Messages in this thread | | | Date | Tue, 17 Jul 2018 09:19:42 +0200 | From | Ingo Molnar <> | Subject | Re: [GIT PULL rcu/next] RCU commits for 4.19 |
| |
* Paul E. McKenney <paulmck@linux.vnet.ibm.com> wrote:
> Hello, Ingo! > > This pull request contains the following changes: > > 1. An optimization and a fix for RCU expedited grace periods, with > the fix being from Boqun Feng. > > http://lkml.kernel.org/r/20180625224308.GA10064@linux.vnet.ibm.com > > 2. Miscellaneous fixes, including a lockdep-annotation fix from > Boqun Feng. > > http://lkml.kernel.org/r/20180625230410.GA13250@linux.vnet.ibm.com > > 3. SRCU updates. > > http://lkml.kernel.org/r/20180625224734.GA10860@linux.vnet.ibm.com > > 4. Updates to rcutorture and associated scripting. > > http://lkml.kernel.org/r/20180625225735.GA11793@linux.vnet.ibm.com > > 5. Introduce grace-period sequence numbers to the RCU-bh, RCU-preempt, > and RCU-sched flavors, replacing the old ->gpnum and ->completed > pair of fields. This change allows lockless code to obtain the > complete grace-period state with a single READ_ONCE(), which is > needed to maintain tolerable lock contention during the upcoming > consolidation of the three RCU flavors. Note that grace-period > sequence numbers are already used by rcu_barrier(), expedited > RCU grace periods, and SRCU, and are thus already heavily used > and well-tested. Joel Fernandes contributed a number of excellent > fixes and improvements. > > http://lkml.kernel.org/r/20180626000841.GA15696@linux.vnet.ibm.com > > 6. Clean up some grace-period-reporting loose ends, including > improving the handling of quiescent states from offline CPUs > and fixing some false-positive WARN_ON_ONCE() invocations. > (Strictly speaking, the WARN_ON_ONCE() invocations were quite > correct, but their invariants were (harmlessly) violated by the > earlier sloppy handling of quiescent states from offline CPUs.) > In addition, improve grace-period forward-progress guarantees so > as to allow removal of fail-safe checks that required otherwise > needless lock acquisitions. Finally, add more diagnostics to > help debug the upcoming consolidation of the RCU-bh, RCU-preempt, > and RCU-sched flavors. > > http://lkml.kernel.org/r/20180626002052.GA24146@linux.vnet.ibm.com > > 7. Additional miscellaneous fixes, including those contributed by > Byungchul Park, Mauro Carvalho Chehab, Joe Perches, Joel Fernandes, > Steven Rostedt, Andrea Parri, and Neil Brown. > > http://lkml.kernel.org/r/20180626003448.GA26209@linux.vnet.ibm.com > > 8. Additional torture-test changes, including several contributed by > Arnd Bergmann and Joel Fernandes. > > http://lkml.kernel.org/r/20180626005205.GA28179@linux.vnet.ibm.com > > The layout of this pull request is unusual due to the large footprint > of the commits in #5 and #6. Therefore #1-#4 are branches on top of > v4.18-rc1, #5 and #6 are stacked on top of each other based off of the > merge point for #1-#4, and #7 and #8 are branches on top of #6. > > Although I merged this series with v4.18-rc4 without conflicts and tested > the results, Stephen Rothwell has reported a couple of conflicts with > this series in the course of his -next integration-testing efforts and > has provided the following resolutions: > > http://lkml.kernel.org/r/20180514134636.61831bcc@canb.auug.org.au > http://lkml.kernel.org/r/20180622122717.5f475829@canb.auug.org.au > > Both of these resolutions look good to me. > > This pull request increases the size of RCU by not quite 500 lines. > Of this, about 200 is due to adding test code to rcutorture in > order to adequately test the consolidated RCU-bh, RCU-preempt, and > RCU-sched flavors. The purpose of adding this test code early is to > test it against RCU-sched, which happens to already support compound RCU > read-side critical sections via rcu_read_lock_sched(), preempt_disable(), > local_bh_disable(), and local_irq_save(). The remainder are primarily > due to the addition of diagnostics and the fixing of quiescent-state and > grace-period reporting. This size increase is temporary: The overall > effect of this set of cleanups, the consolidation of the three RCU > flavors, and additional cleanups will be to -decrease- RCU's line count > by a couple of hundred lines. > > All of these changes have been subjected to 0day Test Robot and -next > testing, and are available in the git repository at: > > git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git for-mingo > > for you to fetch changes up to 18952651dae8efcc6d565c97f8fe5629b399cb3e: > > Merge branches 'fixes1.2018.07.12b' and 'torture1.2018.07.12b' into HEAD (2018-07-12 15:42:41 -0700) > > ---------------------------------------------------------------- > Andrea Parri (1): > doc: Update synchronize_rcu() definition in whatisRCU.txt > > Arnd Bergmann (1): > rcutorture: Use monotonic timestamp for stall detection > > Boqun Feng (2): > rcu: Use the proper lockdep annotation in dump_blkd_tasks() > rcu: Make expedited GPs handle CPU 0 being offline > > Byungchul Park (2): > rcu: Improve rcu_note_voluntary_context_switch() reporting > rcu: Check the range of jiffies_till_{first,next}_fqs when setting them > > Joe Perches (1): > rcu: Use pr_fmt to prefix "rcu: " to logging output > > Joel Fernandes (1): > rcu: Rename the grace-period-request variables and parameters > > Joel Fernandes (Google) (10): > rcu: Produce last "CleanupMore" trace only if late-breaking request > rcu: Fix cpustart tracepoint gp_seq number > rcu: Use better variable names in funnel locking loop > rcu: Identify grace period is in progress as we advance up the tree > rcu: Add comment documenting how rcu_seq_snap works > rcu: Assign higher prio to RCU threads if rcutorture is built-in > rcutorture: Disable RT throttling for boost tests > rcutorture: Make boost test more robust > rcutorture: Add support to detect if boost kthread prio is too low > rcutorture: Fix rcu_barrier successes counter > > Mauro Carvalho Chehab (1): > rcu: rcupdate.h: Get rid of Sphinx warnings at rcu_pointer_handoff() > > NeilBrown (1): > rculist: Improve documentation for list_for_each_entry_from_rcu() > > Paul E. McKenney (98): > rcu: Make expedited grace period use direct call on last leaf > srcu: Fix typos in __call_srcu() header comment > srcu: Document that srcu_funnel_gp_start() implies srcu_funnel_exp_start() > srcu: Add address of first callback to rcutorture output > srcu: Introduce srcu_read_{un,}lock_notrace() > torture: Use a single build directory for torture scenarios > rcutorture: Remove obsolete TREE08-T.boot file > rcutorture: Make kvm-find-errors.sh find close calls > torture: Make online/offline messages appear only for verbose=2 > torture: Keep old-school dmesg format > rcu: Add debugging info to assertion > rcu: Exclude near-simultaneous RCU CPU stall warnings > rcu: Diagnostics for grace-period startup hangs > rcu: Make rcu_gp_cleanup() write only once to ->gp_flags > Merge branches 'expedited.2018.07.12a', 'fixes.2018.07.12a', 'srcu.2018.06.25b' and 'torture.2018.06.25b' into HEAD > rcu: Introduce grace-period sequence numbers > rcu: Move rcu_gp_slow() to ->gp_seq > rcu: Make rcutorture's batches-completed API use ->gp_seq > rcu: Convert rcu_check_gp_kthread_starvation() to GP sequence number > rcu: Make quiescent-state reporting use ->gp_seq > rcu: Convert conditional grace-period primitives to ->gp_seq > rcu: Move RCU's grace-period-change code to ->gp_seq > rcu: Convert rcu_gpnum_ovf() to ->gp_seq > rcu: Move rcu_implicit_dynticks_qs() to ->gp_seq > rcu: Move rcu_try_advance_all_cbs() to ->gp_seq > rcu: Move rcu_nocb_gp_get() to ->gp_seq > rcu: Move rcu_gp_in_progress() to ->gp_seq > rcu: Convert ->rcu_iw_gpnum to ->gp_seq > rcu: Convert ->completedqs to ->gp_seq > rcu: Convert grace-period requests to ->gp_seq > rcu: Make RCU CPU stall warnings use ->gp_seq > rcutorture: Convert rcutorture_get_gp_data() to ->gp_seq > rcu: Move from ->need_future_gp[] to ->gp_seq_needed > rcu: Make rcu_nocb_wait_gp() check if GP already requested > rcu: Convert rcu_grace_period tracepoint to gp_seq > rcu: Convert rcu_future_grace_period tracepoint to gp_seq > rcu: Convert rcu_grace_period_init tracepoint to gp_seq > rcu: Convert rcu_preempt_task tracepoint to ->gp_seq > rcu: Convert rcu_unlock_preempted_task tracepoint to ->gp_seq > rcu: Convert rcu_quiescent_state_report tracepoint to ->gp_seq > rcu: Convert rcu_fqs tracepoint to ->gp_seq > rcu: Remove ->gpnum and ->completed > rcu: Make simple callback acceleration refer to rdp->gp_seq_needed > doc: Update data-structure documentation for ->gp_seq > doc: Update memory-ordering documentation for ->gp-seq > doc: Update RCU CPU stall-warning documentation > rcu: Don't funnel-lock above leaf node if GP in progress > rcu: Make rcu_start_this_gp() check for grace period already started > rcutorture: Correctly handle grace-period sequence wrap > rcu: Regularize resetting of rcu_data wrap indicator > rcu: Clean up handling of tasks blocked across full-rcu_node offline > rcu: Fix an obsolete ->qsmaskinit comment > rcu: Make rcu_init_new_rnp() stop upon already-set bit > rcu: Make rcu_report_unblock_qs_rnp() warn on violated preconditions > rcu: Fix typo and add additional debug > rcu: Replace smp_wmb() with smp_store_release() for stall check > rcu: Prevent useless FQS scan after all CPUs have checked in > rcu: Suppress false-positive offline-CPU lockdep-RCU splat > rcu: Suppress false-positive preempted-task splats > rcu: Suppress more involved false-positive preempted-task splats > rcu: Suppress false-positive splats from mid-init task resume > rcu: Fix grace-period hangs from mid-init task resume > rcu: Fix grace-period hangs due to race with CPU offline > rcu: Add RCU-preempt check for waiting on newly onlined CPU > rcu: Move grace-period pre-init delay after pre-init > rcu: Remove failsafe check for lost quiescent state > rcu: Remove CPU-hotplug failsafe from force-quiescent-state code path > rcu: Add up-tree information to dump_blkd_tasks() diagnostics > rcu: Add CPU online/offline state to dump_blkd_tasks() > rcu: Record ->gp_state for both phases of grace-period initialization > rcu: Add diagnostics for offline CPUs failing to report QS > rcu: Make rcu_read_unlock_special() static > rcu: Improve RCU-tasks naming and comments > rcu: Mark task as .need_qs less aggressively > rcu: Inline rcu_dynticks_momentary_idle() into its sole caller > rcu: Clarify and correct the rcu_preempt_qs() header comment > rcu: Remove unused rcu_kick_nohz_cpu() function > rcu: Remove unused local variable "cpu" > rcu: Remove "inline" from panic_on_rcu_stall() and rcu_blocking_is_gp() > rcu: Remove "inline" from rcu_torture_print_module_parms() > rcu: Remove "inline" from rcu_perf_print_module_parms() > rcu: Remove __maybe_unused from rcu_cpu_has_callbacks() > rcu: Use RCU CPU stall timeout for rcu_check_gp_start_stall() > rcu: Add diagnostics for rcutorture writer stall warning > rcu: Make rcu_seq_diff() more exact > MAINTAINERS: Update RCU, SRCU, and TORTURE-TEST entries > rcu: Print stall-warning NMI dyntick state in hexadecimal > srcu: Add grace-period number to rcutorture statistics printout > rcutorture: Change units of onoff_interval to jiffies > rcu: Remove rcutorture test version and sequence number > rcuperf: Remove unused torturing_tasks() function > rcutorture: Extract common code from rcu_torture_reader() > rcutorture: Use atomic increment for n_rcu_torture_timers > rcutorture: Use per-CPU random state for rcu_torture_timer() > rcutorture: Make rcu_torture_timer() use rcu_torture_one_read() > rcutorture: Handle extended read-side critical sections > rcutorture: Emphasize testing of single reader protection type > Merge branches 'fixes1.2018.07.12b' and 'torture1.2018.07.12b' into HEAD > > Steven Rostedt (VMware) (2): > rcu: Speed up calling of RCU tasks callbacks > rcu: Add comment to the last sleep in the rcu tasks loop > > .../Design/Data-Structures/Data-Structures.html | 118 +-- > .../Memory-Ordering/Tree-RCU-Memory-Ordering.html | 22 +- > .../Design/Memory-Ordering/TreeRCU-gp-cleanup.svg | 123 +-- > .../Design/Memory-Ordering/TreeRCU-gp-init-1.svg | 16 +- > .../Design/Memory-Ordering/TreeRCU-gp-init-3.svg | 56 +- > .../RCU/Design/Memory-Ordering/TreeRCU-gp.svg | 237 ++--- > .../RCU/Design/Memory-Ordering/TreeRCU-qs.svg | 12 +- > Documentation/RCU/stallwarn.txt | 24 +- > Documentation/RCU/whatisRCU.txt | 18 +- > Documentation/admin-guide/kernel-parameters.txt | 4 +- > MAINTAINERS | 9 +- > include/linux/rculist.h | 19 +- > include/linux/rcupdate.h | 20 +- > include/linux/rcutiny.h | 2 +- > include/linux/srcu.h | 17 + > include/linux/torture.h | 4 +- > include/trace/events/rcu.h | 112 ++- > kernel/locking/locktorture.c | 5 +- > kernel/rcu/rcu.h | 104 +- > kernel/rcu/rcuperf.c | 57 +- > kernel/rcu/rcutorture.c | 462 ++++++--- > kernel/rcu/srcutree.c | 39 +- > kernel/rcu/tiny.c | 4 +- > kernel/rcu/tree.c | 1019 +++++++++++--------- > kernel/rcu/tree.h | 71 +- > kernel/rcu/tree_exp.h | 14 +- > kernel/rcu/tree_plugin.h | 176 ++-- > kernel/rcu/update.c | 45 +- > kernel/torture.c | 15 +- > .../testing/selftests/rcutorture/bin/configinit.sh | 26 +- > .../testing/selftests/rcutorture/bin/kvm-build.sh | 11 +- > .../selftests/rcutorture/bin/kvm-recheck-rcu.sh | 1 + > .../selftests/rcutorture/bin/kvm-recheck.sh | 1 + > .../selftests/rcutorture/bin/kvm-test-1-run.sh | 5 +- > tools/testing/selftests/rcutorture/bin/kvm.sh | 2 +- > .../selftests/rcutorture/bin/parse-console.sh | 7 + > .../selftests/rcutorture/configs/rcu/TREE03.boot | 4 +- > .../selftests/rcutorture/configs/rcu/TREE08-T.boot | 1 - > .../rcutorture/configs/rcu/ver_functions.sh | 2 +- > 39 files changed, 1675 insertions(+), 1209 deletions(-) > delete mode 100644 tools/testing/selftests/rcutorture/configs/rcu/TREE08-T.boot
Pulled into tip:core/rcu, thanks a lot Paul!
Ingo
| |