Re: [tip:timers/core 2/58] kernel/ time/posix-cpu-timers.c:1307:9: sp arse: sparse: incorrect type in ar gument 1 (different address spaces)
Bradley Morgan <[email protected]>
| Newsgroups | dev.linux.lists.oe-kbuild-all,org.kernel.vger.linux-kernel |
|---|---|
| Message-ID | <[email protected]> |
On 12 August 2026 20:42:03 BST, Bradley Morgan <[email protected]> wrote: >On 12 August 2026 20:40:00 BST, kernel test robot <[email protected]> wrote: >>tree: https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git >timers/core >>head: 4fa377c19e111c539a530a8200996b911ceff9ff >>commit: a73d7f98e41a96d6e1bcb0e731ab185d9d67878e [2/58] posix-cpu-timers: >Don't abuse lock_task_sighand() in handle_posix_cpu_timers() >>config: microblaze-randconfig-r122-20260812 >(https://download.01.org/0day-ci/archive/20260813/[email protected]/config) >>compiler: microblaze-linux-gcc (GCC) 11.5.0 >>sparse: v0.6.5-rc1 >>reproduce (this is a W=1 build): >>(https://download.01.org/0day-ci/archive/20260813/[email protected]/reproduce) >> >>If you fix the issue in a separate patch/commit (i.e. not just a new >>version of >>the same patch/commit), kindly add following tags >>| Reported-by: kernel test robot <[email protected]> >>| Closes: >>https://lore.kernel.org/oe-kbuild-all/[email protected]/ >> >>sparse warnings: (new ones prefixed by >>) >>>> kernel/time/posix-cpu-timers.c:1307:9: sparse: sparse: incorrect type >>in argument 1 (different address spaces) @@ expected struct spinlock >>[usertype] *lock @@ got struct spinlock [noderef] __rcu * @@ >> kernel/time/posix-cpu-timers.c:1307:9: sparse: expected struct >spinlock [usertype] *lock >> kernel/time/posix-cpu-timers.c:1307:9: sparse: got struct spinlock >[noderef] __rcu * >> kernel/time/posix-cpu-timers.c:1367:36: sparse: sparse: incorrect type >in argument 1 (different address spaces) @@ expected struct spinlock >[usertype] *lock @@ got struct spinlock [noderef] __rcu * @@ >> kernel/time/posix-cpu-timers.c:1367:36: sparse: expected struct >spinlock [usertype] *lock >> kernel/time/posix-cpu-timers.c:1367:36: sparse: got struct >spinlock [noderef] __rcu * >> kernel/time/posix-cpu-timers.c: note: in included file: >> include/linux/sched/signal.h:749:37: sparse: sparse: incorrect type in >argument 1 (different address spaces) @@ expected struct spinlock >[usertype] *lock @@ got struct spinlock [noderef] __rcu * @@ >> include/linux/sched/signal.h:749:37: sparse: expected struct >spinlock [usertype] *lock >> include/linux/sched/signal.h:749:37: sparse: got struct spinlock >[noderef] __rcu * >> include/linux/sched/signal.h:749:37: sparse: sparse: incorrect type in >argument 1 (different address spaces) @@ expected struct spinlock >[usertype] *lock @@ got struct spinlock [noderef] __rcu * @@ >> include/linux/sched/signal.h:749:37: sparse: expected struct >spinlock [usertype] *lock >> include/linux/sched/signal.h:749:37: sparse: got struct spinlock >[noderef] __rcu * >> include/linux/sched/signal.h:749:37: sparse: sparse: incorrect type in >argument 1 (different address spaces) @@ expected struct spinlock >[usertype] *lock @@ got struct spinlock [noderef] __rcu * @@ >> include/linux/sched/signal.h:749:37: sparse: expected struct >spinlock [usertype] *lock >> include/linux/sched/signal.h:749:37: sparse: got struct spinlock >[noderef] __rcu * >> include/linux/sched/signal.h:749:37: sparse: sparse: incorrect type in >argument 1 (different address spaces) @@ expected struct spinlock >[usertype] *lock @@ got struct spinlock [noderef] __rcu * @@ >> include/linux/sched/signal.h:749:37: sparse: expected struct >spinlock [usertype] *lock >> include/linux/sched/signal.h:749:37: sparse: got struct spinlock >[noderef] __rcu * >> include/linux/sched/signal.h:749:37: sparse: sparse: incorrect type in >argument 1 (different address spaces) @@ expected struct spinlock >[usertype] *lock @@ got struct spinlock [noderef] __rcu * @@ >> include/linux/sched/signal.h:749:37: sparse: expected struct >spinlock [usertype] *lock >> include/linux/sched/signal.h:749:37: sparse: got struct spinlock >[noderef] __rcu * >> >>vim +1307 kernel/time/posix-cpu-timers.c >> >> 1296 >> 1297 static void handle_posix_cpu_timers(struct task_struct *tsk) >> 1298 { >> 1299 struct k_itimer *timer, *next; >> 1300 unsigned long flags, start; >> 1301 LIST_HEAD(firing); >> 1302 >> 1303 /* >> 1304 * tsk is current and ->sighand is stable, see the >> 1305 * tsk->exit_state check in run_posix_cpu_timers() >> 1306 */ >>> 1307 spin_lock_irqsave(&tsk->sighand->siglock, flags); >> 1308 >> 1309 do { >> 1310 /* >> 1311 * On RT locking sighand lock does not disable interrupts, >> 1312 * so this needs to be careful vs. ticks. Store the current >> 1313 * jiffies value. >> 1314 */ >> 1315 start = READ_ONCE(jiffies); >> 1316 barrier(); >> 1317 >> 1318 /* >> 1319 * Here we take off tsk->signal->cpu_timers[N] and >> 1320 * tsk->cpu_timers[N] all the timers that are firing, and >> 1321 * put them on the firing list. >> 1322 */ >> 1323 check_thread_timers(tsk, &firing); >> 1324 >> 1325 check_process_timers(tsk, &firing); >> 1326 >> 1327 /* >> 1328 * The above timer checks have updated the expiry cache and >> 1329 * because nothing can have queued or modified timers after >> 1330 * sighand lock was taken above it is guaranteed to be >> 1331 * consistent. So the next timer interrupt fastpath check >> 1332 * will find valid data. >> 1333 * >> 1334 * If timer expiry runs in the timer interrupt context then >> 1335 * the loop is not relevant as timers will be directly >> 1336 * expired in interrupt context. The stub function below >> 1337 * returns always true which allows the compiler to >> 1338 * optimize the loop out. >> 1339 * >> 1340 * If timer expiry is deferred to task work context then >> 1341 * the following rules apply: >> 1342 * >> 1343 * - On !RT kernels no tick can have happened on this CPU >> 1344 * after sighand lock was acquired because interrupts are >> 1345 * disabled. So reenabling task work before dropping >> 1346 * sighand lock and reenabling interrupts is race free. >> 1347 * >> 1348 * - On RT kernels ticks might have happened but the tick >> 1349 * work ignored posix CPU timer handling because the >> 1350 * CPUTIMERS_WORK_SCHEDULED bit is set. Reenabling work >> 1351 * must be done very carefully including a check whether >> 1352 * ticks have happened since the start of the timer >> 1353 * expiry checks. posix_cpu_timers_enable_work() takes >> 1354 * care of that and eventually lets the expiry checks >> 1355 * run again. >> 1356 */ >> 1357 } while (!posix_cpu_timers_enable_work(tsk, start)); >> 1358 >> 1359 /* >> 1360 * We must release sighand lock before taking any timer's lock. >> 1361 * There is a potential race with timer deletion here, as the >> 1362 * siglock now protects our private firing list. We have set >> 1363 * the firing flag in each timer, so that a deletion attempt >> 1364 * that gets the timer lock before we do will give it up and >> 1365 * spin until we've taken care of that timer below. >> 1366 */ >> 1367 spin_unlock_irqrestore(&tsk->sighand->siglock, flags); >> 1368 >> 1369 /* >> 1370 * Now that all the timers on our list have the firing flag, >> 1371 * no one will touch their list entries but us. We'll take >> 1372 * each timer's lock before clearing its firing flag, so no >> 1373 * timer call will interfere. >> 1374 */ >> 1375 list_for_each_entry_safe(timer, next, &firing, it.cpu.elist) { >> 1376 bool cpu_firing; >> 1377 >> 1378 /* >> 1379 * spin_lock() is sufficient here even independent of the >> 1380 * expiry context. If expiry happens in hard interrupt >> 1381 * context it's obvious. For task work context it's safe >> 1382 * because all other operations on timer::it_lock happen in >> 1383 * task context (syscall or exit). >> 1384 */ >> 1385 spin_lock(&timer->it_lock); >> 1386 list_del_init(&timer->it.cpu.elist); >> 1387 cpu_firing = timer->it.cpu.firing; >> 1388 timer->it.cpu.firing = false; >> 1389 /* >> 1390 * If the firing flag is cleared then this raced with a >> 1391 * timer rearm/delete operation. So don't generate an >> 1392 * event. >> 1393 */ >> 1394 if (likely(cpu_firing)) >> 1395 cpu_timer_fire(timer); >> 1396 /* See posix_cpu_timer_wait_running() */ >> 1397 rcu_assign_pointer(timer->it.cpu.handling, NULL); >> 1398 spin_unlock(&timer->it_lock); >> 1399 } >> 1400 } >> 1401 >> >>-- >>0-DAY CI Kernel Test Service >>https://github.com/intel/lkp-tests/wiki >> > > >Oleg already addressed to something like this, I think he said to nuke >__rcu, which idm looking at and submitting. > >Thanks! Suggested patch, Oleg, what do you think? From 5e22f00b181c79202bef233d9266a18a4b7abaf9 Mon Sep 17 00:00:00 2001 From: Bradley Morgan <[email protected]> Date: Wed, 12 Aug 2026 23:52:12 +0000 Subject: [PATCH] signal: Remove the __rcu annotation from task_struct->sighand sighand is not protected by RCU. It is protected by sighand->siglock, and by tasklist_lock on the exit path, so the __rcu annotation on task_struct->sighand describes neither, and sparse complains about every direct ->siglock use because of it. That is exactly what the kbuild robot is reporting now for posix-cpu-timers. What actually keeps lock_task_sighand() safe is SLAB_TYPESAFE_BY_RCU plus the siglock recheck, not this annotation. Oleg suggested just dropping the annotation, so that is what this does, and the few rcu_* accessors get switched over to the READ_ONCE/WRITE_ONCE/smp_store_release they already expand to. This cannot really be split... with the annotation gone the leftover rcu_* sites are sparse errors, and converting them first while the annotation stays leaves warnings, so there is no file by file split that keeps every commit clean. With CONFIG_PROVE_LOCKING off the generated code is identical, with it on the only difference is the RCU lockdep assertions that came with the removed accessors. Reported-by: kernel test robot <[email protected]> Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/ Suggested-by: Oleg Nesterov <[email protected]> Link: https://lore.kernel.org/r/[email protected] Fixes: 913292c97d75 ("sched.h: Annotate sighand_struct with __rcu") Signed-off-by: Bradley Morgan <[email protected]> --- fs/exec.c | 7 ++++ +- include/linux/sched.h | 2 +- kernel/exit.c | 3 +-- kernel/fork.c | 2 +- kernel/signal.c | 6 +++--- net/sunrpc/svc.c | 2 +- security/selinux/hooks.c | 4 ++-- 7 files changed, 15 insertions(+), 11 deletions(-) diff --git a/fs/exec.c b/fs/exec.c index a14f28b15607..9e3d703329ab 100644 --- a/fs/exec.c +++ b/fs/exec.c @@ -1078,7 +1078,12 @@ static int unshare_sighand(struct task_struct *me) spin_lock(&oldsighand->siglock); memcpy(newsighand->action, oldsighand->action, sizeof(newsighand->action)); - rcu_assign_pointer(me->sighand, newsighand); + /* + * Pairs with the READ_ONCE() in lock_task_sighand(): the + * ->action copy above has to be visible before anyone can + * see the new ->sighand. + */ + smp_store_release(&me->sighand, newsighand); spin_unlock(&oldsighand->siglock); write_unlock_irq(&tasklist_lock); diff --git a/include/linux/sched.h b/include/linux/sched.h index 499943987c1a..ef5bcc8caf74 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -1216,7 +1216,7 @@ struct task_struct { /* Signal handlers: */ struct signal_struct *signal; - struct sighand_struct __rcu *sighand; + struct sighand_struct *sighand; sigset_t blocked; sigset_t real_blocked; /* Restored if set_restore_sigmask() was used: */ diff --git a/kernel/exit.c b/kernel/exit.c index 182c06671c78..45a9354a6a77 100644 --- a/kernel/exit.c +++ b/kernel/exit.c @@ -163,8 +163,7 @@ static void __exit_signal(struct release_task_post *post, struct task_struct *ts struct tty_struct *tty; u64 utime, stime; - sighand = rcu_dereference_check(tsk->sighand, - lockdep_tasklist_lock_is_held()); + sighand = READ_ONCE(tsk->sighand); spin_lock(&sighand->siglock); #ifdef CONFIG_POSIX_TIMERS diff --git a/kernel/fork.c b/kernel/fork.c index 86b6351ed120..202915e85b4e 100644 --- a/kernel/fork.c +++ b/kernel/fork.c @@ -1702,7 +1702,7 @@ static int copy_sighand(u64 clone_flags, struct task_struct *tsk) return 0; } sig = kmem_cache_alloc(sighand_cachep, GFP_KERNEL); - RCU_INIT_POINTER(tsk->sighand, sig); + WRITE_ONCE(tsk->sighand, sig); if (!sig) return -ENOMEM; diff --git a/kernel/signal.c b/kernel/signal.c index a5e15bf09d31..04cfefe86208 100644 --- a/kernel/signal.c +++ b/kernel/signal.c @@ -1354,7 +1354,7 @@ struct sighand_struct *lock_task_sighand(struct task_struct *tsk, rcu_read_lock(); for (;;) { - sighand = rcu_dereference(tsk->sighand); + sighand = READ_ONCE(tsk->sighand); if (unlikely(sighand == NULL)) { /* * Pairs with the smp_store_release() in @@ -1378,7 +1378,7 @@ struct sighand_struct *lock_task_sighand(struct task_struct *tsk, * must see ->sighand == NULL. */ spin_lock_irqsave(&sighand->siglock, *flags); - if (likely(sighand == rcu_access_pointer(tsk->sighand))) + if (likely(sighand == READ_ONCE(tsk->sighand))) break; spin_unlock_irqrestore(&sighand->siglock, *flags); } @@ -1393,7 +1393,7 @@ void lockdep_assert_task_sighand_held(struct task_struct *task) struct sighand_struct *sighand; rcu_read_lock(); - sighand = rcu_dereference(task->sighand); + sighand = READ_ONCE(task->sighand); if (sighand) lockdep_assert_held(&sighand->siglock); else diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c index 05c3e6e2f659..7c681b9f829b 100644 --- a/net/sunrpc/svc.c +++ b/net/sunrpc/svc.c @@ -1276,7 +1276,7 @@ static void svc_unregister(const struct svc_serv *serv, struct net *net) } rcu_read_lock(); - sighand = rcu_dereference(current->sighand); + sighand = READ_ONCE(current->sighand); spin_lock_irqsave(&sighand->siglock, flags); recalc_sigpending(); spin_unlock_irqrestore(&sighand->siglock, flags); diff --git a/security/selinux/hooks.c b/security/selinux/hooks.c index 035aaf113d1d..d6423a05ca6c 100644 --- a/security/selinux/hooks.c +++ b/security/selinux/hooks.c @@ -2586,7 +2586,7 @@ static void selinux_bprm_committed_creds(const struct linux_binprm *bprm) if (rc) { clear_itimer(); - spin_lock_irq(&unrcu_pointer(current->sighand)->siglock); + spin_lock_irq(¤t->sighand->siglock); if (!fatal_signal_pending(current)) { flush_sigqueue(¤t->pending); flush_sigqueue(¤t->signal->shared_pending); @@ -2594,7 +2594,7 @@ static void selinux_bprm_committed_creds(const struct linux_binprm *bprm) sigemptyset(¤t->blocked); recalc_sigpending(); } - spin_unlock_irq(&unrcu_pointer(current->sighand)->siglock); + spin_unlock_irq(¤t->sighand->siglock); } /* Wake up the parent if it is waiting so that it can recheck -- 2.47.3 I'm feeling it's a bit, broad, but sparse still cries if I just remove the annotation from sched.h.. honestly this is odd. Thanks!