Re: [PATCH 2/2] refscale: prevent buffer to pr_alert() being too long

[Paul E. McKenney]

On Sun, Oct 24, 2021 at 10:46:16PM +0800, Li, Zhijian wrote:
> 
> on 2021/10/23 7:15, Paul E. McKenney wrote:
> > On Fri, Oct 22, 2021 at 06:51:11PM +0800, Li Zhijian wrote:
> > > 0Day/LKP observed that the refscale results become incompleted
> > > when a larger nruns(such as 300) is specified.
> > > It seems that printk() can accept < 1024 buffer at once.
> > > Print the buffer if its length exceeds 800 simply.
> > > 
> > > CC: Philip Li <philip.li@xxxxxxxxx>
> > > Reported-by: kernel test robot <lkp@xxxxxxxxx>
> > > Signed-off-by: Li Zhijian <lizhijian@xxxxxxxxxxxxxx>
> > Good catch!  A couple of questions below.
> > 
> > 						Thanx, Paul
> > 
> > > ---
> > >   kernel/rcu/refscale.c | 23 +++++++++++++----------
> > >   1 file changed, 13 insertions(+), 10 deletions(-)
> > > 
> > > diff --git a/kernel/rcu/refscale.c b/kernel/rcu/refscale.c
> > > index 2cbe2a2ba387..b1b9052010fd 100644
> > > --- a/kernel/rcu/refscale.c
> > > +++ b/kernel/rcu/refscale.c
> > > @@ -604,7 +604,7 @@ static u64 process_durations(int n)
> > >   	char *buf;
> > >   	u64 sum = 0;
> > > -	buf = kmalloc(128 + nreaders * 32, GFP_KERNEL);
> > > +	buf = kmalloc(64 * 20, GFP_KERNEL);
> > This allocation (and the one below) is 1280 bytes rather than
> > 1024 bytes.  Why the extra couple hundred bytes?
> 
> Nothing special, so let's change 1024 or (800 + 64 ),which is sufficent as well ?

Either works.  No problem having a few extra bytes, but 200 seemed a bit
excessive.  ;-)

> > >   	if (!buf)
> > >   		return 0;
> > >   	buf[0] = 0;
> > > @@ -617,13 +617,15 @@ static u64 process_durations(int n)
> > >   		if (i % 5 == 0)
> > >   			strcat(buf, "\n");
> > > +		if (strlen(buf) > 800) {
> > > +			pr_alert("%s", buf);
> > Does the tools/testing/selftests/rcutorture/bin/kvm-recheck-refscale.sh
> > script also require changes to handle the partial lines?
> 
> I missed that, i will take a look at it

Looking forward to seeing what you come up with.

							Thanx, Paul

> Thanks
> 
> Zhijian
> 
> 
> > 
> > Same for the later comparison against 800.
> > 
> > > +			buf[0] = 0;
> > > +		}
> > >   		strcat(buf, buf1);
> > >   		sum += rt->last_duration_ns;
> > >   	}
> > > -	strcat(buf, "\n");
> > > -
> > > -	SCALEOUT("%s\n", buf);
> > > +	pr_alert("%s\n", buf);
> > >   	kfree(buf);
> > >   	return sum;
> > > @@ -648,7 +650,7 @@ static int main_func(void *arg)
> > >   	VERBOSE_SCALEOUT("main_func task started");
> > >   	result_avg = kzalloc(nruns * sizeof(*result_avg), GFP_KERNEL);
> > > -	buf = kzalloc(64 + nruns * 32, GFP_KERNEL);
> > > +	buf = kzalloc(64 * 20, GFP_KERNEL);
> > >   	if (!result_avg || !buf) {
> > >   		VERBOSE_SCALEOUT_ERRSTRING("out of memory");
> > >   		errexit = true;
> > > @@ -701,10 +703,7 @@ static int main_func(void *arg)
> > >   	if (errexit)
> > >   		goto err;
> > > -	buf[0] = 0;
> > > -	strcat(buf, "\n");
> > > -	strcat(buf, "Runs\tTime(ns)\n");
> > > -
> > > +	pr_alert("Runs\tTime(ns)\n");
> > >   	for (exp = 0; exp < nruns; exp++) {
> > >   		u64 avg;
> > >   		u32 rem;
> > > @@ -712,9 +711,13 @@ static int main_func(void *arg)
> > >   		avg = div_u64_rem(result_avg[exp], 1000, &rem);
> > >   		sprintf(buf1, "%d\t%llu.%03u\n", exp + 1, avg, rem);
> > >   		strcat(buf, buf1);
> > > +		if (strlen(buf) > 800) {
> > > +			pr_alert("%s", buf);
> > > +			buf[0] = 0;
> > > +		}
> > >   	}
> > > -	SCALEOUT("%s", buf);
> > > +	pr_alert("%s", buf);
> > >   err:
> > >   	// This will shutdown everything including us.
> > > -- 
> > > 2.33.0
> > > 
> > > 
> > > 
> > 
> 
>