch_acpi_dev() instead of acpi_bus_get_device()
ACPI: scan: Introduce acpi_fetch_acpi_dev()
ACPI: scan: Do not add device IDs from _CID if _HID is not valid
* acpi-pm:
ACPI: PM: Remove redundant cache flushing
ACPI: PM: Avoid CPU cache flush when entering S4
* acpi-power:
ACPI: PM: Emit debug messages when enabling/disabling wakeup power
* acpi-pci:
PCI/ACPI: Fix acpi_pci_osc_control_set() kernel-doc comment
Print additional debug messages when enabling and disabling wakeup
power for an ACPI device object to facilitate more fine-grained
debugging of problems in that area.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Fix a device wakeup power reference counting error introduced by
commit a2d7b2e004 ("ACPI: PM: Fix sharing of wakeup power
resources") because of a coding mistake.
Fixes: a2d7b2e004 ("ACPI: PM: Fix sharing of wakeup power resources")
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If acpi_device_sleep_wake() called by acpi_enable_wakeup_device_power()
returns an error which means that the evaluation of either _DWS or
_PSW has failed, turn off all of the device's wakeup power resources
to be consistent with the clearing of dev->wakeup.prepare_count.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If an ACPI wakeup power resource is shared between multiple devices,
it may not be managed correctly.
Suppose, for example, that two devices, A and B, share a wakeup power
resource P whose wakeup_enabled flag is 0 initially. Next, suppose
that wakeup power is enabled for A and B, in this order, and disabled
for B. When wakeup power is enabled for A, P will be turned on and
its wakeup_enabled flag will be set. Next, when wakeup power is
enabled for B, P will not be touched, because its wakeup_enabled flag
is set. Now, when wakeup power is disabled for B, P will be turned
off which is incorrect, because A will still need P in order to signal
wakeup.
Moreover, if wakeup power is enabled for A and then disabled for B,
the latter will cause P to be turned off incorrectly (it will be still
needed by A), because acpi_disable_wakeup_device_power() is allowed
to manipulate power resources when the wakeup.prepare_count counter
of the given device is 0.
While the first issue could be addressed by changing the
wakeup_enabled power resource flag into a counter, addressing the
second one requires modifying acpi_disable_wakeup_device_power() to
do nothing when the target device's wakeup.prepare_count reference
counter is zero and that would cause the new counter to be redundant.
Namely, if acpi_disable_wakeup_device_power() is modified as per the
above, every change of the new counter following a wakeup.prepare_count
change would be reflected by the analogous change of the main reference
counter of the given power resource.
Accordingly, modify acpi_disable_wakeup_device_power() to do nothing
when the target device's wakeup.prepare_count reference counter is
zero and drop the power resource wakeup_enabled flag altogether.
While at it, ensure that all of the power resources that can be
turned off will be turned off when disabling device wakeup due to
a power resource manipulation error, to prevent energy from being
wasted.
Fixes: b5d667eb39 ("ACPI / PM: Take unusual configurations of power resources into account")
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If an ACPI power resource is found to be "on" during the
initialization of the list of wakeup power resources of a device,
it is reference counted and its wakeup_enabled flag is set, which is
problematic if the deivce in question is the only user of the given
power resource, it is never runtime-suspended and it is not allowed
to wake up the system from sleep, because in that case the given
power resource will stay "on" until the system reboots and energy
will be wasted.
It is better to simply turn off wakeup power resources that are "on"
during the initialization unless their reference counters are not
zero, because that may be the only opportunity to prevent them from
staying in the "on" state all the time.
Fixes: b5d667eb39 ("ACPI / PM: Take unusual configurations of power resources into account")
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
To avoid situations in which the actual states of certain ACPI power
resources are not known just because they have never been referenced
by any device configuration objects, check the initial states of all
power resources as soon as they are found in the ACPI namespace (and
fall back to turning them on if the state check fails).
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Tested-by: Andreas K. Huettel <andreas.huettel@ur.de>
Drop the name field (that only is used in diagnostic messages) from
struct acpi_power_resource and use the name of the power resource
device object instead of it.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Use acpi_handle_debug() to print diagnostic messages regarding ACPI
power resources so as to make it easier to correlate the kernel
messages with the power resource objects in the ACPI namespace that
they are about.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
