[PATCH V2 0/2] mm/damon: A few fixup with lru_sort

[PATCH V2 0/2] mm/damon: A few fixup with lru_sort

[Xin Hao]

v1:
https://lore.kernel.org/damon/20220818082538.67825-1-xhao@linux.alibaba.com/


Xin Hao (2):
  mm/damon/lru_sort: Move target memory region check to head of func
  mm/damon/lru_sort: Remove struct of damon_lru_sort_ram_walk_arg

 mm/damon/lru_sort.c | 41 +++++++++++++++++------------------------
 1 file changed, 17 insertions(+), 24 deletions(-)

--
2.27.0

[PATCH V2 1/2] mm/damon/lru_sort: Move target memory region check to head of func

[Xin Hao]

In damon_lru_sort_apply_parameters(), if "monitor_region_start"
and "monitor_region_end" is not a valid physical address range,
There no need to run the remainder codes in it.

Signed-off-by: Xin Hao <xhao@linux.alibaba.com>
---
 mm/damon/lru_sort.c | 13 +++++++------
 1 file changed, 7 insertions(+), 6 deletions(-)

diff --git a/mm/damon/lru_sort.c b/mm/damon/lru_sort.c
index 9de6f00a71c5..ac50dca026f9 100644
--- a/mm/damon/lru_sort.c
+++ b/mm/damon/lru_sort.c
@@ -378,6 +378,13 @@ static int damon_lru_sort_apply_parameters(void)
 	unsigned int hot_thres, cold_thres;
 	int err = 0;
 
+	if (monitor_region_start > monitor_region_end)
+		return -EINVAL;
+	if (!monitor_region_start && !monitor_region_end &&
+			!get_monitoring_region(&monitor_region_start,
+				&monitor_region_end))
+		return -EINVAL;
+
 	err = damon_set_attrs(ctx, sample_interval, aggr_interval, 0,
 			min_nr_regions, max_nr_regions);
 	if (err)
@@ -401,12 +408,6 @@ static int damon_lru_sort_apply_parameters(void)
 		return -ENOMEM;
 	damon_add_scheme(ctx, scheme);
 
-	if (monitor_region_start > monitor_region_end)
-		return -EINVAL;
-	if (!monitor_region_start && !monitor_region_end &&
-			!get_monitoring_region(&monitor_region_start,
-				&monitor_region_end))
-		return -EINVAL;
 	addr_range.start = monitor_region_start;
 	addr_range.end = monitor_region_end;
 	return damon_set_regions(target, &addr_range, 1);
-- 
2.27.0

Re: [PATCH V2 1/2] mm/damon/lru_sort: Move target memory region check to head of func

[SeongJae Park]

Hi Xin,


On Thu, 18 Aug 2022 18:57:31 +0800 Xin Hao <xhao@linux.alibaba.com> wrote:

> In damon_lru_sort_apply_parameters(), if "monitor_region_start"
> and "monitor_region_end" is not a valid physical address range,
> There no need to run the remainder codes in it.

The function, 'damon_lru_sort_apply_parameters()', checks validity of
parameters and construct the DAMON context one by one.  For example,
'damon_set_attrs()' returns an error if the parameters are invalid.  So the
intended flow is,

1. check DAMON attributes parameters,
2. apply DAMON attributes parameters,
3. check scheme parameters,
4. apply scheme parameters,
5. check target region parameters, and
6. apply target region parameters.

Therefore what this patch does is making the target regions validity check to
be done earlier than validity checks of other parameters.  There is no special
reason to check the region earlier than others.  Also, this change makes the
flow of the function a little bit weird in my humble opinion, as the flow will
be

1. check target region parameters,
2. check DAMON attributes parameters,
3. apply DAMON attributes parameters,
4. check scheme parameters,
5. apply scheme parameters, and
6. apply target region parameters.

So I'd argue this patch seems not really needed, sorry.


Thanks,
SJ

[...]

Re: [PATCH V2 1/2] mm/damon/lru_sort: Move target memory region check to head of func

[haoxin]

在 2022/8/19 上午1:11, SeongJae Park 写道:
> Hi Xin,
>
>
> On Thu, 18 Aug 2022 18:57:31 +0800 Xin Hao <xhao@linux.alibaba.com> wrote:
>
>> In damon_lru_sort_apply_parameters(), if "monitor_region_start"
>> and "monitor_region_end" is not a valid physical address range,
>> There no need to run the remainder codes in it.
> The function, 'damon_lru_sort_apply_parameters()', checks validity of
> parameters and construct the DAMON context one by one.  For example,
> 'damon_set_attrs()' returns an error if the parameters are invalid.  So the
> intended flow is,
>
> 1. check DAMON attributes parameters,
> 2. apply DAMON attributes parameters,
> 3. check scheme parameters,
> 4. apply scheme parameters,
> 5. check target region parameters, and
> 6. apply target region parameters.
>
> Therefore what this patch does is making the target regions validity check to
> be done earlier than validity checks of other parameters.  There is no special
> reason to check the region earlier than others.  Also, this change makes the
> flow of the function a little bit weird in my humble opinion, as the flow will
> be
>
> 1. check target region parameters,
> 2. check DAMON attributes parameters,
> 3. apply DAMON attributes parameters,
> 4. check scheme parameters,
> 5. apply scheme parameters, and
> 6. apply target region parameters.

Ok, understand what you mean,   my fix looks ugly,  buy any apply above 
are not not necessary if one of them checks failed, why not check all 
fisrt and then apply them, like this:

1. check target region parameters,

2. check DAMON attributes parameters,

3. check scheme parameters,

>
> So I'd argue this patch seems not really needed, sorry.
>
>
> Thanks,
> SJ
>
> [...]

Re: [PATCH V2 1/2] mm/damon/lru_sort: Move target memory region check to head of func

[SeongJae Park]

Hi Xin,

> 
> 在 2022/8/19 上午1:11, SeongJae Park 写道:
> > Hi Xin,
> >
> >
> > On Thu, 18 Aug 2022 18:57:31 +0800 Xin Hao <xhao@linux.alibaba.com> wrote:
> >
> >> In damon_lru_sort_apply_parameters(), if "monitor_region_start"
> >> and "monitor_region_end" is not a valid physical address range,
> >> There no need to run the remainder codes in it.
> > The function, 'damon_lru_sort_apply_parameters()', checks validity of
> > parameters and construct the DAMON context one by one.  For example,
> > 'damon_set_attrs()' returns an error if the parameters are invalid.  So the
> > intended flow is,
> >
> > 1. check DAMON attributes parameters,
> > 2. apply DAMON attributes parameters,
> > 3. check scheme parameters,
> > 4. apply scheme parameters,
> > 5. check target region parameters, and
> > 6. apply target region parameters.
> >
> > Therefore what this patch does is making the target regions validity check to
> > be done earlier than validity checks of other parameters.  There is no special
> > reason to check the region earlier than others.  Also, this change makes the
> > flow of the function a little bit weird in my humble opinion, as the flow will
> > be
> >
> > 1. check target region parameters,
> > 2. check DAMON attributes parameters,
> > 3. apply DAMON attributes parameters,
> > 4. check scheme parameters,
> > 5. apply scheme parameters, and
> > 6. apply target region parameters.
> 
> Ok, understand what you mean,   my fix looks ugly,  buy any apply above 
> are not not necessary if one of them checks failed, why not check all 
> fisrt and then apply them, like this:
> 
> 1. check target region parameters,
> 
> 2. check DAMON attributes parameters,
> 
> 3. check scheme parameters,

The parameter values could be changed by users after the check, so we should
cache those somewhere anyway.  In other words, we cache those in the DAMON
context.  Therefore I think the above works were not totally waste of the time.
Also, because the parameters applying functions like 'damon_set_attrs()' does
the check and applying of the parameters together, I feel like current flow is
natural.

If I'm missing something, please let me know.


Thanks,
SJ

> 
> >
> > So I'd argue this patch seems not really needed, sorry.
> >
> >
> > Thanks,
> > SJ
> >
> > [...]
> 

Re: [PATCH V2 1/2] mm/damon/lru_sort: Move target memory region check to head of func

[haoxin]

在 2022/8/19 上午10:28, SeongJae Park 写道:
> Hi Xin,
>
>> 在 2022/8/19 上午1:11, SeongJae Park 写道:
>>> Hi Xin,
>>>
>>>
>>> On Thu, 18 Aug 2022 18:57:31 +0800 Xin Hao <xhao@linux.alibaba.com> wrote:
>>>
>>>> In damon_lru_sort_apply_parameters(), if "monitor_region_start"
>>>> and "monitor_region_end" is not a valid physical address range,
>>>> There no need to run the remainder codes in it.
>>> The function, 'damon_lru_sort_apply_parameters()', checks validity of
>>> parameters and construct the DAMON context one by one.  For example,
>>> 'damon_set_attrs()' returns an error if the parameters are invalid.  So the
>>> intended flow is,
>>>
>>> 1. check DAMON attributes parameters,
>>> 2. apply DAMON attributes parameters,
>>> 3. check scheme parameters,
>>> 4. apply scheme parameters,
>>> 5. check target region parameters, and
>>> 6. apply target region parameters.
>>>
>>> Therefore what this patch does is making the target regions validity check to
>>> be done earlier than validity checks of other parameters.  There is no special
>>> reason to check the region earlier than others.  Also, this change makes the
>>> flow of the function a little bit weird in my humble opinion, as the flow will
>>> be
>>>
>>> 1. check target region parameters,
>>> 2. check DAMON attributes parameters,
>>> 3. apply DAMON attributes parameters,
>>> 4. check scheme parameters,
>>> 5. apply scheme parameters, and
>>> 6. apply target region parameters.
>> Ok, understand what you mean,   my fix looks ugly,  buy any apply above
>> are not not necessary if one of them checks failed, why not check all
>> fisrt and then apply them, like this:
>>
>> 1. check target region parameters,
>>
>> 2. check DAMON attributes parameters,
>>
>> 3. check scheme parameters,
> The parameter values could be changed by users after the check, so we should
> cache those somewhere anyway.  In other words, we cache those in the DAMON
> context.  Therefore I think the above works were not totally waste of the time.
> Also, because the parameters applying functions like 'damon_set_attrs()' does
> the check and applying of the parameters together, I feel like current flow is
> natural.

Ok,  Thank you for your detailed  explain, just keep it. but there still 
a problem in damon_lru_sort_apply_parameters

if (!monitor_region_start && !monitor_region_end &&
		!get_monitoring_region(&monitor_region_start,
			&monitor_region_end))

if (!monitor_region_start || !monitor_region_end ||
		!get_monitoring_region(&monitor_region_start,
			&monitor_region_end))

the '&&' should fix to '||',  anyone checks fail, it should return ?






>

Re: [PATCH V2 1/2] mm/damon/lru_sort: Move target memory region check to head of func

[SeongJae Park]

On Fri, 19 Aug 2022 10:47:58 +0800 haoxin <xhao@linux.alibaba.com> wrote:

> 
> 在 2022/8/19 上午10:28, SeongJae Park 写道:
> > Hi Xin,
> >
> >> 在 2022/8/19 上午1:11, SeongJae Park 写道:
> >>> Hi Xin,
> >>>
> >>>
> >>> On Thu, 18 Aug 2022 18:57:31 +0800 Xin Hao <xhao@linux.alibaba.com> wrote:
> >>>
> >>>> In damon_lru_sort_apply_parameters(), if "monitor_region_start"
> >>>> and "monitor_region_end" is not a valid physical address range,
> >>>> There no need to run the remainder codes in it.
> >>> The function, 'damon_lru_sort_apply_parameters()', checks validity of
> >>> parameters and construct the DAMON context one by one.  For example,
> >>> 'damon_set_attrs()' returns an error if the parameters are invalid.  So the
> >>> intended flow is,
> >>>
> >>> 1. check DAMON attributes parameters,
> >>> 2. apply DAMON attributes parameters,
> >>> 3. check scheme parameters,
> >>> 4. apply scheme parameters,
> >>> 5. check target region parameters, and
> >>> 6. apply target region parameters.
> >>>
> >>> Therefore what this patch does is making the target regions validity check to
> >>> be done earlier than validity checks of other parameters.  There is no special
> >>> reason to check the region earlier than others.  Also, this change makes the
> >>> flow of the function a little bit weird in my humble opinion, as the flow will
> >>> be
> >>>
> >>> 1. check target region parameters,
> >>> 2. check DAMON attributes parameters,
> >>> 3. apply DAMON attributes parameters,
> >>> 4. check scheme parameters,
> >>> 5. apply scheme parameters, and
> >>> 6. apply target region parameters.
> >> Ok, understand what you mean,   my fix looks ugly,  buy any apply above
> >> are not not necessary if one of them checks failed, why not check all
> >> fisrt and then apply them, like this:
> >>
> >> 1. check target region parameters,
> >>
> >> 2. check DAMON attributes parameters,
> >>
> >> 3. check scheme parameters,
> > The parameter values could be changed by users after the check, so we should
> > cache those somewhere anyway.  In other words, we cache those in the DAMON
> > context.  Therefore I think the above works were not totally waste of the time.
> > Also, because the parameters applying functions like 'damon_set_attrs()' does
> > the check and applying of the parameters together, I feel like current flow is
> > natural.
> 
> Ok,  Thank you for your detailed  explain, just keep it. but there still 
> a problem in damon_lru_sort_apply_parameters
> 
> if (!monitor_region_start && !monitor_region_end &&
> 		!get_monitoring_region(&monitor_region_start,
> 			&monitor_region_end))
> 
> if (!monitor_region_start || !monitor_region_end ||
> 		!get_monitoring_region(&monitor_region_start,
> 			&monitor_region_end))
> 
> the '&&' should fix to '||',  anyone checks fail, it should return ?

No.  The code is for setting the monitoring region as the biggest System RAM
resource only if the user didn't set both 'monitor_region_start' and
'monitor_region_end'.


Thanks,
SJ

[PATCH V2 2/2] mm/damon/lru_sort: Remove struct of damon_lru_sort_ram_walk_arg

[Xin Hao]

The struct of 'damon_lru_sort_ram_walk_arg' is the same with struct of
'damon_addr_range', so, there no need to redefine it, just use struct of
'damon_addr_range' instead.

Signed-off-by: Xin Hao <xhao@linux.alibaba.com>
---
 mm/damon/lru_sort.c | 34 +++++++++++++---------------------
 1 file changed, 13 insertions(+), 21 deletions(-)

diff --git a/mm/damon/lru_sort.c b/mm/damon/lru_sort.c
index ac50dca026f9..a3674532fa67 100644
--- a/mm/damon/lru_sort.c
+++ b/mm/damon/lru_sort.c
@@ -257,18 +257,13 @@ module_param(nr_cold_quota_exceeds, ulong, 0400);
 static struct damon_ctx *ctx;
 static struct damon_target *target;

-struct damon_lru_sort_ram_walk_arg {
-	unsigned long start;
-	unsigned long end;
-};
-
 static int walk_system_ram(struct resource *res, void *arg)
 {
-	struct damon_lru_sort_ram_walk_arg *a = arg;
+	struct damon_addr_range *r = arg;

-	if (a->end - a->start < resource_size(res)) {
-		a->start = res->start;
-		a->end = res->end;
+	if (r->end - r->start < resource_size(res)) {
+		r->start = res->start;
+		r->end = res->end;
 	}
 	return 0;
 }
@@ -277,16 +272,12 @@ static int walk_system_ram(struct resource *res, void *arg)
  * Find biggest 'System RAM' resource and store its start and end address in
  * @start and @end, respectively.  If no System RAM is found, returns false.
  */
-static bool get_monitoring_region(unsigned long *start, unsigned long *end)
+static bool get_monitoring_region(struct damon_addr_range *range)
 {
-	struct damon_lru_sort_ram_walk_arg arg = {};
-
-	walk_system_ram_res(0, ULONG_MAX, &arg, walk_system_ram);
-	if (arg.end <= arg.start)
+	walk_system_ram_res(0, ULONG_MAX, range, walk_system_ram);
+	if (range->end <= range->start)
 		return false;

-	*start = arg.start;
-	*end = arg.end;
 	return true;
 }

@@ -380,9 +371,12 @@ static int damon_lru_sort_apply_parameters(void)

 	if (monitor_region_start > monitor_region_end)
 		return -EINVAL;
-	if (!monitor_region_start && !monitor_region_end &&
-			!get_monitoring_region(&monitor_region_start,
-				&monitor_region_end))
+	if (!monitor_region_end)
+		return -EINVAL;
+
+	addr_range.start = monitor_region_start;
+	addr_range.end = monitor_region_end;
+	if (!get_monitoring_region(&addr_range))
 		return -EINVAL;

 	err = damon_set_attrs(ctx, sample_interval, aggr_interval, 0,
@@ -408,8 +402,6 @@ static int damon_lru_sort_apply_parameters(void)
 		return -ENOMEM;
 	damon_add_scheme(ctx, scheme);

-	addr_range.start = monitor_region_start;
-	addr_range.end = monitor_region_end;
 	return damon_set_regions(target, &addr_range, 1);
 }

--
2.27.0

Re: [PATCH V2 2/2] mm/damon/lru_sort: Remove struct of damon_lru_sort_ram_walk_arg

[SeongJae Park]

Hi Xin,

On Thu, 18 Aug 2022 18:57:32 +0800 Xin Hao <xhao@linux.alibaba.com> wrote:

> The struct of 'damon_lru_sort_ram_walk_arg' is the same with struct of
> 'damon_addr_range', so, there no need to redefine it, just use struct of
> 'damon_addr_range' instead.

Reducing code is always good, thanks.  However, I think the type of the 'start'
and 'end' fields of 'struct damon_addr_range' might be changed in a future.
It's very unlikely, though.  Also, we might add some more fields to the struct
in a future.  After all, the purpose of 'struct damon_addr_range' is not saving
the 'start' and 'end' fields of 'struct resource'.  I'd like to avoid making
any possible dependency here, sorry.


Thanks,
SJ

[...]

Re: [PATCH V2 2/2] mm/damon/lru_sort: Remove struct of damon_lru_sort_ram_walk_arg

[haoxin]

在 2022/8/19 上午1:23, SeongJae Park 写道:
> Hi Xin,
>
> On Thu, 18 Aug 2022 18:57:32 +0800 Xin Hao <xhao@linux.alibaba.com> wrote:
>
>> The struct of 'damon_lru_sort_ram_walk_arg' is the same with struct of
>> 'damon_addr_range', so, there no need to redefine it, just use struct of
>> 'damon_addr_range' instead.
> Reducing code is always good, thanks.  However, I think the type of the 'start'
> and 'end' fields of 'struct damon_addr_range' might be changed in a future.
OK,  get it.
> It's very unlikely, though.  Also, we might add some more fields to the struct
> in a future.  After all, the purpose of 'struct damon_addr_range' is not saving
> the 'start' and 'end' fields of 'struct resource'.  I'd like to avoid making
> any possible dependency here, sorry.
>
>
> Thanks,
> SJ
>
> [...]