Currently enabling CONFIG_CMA_DEBUG enables DEBUG preprocessor macro.
If DEBUG is defined, it's equivalent to a printk with KERN_DEBUG loglevel
flooding the dmesg buffer with pr_debug prints from mm/cma driver and from
included files as well. This results in excessive amount of CMA logging and
also might distract the debug teams with unrelated KERN_DEBUG prints.One of
the ways engineers currently tackle this problem is by passing loglevel=N
though commandline to suppress KERN_DEBUG messages. This approach can
sometimes become tiresome due to its repetitive nature.
This patch proposes an alternative approach by introducing a simple new
config CONFIG_CMA_ALLOC_DEBUG which only shows the cma bit allocation
status in case of cma failure and do not enable DEBUG preprocessor macro
from CONFIG_CMA_DEBUG avoiding excessive CMA logging from pr_debug.
Engineers and tech teams seeking only for bitmap status in case of cma
failure can use this simple config instead of worrying about changing
the loglevel or trying other similar workarounds.
Signed-off-by: Bibek Kumar Patro <[email protected]>
---
mm/Kconfig | 11 +++++++++++
mm/cma.c | 2 +-
2 files changed, 12 insertions(+), 1 deletion(-)
diff --git a/mm/Kconfig b/mm/Kconfig
index 09130434e30d..ad2a20576a65 100644
--- a/mm/Kconfig
+++ b/mm/Kconfig
@@ -891,12 +891,23 @@ config CMA
config CMA_DEBUG
bool "CMA debug messages (DEVELOPMENT)"
depends on DEBUG_KERNEL && CMA
+ select CMA_ALLOC_DEBUG
help
Turns on debug messages in CMA. This produces KERN_DEBUG
messages for every CMA call as well as various messages while
processing calls such as dma_alloc_from_contiguous().
This option does not affect warning and error messages.
+config CMA_ALLOC_DEBUG
+ bool "CMA bitmap status upon failure"
+ depends on CMA
+ help
+ Turns on bitmap dump in case of cma failures. This prints
+ current status of CMA bit upon failures and will appear
+ when cma_alloc() return no_warn argument as false and won't
+ flood the dmesg buffer with unrelated KERN_DEBUG prints. No
+ need to explicitly enable this if CMA_DEBUG is enabled.
+
config CMA_DEBUGFS
bool "CMA debugfs interface"
depends on CMA && DEBUG_FS
diff --git a/mm/cma.c b/mm/cma.c
index a4cfe995e11e..0ead392769b3 100644
--- a/mm/cma.c
+++ b/mm/cma.c
@@ -383,7 +383,7 @@ int __init cma_declare_contiguous_nid(phys_addr_t base,
return ret;
}
-#ifdef CONFIG_CMA_DEBUG
+#ifdef CONFIG_CMA_ALLOC_DEBUG
static void cma_debug_show_areas(struct cma *cma)
{
unsigned long next_zero_bit, next_set_bit, nr_zero;
--
2.17.1
v2 was added by mistake. It's first submission of the patch. Please
ignore the v2.
regards,
Bibek
On 8/9/2023 6:46 PM, Bibek Kumar Patro wrote:
> Currently enabling CONFIG_CMA_DEBUG enables DEBUG preprocessor macro.
> If DEBUG is defined, it's equivalent to a printk with KERN_DEBUG loglevel
> flooding the dmesg buffer with pr_debug prints from mm/cma driver and from
> included files as well. This results in excessive amount of CMA logging and
> also might distract the debug teams with unrelated KERN_DEBUG prints.One of
> the ways engineers currently tackle this problem is by passing loglevel=N
> though commandline to suppress KERN_DEBUG messages. This approach can
> sometimes become tiresome due to its repetitive nature.
> This patch proposes an alternative approach by introducing a simple new
> config CONFIG_CMA_ALLOC_DEBUG which only shows the cma bit allocation
> status in case of cma failure and do not enable DEBUG preprocessor macro
> from CONFIG_CMA_DEBUG avoiding excessive CMA logging from pr_debug.
> Engineers and tech teams seeking only for bitmap status in case of cma
> failure can use this simple config instead of worrying about changing
> the loglevel or trying other similar workarounds.
>
> Signed-off-by: Bibek Kumar Patro <[email protected]>
> ---
> mm/Kconfig | 11 +++++++++++
> mm/cma.c | 2 +-
> 2 files changed, 12 insertions(+), 1 deletion(-)
>
> diff --git a/mm/Kconfig b/mm/Kconfig
> index 09130434e30d..ad2a20576a65 100644
> --- a/mm/Kconfig
> +++ b/mm/Kconfig
> @@ -891,12 +891,23 @@ config CMA
> config CMA_DEBUG
> bool "CMA debug messages (DEVELOPMENT)"
> depends on DEBUG_KERNEL && CMA
> + select CMA_ALLOC_DEBUG
> help
> Turns on debug messages in CMA. This produces KERN_DEBUG
> messages for every CMA call as well as various messages while
> processing calls such as dma_alloc_from_contiguous().
> This option does not affect warning and error messages.
>
> +config CMA_ALLOC_DEBUG
> + bool "CMA bitmap status upon failure"
> + depends on CMA
> + help
> + Turns on bitmap dump in case of cma failures. This prints
> + current status of CMA bit upon failures and will appear
> + when cma_alloc() return no_warn argument as false and won't
> + flood the dmesg buffer with unrelated KERN_DEBUG prints. No
> + need to explicitly enable this if CMA_DEBUG is enabled.
> +
> config CMA_DEBUGFS
> bool "CMA debugfs interface"
> depends on CMA && DEBUG_FS
> diff --git a/mm/cma.c b/mm/cma.c
> index a4cfe995e11e..0ead392769b3 100644
> --- a/mm/cma.c
> +++ b/mm/cma.c
> @@ -383,7 +383,7 @@ int __init cma_declare_contiguous_nid(phys_addr_t base,
> return ret;
> }
>
> -#ifdef CONFIG_CMA_DEBUG
> +#ifdef CONFIG_CMA_ALLOC_DEBUG
> static void cma_debug_show_areas(struct cma *cma)
> {
> unsigned long next_zero_bit, next_set_bit, nr_zero;
> --
> 2.17.1
>
On Wed, 9 Aug 2023 18:46:40 +0530 Bibek Kumar Patro <[email protected]> wrote:
> Currently enabling CONFIG_CMA_DEBUG enables DEBUG preprocessor macro.
> If DEBUG is defined, it's equivalent to a printk with KERN_DEBUG loglevel
> flooding the dmesg buffer with pr_debug prints from mm/cma driver and from
> included files as well. This results in excessive amount of CMA logging and
> also might distract the debug teams with unrelated KERN_DEBUG prints.One of
> the ways engineers currently tackle this problem is by passing loglevel=N
> though commandline to suppress KERN_DEBUG messages. This approach can
> sometimes become tiresome due to its repetitive nature.
> This patch proposes an alternative approach by introducing a simple new
> config CONFIG_CMA_ALLOC_DEBUG which only shows the cma bit allocation
> status in case of cma failure and do not enable DEBUG preprocessor macro
> from CONFIG_CMA_DEBUG avoiding excessive CMA logging from pr_debug.
> Engineers and tech teams seeking only for bitmap status in case of cma
> failure can use this simple config instead of worrying about changing
> the loglevel or trying other similar workarounds.
Would it be better to control this at runtime? With a /proc or /sys tunable?
On 8/10/2023 10:24 PM, Andrew Morton wrote:
> On Wed, 9 Aug 2023 18:46:40 +0530 Bibek Kumar Patro <[email protected]> wrote:
>
>> Currently enabling CONFIG_CMA_DEBUG enables DEBUG preprocessor macro.
>> If DEBUG is defined, it's equivalent to a printk with KERN_DEBUG loglevel
>> flooding the dmesg buffer with pr_debug prints from mm/cma driver and from
>> included files as well. This results in excessive amount of CMA logging and
>> also might distract the debug teams with unrelated KERN_DEBUG prints.One of
>> the ways engineers currently tackle this problem is by passing loglevel=N
>> though commandline to suppress KERN_DEBUG messages. This approach can
>> sometimes become tiresome due to its repetitive nature.
>> This patch proposes an alternative approach by introducing a simple new
>> config CONFIG_CMA_ALLOC_DEBUG which only shows the cma bit allocation
>> status in case of cma failure and do not enable DEBUG preprocessor macro
>> from CONFIG_CMA_DEBUG avoiding excessive CMA logging from pr_debug.
>> Engineers and tech teams seeking only for bitmap status in case of cma
>> failure can use this simple config instead of worrying about changing
>> the loglevel or trying other similar workarounds.
>
> Would it be better to control this at runtime? With a /proc or /sys tunable?
Currently it's being controlled at runtime by changing the
/proc/sys/kernel/printk tunable or through loglevel value in cmdline but
issue faced by engineers in both these approach is these tunable value
would reset every time on reboot and won't retain the set value. So
these approaches are being used as workarounds only as of now.
Also another issue with the earlier CMA_DEBUG config is the text code
size might increase (It might be minuscule sometimes but will happen)
due to all pr_debug in the code.
Bibek,
On Thu, Aug 10, 2023 at 09:54:51AM -0700, Andrew Morton wrote:
> On Wed, 9 Aug 2023 18:46:40 +0530 Bibek Kumar Patro <[email protected]> wrote:
>
> > Currently enabling CONFIG_CMA_DEBUG enables DEBUG preprocessor macro.
> > If DEBUG is defined, it's equivalent to a printk with KERN_DEBUG loglevel
> > flooding the dmesg buffer with pr_debug prints from mm/cma driver and from
> > included files as well. This results in excessive amount of CMA logging and
> > also might distract the debug teams with unrelated KERN_DEBUG prints.One of
> > the ways engineers currently tackle this problem is by passing loglevel=N
> > though commandline to suppress KERN_DEBUG messages. This approach can
> > sometimes become tiresome due to its repetitive nature.
> > This patch proposes an alternative approach by introducing a simple new
> > config CONFIG_CMA_ALLOC_DEBUG which only shows the cma bit allocation
> > status in case of cma failure and do not enable DEBUG preprocessor macro
> > from CONFIG_CMA_DEBUG avoiding excessive CMA logging from pr_debug.
> > Engineers and tech teams seeking only for bitmap status in case of cma
> > failure can use this simple config instead of worrying about changing
> > the loglevel or trying other similar workarounds.
>
> Would it be better to control this at runtime? With a /proc or /sys tunable?
IIUC, Andrew is asking/suggesting that would a runtime knob be useful
here?
For ex:
/proc/sys/vm/cma_dump_on_failure : This needs to be registerd in the
cma.c . grep for "register_sysctl_init" in mm code. We can set this to
true by default in case CMA_DEBUG is enabled so that current users are
not affected.
/sys/kernel/mm/cma/bitmap_dump_on_failure : This needs to stay in
cma_sysfs.c code
Thanks,
Pavan