| Message ID | 20191104142745.14722-6-elver@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Add Kernel Concurrency Sanitizer (KCSAN) | expand |
On Tue, 5 Nov 2019 at 12:35, kbuild test robot <lkp@intel.com> wrote: > > Hi Marco, > > I love your patch! Perhaps something to improve: > > [auto build test WARNING on linus/master] > [also build test WARNING on v5.4-rc6] > [cannot apply to next-20191031] > [if your patch is applied to the wrong git tree, please drop us a note to help > improve the system. BTW, we also suggest to use '--base' option to specify the > base tree in git format-patch, please see https://stackoverflow.com/a/37406982] > > url: https://github.com/0day-ci/linux/commits/Marco-Elver/Add-Kernel-Concurrency-Sanitizer-KCSAN/20191105-002542 > base: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git a99d8080aaf358d5d23581244e5da23b35e340b9 > reproduce: > # apt-get install sparse > # sparse version: v0.6.1-6-g57f8611-dirty > make ARCH=x86_64 allmodconfig > make C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__' > > If you fix the issue, kindly add following tag > Reported-by: kbuild test robot <lkp@intel.com> > > > sparse warnings: (new ones prefixed by >>) > > >> include/linux/rcupdate.h:651:9: sparse: sparse: context imbalance in 'thread_group_cputime' - different lock contexts for basic block This is a problem with sparse. Without the patch series this warning is also generated, but sparse seems to attribute it to the right file: kernel/sched/cputime.c:316:17: sparse: warning: context imbalance in 'thread_group_cputime' - different lock contexts for basic block Without the patch series, I observe that sparse also generates 5 warnings that it attributes to include/linux/rcupdate.h ("different lock contexts for basic block") but the actual function is in a different file. In the function thread_group_cputime in kernel/sched/cputime.c, what seems to happen is that a seq-reader critical section is contained within an RCU reader critical section (sparse seems unhappy with this pattern to begin with). The KCSAN patches add annotations to seqlock.h which seems to somehow affect sparse to attribute the problem in thread_group_cputime to rcupdate.h. Note that, the config does not even enable KCSAN and all the annotations are no-ops (empty inline functions). So I do not think that I can change this patch to make sparse happy here, since this problem already existed, only sparse somehow decided to attribute the problem to rcupdate.h instead of cputime.c due to subtle changes in the code. Thanks, -- Marco > vim +/thread_group_cputime +651 include/linux/rcupdate.h > > ^1da177e4c3f41 Linus Torvalds 2005-04-16 603 > ^1da177e4c3f41 Linus Torvalds 2005-04-16 604 /* > ^1da177e4c3f41 Linus Torvalds 2005-04-16 605 * So where is rcu_write_lock()? It does not exist, as there is no > ^1da177e4c3f41 Linus Torvalds 2005-04-16 606 * way for writers to lock out RCU readers. This is a feature, not > ^1da177e4c3f41 Linus Torvalds 2005-04-16 607 * a bug -- this property is what provides RCU's performance benefits. > ^1da177e4c3f41 Linus Torvalds 2005-04-16 608 * Of course, writers must coordinate with each other. The normal > ^1da177e4c3f41 Linus Torvalds 2005-04-16 609 * spinlock primitives work well for this, but any other technique may be > ^1da177e4c3f41 Linus Torvalds 2005-04-16 610 * used as well. RCU does not care how the writers keep out of each > ^1da177e4c3f41 Linus Torvalds 2005-04-16 611 * others' way, as long as they do so. > ^1da177e4c3f41 Linus Torvalds 2005-04-16 612 */ > 3d76c082907e8f Paul E. McKenney 2009-09-28 613 > 3d76c082907e8f Paul E. McKenney 2009-09-28 614 /** > ca5ecddfa8fcbd Paul E. McKenney 2010-04-28 615 * rcu_read_unlock() - marks the end of an RCU read-side critical section. > 3d76c082907e8f Paul E. McKenney 2009-09-28 616 * > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 617 * In most situations, rcu_read_unlock() is immune from deadlock. > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 618 * However, in kernels built with CONFIG_RCU_BOOST, rcu_read_unlock() > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 619 * is responsible for deboosting, which it does via rt_mutex_unlock(). > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 620 * Unfortunately, this function acquires the scheduler's runqueue and > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 621 * priority-inheritance spinlocks. This means that deadlock could result > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 622 * if the caller of rcu_read_unlock() already holds one of these locks or > ec84b27f9b3b56 Anna-Maria Gleixner 2018-05-25 623 * any lock that is ever acquired while holding them. > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 624 * > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 625 * That said, RCU readers are never priority boosted unless they were > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 626 * preempted. Therefore, one way to avoid deadlock is to make sure > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 627 * that preemption never happens within any RCU read-side critical > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 628 * section whose outermost rcu_read_unlock() is called with one of > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 629 * rt_mutex_unlock()'s locks held. Such preemption can be avoided in > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 630 * a number of ways, for example, by invoking preempt_disable() before > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 631 * critical section's outermost rcu_read_lock(). > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 632 * > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 633 * Given that the set of locks acquired by rt_mutex_unlock() might change > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 634 * at any time, a somewhat more future-proofed approach is to make sure > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 635 * that that preemption never happens within any RCU read-side critical > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 636 * section whose outermost rcu_read_unlock() is called with irqs disabled. > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 637 * This approach relies on the fact that rt_mutex_unlock() currently only > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 638 * acquires irq-disabled locks. > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 639 * > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 640 * The second of these two approaches is best in most situations, > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 641 * however, the first approach can also be useful, at least to those > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 642 * developers willing to keep abreast of the set of locks acquired by > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 643 * rt_mutex_unlock(). > f27bc4873fa8b7 Paul E. McKenney 2014-05-04 644 * > 3d76c082907e8f Paul E. McKenney 2009-09-28 645 * See rcu_read_lock() for more information. > 3d76c082907e8f Paul E. McKenney 2009-09-28 646 */ > bc33f24bdca8b6 Paul E. McKenney 2009-08-22 647 static inline void rcu_read_unlock(void) > bc33f24bdca8b6 Paul E. McKenney 2009-08-22 648 { > f78f5b90c4ffa5 Paul E. McKenney 2015-06-18 649 RCU_LOCKDEP_WARN(!rcu_is_watching(), > bde23c6892878e Heiko Carstens 2012-02-01 650 "rcu_read_unlock() used illegally while idle"); > bc33f24bdca8b6 Paul E. McKenney 2009-08-22 @651 __release(RCU); > bc33f24bdca8b6 Paul E. McKenney 2009-08-22 652 __rcu_read_unlock(); > d24209bb689e2c Paul E. McKenney 2015-01-21 653 rcu_lock_release(&rcu_lock_map); /* Keep acq info for rls diags. */ > bc33f24bdca8b6 Paul E. McKenney 2009-08-22 654 } > ^1da177e4c3f41 Linus Torvalds 2005-04-16 655 > > :::::: The code at line 651 was first introduced by commit > :::::: bc33f24bdca8b6e97376e3a182ab69e6cdefa989 rcu: Consolidate sparse and lockdep declarations in include/linux/rcupdate.h > > :::::: TO: Paul E. McKenney <paulmck@linux.vnet.ibm.com> > :::::: CC: Ingo Molnar <mingo@elte.hu> > > --- > 0-DAY kernel test infrastructure Open Source Technology Center > https://lists.01.org/pipermail/kbuild-all Intel Corporation > > -- > You received this message because you are subscribed to the Google Groups "kasan-dev" group. > To unsubscribe from this group and stop receiving emails from it, send an email to kasan-dev+unsubscribe@googlegroups.com. > To view this discussion on the web visit https://groups.google.com/d/msgid/kasan-dev/201911051950.7sv6Mqoe%25lkp%40intel.com.
diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h index bcf4cf26b8c8..61232bc223fd 100644 --- a/include/linux/seqlock.h +++ b/include/linux/seqlock.h @@ -37,8 +37,24 @@ #include <linux/preempt.h> #include <linux/lockdep.h> #include <linux/compiler.h> +#include <linux/kcsan.h> #include <asm/processor.h> +/* + * The seqlock interface does not prescribe a precise sequence of read + * begin/retry/end. For readers, typically there is a call to + * read_seqcount_begin() and read_seqcount_retry(), however, there are more + * esoteric cases which do not follow this pattern. + * + * As a consequence, we take the following best-effort approach for raw usage + * via seqcount_t under KCSAN: upon beginning a seq-reader critical section, + * pessimistically mark then next KCSAN_SEQLOCK_REGION_MAX memory accesses as + * atomics; if there is a matching read_seqcount_retry() call, no following + * memory operations are considered atomic. Usage of seqlocks via seqlock_t + * interface is not affected. + */ +#define KCSAN_SEQLOCK_REGION_MAX 1000 + /* * Version using sequence counter only. * This can be used when code has its own mutex protecting the @@ -115,6 +131,7 @@ static inline unsigned __read_seqcount_begin(const seqcount_t *s) cpu_relax(); goto repeat; } + kcsan_atomic_next(KCSAN_SEQLOCK_REGION_MAX); return ret; } @@ -131,6 +148,7 @@ static inline unsigned raw_read_seqcount(const seqcount_t *s) { unsigned ret = READ_ONCE(s->sequence); smp_rmb(); + kcsan_atomic_next(KCSAN_SEQLOCK_REGION_MAX); return ret; } @@ -183,6 +201,7 @@ static inline unsigned raw_seqcount_begin(const seqcount_t *s) { unsigned ret = READ_ONCE(s->sequence); smp_rmb(); + kcsan_atomic_next(KCSAN_SEQLOCK_REGION_MAX); return ret & ~1; } @@ -202,7 +221,8 @@ static inline unsigned raw_seqcount_begin(const seqcount_t *s) */ static inline int __read_seqcount_retry(const seqcount_t *s, unsigned start) { - return unlikely(s->sequence != start); + kcsan_atomic_next(0); + return unlikely(READ_ONCE(s->sequence) != start); } /** @@ -225,6 +245,7 @@ static inline int read_seqcount_retry(const seqcount_t *s, unsigned start) static inline void raw_write_seqcount_begin(seqcount_t *s) { + kcsan_nestable_atomic_begin(); s->sequence++; smp_wmb(); } @@ -233,6 +254,7 @@ static inline void raw_write_seqcount_end(seqcount_t *s) { smp_wmb(); s->sequence++; + kcsan_nestable_atomic_end(); } /** @@ -271,9 +293,11 @@ static inline void raw_write_seqcount_end(seqcount_t *s) */ static inline void raw_write_seqcount_barrier(seqcount_t *s) { + kcsan_nestable_atomic_begin(); s->sequence++; smp_wmb(); s->sequence++; + kcsan_nestable_atomic_end(); } static inline int raw_read_seqcount_latch(seqcount_t *s) @@ -398,7 +422,9 @@ static inline void write_seqcount_end(seqcount_t *s) static inline void write_seqcount_invalidate(seqcount_t *s) { smp_wmb(); + kcsan_nestable_atomic_begin(); s->sequence+=2; + kcsan_nestable_atomic_end(); } typedef struct { @@ -430,11 +456,21 @@ typedef struct { */ static inline unsigned read_seqbegin(const seqlock_t *sl) { - return read_seqcount_begin(&sl->seqcount); + unsigned ret = read_seqcount_begin(&sl->seqcount); + + kcsan_atomic_next(0); /* non-raw usage, assume closing read_seqretry */ + kcsan_flat_atomic_begin(); + return ret; } static inline unsigned read_seqretry(const seqlock_t *sl, unsigned start) { + /* + * Assume not nested: read_seqretry may be called multiple times when + * completing read critical section. + */ + kcsan_flat_atomic_end(); + return read_seqcount_retry(&sl->seqcount, start); }
Since seqlocks in the Linux kernel do not require the use of marked atomic accesses in critical sections, we teach KCSAN to assume such accesses are atomic. KCSAN currently also pretends that writes to `sequence` are atomic, although currently plain writes are used (their corresponding reads are READ_ONCE). Further, to avoid false positives in the absence of clear ending of a seqlock reader critical section (only when using the raw interface), KCSAN assumes a fixed number of accesses after start of a seqlock critical section are atomic. === Commentary on design around absence of clear begin/end markings === Seqlock usage via seqlock_t follows a predictable usage pattern, where clear critical section begin/end is enforced. With subtle special cases for readers needing to be flat atomic regions, e.g. because usage such as in: - fs/namespace.c:__legitimize_mnt - unbalanced read_seqretry - fs/dcache.c:d_walk - unbalanced need_seqretry But, anything directly accessing seqcount_t seems to be unpredictable. Filtering for usage of read_seqcount_retry not following 'do { .. } while (read_seqcount_retry(..));': $ git grep 'read_seqcount_retry' | grep -Ev 'while \(|seqlock.h|Doc|\* ' => about 1/3 of the total read_seqcount_retry usage. Just looking at fs/namei.c, we conclude that it is non-trivial to prescribe and migrate to an interface that would force clear begin/end seqlock markings for critical sections. As such, we concluded that the best design currently, is to simply ensure that KCSAN works well with the existing code. Signed-off-by: Marco Elver <elver@google.com> --- v3: * Remove comment from raw_seqcount_barrier that should have been in next patch. * Renamed kcsan_{nestable,flat}_atomic_{begin,end} * Elaborate why clear begin/end cannot be enforced easily. --- include/linux/seqlock.h | 40 ++++++++++++++++++++++++++++++++++++++-- 1 file changed, 38 insertions(+), 2 deletions(-)