On Wed, Jan 08, 2025 at 01:55:12PM +0800, Z qiang wrote:
> >
> > On Tue, Jan 07, 2025 at 05:49:42PM +0800, Z qiang wrote:
> > > >
> > > > On Mon, Jan 06, 2025 at 11:24:54AM +0800, Z qiang wrote:
> > > > > >
> > > > > > On Fri, Jan 03, 2025 at 01:50:50PM +0800, Z qiang wrote:
> > > > > > > >
> > > > > > > > On Thu, Jan 02, 2025 at 01:41:16PM +0800, Oliver Sang wrote:
> > > > > > > > > hi, Zqiang, hi, Paul,
> > > > > > > > >
> > > > > > > > > On Tue, Dec 31, 2024 at 08:19:44AM -0800, Paul E. McKenney
> > > > > > > > > wrote:
> > > > > > > > > > On Tue, Dec 31, 2024 at 03:16:25PM +0800, Z qiang wrote:
> > > > > > > > >
> > > > > > > > > [...]
> > > > > > > > >
> > > > > > > > > > > >
> > > > > > > > > > > > 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-lkp/[email protected]
> > > > > > > > > > > >
> > > > > > > > > > >
> > > > > > > > > > > Please try the following modifications:
> > > > > > > > >
> > > > > > > > > yes, the following modifications fix the issues we found for
> > > > > > > > > 9216c28c6a. thanks!
> > > > > > > > >
> > > > > > > > > Tested-by: kernel test robot <[email protected]>
> > > > > > > >
> > > > > > > > Thank you, and I will apply this during my next rebase.
> > > > > > > >
> > > > > > > > Zqiang, unless you tell me otherwise, I will add your
> > > > > > > > Co-developed-by
> > > > > > > > and Signed-off-by.
> > > > > > > >
> > > > > > > > Either way, Happy Square New Year!
> > > > > > >
> > > > > > > Thank you, Happy New Year! :)
> > > > > >
> > > > > > And here you go! Please let me know of anything I missed.
> > > > > >
> > > > > > And thank you for saving me the time it would have taken to track
> > > > > > this one down. ;-)
> > > > > >
> > > > > > Thanx, Paul
> > > > > >
> > > > > > ------------------------------------------------------------------------
> > > > > >
> > > > > > commit 71c893e6d1857d1e4ea37aec557d734a560fdb39
> > > > > > Author: Paul E. McKenney <[email protected]>
> > > > > > Date: Thu Dec 19 16:08:54 2024 -0800
> > > > > >
> > > > > > srcu: Make SRCU readers use ->srcu_ctrs for counter selection
> > > > > >
> > > > > > This commit causes SRCU readers to use ->srcu_ctrs for counter
> > > > > > selection instead of ->srcu_idx. This takes another step
> > > > > > towards
> > > > > > array-indexing-free SRCU readers.
> > > > > >
> > > > > > [ paulmck: Apply kernel test robot feedback. ]
> > > > > >
> > > > > > Co-developed-by: Z qiang <[email protected]>
> > > > > > Signed-off-by: Z qiang <[email protected]>
> > > > > > Signed-off-by: Paul E. McKenney <[email protected]>
> > > > > > Tested-by: kernel test robot <[email protected]>
> > > > > >
> > > > > > diff --git a/include/linux/srcutree.h b/include/linux/srcutree.h
> > > > > > index c794d599db5c1..1b01ced61a45b 100644
> > > > > > --- a/include/linux/srcutree.h
> > > > > > +++ b/include/linux/srcutree.h
> > > > > > @@ -101,6 +101,7 @@ struct srcu_usage {
> > > > > > */
> > > > > > struct srcu_struct {
> > > > > > unsigned int srcu_idx; /* Current rdr
> > > > > > array element. */
> > > > > > + struct srcu_ctr __percpu *srcu_ctrp;
> > > > > > struct srcu_data __percpu *sda; /* Per-CPU
> > > > > > srcu_data array. */
> > > > > > struct lockdep_map dep_map;
> > > > > > struct srcu_usage *srcu_sup; /* Update-side
> > > > > > data. */
> > > > > > @@ -167,6 +168,7 @@ struct srcu_struct {
> > > > > > #define __SRCU_STRUCT_INIT(name, usage_name, pcpu_name)
> > > > > > \
> > > > > > {
> > > > > > \
> > > > > > .sda = &pcpu_name,
> > > > > > \
> > > > > > + .srcu_ctrp = &pcpu_name.srcu_ctrs[0],
> > > > > > \
> > > > > > __SRCU_STRUCT_INIT_COMMON(name, usage_name)
> > > > > > \
> > > > > > }
> > > > > >
> > > > > > @@ -222,13 +224,12 @@ void srcu_torture_stats_print(struct
> > > > > > srcu_struct *ssp, char *tt, char *tf);
> > > > > > */
> > > > > > static inline int __srcu_read_lock_lite(struct srcu_struct *ssp)
> > > > > > {
> > > > > > - int idx;
> > > > > > + struct srcu_ctr __percpu *scp = READ_ONCE(ssp->srcu_ctrp);
> > > > > >
> > > > > > RCU_LOCKDEP_WARN(!rcu_is_watching(), "RCU must be watching
> > > > > > srcu_read_lock_lite().");
> > > > > > - idx = READ_ONCE(ssp->srcu_idx) & 0x1;
> > > > > > - this_cpu_inc(ssp->sda->srcu_ctrs[idx].srcu_locks.counter);
> > > > > > /* Y */
> > > > > > + this_cpu_inc(scp->srcu_locks.counter); /* Y */
> > > > > > barrier(); /* Avoid leaking the critical section. */
> > > > > > - return idx;
> > > > > > + return scp - &ssp->sda->srcu_ctrs[0];
> > > > > > }
> > > > > >
> > > > > > /*
> > > > > > diff --git a/kernel/rcu/srcutree.c b/kernel/rcu/srcutree.c
> > > > > > index d7ee2f345e192..3bf7f41ad72b8 100644
> > > > > > --- a/kernel/rcu/srcutree.c
> > > > > > +++ b/kernel/rcu/srcutree.c
> > > > > > @@ -253,8 +253,10 @@ static int init_srcu_struct_fields(struct
> > > > > > srcu_struct *ssp, bool is_static)
> > > > > > atomic_set(&ssp->srcu_sup->srcu_barrier_cpu_cnt, 0);
> > > > > > INIT_DELAYED_WORK(&ssp->srcu_sup->work, process_srcu);
> > > > > > ssp->srcu_sup->sda_is_static = is_static;
> > > > > > - if (!is_static)
> > > > > > + if (!is_static) {
> > > > > > ssp->sda = alloc_percpu(struct srcu_data);
> > > > > > + ssp->srcu_ctrp = &ssp->sda->srcu_ctrs[0];
> > > > > > + }
> > > > > > if (!ssp->sda)
> > > > > > goto err_free_sup;
> > > > >
> > > > > ssp->srcu_ctrp should be assigned a value under the condition that
> > > > > ssp->sda is allocated successfully.
> > > > >
> > > > > - if (!is_static) {
> > > > > + if (!is_static)
> > > > > ssp->sda = alloc_percpu(struct srcu_data);
> > > > > - ssp->srcu_ctrp = &ssp->sda->srcu_ctrs[0];
> > > > > - }
> > > > > if (!ssp->sda)
> > > > > goto err_free_sup;
> > > > > + if (!is_static)
> > > > > + ssp->srcu_ctrp = &ssp->sda->srcu_ctrs[0];
> > > >
> > > > It doesn't hurt to assign NULL to ssp->srcu_ctrp on allocation failure.
> > > >
> > > > Or am I missing something here?
> > >
> > > No, I just think in case of failed allocation, there is no need to set
> > > srcu_ctrp. :)
> >
> > True enough, but you are adding an extra "if", which is not free
> > from either an execution-time viewpoint (not an issue here) or from a
> > complexity viewpoint. I am not ruling your change out, but neither am
> > I convinced.
> >
> > Please see below for the current state, which passes light testing even
> > given proper testing of _nmisafe(). Which I was failing to do before.
> >
> > Thoughts?
> >
> > Thanx, Paul
> >
> > ------------------------------------------------------------------------
> >
> > commit e8964cf2179673adc54aae38b62361fd0fdf89ff
> > Author: Paul E. McKenney <[email protected]>
> > Date: Thu Dec 19 16:08:54 2024 -0800
> >
> > srcu: Make SRCU readers use ->srcu_ctrs for counter selection
> >
> > This commit causes SRCU readers to use ->srcu_ctrs for counter
> > selection instead of ->srcu_idx. This takes another step towards
> > array-indexing-free SRCU readers.
> >
> > [ paulmck: Apply kernel test robot feedback. ]
> >
> > Co-developed-by: Z qiang <[email protected]>
> > Signed-off-by: Z qiang <[email protected]>
> > Signed-off-by: Paul E. McKenney <[email protected]>
> > Tested-by: kernel test robot <[email protected]>
> >
> > diff --git a/include/linux/srcutree.h b/include/linux/srcutree.h
> > index c794d599db5c1..1b01ced61a45b 100644
> > --- a/include/linux/srcutree.h
> > +++ b/include/linux/srcutree.h
> > @@ -101,6 +101,7 @@ struct srcu_usage {
> > */
> > struct srcu_struct {
> > unsigned int srcu_idx; /* Current rdr array
> > element. */
> > + struct srcu_ctr __percpu *srcu_ctrp;
> > struct srcu_data __percpu *sda; /* Per-CPU srcu_data array.
> > */
> > struct lockdep_map dep_map;
> > struct srcu_usage *srcu_sup; /* Update-side data. */
> > @@ -167,6 +168,7 @@ struct srcu_struct {
> > #define __SRCU_STRUCT_INIT(name, usage_name, pcpu_name)
> > \
> > {
> > \
> > .sda = &pcpu_name,
> > \
> > + .srcu_ctrp = &pcpu_name.srcu_ctrs[0],
> > \
> > __SRCU_STRUCT_INIT_COMMON(name, usage_name)
> > \
> > }
> >
> > @@ -222,13 +224,12 @@ void srcu_torture_stats_print(struct srcu_struct
> > *ssp, char *tt, char *tf);
> > */
> > static inline int __srcu_read_lock_lite(struct srcu_struct *ssp)
> > {
> > - int idx;
> > + struct srcu_ctr __percpu *scp = READ_ONCE(ssp->srcu_ctrp);
> >
> > RCU_LOCKDEP_WARN(!rcu_is_watching(), "RCU must be watching
> > srcu_read_lock_lite().");
> > - idx = READ_ONCE(ssp->srcu_idx) & 0x1;
> > - this_cpu_inc(ssp->sda->srcu_ctrs[idx].srcu_locks.counter); /* Y */
> > + this_cpu_inc(scp->srcu_locks.counter); /* Y */
> > barrier(); /* Avoid leaking the critical section. */
> > - return idx;
> > + return scp - &ssp->sda->srcu_ctrs[0];
> > }
> >
> > /*
> > diff --git a/kernel/rcu/srcutree.c b/kernel/rcu/srcutree.c
> > index d7ee2f345e192..308cc7b5098c5 100644
> > --- a/kernel/rcu/srcutree.c
> > +++ b/kernel/rcu/srcutree.c
> > @@ -253,8 +253,10 @@ static int init_srcu_struct_fields(struct srcu_struct
> > *ssp, bool is_static)
> > atomic_set(&ssp->srcu_sup->srcu_barrier_cpu_cnt, 0);
> > INIT_DELAYED_WORK(&ssp->srcu_sup->work, process_srcu);
> > ssp->srcu_sup->sda_is_static = is_static;
> > - if (!is_static)
> > + if (!is_static) {
> > ssp->sda = alloc_percpu(struct srcu_data);
> > + ssp->srcu_ctrp = &ssp->sda->srcu_ctrs[0];
> > + }
> > if (!ssp->sda)
> > goto err_free_sup;
> > init_srcu_struct_data(ssp);
> > @@ -742,12 +744,11 @@ EXPORT_SYMBOL_GPL(__srcu_check_read_flavor);
> > */
> > int __srcu_read_lock(struct srcu_struct *ssp)
> > {
> > - int idx;
> > + struct srcu_ctr __percpu *scp = READ_ONCE(ssp->srcu_ctrp);
> >
> > - idx = READ_ONCE(ssp->srcu_idx) & 0x1;
> > - this_cpu_inc(ssp->sda->srcu_ctrs[idx].srcu_locks.counter);
> > + this_cpu_inc(scp->srcu_locks.counter);
> > smp_mb(); /* B */ /* Avoid leaking the critical section. */
> > - return idx;
> > + return scp - &ssp->sda->srcu_ctrs[0];
> > }
> > EXPORT_SYMBOL_GPL(__srcu_read_lock);
> >
> > @@ -772,13 +773,12 @@ EXPORT_SYMBOL_GPL(__srcu_read_unlock);
> > */
> > int __srcu_read_lock_nmisafe(struct srcu_struct *ssp)
> > {
> > - int idx;
> > - struct srcu_data *sdp = raw_cpu_ptr(ssp->sda);
> > + struct srcu_ctr __percpu *scpp = READ_ONCE(ssp->srcu_ctrp);
> > + struct srcu_ctr *scp = this_cpu_ptr(scpp);
>
> Whether raw_cpu_ptr() should be used, otherwise it will appear:
>
> BUG: using smp_processor_id() in preemptible [00000000] code: init/1
> [ 7.324304][ T1] caller is debug_smp_processor_id+0x1b/0x30
> [ 7.324308][ T1] CPU: 0 UID: 0 PID: 1 Comm: init Not tainted
> 6.13.0-rc2-yoctodev-standard+ #99
> 10b660623b999b402f7cd2198fe7d4ebd36ef1d0
> [ 7.324311][ T1] Hardware name: QEMU Standard PC (Q35 + ICH9,
> 2009), BIOS rel-1.16.2-0-gea1b7a073390-prebuilt.qemu.org 04/01/2014
> [ 7.324314][ T1] Call Trace:
> [ 7.324315][ T1] <TASK>
> [ 7.324319][ T1] dump_stack_lvl+0xbb/0xd0
> [ 7.324326][ T1] dump_stack+0x14/0x20
> [ 7.324330][ T1] check_preemption_disabled+0xc7/0xf0
> [ 7.324340][ T1] debug_smp_processor_id+0x1b/0x30
> [ 7.324343][ T1] __srcu_read_lock_nmisafe+0x39/0xd0
> [ 7.324353][ T1] console_flush_all+0x18c/0xbf0
> ........
You are quite right! Fixing, thank you!
Ah, and the reason that the SRCU-P scenario did not catch this is that
I didn't force CONFIG_ARCH_HAS_NMI_SAFE_THIS_CPU_OPS=y.
Thanx, Paul
> Thanks
> Zqiang
>
> >
> > - idx = READ_ONCE(ssp->srcu_idx) & 0x1;
> > - atomic_long_inc(&sdp->srcu_ctrs[idx].srcu_locks);
> > + atomic_long_inc(&scp->srcu_locks);
> > smp_mb__after_atomic(); /* B */ /* Avoid leaking the critical
> > section. */
> > - return idx;
> > + return scpp - &ssp->sda->srcu_ctrs[0];
> > }
> > EXPORT_SYMBOL_GPL(__srcu_read_lock_nmisafe);
> >
> > @@ -1152,6 +1152,8 @@ static void srcu_flip(struct srcu_struct *ssp)
> > smp_mb(); /* E */ /* Pairs with B and C. */
> >
> > WRITE_ONCE(ssp->srcu_idx, ssp->srcu_idx + 1); // Flip the counter.
> > + WRITE_ONCE(ssp->srcu_ctrp,
> > + &ssp->sda->srcu_ctrs[!(ssp->srcu_ctrp -
> > &ssp->sda->srcu_ctrs[0])]);
> >
> > /*
> > * Ensure that if the updater misses an __srcu_read_unlock()
> > @@ -2000,6 +2002,7 @@ static int srcu_module_coming(struct module *mod)
> > ssp->sda = alloc_percpu(struct srcu_data);
> > if (WARN_ON_ONCE(!ssp->sda))
> > return -ENOMEM;
> > + ssp->srcu_ctrp = &ssp->sda->srcu_ctrs[0];
> > }
> > return 0;
> > }
