mm/damon: Fix some little bugs in using DAMON
Subsystems (2)
v2 cover mm/damon: Fix some little bugs in using DAMON
V1 -> V2
Fix patch1 compile problem.
Add review by SeongJae Park in patch2
V1:
https://lkml.org/lkml/2021/11/20/191
Xin Hao (2):
mm/damon/dbgfs: Modify Damon dbfs interface dependency in Kconfig mm/damon: move damon_rand() definition into damon.h
include/linux/damon.h | 16 ++++++++++++++++ mm/damon/Kconfig | 2 +- mm/damon/core.c | 4 ---- mm/damon/prmtv-common.h | 4 ---- 4 files changed, 17 insertions(+), 9 deletions(-)
--
2.31.0
PATCH v2 1/2 mm/damon/dbgfs: Modify Damon dbfs interface dependency in Kconfig
If you want to support "DAMON_DBGFS" in config file, it only depends on any one of "DAMON_VADDR" and "DAMON_PADDR", and sometimes we just want to use damon virtual address function, but it is unreasonable to include "DAMON_PADDR" in config file which cause the damon/paddr.c be compiled, so there fix it.
--- include/linux/damon.h | 12 ++++++++++++ mm/damon/Kconfig | 2 +- 2 files changed, 13 insertions(+), 1 deletion(-)
include/linux/damon.h +12 -0
@@ -463,11 +463,23 @@ int damon_stop(struct damon_ctx **ctxs, int nr_ctxs);#ifdef CONFIG_DAMON_VADDR void damon_va_set_primitives (struct damon_ctx *ctx);bool damon_va_target_valid (void *t); +#else +static inline void damon_va_set_primitives (struct damon_ctx *ctx) {} +static inline bool damon_va_target_valid (void *t) +{ +return false; +}#endif /* CONFIG_DAMON_VADDR */ #ifdef CONFIG_DAMON_PADDR void damon_pa_set_primitives (struct damon_ctx *ctx);bool damon_pa_target_valid (void *t); +#else +static inline void damon_pa_set_primitives (struct damon_ctx *ctx) {} +static inline bool damon_pa_target_valid (void *t) +{ +return false; +}#endif /* CONFIG_DAMON_PADDR */ #endif /* _DAMON_H */
mm/damon/Kconfig +1 -1
@@ -54,7 +54,7 @@ config DAMON_VADDR_KUNIT_TEST config DAMON_DBGFS bool "DAMON debugfs interface" - depends on DAMON_VADDR && DAMON_PADDR && DEBUG_FS + depends on DAMON_VADDR || DAMON_PADDR && DEBUG_FS help This builds the debugfs interface for DAMON. The user space admins can use the interface for arbitrary data access monitoring.
Signature · 1 line
2.31.0
PATCH v2 2/2 mm/damon: move damon_rand() definition into damon.h
In reply to Xin Haodamon_rand() is called in three files:damon/core.c, damon/ paddr.c, damon/vaddr.c, i think there is no need to redefine this twice, So move it to damon.h will be a good choice.
--- include/linux/damon.h | 4 ++++ mm/damon/core.c | 4 ---- mm/damon/prmtv-common.h | 4 ---- 3 files changed, 4 insertions(+), 8 deletions(-)
include/linux/damon.h +4 -0
@@ -11,12 +11,16 @@#include <linux/mutex.h> #include <linux/time64.h> #include <linux/types.h> +#include <linux/random.h> /* Minimal region size. Every damon_region is aligned by this. */ #define DAMON_MIN_REGION PAGE_SIZE/* Max priority score for DAMON-based operation schemes */ #define DAMOS_MAX_SCORE (99) +/* Get a random number in [l, r) */ +#define damon_rand (l, r) (l + prandom_u32_max(r - l)) +/** * struct damon_addr_range - Represents an address region of [@start, @end). * @start: Start address of the region (inclusive).
mm/damon/core.c +0 -4
@@ -11,7 +11,6 @@#include <linux/delay.h> #include <linux/kthread.h> #include <linux/mm.h> -#include <linux/random.h> #include <linux/slab.h> #include <linux/string.h>
@@ -23,9 +22,6 @@#define DAMON_MIN_REGION 1#endif -/* Get a random number in [l, r) */ -#define damon_rand (l, r) (l + prandom_u32_max(r - l)) -static DEFINE_MUTEX (damon_lock);static int nr_running_ctxs;
mm/damon/prmtv-common.h +0 -4
@@ -6,10 +6,6 @@*/ #include <linux/damon.h> -#include <linux/random.h> - -/* Get a random number in [l, r) */ -#define damon_rand (l, r) (l + prandom_u32_max(r - l))struct page *damon_get_page (unsigned long pfn);
--
2.31.0
SeongJae Park
maintainerRe: [PATCH V2 1/2] mm/damon/dbgfs: Modify Damon dbfs interface dependency in Kconfig
In reply to Xin HaoQuoted 4 lines · On Sun, 21 Nov 2021 22:07:04 +0800 Xin Hao <[email protected]> wrote:
If you want to support "
DAMON_DBGFS" in config file, it only depends on any one of "DAMON_VADDR" and "DAMON_PADDR", and sometimes we just want to use damon virtual address function, but it is unreasonable to include "DAMON_PADDR" in config file which cause thedamon/paddr.cbe compiled, so there fix it.
Seems the above lines are not well wrapped[1].
[1] https://docs.kernel.org/process/submitting-patches.html?highlight=75#the-canonical-patch-format
Quoted 40 lines
Signed-off-by Xin Hao <[email protected]>--- include/linux/damon.h | 12 ++++++++++++ mm/damon/Kconfig | 2 +- 2 files changed, 13 insertions(+), 1 deletion(-)include/linux/damon.h +12
-0@@ -463,11 +463,23 @@ int damon_stop(struct damon_ctx **ctxs, int nr_ctxs);#ifdef CONFIG_DAMON_VADDR void damon_va_set_primitives (struct damon_ctx *ctx);bool damon_va_target_valid (void *t); +#else +static inline void damon_va_set_primitives (struct damon_ctx *ctx) {} +static inline bool damon_va_target_valid (void *t) +{ +return false; +}#endif /* CONFIG_DAMON_VADDR */ #ifdef CONFIG_DAMON_PADDR void damon_pa_set_primitives (struct damon_ctx *ctx);bool damon_pa_target_valid (void *t); +#else +static inline void damon_pa_set_primitives (struct damon_ctx *ctx) {} +static inline bool damon_pa_target_valid (void *t) +{ +return false; +}#endif /* CONFIG_DAMON_PADDR */ #endif /* _DAMON_H */ mm/damon/Kconfig +0
-0
mm/damon/Kconfig config DAMON_VADDR_KUNIT_TEST · line 57
Quoted 5 lines
mm/damon/Kconfig +1
-1@@ -54,7 +54,7 @@ config DAMON_VADDR_KUNIT_TEST config DAMON_DBGFS bool "DAMON debugfs interface" - depends on DAMON_VADDR && DAMON_PADDR && DEBUG_FS + depends on DAMON_VADDR || DAMON_PADDR && DEBUG_FS
Due to the lack of parentheses, this allows enabling DAMON_DBGFS while DEBUG_FS is disabled.
Sorry to say this but this makes me unsure if this patch is sufficiently tested. I am also unsure if this patch is carefully handling all possible corner cases in appropriate ways. For example, on kernels that doesn't having 'CONFIG_DAMON_PADDR=y', this patch would still make
'echo paddr > $debugfs/damon/target_ids' success. Is that what you are intending? And, have you confirmed that will never make any issue, both in kernel space and the user space? Could the behavioral change make the user space confused?
Basically, reducing the dependency is a good idea. But, IMHO, the benefit seems not so big. After all, efficiency was not a first goal of DAMON debugfs interface. That's why it is implemented on debugfs rather than its own specialized/optimized file system. Better way for the efficiency of the user space interface might be inventing such DAMON's own efficient file system, like tracepoints had its interface on debugfs but now uses its own file system, tracefs.
So, unless you make me believe the benefit is big enough and all possible corner cases are well handled and sufficiently tested, I'm sorry but I would not be convinced with this change.
Thanks,
SJ
Quoted 5 lines
mm/damon/Kconfig +0
-0help This builds the debugfs interface for DAMON. The user space admins can use the interface for arbitrary data access monitoring.Signature · 1 line
2.31.0
Re: [PATCH V2 1/2] mm/damon/dbgfs: Modify Damon dbfs interface dependency in Kconfig
Hi Park:
Quoted 57 lines · SeongJae Park
Quoted 4 lines · On Sun, 21 Nov 2021 22:07:04 +0800 Xin Hao <[email protected]> wrote:
If you want to support "
DAMON_DBGFS" in config file, it only depends on any one of "DAMON_VADDR" and "DAMON_PADDR", and sometimes we just want to use damon virtual address function, but it is unreasonable to include "DAMON_PADDR" in config file which cause thedamon/paddr.cbe compiled, so there fix it.Seems the above lines are not well wrapped[1].
[1] https://docs.kernel.org/process/submitting-patches.html?highlight=75#the-canonical-patch-format
Quoted 39 lines
Signed-off-by Xin Hao <[email protected]>--- include/linux/damon.h | 12 ++++++++++++ mm/damon/Kconfig | 2 +- 2 files changed, 13 insertions(+), 1 deletion(-)include/linux/damon.h +12
-0@@ -463,11 +463,23 @@ int damon_stop(struct damon_ctx **ctxs, int nr_ctxs);#ifdef CONFIG_DAMON_VADDR void damon_va_set_primitives (struct damon_ctx *ctx);bool damon_va_target_valid (void *t); +#else +static inline void damon_va_set_primitives (struct damon_ctx *ctx) {} +static inline bool damon_va_target_valid (void *t) +{ +return false; +}#endif /* CONFIG_DAMON_VADDR */ #ifdef CONFIG_DAMON_PADDR void damon_pa_set_primitives (struct damon_ctx *ctx);bool damon_pa_target_valid (void *t); +#else +static inline void damon_pa_set_primitives (struct damon_ctx *ctx) {} +static inline bool damon_pa_target_valid (void *t) +{ +return false; +}#endif /* CONFIG_DAMON_PADDR */ #endif /* _DAMON_H */ mm/damon/Kconfig +0
-0mm/damon/Kconfig config DAMON_VADDR_KUNIT_TEST · line 57
Quoted 5 lines
mm/damon/Kconfig +1
-1@@ -54,7 +54,7 @@ config DAMON_VADDR_KUNIT_TEST config DAMON_DBGFS bool "DAMON debugfs interface" - depends on DAMON_VADDR && DAMON_PADDR && DEBUG_FS + depends on DAMON_VADDR || DAMON_PADDR && DEBUG_FSDue to the lack of parentheses, this allows enabling
DAMON_DBGFSwhileDEBUG_FSis disabled.
Quoted 6 lines
Sorry to say this but this makes me unsure if this patch is sufficiently tested. I am also unsure if this patch is carefully handling all possible corner cases in appropriate ways. For example, on kernels that doesn't having '
CONFIG_DAMON_PADDR=y', this patch would still make
'echo paddr > $debugfs/damon/target_ids' success. Is that what you are intending?
Yes, you are all right, I have not tested as above, sorry, i did not fully consider the impact on the code after doing this,
Please ignore this patch.
Quoted 12 lines
And, have you confirmed that will never make any issue, both in kernel space and the user space? Could the behavioral change make the user space confused?
Basically, reducing the dependency is a good idea. But, IMHO, the benefit seems not so big. After all, efficiency was not a first goal of DAMON debugfs interface. That's why it is implemented on debugfs rather than its own specialized/optimized file system. Better way for the efficiency of the user space interface might be inventing such DAMON's own efficient file system, like tracepoints had its interface on debugfs but now uses its own file system, tracefs.
Quoted 3 lines
So, unless you make me believe the benefit is big enough and all possible corner cases are well handled and sufficiently tested, I'm sorry but I would not be convinced with this change.
I think it makes sense to separate PADDR & VADDR, users can freely choose dbfs + vaddr or dbfs + paddr;
Doing this can make the code look like it doesn’t mix together,i will continue to try to separate this part
of the code later.
Quoted 10 lines
Thanks,
SJQuoted 5 lines
mm/damon/Kconfig +0
-0help This builds the debugfs interface for DAMON. The user space admins can use the interface for arbitrary data access monitoring.Signature · 1 line
2.31.0
Signature · 2 lines
Best Regards! Xin Hao