----- On Jul 17, 2015, at 8:48 AM, Nikolay Borisov n.borisov@xxxxxxxxxxxxxx wrote: > On 07/16/2015 11:00 PM, Mathieu Desnoyers wrote: >> Expose a new system call allowing threads to register a userspace memory >> area where to store the current CPU number. Scheduler migration sets the >> TIF_NOTIFY_RESUME flag on the current thread. Upon return to user-space, >> a notify-resume handler updates the current CPU value within that >> user-space memory area. >> >> This getcpu cache is an alternative to the sched_getcpu() vdso which has >> a few benefits: >> - It is faster to do a memory read that to call a vDSO, >> - This cache value can be read from within an inline assembly, which >> makes it a useful building block for restartable sequences. >> >> This approach is inspired by Paul Turner and Andrew Hunter's work >> on percpu atomics, which lets the kernel handle restart of critical >> sections: >> Ref.: >> * https://lkml.org/lkml/2015/6/24/665 >> * https://lwn.net/Articles/650333/ >> * >> http://www.linuxplumbersconf.org/2013/ocw/system/presentations/1695/original/LPC%20-%20PerCpu%20Atomics.pdf >> >> Benchmarking sched_getcpu() vs tls cache approach. Getting the >> current CPU number: >> >> - With Linux vdso: 12.7 ns >> - With TLS-cached cpu number: 0.3 ns >> >> The system call can be extended by registering a larger structure in >> the future. >> [...] >> +/* >> + * sys_thread_local_abi - setup thread-local ABI for caller thread >> + */ >> +SYSCALL_DEFINE3(thread_local_abi, struct thread_local_abi __user *, tlap, >> + size_t, len, int, flags) >> +{ >> + size_t minlen; >> + >> + if (flags) >> + return -EINVAL; >> + if (current->thread_local_abi && tlap) >> + return -EBUSY; >> + /* Agree on the intersection of userspace and kernel features */ >> + minlen = min_t(size_t, len, sizeof(struct thread_local_abi)); >> + current->thread_local_abi_len = minlen; >> + current->thread_local_abi = tlap; >> + if (!tlap) >> + return 0; >> + /* >> + * Migration checks ->thread_local_abi to see if notify_resume >> + * flag should be set. Therefore, we need to ensure that >> + * the scheduler sees ->thread_local_abi before we update its content. >> + */ >> + barrier(); /* Store thread_local_abi before update content */ >> + if (getcpu_cache_active(current)) { > > Just checking whether my understanding of the code is correct, but this > 'if' is necessary in case we have been moved to a different CPU after > the store of the thread_local_abi? No, this is not correct. Currently, only the getcpu_cache feature is implemented, but if struct thread_local_abi eventually grows with more fields, userspace could call the kernel with a "len" argument that does not cover some of the features. Therefore, the generic way to check whether getcpu_cache is implemented by the current thread is to call "getcpu_cache_active()". If it is enabled, then we need to update the getcpu_cache content for the current thread. The barrier() above is required because we want to store thread_local_abi (and thread_local_abi_len) before we get the current CPU number and store it into the getcpu_cache, because we could be migrated by the scheduler with CONFIG_PREEMPT=y at any point between the moment we read the current CPU number within getcpu_cache_update() and resume userspace. Having thread_local_abi and thread_local_abi_len set before fetching the current CPU number ensures that the scheduler will succeed its own getcpu_cache_active() check, and will therefore raise the resume notifier flag upon migration, which will then fix the CPU number before resuming to userspace. Thanks, Mathieu > >> + if (getcpu_cache_update(current)) >> + return -EFAULT; >> + } >> + return minlen; >> +} -- Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com -- To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html