Re: [PATCH v6 3/9] s390/crypto: Fix use of mutex in atomic context in PAES

[Harald Freudenberger <freude@linux.ibm.com>]

On 2026-08-21 10:56, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential
> issue(s) to consider:
> 
> Pre-existing issues:
> - [High] The `aes_s390.c` module contains the exact same locking bug
> in atomic context that this patch fixes in `paes_s390.c`. The
> `ctr_aes_crypt()` function uses `mutex_trylock(&ctrblk_lock)` and
> subsequently calls `mutex_unlock(&ctrblk_lock)`. As the patch's commit
> message notes, mutexes are not safe to use in atomic or interrupt
> context. Synchronous skcipher algorithms like `ctr-aes-s390` can be
> invoked directly from `softirq` context (e.g., by the IPsec stack).
> Calling `mutex_unlock()` in `softirq` is a strict violation of kernel
> locking rules.
> --
> 
> --- Patch [3]: [PATCH v6 3/9] s390/crypto: Fix use of mutex in atomic
> context in PAES ---
> commit 425e8f1d3532a176433992d286f917e0fd14ff9e
> Author: Harald Freudenberger <freude@linux.ibm.com>
> 
> s390/crypto: Fix use of mutex in atomic context in PAES
> 
> This commit fixes a locking issue in the PAES CTR implementation where 
> a
> mutex was used in atomic or interrupt context. It replaces the mutex 
> with
> a binary semaphore which is safe to use in such environments.
> 
>> diff --git a/arch/s390/crypto/paes_s390.c 
>> b/arch/s390/crypto/paes_s390.c
>> --- a/arch/s390/crypto/paes_s390.c
>> +++ b/arch/s390/crypto/paes_s390.c
> [ ... ]
>> @@ -45,7 +45,7 @@ module_param_named(clrkey, pkey_clrkey_allowed, 
>> bool, 0444);
>>  MODULE_PARM_DESC(clrkey, "Allow clear key material (default N)");
>> 
>>  static u8 *ctrblk;
>> -static DEFINE_MUTEX(ctrblk_lock);
>> +static DEFINE_SEMAPHORE(ctrblk_sem, 1);
> 
> [Severity: High]
> This isn't a bug introduced by this patch, but 
> arch/s390/crypto/aes_s390.c
> appears to have the exact same bug that this patch is fixing here.
> 
> Looking at ctr_aes_crypt() in aes_s390.c:
> 
> 	locked = mutex_trylock(&ctrblk_lock);
> 
> 	ret = skcipher_walk_virt(&walk, req, false);
> 	while ((nbytes = walk.nbytes) >= AES_BLOCK_SIZE) {
> 		...
> 	}
> 	if (locked)
> 		mutex_unlock(&ctrblk_lock);
> 
> Since synchronous skcipher algorithms like ctr-aes-s390 can be invoked 
> from
> softirq context (such as by the IPsec stack), would calling 
> mutex_unlock()
> here trigger a lockdep splat or crash, especially on PREEMPT_RT where
> mutex_unlock() must be called from process context?
> 
> Does aes_s390.c also need to be converted to use a binary semaphore?

Allready addressed and fixed with the aes_s390 patch series.