| Message ID | 20230809100754.3094517-3-shikemeng@huaweicloud.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [1/2] mm/page_alloc: remove track of active PCP lists range in bulk free | expand |
Hi Kemeng, Since I am discussing the other patch in this series, I might just commend on this one as well. On Wed, Aug 09, 2023 at 06:07:54PM +0800, Kemeng Shi wrote: > We get batch from pcp and just pass it to nr_pcp_free immediately. Get > batch from pcp inside nr_pcp_free to remove unnecessary parameter batch. > > Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com> > --- > mm/page_alloc.c | 8 +++----- > 1 file changed, 3 insertions(+), 5 deletions(-) > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c > index 1ddcb2707d05..bb1d14e806ad 100644 > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -2376,10 +2376,10 @@ static bool free_unref_page_prepare(struct page *page, unsigned long pfn, > return true; > } > > -static int nr_pcp_free(struct per_cpu_pages *pcp, int high, int batch, > - bool free_high) > +static int nr_pcp_free(struct per_cpu_pages *pcp, int high, bool free_high) > { > int min_nr_free, max_nr_free; > + int batch = READ_ONCE(pcp->batch); Because nr_pcp_free is static and has only one caller. This function gets inlined at the caller's side. I verified that on X86_64 compiled code. So this fix in my opinion is not worthwhile to fix. It will produce the same machine code. One minor side effect is that it will hide the commit under it in "git blame". Chris
on 8/16/2023 1:46 AM, Chris Li wrote: > Hi Kemeng, > > Since I am discussing the other patch in this series, I might just commend on this one > as well. > > On Wed, Aug 09, 2023 at 06:07:54PM +0800, Kemeng Shi wrote: >> We get batch from pcp and just pass it to nr_pcp_free immediately. Get >> batch from pcp inside nr_pcp_free to remove unnecessary parameter batch. >> >> Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com> >> --- >> mm/page_alloc.c | 8 +++----- >> 1 file changed, 3 insertions(+), 5 deletions(-) >> >> diff --git a/mm/page_alloc.c b/mm/page_alloc.c >> index 1ddcb2707d05..bb1d14e806ad 100644 >> --- a/mm/page_alloc.c >> +++ b/mm/page_alloc.c >> @@ -2376,10 +2376,10 @@ static bool free_unref_page_prepare(struct page *page, unsigned long pfn, >> return true; >> } >> >> -static int nr_pcp_free(struct per_cpu_pages *pcp, int high, int batch, >> - bool free_high) >> +static int nr_pcp_free(struct per_cpu_pages *pcp, int high, bool free_high) >> { >> int min_nr_free, max_nr_free; >> + int batch = READ_ONCE(pcp->batch); > > Because nr_pcp_free is static and has only one caller. This function gets inlined > at the caller's side. I verified that on X86_64 compiled code. > > So this fix in my opinion is not worthwhile to fix. It will produce the same > machine code. One minor side effect is that it will hide the commit under it > in "git blame". > Hi Chris, thank for the reply. Except to reduce argument to pass, this patch also tries make code look little cleaner. I think it's always better to reduce variable scope and keep relevant code tight. In this case, we know batch is from per_cpu_pages during reading nr_pcp_free alone rather than search caller to find it out. And more callers of nr_pcp_free in fulture is free from pass pcp->batch. And so on. Anyway, this patch definely gains a little without lost in my opinion.:) With it makes sense to you. > Chris >
diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 1ddcb2707d05..bb1d14e806ad 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -2376,10 +2376,10 @@ static bool free_unref_page_prepare(struct page *page, unsigned long pfn, return true; } -static int nr_pcp_free(struct per_cpu_pages *pcp, int high, int batch, - bool free_high) +static int nr_pcp_free(struct per_cpu_pages *pcp, int high, bool free_high) { int min_nr_free, max_nr_free; + int batch = READ_ONCE(pcp->batch); /* Free everything if batch freeing high-order pages. */ if (unlikely(free_high)) @@ -2446,9 +2446,7 @@ static void free_unref_page_commit(struct zone *zone, struct per_cpu_pages *pcp, high = nr_pcp_high(pcp, zone, free_high); if (pcp->count >= high) { - int batch = READ_ONCE(pcp->batch); - - free_pcppages_bulk(zone, nr_pcp_free(pcp, high, batch, free_high), pcp, pindex); + free_pcppages_bulk(zone, nr_pcp_free(pcp, high, free_high), pcp, pindex); } }
We get batch from pcp and just pass it to nr_pcp_free immediately. Get batch from pcp inside nr_pcp_free to remove unnecessary parameter batch. Signed-off-by: Kemeng Shi <shikemeng@huaweicloud.com> --- mm/page_alloc.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-)