lkml.org 
[lkml]   [2019]   [Oct]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH] Remove GP_REPLAY state from rcu_sync
Hi "Joel,

I love your patch! Perhaps something to improve:

[auto build test WARNING on rcu/dev]
[cannot apply to v5.4-rc1 next-20191004]
[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/Joel-Fernandes-Google/Remove-GP_REPLAY-state-from-rcu_sync/20191005-024257
base: https://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git dev
config: x86_64-randconfig-b002-201939 (attached as .config)
compiler: gcc-7 (Debian 7.4.0-13) 7.4.0
reproduce:
# save the attached .config to linux build tree
make ARCH=x86_64

If you fix the issue, kindly add following tag
Reported-by: kbuild test robot <lkp@intel.com>

All warnings (new ones prefixed by >>):

In file included from include/linux/export.h:44:0,
from include/linux/linkage.h:7,
from include/linux/kernel.h:8,
from include/linux/list.h:9,
from include/linux/wait.h:7,
from include/linux/rcu_sync.h:13,
from kernel//rcu/sync.c:10:
kernel//rcu/sync.c: In function 'rcu_sync_dtor':
kernel//rcu/sync.c:187:23: error: 'GP_REPLAY' undeclared (first use in this function)
if (rsp->gp_state == GP_REPLAY)
^
include/linux/compiler.h:58:52: note: in definition of macro '__trace_if_var'
#define __trace_if_var(cond) (__builtin_constant_p(cond) ? (cond) : __trace_if_value(cond))
^~~~
>> kernel//rcu/sync.c:187:2: note: in expansion of macro 'if'
if (rsp->gp_state == GP_REPLAY)
^~
kernel//rcu/sync.c:187:23: note: each undeclared identifier is reported only once for each function it appears in
if (rsp->gp_state == GP_REPLAY)
^
include/linux/compiler.h:58:52: note: in definition of macro '__trace_if_var'
#define __trace_if_var(cond) (__builtin_constant_p(cond) ? (cond) : __trace_if_value(cond))
^~~~
>> kernel//rcu/sync.c:187:2: note: in expansion of macro 'if'
if (rsp->gp_state == GP_REPLAY)
^~

vim +/if +187 kernel//rcu/sync.c

cc44ca848f5e51 Oleg Nesterov 2015-08-21 @10 #include <linux/rcu_sync.h>
cc44ca848f5e51 Oleg Nesterov 2015-08-21 11 #include <linux/sched.h>
cc44ca848f5e51 Oleg Nesterov 2015-08-21 12
6d1a4c2dfe7bb0 Joel Fernandes 2019-10-04 13 enum { GP_IDLE = 0, GP_ENTER, GP_PASSED, GP_EXIT };
cc44ca848f5e51 Oleg Nesterov 2015-08-21 14
cc44ca848f5e51 Oleg Nesterov 2015-08-21 15 #define rss_lock gp_wait.lock
cc44ca848f5e51 Oleg Nesterov 2015-08-21 16
cc44ca848f5e51 Oleg Nesterov 2015-08-21 17 /**
cc44ca848f5e51 Oleg Nesterov 2015-08-21 18 * rcu_sync_init() - Initialize an rcu_sync structure
cc44ca848f5e51 Oleg Nesterov 2015-08-21 19 * @rsp: Pointer to rcu_sync structure to be initialized
cc44ca848f5e51 Oleg Nesterov 2015-08-21 20 */
95bf33b55ff446 Oleg Nesterov 2019-04-23 21 void rcu_sync_init(struct rcu_sync *rsp)
cc44ca848f5e51 Oleg Nesterov 2015-08-21 22 {
cc44ca848f5e51 Oleg Nesterov 2015-08-21 23 memset(rsp, 0, sizeof(*rsp));
cc44ca848f5e51 Oleg Nesterov 2015-08-21 24 init_waitqueue_head(&rsp->gp_wait);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 25 }
cc44ca848f5e51 Oleg Nesterov 2015-08-21 26
3942a9bd7b5842 Peter Zijlstra 2016-08-11 27 /**
27fdb35fe99011 Paul E. McKenney 2017-10-19 28 * rcu_sync_enter_start - Force readers onto slow path for multiple updates
27fdb35fe99011 Paul E. McKenney 2017-10-19 29 * @rsp: Pointer to rcu_sync structure to use for synchronization
27fdb35fe99011 Paul E. McKenney 2017-10-19 30 *
3942a9bd7b5842 Peter Zijlstra 2016-08-11 31 * Must be called after rcu_sync_init() and before first use.
3942a9bd7b5842 Peter Zijlstra 2016-08-11 32 *
3942a9bd7b5842 Peter Zijlstra 2016-08-11 33 * Ensures rcu_sync_is_idle() returns false and rcu_sync_{enter,exit}()
3942a9bd7b5842 Peter Zijlstra 2016-08-11 34 * pairs turn into NO-OPs.
3942a9bd7b5842 Peter Zijlstra 2016-08-11 35 */
3942a9bd7b5842 Peter Zijlstra 2016-08-11 36 void rcu_sync_enter_start(struct rcu_sync *rsp)
3942a9bd7b5842 Peter Zijlstra 2016-08-11 37 {
3942a9bd7b5842 Peter Zijlstra 2016-08-11 38 rsp->gp_count++;
3942a9bd7b5842 Peter Zijlstra 2016-08-11 39 rsp->gp_state = GP_PASSED;
3942a9bd7b5842 Peter Zijlstra 2016-08-11 40 }
3942a9bd7b5842 Peter Zijlstra 2016-08-11 41
cc44ca848f5e51 Oleg Nesterov 2015-08-21 42
89da3b94bb9741 Oleg Nesterov 2019-04-25 43 static void rcu_sync_func(struct rcu_head *rhp);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 44
89da3b94bb9741 Oleg Nesterov 2019-04-25 45 static void rcu_sync_call(struct rcu_sync *rsp)
89da3b94bb9741 Oleg Nesterov 2019-04-25 46 {
89da3b94bb9741 Oleg Nesterov 2019-04-25 47 call_rcu(&rsp->cb_head, rcu_sync_func);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 48 }
cc44ca848f5e51 Oleg Nesterov 2015-08-21 49
cc44ca848f5e51 Oleg Nesterov 2015-08-21 50 /**
cc44ca848f5e51 Oleg Nesterov 2015-08-21 51 * rcu_sync_func() - Callback function managing reader access to fastpath
27fdb35fe99011 Paul E. McKenney 2017-10-19 52 * @rhp: Pointer to rcu_head in rcu_sync structure to use for synchronization
cc44ca848f5e51 Oleg Nesterov 2015-08-21 53 *
89da3b94bb9741 Oleg Nesterov 2019-04-25 54 * This function is passed to call_rcu() function by rcu_sync_enter() and
cc44ca848f5e51 Oleg Nesterov 2015-08-21 55 * rcu_sync_exit(), so that it is invoked after a grace period following the
89da3b94bb9741 Oleg Nesterov 2019-04-25 56 * that invocation of enter/exit.
89da3b94bb9741 Oleg Nesterov 2019-04-25 57 *
89da3b94bb9741 Oleg Nesterov 2019-04-25 58 * If it is called by rcu_sync_enter() it signals that all the readers were
89da3b94bb9741 Oleg Nesterov 2019-04-25 59 * switched onto slow path.
89da3b94bb9741 Oleg Nesterov 2019-04-25 60 *
89da3b94bb9741 Oleg Nesterov 2019-04-25 61 * If it is called by rcu_sync_exit() it takes action based on events that
cc44ca848f5e51 Oleg Nesterov 2015-08-21 62 * have taken place in the meantime, so that closely spaced rcu_sync_enter()
cc44ca848f5e51 Oleg Nesterov 2015-08-21 63 * and rcu_sync_exit() pairs need not wait for a grace period.
cc44ca848f5e51 Oleg Nesterov 2015-08-21 64 *
cc44ca848f5e51 Oleg Nesterov 2015-08-21 65 * If another rcu_sync_enter() is invoked before the grace period
cc44ca848f5e51 Oleg Nesterov 2015-08-21 66 * ended, reset state to allow the next rcu_sync_exit() to let the
cc44ca848f5e51 Oleg Nesterov 2015-08-21 67 * readers back onto their fastpaths (after a grace period). If both
cc44ca848f5e51 Oleg Nesterov 2015-08-21 68 * another rcu_sync_enter() and its matching rcu_sync_exit() are invoked
cc44ca848f5e51 Oleg Nesterov 2015-08-21 69 * before the grace period ended, re-invoke call_rcu() on behalf of that
cc44ca848f5e51 Oleg Nesterov 2015-08-21 70 * rcu_sync_exit(). Otherwise, set all state back to idle so that readers
cc44ca848f5e51 Oleg Nesterov 2015-08-21 71 * can again use their fastpaths.
cc44ca848f5e51 Oleg Nesterov 2015-08-21 72 */
27fdb35fe99011 Paul E. McKenney 2017-10-19 73 static void rcu_sync_func(struct rcu_head *rhp)
cc44ca848f5e51 Oleg Nesterov 2015-08-21 74 {
27fdb35fe99011 Paul E. McKenney 2017-10-19 75 struct rcu_sync *rsp = container_of(rhp, struct rcu_sync, cb_head);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 76 unsigned long flags;
cc44ca848f5e51 Oleg Nesterov 2015-08-21 77
89da3b94bb9741 Oleg Nesterov 2019-04-25 78 WARN_ON_ONCE(READ_ONCE(rsp->gp_state) == GP_IDLE);
89da3b94bb9741 Oleg Nesterov 2019-04-25 79 WARN_ON_ONCE(READ_ONCE(rsp->gp_state) == GP_PASSED);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 80
cc44ca848f5e51 Oleg Nesterov 2015-08-21 81 spin_lock_irqsave(&rsp->rss_lock, flags);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 82 if (rsp->gp_count) {
cc44ca848f5e51 Oleg Nesterov 2015-08-21 83 /*
89da3b94bb9741 Oleg Nesterov 2019-04-25 84 * We're at least a GP after the GP_IDLE->GP_ENTER transition.
cc44ca848f5e51 Oleg Nesterov 2015-08-21 85 */
89da3b94bb9741 Oleg Nesterov 2019-04-25 86 WRITE_ONCE(rsp->gp_state, GP_PASSED);
89da3b94bb9741 Oleg Nesterov 2019-04-25 87 wake_up_locked(&rsp->gp_wait);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 88 } else {
cc44ca848f5e51 Oleg Nesterov 2015-08-21 89 /*
89da3b94bb9741 Oleg Nesterov 2019-04-25 90 * We're at least a GP after the last rcu_sync_exit(); eveybody
89da3b94bb9741 Oleg Nesterov 2019-04-25 91 * will now have observed the write side critical section.
89da3b94bb9741 Oleg Nesterov 2019-04-25 92 * Let 'em rip!.
cc44ca848f5e51 Oleg Nesterov 2015-08-21 93 */
89da3b94bb9741 Oleg Nesterov 2019-04-25 94 WRITE_ONCE(rsp->gp_state, GP_IDLE);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 95 }
cc44ca848f5e51 Oleg Nesterov 2015-08-21 96 spin_unlock_irqrestore(&rsp->rss_lock, flags);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 97 }
cc44ca848f5e51 Oleg Nesterov 2015-08-21 98
cc44ca848f5e51 Oleg Nesterov 2015-08-21 99 /**
89da3b94bb9741 Oleg Nesterov 2019-04-25 100 * rcu_sync_enter() - Force readers onto slowpath
89da3b94bb9741 Oleg Nesterov 2019-04-25 101 * @rsp: Pointer to rcu_sync structure to use for synchronization
89da3b94bb9741 Oleg Nesterov 2019-04-25 102 *
89da3b94bb9741 Oleg Nesterov 2019-04-25 103 * This function is used by updaters who need readers to make use of
89da3b94bb9741 Oleg Nesterov 2019-04-25 104 * a slowpath during the update. After this function returns, all
89da3b94bb9741 Oleg Nesterov 2019-04-25 105 * subsequent calls to rcu_sync_is_idle() will return false, which
89da3b94bb9741 Oleg Nesterov 2019-04-25 106 * tells readers to stay off their fastpaths. A later call to
89da3b94bb9741 Oleg Nesterov 2019-04-25 107 * rcu_sync_exit() re-enables reader slowpaths.
89da3b94bb9741 Oleg Nesterov 2019-04-25 108 *
89da3b94bb9741 Oleg Nesterov 2019-04-25 109 * When called in isolation, rcu_sync_enter() must wait for a grace
89da3b94bb9741 Oleg Nesterov 2019-04-25 110 * period, however, closely spaced calls to rcu_sync_enter() can
89da3b94bb9741 Oleg Nesterov 2019-04-25 111 * optimize away the grace-period wait via a state machine implemented
89da3b94bb9741 Oleg Nesterov 2019-04-25 112 * by rcu_sync_enter(), rcu_sync_exit(), and rcu_sync_func().
89da3b94bb9741 Oleg Nesterov 2019-04-25 113 */
89da3b94bb9741 Oleg Nesterov 2019-04-25 114 void rcu_sync_enter(struct rcu_sync *rsp)
89da3b94bb9741 Oleg Nesterov 2019-04-25 115 {
89da3b94bb9741 Oleg Nesterov 2019-04-25 116 int gp_state;
89da3b94bb9741 Oleg Nesterov 2019-04-25 117
89da3b94bb9741 Oleg Nesterov 2019-04-25 118 spin_lock_irq(&rsp->rss_lock);
89da3b94bb9741 Oleg Nesterov 2019-04-25 119 gp_state = rsp->gp_state;
89da3b94bb9741 Oleg Nesterov 2019-04-25 120 if (gp_state == GP_IDLE) {
89da3b94bb9741 Oleg Nesterov 2019-04-25 121 WRITE_ONCE(rsp->gp_state, GP_ENTER);
89da3b94bb9741 Oleg Nesterov 2019-04-25 122 WARN_ON_ONCE(rsp->gp_count);
89da3b94bb9741 Oleg Nesterov 2019-04-25 123 /*
89da3b94bb9741 Oleg Nesterov 2019-04-25 124 * Note that we could simply do rcu_sync_call(rsp) here and
89da3b94bb9741 Oleg Nesterov 2019-04-25 125 * avoid the "if (gp_state == GP_IDLE)" block below.
89da3b94bb9741 Oleg Nesterov 2019-04-25 126 *
89da3b94bb9741 Oleg Nesterov 2019-04-25 127 * However, synchronize_rcu() can be faster if rcu_expedited
89da3b94bb9741 Oleg Nesterov 2019-04-25 128 * or rcu_blocking_is_gp() is true.
89da3b94bb9741 Oleg Nesterov 2019-04-25 129 *
89da3b94bb9741 Oleg Nesterov 2019-04-25 130 * Another reason is that we can't wait for rcu callback if
89da3b94bb9741 Oleg Nesterov 2019-04-25 131 * we are called at early boot time but this shouldn't happen.
89da3b94bb9741 Oleg Nesterov 2019-04-25 132 */
89da3b94bb9741 Oleg Nesterov 2019-04-25 133 }
89da3b94bb9741 Oleg Nesterov 2019-04-25 134 rsp->gp_count++;
89da3b94bb9741 Oleg Nesterov 2019-04-25 135 spin_unlock_irq(&rsp->rss_lock);
89da3b94bb9741 Oleg Nesterov 2019-04-25 136
89da3b94bb9741 Oleg Nesterov 2019-04-25 137 if (gp_state == GP_IDLE) {
89da3b94bb9741 Oleg Nesterov 2019-04-25 138 /*
89da3b94bb9741 Oleg Nesterov 2019-04-25 139 * See the comment above, this simply does the "synchronous"
89da3b94bb9741 Oleg Nesterov 2019-04-25 140 * call_rcu(rcu_sync_func) which does GP_ENTER -> GP_PASSED.
89da3b94bb9741 Oleg Nesterov 2019-04-25 141 */
89da3b94bb9741 Oleg Nesterov 2019-04-25 142 synchronize_rcu();
89da3b94bb9741 Oleg Nesterov 2019-04-25 143 rcu_sync_func(&rsp->cb_head);
89da3b94bb9741 Oleg Nesterov 2019-04-25 144 /* Not really needed, wait_event() would see GP_PASSED. */
89da3b94bb9741 Oleg Nesterov 2019-04-25 145 return;
89da3b94bb9741 Oleg Nesterov 2019-04-25 146 }
89da3b94bb9741 Oleg Nesterov 2019-04-25 147
89da3b94bb9741 Oleg Nesterov 2019-04-25 148 wait_event(rsp->gp_wait, READ_ONCE(rsp->gp_state) >= GP_PASSED);
89da3b94bb9741 Oleg Nesterov 2019-04-25 149 }
89da3b94bb9741 Oleg Nesterov 2019-04-25 150
89da3b94bb9741 Oleg Nesterov 2019-04-25 151 /**
89da3b94bb9741 Oleg Nesterov 2019-04-25 152 * rcu_sync_exit() - Allow readers back onto fast path after grace period
cc44ca848f5e51 Oleg Nesterov 2015-08-21 153 * @rsp: Pointer to rcu_sync structure to use for synchronization
cc44ca848f5e51 Oleg Nesterov 2015-08-21 154 *
cc44ca848f5e51 Oleg Nesterov 2015-08-21 155 * This function is used by updaters who have completed, and can therefore
cc44ca848f5e51 Oleg Nesterov 2015-08-21 156 * now allow readers to make use of their fastpaths after a grace period
cc44ca848f5e51 Oleg Nesterov 2015-08-21 157 * has elapsed. After this grace period has completed, all subsequent
cc44ca848f5e51 Oleg Nesterov 2015-08-21 158 * calls to rcu_sync_is_idle() will return true, which tells readers that
cc44ca848f5e51 Oleg Nesterov 2015-08-21 159 * they can once again use their fastpaths.
cc44ca848f5e51 Oleg Nesterov 2015-08-21 160 */
cc44ca848f5e51 Oleg Nesterov 2015-08-21 161 void rcu_sync_exit(struct rcu_sync *rsp)
cc44ca848f5e51 Oleg Nesterov 2015-08-21 162 {
6d1a4c2dfe7bb0 Joel Fernandes 2019-10-04 163 WARN_ON_ONCE(READ_ONCE(rsp->gp_state) < GP_PASSED);
89da3b94bb9741 Oleg Nesterov 2019-04-25 164
cc44ca848f5e51 Oleg Nesterov 2015-08-21 165 spin_lock_irq(&rsp->rss_lock);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 166 if (!--rsp->gp_count) {
89da3b94bb9741 Oleg Nesterov 2019-04-25 167 if (rsp->gp_state == GP_PASSED) {
89da3b94bb9741 Oleg Nesterov 2019-04-25 168 WRITE_ONCE(rsp->gp_state, GP_EXIT);
89da3b94bb9741 Oleg Nesterov 2019-04-25 169 rcu_sync_call(rsp);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 170 }
cc44ca848f5e51 Oleg Nesterov 2015-08-21 171 }
cc44ca848f5e51 Oleg Nesterov 2015-08-21 172 spin_unlock_irq(&rsp->rss_lock);
cc44ca848f5e51 Oleg Nesterov 2015-08-21 173 }
07899a6e5f5613 Oleg Nesterov 2015-08-21 174
07899a6e5f5613 Oleg Nesterov 2015-08-21 175 /**
07899a6e5f5613 Oleg Nesterov 2015-08-21 176 * rcu_sync_dtor() - Clean up an rcu_sync structure
07899a6e5f5613 Oleg Nesterov 2015-08-21 177 * @rsp: Pointer to rcu_sync structure to be cleaned up
07899a6e5f5613 Oleg Nesterov 2015-08-21 178 */
07899a6e5f5613 Oleg Nesterov 2015-08-21 179 void rcu_sync_dtor(struct rcu_sync *rsp)
07899a6e5f5613 Oleg Nesterov 2015-08-21 180 {
89da3b94bb9741 Oleg Nesterov 2019-04-25 181 int gp_state;
07899a6e5f5613 Oleg Nesterov 2015-08-21 182
89da3b94bb9741 Oleg Nesterov 2019-04-25 183 WARN_ON_ONCE(READ_ONCE(rsp->gp_count));
89da3b94bb9741 Oleg Nesterov 2019-04-25 184 WARN_ON_ONCE(READ_ONCE(rsp->gp_state) == GP_PASSED);
07899a6e5f5613 Oleg Nesterov 2015-08-21 185
07899a6e5f5613 Oleg Nesterov 2015-08-21 186 spin_lock_irq(&rsp->rss_lock);
89da3b94bb9741 Oleg Nesterov 2019-04-25 @187 if (rsp->gp_state == GP_REPLAY)

:::::: The code at line 187 was first introduced by commit
:::::: 89da3b94bb97417ca2c5b0ce3a28643819030247 rcu/sync: Simplify the state machine

:::::: TO: Oleg Nesterov <oleg@redhat.com>
:::::: CC: Paul E. McKenney <paulmck@linux.ibm.com>

---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
[unhandled content-type:application/gzip]
\
 
 \ /
  Last update: 2019-10-05 00:04    [W:0.128 / U:2.052 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site