[PATCH 2/3] nptl: Do not always assume set_robust_list availability (BZ 33225)

[Adhemerval Zanella]

> Em 29 de jul. de 2025, à(s) 19:16, Daniel Henrique Barboza <dbarboza@ventanamicro.com> escreveu:
> 
> 
> 
>> On 7/29/25 5:32 PM, Florian Weimer wrote:
>> * Adhemerval Zanella:
>>> The __ASSUME_SET_ROBUST_LIST is used to define and set
>>> __nptl_set_robust_list_avail to advertise process shared robust mutex
>>> support, done at __tls_init_tp (during pthread initialization).
>>> 
>>> Some specific kernel configurations and the qemu-user (for all ABIs) do
>>> not implement the set_robust_list syscal.  Therefore, for the default case
>>> (__ASSUME_SET_ROBUST_LIST being true), the missing support is not
>>> detected, and pthread_mutex_init succeeds where it should fail.  For
>>> instance, the sequence succeeds on qemu-user even when set_robust_list
>>> fails with ENOSYS.
>>> 
>>>   pthread_mutexattr_init(&attr);
>>>   pthread_mutexattr_setpshared(&attr, PTHREAD_PROCESS_SHARED);
>>>   pthread_mutexattr_setrobust(&attr, PTHREAD_MUTEX_ROBUST);
>>> 
>>>   pthread_mutex_init(&mutex, &attr);
>>> 
>>> This patch removes __ASSUME_SET_ROBUST_LIST and always enables
>>> __nptl_set_robust_list_avail detection.
>> I'm not sure this is the right direction.  Is qemu-user compatibility
>> that important to users?
>> Why can't support for set_robust_list be added to qemu-user?  It's
>> probably not easy, but if robust mutex functionality is important to
>> qemu-user users, wouldn't that make more sense?
> 
> Here's the documented reason why QEMU doesn't implement this syscall:
> 
> (QEMU source linux-user/syscall.c)
> 
> #ifdef TARGET_NR_set_robust_list
>    case TARGET_NR_set_robust_list:
>    case TARGET_NR_get_robust_list:
>        /* The ABI for supporting robust futexes has userspace pass
>         * the kernel a pointer to a linked list which is updated by
>         * userspace after the syscall; the list is walked by the kernel
>         * when the thread exits. Since the linked list in QEMU guest
>         * memory isn't a valid linked list for the host and we have
>         * no way to reliably intercept the thread-death event, we can't
>         * support these. Silently return ENOSYS so that guest userspace
>         * falls back to a non-robust futex implementation (which should
>         * be OK except in the corner case of the guest crashing while
>         * holding a mutex that is shared with another process via
>         * shared memory).
>         */
>        return -TARGET_ENOSYS;
> #endif
> 
> Maybe things got easier (this comment is 12+ years old) but the nature of this
> syscall is indeed tricky to deal with.
> 

There a new syscall is discussion, set_roboust_list2 [1], which should handle this shortcoming for all ABIs. But it would require proper support on qemu and it will depent of the underlying kernel version.

In any case I do not see why we can’t proper support detect process-shared mutexes consistent across ABIs that should give better error diagnostics on qemu.

[1] https://lwn.net/Articles/1027135/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://sourceware.org/pipermail/libc-alpha/attachments/20250729/1e17e43a/attachment-0001.htm>