Re: [RFT v4] printk: allow increasing the ring buffer depending on the number of CPUs
From: Luis R. Rodriguez
Date: Mon Jun 16 2014 - 14:03:46 EST
On Mon, Jun 16, 2014 at 06:32:53PM +0200, Petr Mládek wrote:
> On Sat 2014-06-14 11:52:45, Luis R. Rodriguez wrote:
> > From: "Luis R. Rodriguez" <mcgrof@xxxxxxxx>
> >
> > The default size of the ring buffer is too small for machines
> > with a large amount of CPUs under heavy load. What ends up
> > happening when debugging is the ring buffer overlaps and chews
> > up old messages making debugging impossible unless the size is
> > passed as a kernel parameter. An idle system upon boot up will
> > on average spew out only about one or two extra lines but where
> > this really matters is on heavy load and that will vary widely
> > depending on the system and environment.
> >
> > There are mechanisms to help increase the kernel ring buffer
> > for tracing through debugfs, and those interfaces even allow growing
> > the kernel ring buffer per CPU. We also have a static value which
> > can be passed upon boot. Relying on debugfs however is not ideal
> > for production, and relying on the value passed upon bootup is
> > can only used *after* an issue has creeped up. Instead of being
> > reactive this adds a proactive measure which lets you scale the
> > amount of contributions you'd expect to the kernel ring buffer
> > under load by each CPU in the worst case scenario.
> >
> > We use num_possible_cpus() to avoid complexities which could be
> > introduced by dynamically changing the ring buffer size at run
> > time, num_possible_cpus() lets us use the upper limit on possible
> > number of CPUs therefore avoiding having to deal with hotplugging
> > CPUs on and off. This introduces the kernel configuration option
> > LOG_CPU_MIN_BUF_SHIFT which is used to specify the maximum amount
> > of contributions to the kernel ring buffer in the worst case before
> > the kernel ring buffer flips over, the size is specified as a power
> > of 2. The total amount of contributions made by each CPU must be
> > greater than half of the default kernel ring buffer size
> > (1 << LOG_BUF_SHIFT bytes) in order to trigger an increase upon
> > bootup. For example if LOG_BUF_SHIFT is 18 (256 KB) you'd require at
> > least 128 KB contributions by other CPUs in order to trigger an
> > increase. With a LOG_CPU_BUF_SHIFT of 12 (4 KB) you'd require at
> > least anything over > 64 possible CPUs to trigger an increase. If
> > you had 128 possible CPUs your kernel buffer size would be:
> >
> > ((1 << 18) + ((128 - 1) * (1 << 12))) / 1024 = 764 KB
> >
> > This value is ignored when "log_buf_len" kernel parameter is used
> > as it forces the exact size of the ring buffer to an expected value.
> >
> > In order to use num_possible_cpus() we need to make this a late call
> > on the init process but that also means we cannot share __init code,
> > the late_setup_log_buf() provided is therefore very similar to the
> > setup_log_buf() but just modified slightly, in particular since we're
> > at late boot we can now also use kzalloc().
> >
> > Cc: Michal Hocko <mhocko@xxxxxxx>
> > Cc: Petr Mladek <pmladek@xxxxxxx>
> > Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
> > Cc: Joe Perches <joe@xxxxxxxxxxx>
> > Cc: Arun KS <arunks.linux@xxxxxxxxx>
> > Cc: Kees Cook <keescook@xxxxxxxxxxxx>
> > Cc: Davidlohr Bueso <davidlohr@xxxxxx>
> > Cc: Chris Metcalf <cmetcalf@xxxxxxxxxx>
> > Cc: linux-kernel@xxxxxxxxxxxxxxx
> > Signed-off-by: Luis R. Rodriguez <mcgrof@xxxxxxxx>
> > ---
> >
> > This doesn't kfree() at halt but do we need to do it ?
> >
> > Documentation/kernel-parameters.txt | 8 ++++--
> > include/linux/printk.h | 5 ++++
> > init/Kconfig | 45 +++++++++++++++++++++++++++++++-
> > init/main.c | 1 +
> > kernel/printk/printk.c | 51 +++++++++++++++++++++++++++++++++++++
> > 5 files changed, 107 insertions(+), 3 deletions(-)
> >
> > diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
> > index 6eaa9cd..229d031 100644
> > --- a/Documentation/kernel-parameters.txt
> > +++ b/Documentation/kernel-parameters.txt
> > @@ -1685,8 +1685,12 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
> > 7 (KERN_DEBUG) debug-level messages
> >
> > log_buf_len=n[KMG] Sets the size of the printk ring buffer,
> > - in bytes. n must be a power of two. The default
> > - size is set in the kernel config file.
> > + in bytes. n must be a power of two and greater
> > + than the minimal size. The minimal size is defined
> > + by LOG_BUF_SHIFT kernel config parameter. There is
> > + also CONFIG_LOG_CPU_MIN_BUF_SHIFT config parameter
> > + that allows to increase the default size depending on
> > + the number of CPUs. See init/Kconfig for more details.
> >
> > logo.nologo [FB] Disables display of the built-in Linux logo.
> > This may be used to provide more screen space for
> > diff --git a/include/linux/printk.h b/include/linux/printk.h
> > index 319ff7e..315a0b7 100644
> > --- a/include/linux/printk.h
> > +++ b/include/linux/printk.h
> > @@ -161,6 +161,7 @@ extern void wake_up_klogd(void);
> >
> > void log_buf_kexec_setup(void);
> > void __init setup_log_buf(int early);
> > +void late_setup_log_buf(void);
> > void dump_stack_set_arch_desc(const char *fmt, ...);
> > void dump_stack_print_info(const char *log_lvl);
> > void show_regs_print_info(const char *log_lvl);
> > @@ -202,6 +203,10 @@ static inline void setup_log_buf(int early)
> > {
> > }
> >
> > +static inline void late_setup_log_buf(void)
> > +{
> > +}
> > +
> > static inline void dump_stack_set_arch_desc(const char *fmt, ...)
> > {
> > }
> > diff --git a/init/Kconfig b/init/Kconfig
> > index 9d76b99..f0607d0 100644
> > --- a/init/Kconfig
> > +++ b/init/Kconfig
> > @@ -807,7 +807,11 @@ config LOG_BUF_SHIFT
> > range 12 21
> > default 17
> > help
> > - Select kernel log buffer size as a power of 2.
> > + Select the minimal kernel log buffer size as a power of 2.
> > + The final size is affected by LOG_CPU_MIN_BUF_SHIFT config
> > + parameter, see below. Any higher size also might be forced
> > + by "log_buf_len" boot parameter.
> > +
> > Examples:
> > 17 => 128 KB
> > 16 => 64 KB
> > @@ -816,6 +820,45 @@ config LOG_BUF_SHIFT
> > 13 => 8 KB
> > 12 => 4 KB
> >
> > +config LOG_CPU_MIN_BUF_SHIFT
> > + int "CPU kernel log buffer size contribution (13 => 8 KB, 17 => 128KB)"
> > + range 0 21
> > + default 12
> > + depends on SMP
> > + depends on !BASE_SMALL
> > + help
> > + The kernel ring buffer will get additional data logged onto it
> > + when multiple CPUs are supported. Typically the contributions are
> > + only a few lines when idle however under under load this can vary
> > + and in the worst case it can mean losing logging information. You
> > + can use this to set the maximum expected mount of amount of logging
> > + contribution under load by each CPU in the worst case scenario, as
> > + a power of 2. The total amount of contributions made by each CPU
> > + must be greater than half of the default kernel ring buffer size
> > + (1 << LOG_BUF_SHIFT bytes) in order to trigger an increase upon
> > + bootup. For example if LOG_BUF_SHIFT is 18 (256 KB) you're require
> > + at least 128 KB contributions by other CPUs in order to trigger
> > + an increase. With a LOG_CPU_BUF_SHIFT of 12 (4 KB) you'd require
> > + at least anything over > 64 possible CPUs to trigger an increase.
> > + If you had 128 possible CPUs your kernel buffer size would be:
> > +
> > + ((1 << 18) + ((128 - 1) * (1 << 12))) / 1024 = 764 KB
> > +
> > + This value is ignored when "log_buf_len" kernel parameter is used
> > + as it forces the exact size of the ring buffer to an expected value.
> > +
> > + The number of possible CPUs is used for this computation ignoring
> > + hotplugging making the compuation optimal for the the worst case
> > + scenerio while allowing a simple algorithm to be used from bootup.
> > +
> > + Examples shift values and their meaning:
> > + 17 => 128 KB for each CPU
> > + 16 => 64 KB for each CPU
> > + 15 => 32 KB for each CPU
> > + 14 => 16 KB for each CPU
> > + 13 => 8 KB for each CPU
> > + 12 => 4 KB for each CPU
> > +
> > #
> > # Architectures with an unreliable sched_clock() should select this:
> > #
> > diff --git a/init/main.c b/init/main.c
> > index e8ae1fe..8b472c6 100644
> > --- a/init/main.c
> > +++ b/init/main.c
> > @@ -940,6 +940,7 @@ static int __ref kernel_init(void *unused)
> > numa_default_policy();
> >
> > flush_delayed_fput();
> > + late_setup_log_buf();
> >
> > if (ramdisk_execute_command) {
> > ret = run_init_process(ramdisk_execute_command);
> > diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> > index ea2d5f6..9cf919f 100644
> > --- a/kernel/printk/printk.c
> > +++ b/kernel/printk/printk.c
> > @@ -266,6 +266,7 @@ static u32 clear_idx;
> > #define LOG_ALIGN __alignof__(struct printk_log)
> > #endif
> > #define __LOG_BUF_LEN (1 << CONFIG_LOG_BUF_SHIFT)
> > +#define __LOG_CPU_MIN_BUF_LEN (1 << CONFIG_LOG_CPU_MIN_BUF_SHIFT)
> > static char __log_buf[__LOG_BUF_LEN] __aligned(LOG_ALIGN);
> > static char *log_buf = __log_buf;
> > static u32 log_buf_len = __LOG_BUF_LEN;
> > @@ -842,6 +843,12 @@ static int __init log_buf_len_setup(char *str)
> > }
> > early_param("log_buf_len", log_buf_len_setup);
> >
> > +/*
> > + * The log buffer is by default set using static memory unless
> > + * you either set the log_buf_len=n kernel parameter or use a
> > + * LOG_CPU_MIN_BUF_SHIFT that triggers a required increase
> > + * via late_setup_log_buf().
> > + */
> > void __init setup_log_buf(int early)
> > {
> > unsigned long flags;
> > @@ -877,6 +884,50 @@ void __init setup_log_buf(int early)
> > free, (free * 100) / __LOG_BUF_LEN);
> > }
> >
> > +/*
> > + * If you set log_buf_len=n kernel parameter LOG_CPU_MIN_BUF_SHIFT will
> > + * be ignored. LOG_CPU_MIN_BUF_SHIFT is a proactive measure for large
> > + * systems. With a LOG_BUF_SHIFT of 18 and LOG_CPU_MIN_BUF_SHIFT 12 at
> > + * we'd require more than 64 CPUs to trigger an increase from the
> > + * default. Since we need num_possible_cpus() we do this allocation
> > + * later in the boot process.
> > + */
> > +void late_setup_log_buf(void)
> > +{
> > + unsigned long flags;
> > + char *new_log_buf;
> > + int free;
> > + int cpu_extra = (num_possible_cpus() - 1) * __LOG_CPU_MIN_BUF_LEN;
> > + size_t new_log_size;
> > +
> > + /* The log_buf_len=n kernel parameter was passed */
> > + if (log_buf_len > __LOG_BUF_LEN)
> > + return;
> > +
> > + /* Extra CPU contribution penalty not incurred */
> > + if (cpu_extra < __LOG_BUF_LEN / 2)
> > + return;
> > +
> > + new_log_size = __LOG_BUF_LEN + cpu_extra;
> > + new_log_buf = kzalloc(new_log_size, GFP_KERNEL);
> > +
> > + if (unlikely(!new_log_buf)) {
> > + pr_err("log_buf_len: %ld bytes not available\n", new_log_size);
> > + return;
> > + }
> > +
> > + raw_spin_lock_irqsave(&logbuf_lock, flags);
> > + log_buf_len = new_log_size;
> > + log_buf = new_log_buf;
> > + free = __LOG_BUF_LEN - log_next_idx;
>
> I wonder if the ring buffer could be already rotated at this late boot
> stage.
I was a bit concerned over this so I did more digging just now. The earliest
num_possible_cpus() should have been available to us is right after
setup_arch() which is where all archs deal with prefilling the possible CPU map
with set_cpu_possible() for each detected valid possible CPU. For example on
x86 and ia64 this is done via prefill_possible_map(). Note that the current
placement of setup_log_buf() is after setup_arch() *but* there is a special
excemption for x86 -- its the only architecture that makes use of calling
calls setup_log_buf(1) for an early setup on its own setup_arch() before
it sets the number of possible CPU map via prefill_possible_map(). This would
explain why I ran into the issue with the original patch. Even when early
is used the later non-early call will reallocate the buffer for x86 though
so I think I can fix the original patch then to only account for
num_possible_cpus() if the call is not early -- and I just document the
requirement on setup_arch() on setup_log_buf().
This should take care of the concern you've expressed, I'll respin.
> This "free" space computation is valid only when
> "log_first_idx" is still zero. As simple solution would be to use:
>
> int free = 0;
>
> if (log_first_idx == 0)
> free = __LOG_BUF_LEN - log_next_idx;
OK thanks, will amend, but now that I think we can move this back to __init
this should not be an issue and we can share all the code.
> > + memcpy(log_buf, __log_buf, __LOG_BUF_LEN);
> > + raw_spin_unlock_irqrestore(&logbuf_lock, flags);
> > +
> > + pr_info("big CPU log_buf_len: %d\n", log_buf_len);
> > + pr_info("big CPU early log buf free: %d(%d%%)\n",
> > + free, (free * 100) / __LOG_BUF_LEN);
> > +}
> > +
> > static bool __read_mostly ignore_loglevel;
> >
> > static int __init ignore_loglevel_setup(char *str)
>
> Otherwise, it looks fine to me. I wanted to do some testing but I have
> got stucked by some other problems. I hope that I'll find some time
> tomorrow.
Thanks, I'll respin shortly.
Luis
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/