In the export_store function, the error handling paths followed a
successful vinput_alloc_vdevice call are missing a corresponding
input_free_device call. Since vinput_alloc_vdevice internally calls
input_allocate_device, and input_register_device has not been called
yet, input_free_device should be used to properly free the allocated
input_device struct in this scenario[1].
[1] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/input/input.c#n2094
In the export_store function, when vinput_register_vdevice fails, the
current error handling process calls device_unregister followed by
vinput_destroy_vdevice. However, the device_unregister here triggers its
release function, which already includes a call to
vinput_destroy_vdevice. As a result, vinput_destroy_vdevice is called
twice. Since vinput_destroy_vdevice contains module_put, this double
call must be avoided to prevent potential reference count issues.
After class_register failure, the wrong function class_unregister was
used. The correct cleanup function should be unregister_chrdev, as
register_chrdev was used before class_register.
The function pointer type of myvariable_store was incompatible with the
store member of struct kobj_attribute, leading to a compilation error
when the type conversion (void *) was omitted. This patch corrects the
function declaration and eliminates the need for type conversion.
As myvariable is declared as an integer, it is meaningless that putting
u after %d in the format specifier of sscanf for scanning an integer.
This patch removes the u in the format specifier for avoiding
misleading.
The procfile_write prints the content what user writes into. However,
when the content size is greater than or equal to PROCFS_MAX_SIZE,
procfile_write will print nothing, because the index for appending the
tail NULL character will be modulo to 0, which is an off-by-one error.
This fixes the problem by changing the upper bound of procfs_buffer_size
to (PROCFS_MAX_SIZE - 1), leaving one byte for NULL character. After
the change, we can discard the modulo because the range of
procfs_buffer_size is already between 0 and (PROCFS_MAX_SIZE - 1).
There is a subtle bug that if the atomic flag changes
between the time it was checked and the second time
it was checked, sleep.c would potentially block a
process that had specified O_NONBLOCK. This fixes
the bug by using atomic_cmpxchg instead of atomic_read.
For x86 architecture, the system call table cannot be used to invoke
a system call after commit 1e3ad78 since v6.9. This commit has been
backported to long term stable kernels, like v5.15.154+, v6.1.85+,
v6.6.26+ and v6.8.5+[1]. In this case, thanks to Kprobes, a hook can be
used instead on the system call entry to intercept the system call.
[1] https://stackoverflow.com/a/78607015
Co-authored-by: Hao Dong <hao.dong.work@outlook.com>
When raw_copy_to_user() failed in vinput_read(), the function would set
'count' to -EFAULT and then subtract EFAULT from '*offset'. However,
modifying '*offset' on raw_copy_to_user() failure was incorrect. Fix
this behavior by changing count = -EFAULT to return -EFAULT.
When kzalloc fails to allocate memory and returns NULL, it leads to a
NULL pointer dereference error later on. Add a check for the return
value of kzalloc. When kzalloc fails to allocate memory, it prints an
error message and returns ERR_PTR(-ENOMEM).
The last parameter of request_threaded_irq must be a globally unique
cookie[1]. Usually this would be the device struct received by probe().
Since we are not using driver model, pass the gpio structs instead.
[1] https://docs.kernel.org/core-api/genericirq.html
The past content in the crypto section lacks informative descriptions,
and there should be a proper procedure to demonstrate how Linux
cryptography works. Due to poor maintenance, let's drop the section.
The code 'case -EINPROGRESS || -EBUSY: ' is the same as
'case -115 || -16 :' at compiler time, as both error code are
implemented with macro like '#define EBUSY 16'.
The code above is essentially the same as 'case 1:'. In C, there is no
real boolean value. Boolean-like value will be converted to 1 or 0.
It does not matter too much if the '-EINPROGRESS || -EBUSY' is
calculated at build time or at runtime. In both case, it will compare
the 'rc' with 1 in the switch expression. It will not compare the
'rc' with any real error code number. When the code is really '-EBUSY',
the execution will fallback to the default branch.
And in practice, most of the compilers will do this simple compile-time
static calculation, and generate code like
static int test_skcipher_result(struct skcipher_def *sk, int rc)
{
switch (rc) {
case 0:
break;
case 1:
rc = wait_for_completion_interruptible(&sk->result.completion);
/* code removed for conciseness */
break;
default:
pr_info("skcipher encrypt returned with %d result %d\n", rc,
sk->result.err);
break;
}
init_completion(&sk->result.completion);
return rc;
}
sys_call_table is already declared in arch/x86/include/asm/syscall.h but of
cource not exported by the kernel.
before this commit, gcc complains as follows:
/usr/src/linux-headers-6.1.0-16-common/arch/x86/include/asm/syscall.h:21:29:
note: previous declaration of 'sys_call_table' with type 'long int (*
const[])(const struct pt_regs *)'
21 | extern const sys_call_ptr_t sys_call_table[];
In Debian, the name syscall conflicts with this patch:
in debian kernel source tree:
debian/patches/features/x86/x86-make-x32-syscall-support-conditional.patch
mailing list url:
https://lore.kernel.org/lkml/1415245982.3398.53.camel@decadent.org.uk/T/#u
which introduces a parameter named syscall.x32. So change our name.
schedule_work adds work to global workqueue. In this example, we create a local workqueue. Use the local workqueue by calling queue_work(), instead of putting work on the global workqueue.
dmesg only flushes when it encounter a newline. Without a newline, the line
is held in memory pending another printk. In this particular example
(example_atomic.c), the last pr_info in atomic_bitwise() prints when
another printk happens (either by another module, or __exit for this
module.
This can be confusing to new learner. This patch adds a newline to the last
pr_info forcing dmesg to print to the screen when the module is loaded.
Two struct completion(s) are encapsulated within another 'struct machine'.
Simplify the code by removing the outer struct and let the struct
completion(s) be self-standing.
Update description in tex to match code.
From v6.4, class_create() does not have module *
See commit 11ba11 :
> driver core: class: remove module * from class_create()
>
> The module pointer in class_create() never actually did anything,
> and it shouldn't have been requred to be set as a parameter even
> if it did something. So just remove it and fix up all callers of the
> function in the kernel tree at the same time.
This patch introduces static key technique and adds a kernel module
to demostrate how to use static key to optimize an almost unlikely
branch, the module can interact with userspace thru a character device.
Signed-off-by: Peter Lin <peterlin@qilai.dev>
Since v5.17-rc1, particularly after the commit cead1855266 ("exit:
Rename complete_and_exit to kthread_complete_and_exit"),
complete_and_exit() is renamed to kthread_complete_and_exit().
Close#188
The rule of thumb is to include the headers we are the direct user of.
In particular, if we need an atomic API, we include <linux/atomic.h>.
On the other hand we should not use headers for no reason. In particular,
if we are not doing any IRQ job, why is the <linux/irq.h> included?
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
