On 08/17/11 12:10, Sankara Muthukrishnan wrote:
> On Wed, Aug 17, 2011 at 11:54 AM, Sankara Muthukrishnan
> <sankara.m@xxxxxxxxx> wrote:
>> The cyclic test results are not acceptable (attached the output of
>> cyclictest). I am seeing maximum latency of 757 us on core-0 and 1905
>> us on core-1. Are you (Is anyone) seeing better results on latency
>> with SMP kernel with RT patch on ARM (OMAP4 processors)? I have seen
>> ~25 us of latency with single core kernel (though I have used a
>> different kernel configuration).
>>
>
> Never ever mind and sorry for the confusion. I did not realize that
> "Kernel debugging" was enabled in the config Frank had attached. With
> kernel debugging disabled, I see max latencies less than 20 us on both
> cores on panda board with cyclictest (cyclictest -l1000000 -m -Sp99
> -i200 -h60 -q -n) in my quick run and it is exciting to see these
> numbers. Thanks Frank.
>
> The only other outstanding problem as of now is that the RT kernel
> hanging during boot on panda board (boots sometimes after several
> resets and/or power-cycles).
You didn't mention whether you are getting any console messages when
the boot hangs. If you aren't getting any messages, then the attached
patches __might__ allow some messages to get out. You also need to
enable CONFIG_DEBUG_LL and CONFIG_EARLY_PRINTK, and also add
earlyprintk to the kernel boot command line.
(I need to clean up the patches a bit before even thinking about
whether they should be submitted.)
-Frank
WARNING: this is a hack to support debugging.
For CONFIG_PREEMPT_RT_FULL console_trylock_for_printk() will not attempt to
lock the console if:
!early_boot_irqs_disabled && !irqs_disabled_flags(flags) && !preempt_count()
thus vprintk() will not print out the buffers.
Allow console_trylock_for_printk() attempt to lock the console if
sysrq_in_progress.
This hack patch sets oops_in_progress so that sysrq output can maybe get
delivered to the console without delay.
Using oops_in_progress is important because that also affects the locking
methods used by the serial driver.
---
kernel/printk.c | 5 3 + 2 - 0 !
1 file changed, 3 insertions(+), 2 deletions(-)
Index: b/kernel/printk.c
===================================================================
--- a/kernel/printk.c
+++ b/kernel/printk.c
@@ -824,8 +824,9 @@ static int console_trylock_for_printk(un
__releases(&logbuf_lock)
{
#ifdef CONFIG_PREEMPT_RT_FULL
- int lock = !early_boot_irqs_disabled && !irqs_disabled_flags(flags) &&
- !preempt_count();
+ int lock = oops_in_progress ||
+ (!early_boot_irqs_disabled && !irqs_disabled_flags(flags) &&
+ !preempt_count());
#else
int lock = 1;
#endif
---
drivers/tty/sysrq.c | 3 3 + 0 - 0 !
1 file changed, 3 insertions(+)
Index: b/drivers/tty/sysrq.c
===================================================================
--- a/drivers/tty/sysrq.c
+++ b/drivers/tty/sysrq.c
@@ -499,6 +499,8 @@ void __handle_sysrq(int key, bool check_
int i;
unsigned long flags;
+int save_oops_in_progress = oops_in_progress;
+oops_in_progress = 1;
spin_lock_irqsave(&sysrq_key_table_lock, flags);
/*
* Raise the apparent loglevel to maximum so that the sysrq header
@@ -542,6 +544,7 @@ void __handle_sysrq(int key, bool check_
console_loglevel = orig_log_level;
}
spin_unlock_irqrestore(&sysrq_key_table_lock, flags);
+oops_in_progress = save_oops_in_progress;
}
void handle_sysrq(int key)
---
kernel/printk.c | 2 1 + 1 - 0 !
1 file changed, 1 insertion(+), 1 deletion(-)
Index: b/kernel/printk.c
===================================================================
--- a/kernel/printk.c
+++ b/kernel/printk.c
@@ -66,7 +66,7 @@ int console_printk[4] = {
* Low level drivers may need that to know if they can schedule in
* their unblank() callback or not. So let's export it.
*/
-int oops_in_progress;
+int oops_in_progress = 1;
EXPORT_SYMBOL(oops_in_progress);
/*
