mm/page_alloc: free order-0 pages through PCP in page_frag_free()

We have multiple places of freeing a page, most of them doing similar things and a common function can be used to reduce code duplicate.

It also avoids bug fixed in one function and left in another.

mm/page_alloc.c +0 -0

index 91a9a6af41a2..2b330296e92a 100644

We have multiple places of freeing a page, most of them doing similar things and a common function can be used to reduce code duplicate.

It also avoids bug fixed in one function but left in another.

Signed-off-by Aaron Lu <[email protected]>

---
v2: move comments close to code as suggested by Dave.

 mm/page_alloc.c | 36 ++++++++++++++++--------------------
 1 file changed, 16 insertions(+), 20 deletions(-)

mm/page_alloc.c +16 -20

index 91a9a6af41a2..4faf6b7bf225 100644
@@ -4425,9 +4425,17 @@ unsigned long get_zeroed_page(gfp_t gfp_mask)
 }
 EXPORT_SYMBOL(get_zeroed_page);
 
-void __free_pages(struct page *page, unsigned int order)
+static inline void free_the_page(struct page *page, unsigned int order, int nr)
 {
-	if (put_page_testzero(page)) {
+	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
+
+	/*
+	 * Free a page by reducing its ref count by @nr.
+	 * If its refcount reaches 0, then according to its order:
+	 * order0: send to PCP;
+	 * high order: directly send to Buddy.
+	 */
+	if (page_ref_sub_and_test(page, nr)) {
 		if (order == 0)
 			free_unref_page(page);
 		else

@@ -4435,6 +4443,10 @@ void __free_pages(struct page *page, unsigned int order)
 	}
 }
 
+void __free_pages(struct page *page, unsigned int order)
+{
+	free_the_page(page, order, 1);
+}
 EXPORT_SYMBOL(__free_pages);
 
 void free_pages(unsigned long addr, unsigned int order)

@@ -4481,16 +4493,7 @@ static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
 
 void __page_frag_cache_drain(struct page *page, unsigned int count)
 {
-	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
-
-	if (page_ref_sub_and_test(page, count)) {
-		unsigned int order = compound_order(page);
-
-		if (order == 0)
-			free_unref_page(page);
-		else
-			__free_pages_ok(page, order);
-	}
+	free_the_page(page, compound_order(page), count);
 }
 EXPORT_SYMBOL(__page_frag_cache_drain);
 

@@ -4555,14 +4558,7 @@ void page_frag_free(void *addr)
 {
 	struct page *page = virt_to_head_page(addr);
 
-	if (unlikely(put_page_testzero(page))) {
-		unsigned int order = compound_order(page);
-
-		if (order == 0)
-			free_unref_page(page);
-		else
-			__free_pages_ok(page, order);
-	}
+	free_the_page(page, compound_order(page), 1);
 }
 EXPORT_SYMBOL(page_frag_free);
 

Quoted 6 lines · Aaron Lu

We have multiple places of freeing a page, most of them doing similar things and a common function can be used to reduce code duplicate.

It also avoids bug fixed in one function but left in another.

Signed-off-by Aaron Lu <[email protected]>

Acked-by Vlastimil Babka <[email protected]>

I assume there's no arch that would run page_ref_sub_and_test(1) slower than put_page_testzero(), for the critical __free_pages() case?

Quoted 76 lines

---
v2: move comments close to code as suggested by Dave.

 mm/page_alloc.c | 36 ++++++++++++++++--------------------
 1 file changed, 16 insertions(+), 20 deletions(-)

mm/page_alloc.c +16 -20

index 91a9a6af41a2..4faf6b7bf225 100644
@@ -4425,9 +4425,17 @@ unsigned long get_zeroed_page(gfp_t gfp_mask)
 }
 EXPORT_SYMBOL(get_zeroed_page);
 
-void __free_pages(struct page *page, unsigned int order)
+static inline void free_the_page(struct page *page, unsigned int order, int nr)
 {
-	if (put_page_testzero(page)) {
+	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
+
+	/*
+	 * Free a page by reducing its ref count by @nr.
+	 * If its refcount reaches 0, then according to its order:
+	 * order0: send to PCP;
+	 * high order: directly send to Buddy.
+	 */
+	if (page_ref_sub_and_test(page, nr)) {
 		if (order == 0)
 			free_unref_page(page);
 		else

@@ -4435,6 +4443,10 @@ void __free_pages(struct page *page, unsigned int order)
 	}
 }
 
+void __free_pages(struct page *page, unsigned int order)
+{
+	free_the_page(page, order, 1);
+}
 EXPORT_SYMBOL(__free_pages);
 
 void free_pages(unsigned long addr, unsigned int order)

@@ -4481,16 +4493,7 @@ static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
 
 void __page_frag_cache_drain(struct page *page, unsigned int count)
 {
-	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
-
-	if (page_ref_sub_and_test(page, count)) {
-		unsigned int order = compound_order(page);
-
-		if (order == 0)
-			free_unref_page(page);
-		else
-			__free_pages_ok(page, order);
-	}
+	free_the_page(page, compound_order(page), count);
 }
 EXPORT_SYMBOL(__page_frag_cache_drain);
 

@@ -4555,14 +4558,7 @@ void page_frag_free(void *addr)
 {
 	struct page *page = virt_to_head_page(addr);
 
-	if (unlikely(put_page_testzero(page))) {
-		unsigned int order = compound_order(page);
-
-		if (order == 0)
-			free_unref_page(page);
-		else
-			__free_pages_ok(page, order);
-	}
+	free_the_page(page, compound_order(page), 1);
 }
 EXPORT_SYMBOL(page_frag_free);
 

Quoted 9 lines · Vlastimil Babka

Quoted 6 lines · Aaron Lu

We have multiple places of freeing a page, most of them doing similar things and a common function can be used to reduce code duplicate.

It also avoids bug fixed in one function but left in another.

Signed-off-by Aaron Lu <[email protected]>

Acked-by Vlastimil Babka <[email protected]>

Thanks.

Quoted 2 lines

I assume there's no arch that would run page_ref_sub_and_test(1) slower than put_page_testzero(), for the critical __free_pages() case?

Good question.

I followed the non-arch specific calls and found that:
page_ref_sub_and_test() ends up calling atomic_sub_return(i, v) while put_page_testzero() ends up calling atomic_sub_return(1, v). So they should be same for archs that do not have their own implementations.

Back to your question: I don't know either.
If this is deemed unsafe, we can probably keep the ref modify part in their original functions and only take the free part into a common function.

Regards,
Aaron

Quoted 77 lines

Quoted 76 lines

---
v2: move comments close to code as suggested by Dave.

 mm/page_alloc.c | 36 ++++++++++++++++--------------------
 1 file changed, 16 insertions(+), 20 deletions(-)

mm/page_alloc.c +16 -20

index 91a9a6af41a2..4faf6b7bf225 100644
@@ -4425,9 +4425,17 @@ unsigned long get_zeroed_page(gfp_t gfp_mask)
 }
 EXPORT_SYMBOL(get_zeroed_page);
 
-void __free_pages(struct page *page, unsigned int order)
+static inline void free_the_page(struct page *page, unsigned int order, int nr)
 {
-	if (put_page_testzero(page)) {
+	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
+
+	/*
+	 * Free a page by reducing its ref count by @nr.
+	 * If its refcount reaches 0, then according to its order:
+	 * order0: send to PCP;
+	 * high order: directly send to Buddy.
+	 */
+	if (page_ref_sub_and_test(page, nr)) {
 		if (order == 0)
 			free_unref_page(page);
 		else

@@ -4435,6 +4443,10 @@ void __free_pages(struct page *page, unsigned int order)
 	}
 }
 
+void __free_pages(struct page *page, unsigned int order)
+{
+	free_the_page(page, order, 1);
+}
 EXPORT_SYMBOL(__free_pages);
 
 void free_pages(unsigned long addr, unsigned int order)

@@ -4481,16 +4493,7 @@ static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
 
 void __page_frag_cache_drain(struct page *page, unsigned int count)
 {
-	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
-
-	if (page_ref_sub_and_test(page, count)) {
-		unsigned int order = compound_order(page);
-
-		if (order == 0)
-			free_unref_page(page);
-		else
-			__free_pages_ok(page, order);
-	}
+	free_the_page(page, compound_order(page), count);
 }
 EXPORT_SYMBOL(__page_frag_cache_drain);
 

@@ -4555,14 +4558,7 @@ void page_frag_free(void *addr)
 {
 	struct page *page = virt_to_head_page(addr);
 
-	if (unlikely(put_page_testzero(page))) {
-		unsigned int order = compound_order(page);
-
-		if (order == 0)
-			free_unref_page(page);
-		else
-			__free_pages_ok(page, order);
-	}
+	free_the_page(page, compound_order(page), 1);
 }
 EXPORT_SYMBOL(page_frag_free);
 

Quoted 16 lines · Aaron Lu

Quoted 9 lines · Vlastimil Babka

Quoted 6 lines · Aaron Lu

We have multiple places of freeing a page, most of them doing similar things and a common function can be used to reduce code duplicate.

It also avoids bug fixed in one function but left in another.

Signed-off-by Aaron Lu <[email protected]>

Acked-by Vlastimil Babka <[email protected]>

Thanks.

Quoted 2 lines

I assume there's no arch that would run page_ref_sub_and_test(1) slower than put_page_testzero(), for the critical __free_pages() case?

Quoted 6 lines

Good question.

I followed the non-arch specific calls and found that:
page_ref_sub_and_test() ends up calling atomic_sub_return(i, v) while put_page_testzero() ends up calling atomic_sub_return(1, v). So they should be same for archs that do not have their own implementations.

x86 seems to distinguish between DECL and SUBL, see

arch/x86/include/asm/atomic.h although I could not figure out where does e.g. arch_atomic_dec_and_test become atomic_dec_and_test to override the generic implementation.

I don't know if the CPU e.g. executes DECL faster, but objectively it has one parameter less. Maybe it doesn't matter?

Quoted 4 lines

Back to your question: I don't know either.
If this is deemed unsafe, we can probably keep the ref modify part in their original functions and only take the free part into a common function.

I guess you could also employ if (__builtin_constant_p(nr)) in free_the_page(), but the result will be ugly I guess, and maybe not worth it :)

Quoted 77 lines

Quoted 77 lines

Quoted 76 lines

---
v2: move comments close to code as suggested by Dave.

 mm/page_alloc.c | 36 ++++++++++++++++--------------------
 1 file changed, 16 insertions(+), 20 deletions(-)

mm/page_alloc.c +16 -20

index 91a9a6af41a2..4faf6b7bf225 100644
@@ -4425,9 +4425,17 @@ unsigned long get_zeroed_page(gfp_t gfp_mask)
 }
 EXPORT_SYMBOL(get_zeroed_page);
 
-void __free_pages(struct page *page, unsigned int order)
+static inline void free_the_page(struct page *page, unsigned int order, int nr)
 {
-	if (put_page_testzero(page)) {
+	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
+
+	/*
+	 * Free a page by reducing its ref count by @nr.
+	 * If its refcount reaches 0, then according to its order:
+	 * order0: send to PCP;
+	 * high order: directly send to Buddy.
+	 */
+	if (page_ref_sub_and_test(page, nr)) {
 		if (order == 0)
 			free_unref_page(page);
 		else

@@ -4435,6 +4443,10 @@ void __free_pages(struct page *page, unsigned int order)
 	}
 }
 
+void __free_pages(struct page *page, unsigned int order)
+{
+	free_the_page(page, order, 1);
+}
 EXPORT_SYMBOL(__free_pages);
 
 void free_pages(unsigned long addr, unsigned int order)

@@ -4481,16 +4493,7 @@ static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
 
 void __page_frag_cache_drain(struct page *page, unsigned int count)
 {
-	VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
-
-	if (page_ref_sub_and_test(page, count)) {
-		unsigned int order = compound_order(page);
-
-		if (order == 0)
-			free_unref_page(page);
-		else
-			__free_pages_ok(page, order);
-	}
+	free_the_page(page, compound_order(page), count);
 }
 EXPORT_SYMBOL(__page_frag_cache_drain);
 

@@ -4555,14 +4558,7 @@ void page_frag_free(void *addr)
 {
 	struct page *page = virt_to_head_page(addr);
 
-	if (unlikely(put_page_testzero(page))) {
-		unsigned int order = compound_order(page);
-
-		if (order == 0)
-			free_unref_page(page);
-		else
-			__free_pages_ok(page, order);
-	}
+	free_the_page(page, compound_order(page), 1);
 }
 EXPORT_SYMBOL(page_frag_free);
 

We have multiple places of freeing a page, most of them doing similar things and a common function can be used to reduce code duplicate.

It also avoids bug fixed in one function but left in another.

Signed-off-by Aaron Lu <[email protected]>

---
v3: Vlastimil mentioned the possible performance loss by using
page_ref_sub_and_test(page, 1) for put_page_testzero(page), since we aren't sure so be safe by keeping page ref decreasing code as is, only move freeing page part to a common function.

page_frag_free() calls __free_pages_ok() to free the page back to Buddy. This is OK for high order page, but for order-0 pages, it misses the optimization opportunity of using Per-Cpu-Pages and can cause zone lock contention when called frequently.

Paweł Staszewski recently shared his result of 'how Linux kernel handles normal traffic'[1] and from perf data, Jesper Dangaard Brouer found the lock contention comes from page allocator:

mlx5e_poll_tx_cq
|
--16.34%--napi_consume_skb

             |
             |--12.65%--__free_pages_ok
             |          |
             |           --11.86%--free_one_page
             |                     |
             |                     |--10.10%--queued_spin_lock_slowpath
             |                     |
             |                      --0.65%--_raw_spin_lock
             |
             |--1.55%--page_frag_free
             |
              --1.44%--skb_release_data

Jesper explained how it happened: mlx5 driver RX-page recycle mechanism is not effective in this workload and pages have to go through the page allocator. The lock contention happens during mlx5 DMA TX completion cycle. And the page allocator cannot keep up at these speeds.[2]

I thought that __free_pages_ok() are mostly freeing high order pages and thought this is an lock contention for high order pages but Jesper explained in detail that __free_pages_ok() here are actually freeing order-0 pages because mlx5 is using order-0 pages to satisfy its page pool allocation request.[3]

The free path as pointed out by Jesper is:
skb_free_head()
-> skb_free_frag()

    -> skb_free_frag()
      -> page_frag_free()

And the pages being freed on this path are order-0 pages.

Fix this by doing similar things as in __page_frag_cache_drain() - send the being freed page to PCP if it's an order-0 page, or directly to Buddy if it is a high order page.

With this change, Paweł hasn't noticed lock contention yet in his workload and Jesper has noticed a 7% performance improvement using a micro benchmark and lock contention is gone.

[1]: https://www.spinics.net/lists/netdev/msg531362.html
[2]: https://www.spinics.net/lists/netdev/msg531421.html
[3]: https://www.spinics.net/lists/netdev/msg531556.html

Reported-by Paweł Staszewski <[email protected]>

Analysed-by Jesper Dangaard Brouer <[email protected]>

Signed-off-by Aaron Lu <[email protected]>

---
 mm/page_alloc.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

mm/page_alloc.c +8 -2

index ae31839874b8..91a9a6af41a2 100644
@@ -4555,8 +4555,14 @@ void page_frag_free(void *addr)
 {
 	struct page *page = virt_to_head_page(addr);
 
-	if (unlikely(put_page_testzero(page)))
-		__free_pages_ok(page, compound_order(page));
+	if (unlikely(put_page_testzero(page))) {
+		unsigned int order = compound_order(page);
+
+		if (order == 0)
+			free_unref_page(page);
+		else
+			__free_pages_ok(page, order);
+	}
 }
 EXPORT_SYMBOL(page_frag_free);
 

page_frag_free() calls __free_pages_ok() to free the page back to Buddy. This is OK for high order page, but for order-0 pages, it misses the optimization opportunity of using Per-Cpu-Pages and can cause zone lock contention when called frequently.

[1]: https://www.spinics.net/lists/netdev/msg531362.html
[2]: https://www.spinics.net/lists/netdev/msg531421.html
[3]: https://www.spinics.net/lists/netdev/msg531556.html

Reported-by Paweł Staszewski <[email protected]>

Analysed-by Jesper Dangaard Brouer <[email protected]>

Signed-off-by Aaron Lu <[email protected]>

page_frag_free() calls __free_pages_ok() to free the page back to Buddy. This is OK for high order page, but for order-0 pages, it misses the optimization opportunity of using Per-Cpu-Pages and can cause zone lock contention when called frequently.

Paweł Staszewski recently shared his result of 'how Linux kernel handles normal traffic'[1] and from perf data, Jesper Dangaard Brouer found the lock contention comes from page allocator:

mlx5e_poll_tx_cq
|
--16.34%--napi_consume_skb

             |
             |--12.65%--__free_pages_ok
             |          |
             |           --11.86%--free_one_page
             |                     |
             |                     |--10.10%--queued_spin_lock_slowpath
             |                     |
             |                      --0.65%--_raw_spin_lock
             |
             |--1.55%--page_frag_free
             |
              --1.44%--skb_release_data

Jesper explained how it happened: mlx5 driver RX-page recycle mechanism is not effective in this workload and pages have to go through the page allocator. The lock contention happens during mlx5 DMA TX completion cycle. And the page allocator cannot keep up at these speeds.[2]

I thought that __free_pages_ok() are mostly freeing high order pages and thought this is an lock contention for high order pages but Jesper explained in detail that __free_pages_ok() here are actually freeing order-0 pages because mlx5 is using order-0 pages to satisfy its page pool allocation request.[3]

The free path as pointed out by Jesper is:
skb_free_head()
-> skb_free_frag()

    -> skb_free_frag()

      -> page_frag_free()  

And the pages being freed on this path are order-0 pages.

Fix this by doing similar things as in __page_frag_cache_drain() - send the being freed page to PCP if it's an order-0 page, or directly to Buddy if it is a high order page.

With this change, Paweł hasn't noticed lock contention yet in his workload and Jesper has noticed a 7% performance improvement using a micro benchmark and lock contention is gone.

[1]: https://www.spinics.net/lists/netdev/msg531362.html
[2]: https://www.spinics.net/lists/netdev/msg531421.html
[3]: https://www.spinics.net/lists/netdev/msg531556.html

Reported-by Paweł Staszewski <[email protected]>

Analysed-by Jesper Dangaard Brouer <[email protected]>

Signed-off-by Aaron Lu <[email protected]>

---

 mm/page_alloc.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

mm/page_alloc.c +8 -2

index ae31839874b8..91a9a6af41a2 100644
@@ -4555,8 +4555,14 @@ void page_frag_free(void *addr)
 {
 	struct page *page = virt_to_head_page(addr);
 
-	if (unlikely(put_page_testzero(page)))
-		__free_pages_ok(page, compound_order(page));
+	if (unlikely(put_page_testzero(page))) {
+		unsigned int order = compound_order(page);
+
+		if (order == 0)
+			free_unref_page(page);
+		else
+			__free_pages_ok(page, order);
+	}
 }
 EXPORT_SYMBOL(page_frag_free);
 

page_frag_free() calls __free_pages_ok() to free the page back to Buddy. This is OK for high order page, but for order-0 pages, it misses the optimization opportunity of using Per-Cpu-Pages and can cause zone lock contention when called frequently.

Paweł Staszewski recently shared his result of 'how Linux kernel handles normal traffic'[1] and from perf data, Jesper Dangaard Brouer found the lock contention comes from page allocator:

mlx5e_poll_tx_cq
|
--16.34%--napi_consume_skb

             |
             |--12.65%--__free_pages_ok
             |          |
             |           --11.86%--free_one_page
             |                     |
             |                     |--10.10%--queued_spin_lock_slowpath
             |                     |
             |                      --0.65%--_raw_spin_lock
             |
             |--1.55%--page_frag_free
             |
              --1.44%--skb_release_data

Jesper explained how it happened: mlx5 driver RX-page recycle mechanism is not effective in this workload and pages have to go through the page allocator. The lock contention happens during mlx5 DMA TX completion cycle. And the page allocator cannot keep up at these speeds.[2]

I thought that __free_pages_ok() are mostly freeing high order pages and thought this is an lock contention for high order pages but Jesper explained in detail that __free_pages_ok() here are actually freeing order-0 pages because mlx5 is using order-0 pages to satisfy its page pool allocation request.[3]

The free path as pointed out by Jesper is:
skb_free_head()
-> skb_free_frag()

    -> skb_free_frag()
      -> page_frag_free()

And the pages being freed on this path are order-0 pages.

Fix this by doing similar things as in __page_frag_cache_drain() - send the being freed page to PCP if it's an order-0 page, or directly to Buddy if it is a high order page.

With this change, Paweł hasn't noticed lock contention yet in his workload and Jesper has noticed a 7% performance improvement using a micro benchmark and lock contention is gone.

page_frag_free() calls __free_pages_ok() to free the page back to Buddy. This is OK for high order page, but for order-0 pages, it misses the optimization opportunity of using Per-Cpu-Pages and can cause zone lock contention when called frequently.

Paweł Staszewski recently shared his result of 'how Linux kernel handles normal traffic'[1] and from perf data, Jesper Dangaard Brouer found the lock contention comes from page allocator:

mlx5e_poll_tx_cq
|
--16.34%--napi_consume_skb

             |
             |--12.65%--__free_pages_ok
             |          |
             |           --11.86%--free_one_page
             |                     |
             |                     |--10.10%--queued_spin_lock_slowpath
             |                     |
             |                      --0.65%--_raw_spin_lock
             |
             |--1.55%--page_frag_free
             |
              --1.44%--skb_release_data

Jesper explained how it happened: mlx5 driver RX-page recycle mechanism is not effective in this workload and pages have to go through the page allocator. The lock contention happens during mlx5 DMA TX completion cycle. And the page allocator cannot keep up at these speeds.[2]

I thought that __free_pages_ok() are mostly freeing high order pages and thought this is an lock contention for high order pages but Jesper explained in detail that __free_pages_ok() here are actually freeing order-0 pages because mlx5 is using order-0 pages to satisfy its page pool allocation request.[3]

The free path as pointed out by Jesper is:
skb_free_head()
-> skb_free_frag()

    -> skb_free_frag()
      -> page_frag_free()

And the pages being freed on this path are order-0 pages.

Fix this by doing similar things as in __page_frag_cache_drain() - send the being freed page to PCP if it's an order-0 page, or directly to Buddy if it is a high order page.

With this change, Paweł hasn't noticed lock contention yet in his workload and Jesper has noticed a 7% performance improvement using a micro benchmark and lock contention is gone.

[1]: https://www.spinics.net/lists/netdev/msg531362.html
[2]: https://www.spinics.net/lists/netdev/msg531421.html
[3]: https://www.spinics.net/lists/netdev/msg531556.html

Reported-by Paweł Staszewski <[email protected]>

Analysed-by Jesper Dangaard Brouer <[email protected]>

Signed-off-by Aaron Lu <[email protected]>

---
 mm/page_alloc.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

mm/page_alloc.c +1 -2

index ae31839874b8..91a9a6af41a2 100644
@@ -4555,8 +4555,14 @@ void page_frag_free(void *addr)
 {
        struct page *page = virt_to_head_page(addr);
 
-       if (unlikely(put_page_testzero(page)))
-               __free_pages_ok(page, compound_order(page));
+       if (unlikely(put_page_testzero(page))) {

Quoted 73 lines

page_frag_free() calls __free_pages_ok() to free the page back to Buddy. This is OK for high order page, but for order-0 pages, it misses the optimization opportunity of using Per-Cpu-Pages and can cause zone lock contention when called frequently.

PaweA? Staszewski recently shared his result of 'how Linux kernel handles normal traffic'[1] and from perf data, Jesper Dangaard Brouer found the lock contention comes from page allocator:

   mlx5e_poll_tx_cq
   |
    --16.34%--napi_consume_skb
              |
              |--12.65%--__free_pages_ok
              |          |
              |           --11.86%--free_one_page
              |                     |
              |                     |--10.10%--queued_spin_lock_slowpath
              |                     |
              |                      --0.65%--_raw_spin_lock
              |
              |--1.55%--page_frag_free
              |
               --1.44%--skb_release_data

Jesper explained how it happened: mlx5 driver RX-page recycle mechanism is not effective in this workload and pages have to go through the page allocator. The lock contention happens during mlx5 DMA TX completion cycle. And the page allocator cannot keep up at these speeds.[2]

I thought that __free_pages_ok() are mostly freeing high order pages and thought this is an lock contention for high order pages but Jesper explained in detail that __free_pages_ok() here are actually freeing order-0 pages because mlx5 is using order-0 pages to satisfy its page pool allocation request.[3]

The free path as pointed out by Jesper is:
skb_free_head()
-> skb_free_frag()

     -> skb_free_frag()
       -> page_frag_free()

And the pages being freed on this path are order-0 pages.

Fix this by doing similar things as in __page_frag_cache_drain() - send the being freed page to PCP if it's an order-0 page, or directly to Buddy if it is a high order page.

With this change, PaweA? hasn't noticed lock contention yet in his workload and Jesper has noticed a 7% performance improvement using a micro benchmark and lock contention is gone.

[1]: https://www.spinics.net/lists/netdev/msg531362.html
[2]: https://www.spinics.net/lists/netdev/msg531421.html
[3]: https://www.spinics.net/lists/netdev/msg531556.html

Reported-by PaweA? Staszewski <[email protected]>

Analysed-by Jesper Dangaard Brouer <[email protected]>

Signed-off-by Aaron Lu <[email protected]>

---
  mm/page_alloc.c | 10 ++++++++--
  1 file changed, 8 insertions(+), 2 deletions(-)

mm/page_alloc.c +1 -2

index ae31839874b8..91a9a6af41a2 100644
@@ -4555,8 +4555,14 @@ void page_frag_free(void *addr)
  {
         struct page *page = virt_to_head_page(addr);
 
-       if (unlikely(put_page_testzero(page)))
-               __free_pages_ok(page, compound_order(page));
+       if (unlikely(put_page_testzero(page))) {

mm/page_alloc.c void page_frag_free(void *addr) · line 4568

Quoted 10 lines

mm/page_alloc.c +7 -0

+               unsigned int order = compound_order(page);
+
+               if (order == 0)
+                       free_unref_page(page);
+               else
+                       __free_pages_ok(page, order);
+       }
  }
  EXPORT_SYMBOL(page_frag_free);
 

One thing I would suggest for Pawel to try would be to reduce the Tx qdisc size on his transmitting interfaces, Reduce the Tx ring size, and possibly increase the Tx interrupt rate. Ideally we shouldn't have too many packets in-flight and I suspect that is the issue that Pawel is seeing that is leading to the page pool allocator freeing up the memory. I know we like to try to batch things but the issue is processing too many Tx buffers in one batch leads to us eating up too

much memory and causing evictions from the cache. Ideally the Rx and Tx rings and queues should be sized as small as possible while still allowing us to process up to our NAPI budget. Usually I run things with a 128 Rx / 128 Tx setup and then reduce the Tx queue length so we don't have more buffers stored there than we can place in the Tx ring. Then we can avoid the extra thrash of having to pull/push memory into and out of the freelists. Essentially the issue here ends up being another form of buffer bloat.

With that said this change should be mostly harmless and does address the fact that we can have both regular order 0 pages and page frags used for skb->head.

Acked-by Alexander Duyck <[email protected]>

W dniu 05.11.2018 o 16:44, Alexander Duyck pisze:

Quoted 91 lines · Aaron Lu

Quoted 73 lines

page_frag_free() calls __free_pages_ok() to free the page back to Buddy. This is OK for high order page, but for order-0 pages, it misses the optimization opportunity of using Per-Cpu-Pages and can cause zone lock contention when called frequently.

Paweł Staszewski recently shared his result of 'how Linux kernel handles normal traffic'[1] and from perf data, Jesper Dangaard Brouer found the lock contention comes from page allocator:

   mlx5e_poll_tx_cq
   |
    --16.34%--napi_consume_skb
              |
              |--12.65%--__free_pages_ok
              |          |
              |           --11.86%--free_one_page
              |                     |
              |                     |--10.10%--queued_spin_lock_slowpath
              |                     |
              |                      --0.65%--_raw_spin_lock
              |
              |--1.55%--page_frag_free
              |
               --1.44%--skb_release_data

Jesper explained how it happened: mlx5 driver RX-page recycle mechanism is not effective in this workload and pages have to go through the page allocator. The lock contention happens during mlx5 DMA TX completion cycle. And the page allocator cannot keep up at these speeds.[2]

I thought that __free_pages_ok() are mostly freeing high order pages and thought this is an lock contention for high order pages but Jesper explained in detail that __free_pages_ok() here are actually freeing order-0 pages because mlx5 is using order-0 pages to satisfy its page pool allocation request.[3]

The free path as pointed out by Jesper is:
skb_free_head()
-> skb_free_frag()

     -> skb_free_frag()
       -> page_frag_free()

And the pages being freed on this path are order-0 pages.

Fix this by doing similar things as in __page_frag_cache_drain() - send the being freed page to PCP if it's an order-0 page, or directly to Buddy if it is a high order page.

With this change, Paweł hasn't noticed lock contention yet in his workload and Jesper has noticed a 7% performance improvement using a micro benchmark and lock contention is gone.

[1]: https://www.spinics.net/lists/netdev/msg531362.html
[2]: https://www.spinics.net/lists/netdev/msg531421.html
[3]: https://www.spinics.net/lists/netdev/msg531556.html

Reported-by Paweł Staszewski <[email protected]>

Analysed-by Jesper Dangaard Brouer <[email protected]>

Signed-off-by Aaron Lu <[email protected]>

---
  mm/page_alloc.c | 10 ++++++++--
  1 file changed, 8 insertions(+), 2 deletions(-)

mm/page_alloc.c +1 -2

index ae31839874b8..91a9a6af41a2 100644
@@ -4555,8 +4555,14 @@ void page_frag_free(void *addr)
  {
         struct page *page = virt_to_head_page(addr);
 
-       if (unlikely(put_page_testzero(page)))
-               __free_pages_ok(page, compound_order(page));
+       if (unlikely(put_page_testzero(page))) {

mm/page_alloc.c void page_frag_free(void *addr) · line 4568

Quoted 10 lines

mm/page_alloc.c +7 -0

+               unsigned int order = compound_order(page);
+
+               if (order == 0)
+                       free_unref_page(page);
+               else
+                       __free_pages_ok(page, order);
+       }
  }
  EXPORT_SYMBOL(page_frag_free);
 

One thing I would suggest for Pawel to try would be to reduce the Tx qdisc size on his transmitting interfaces, Reduce the Tx ring size, and possibly increase the Tx interrupt rate. Ideally we shouldn't have too many packets in-flight and I suspect that is the issue that Pawel is seeing that is leading to the page pool allocator freeing up the memory. I know we like to try to batch things but the issue is processing too many Tx buffers in one batch leads to us eating up too

Quoted 8 lines

much memory and causing evictions from the cache. Ideally the Rx and Tx rings and queues should be sized as small as possible while still allowing us to process up to our NAPI budget. Usually I run things with a 128 Rx / 128 Tx setup and then reduce the Tx queue length so we don't have more buffers stored there than we can place in the Tx ring. Then we can avoid the extra thrash of having to pull/push memory into and out of the freelists. Essentially the issue here ends up being another form of buffer bloat.

Thanks Aleksandar - yes it can be - but in my scenario setting RX buffer <4096 producing more interface rx drops - and no_rx_buffer on network controller that is receiving more packets
So i need to stick with 3000-4000 on RX - and yes i was trying to lower the TX buff on connectx4 - but that changed nothing before Aaron patch

After Aaron patch - decreasing TX buffer influencing total bandwidth that can be handled by the router/server
Dono why before this patch there was no difference there no matter what i set there there was always page_alloc/slowpath on top in perf

Currently testing RX4096/TX256 - this helps with bandwidth like +10% more bandwidth with less interrupts...

Quoted 114 lines

W dniu 05.11.2018 o 16:44, Alexander Duyck pisze:

Quoted 91 lines · Aaron Lu

Quoted 73 lines

page_frag_free() calls __free_pages_ok() to free the page back to Buddy. This is OK for high order page, but for order-0 pages, it misses the optimization opportunity of using Per-Cpu-Pages and can cause zone lock contention when called frequently.

Paweł Staszewski recently shared his result of 'how Linux kernel handles normal traffic'[1] and from perf data, Jesper Dangaard Brouer found the lock contention comes from page allocator:

    mlx5e_poll_tx_cq
    |
     --16.34%--napi_consume_skb
               |
               |--12.65%--__free_pages_ok
               |          |
               |           --11.86%--free_one_page
               |                     |
               |                     |--10.10%--queued_spin_lock_slowpath
               |                     |
               |                      --0.65%--_raw_spin_lock
               |
               |--1.55%--page_frag_free
               |
                --1.44%--skb_release_data

Jesper explained how it happened: mlx5 driver RX-page recycle mechanism is not effective in this workload and pages have to go through the page allocator. The lock contention happens during mlx5 DMA TX completion cycle. And the page allocator cannot keep up at these speeds.[2]

I thought that __free_pages_ok() are mostly freeing high order pages and thought this is an lock contention for high order pages but Jesper explained in detail that __free_pages_ok() here are actually freeing order-0 pages because mlx5 is using order-0 pages to satisfy its page pool allocation request.[3]

The free path as pointed out by Jesper is:
skb_free_head()

    -> skb_free_frag()
      -> skb_free_frag()
        -> page_frag_free()

And the pages being freed on this path are order-0 pages.

Fix this by doing similar things as in __page_frag_cache_drain() - send the being freed page to PCP if it's an order-0 page, or directly to Buddy if it is a high order page.

With this change, Paweł hasn't noticed lock contention yet in his workload and Jesper has noticed a 7% performance improvement using a micro benchmark and lock contention is gone.

[1]: https://www.spinics.net/lists/netdev/msg531362.html
[2]: https://www.spinics.net/lists/netdev/msg531421.html
[3]: https://www.spinics.net/lists/netdev/msg531556.html

Reported-by Paweł Staszewski <[email protected]>

Analysed-by Jesper Dangaard Brouer <[email protected]>

Signed-off-by Aaron Lu <[email protected]>

---
   mm/page_alloc.c | 10 ++++++++--
   1 file changed, 8 insertions(+), 2 deletions(-)

mm/page_alloc.c +1 -2

index ae31839874b8..91a9a6af41a2 100644
@@ -4555,8 +4555,14 @@ void page_frag_free(void *addr)
   {
          struct page *page = virt_to_head_page(addr);
 
-       if (unlikely(put_page_testzero(page)))
-               __free_pages_ok(page, compound_order(page));
+       if (unlikely(put_page_testzero(page))) {

mm/page_alloc.c void page_frag_free(void *addr) · line 4568

Quoted 10 lines

mm/page_alloc.c +7 -0

+               unsigned int order = compound_order(page);
+
+               if (order == 0)
+                       free_unref_page(page);
+               else
+                       __free_pages_ok(page, order);
+       }
   }
   EXPORT_SYMBOL(page_frag_free);
 

One thing I would suggest for Pawel to try would be to reduce the Tx qdisc size on his transmitting interfaces, Reduce the Tx ring size, and possibly increase the Tx interrupt rate. Ideally we shouldn't have too many packets in-flight and I suspect that is the issue that Pawel is seeing that is leading to the page pool allocator freeing up the memory. I know we like to try to batch things but the issue is processing too many Tx buffers in one batch leads to us eating up too

Quoted 8 lines

much memory and causing evictions from the cache. Ideally the Rx and Tx rings and queues should be sized as small as possible while still allowing us to process up to our NAPI budget. Usually I run things with a 128 Rx / 128 Tx setup and then reduce the Tx queue length so we don't have more buffers stored there than we can place in the Tx ring. Then we can avoid the extra thrash of having to pull/push memory into and out of the freelists. Essentially the issue here ends up being another form of buffer bloat.

Thanks Aleksandar - yes it can be - but in my scenario setting RX buffer <4096 producing more interface rx drops - and no_rx_buffer on network controller that is receiving more packets
So i need to stick with 3000-4000 on RX - and yes i was trying to lower the TX buff on connectx4 - but that changed nothing before Aaron patch

After Aaron patch - decreasing TX buffer influencing total bandwidth that can be handled by the router/server
Dono why before this patch there was no difference there no matter what i set there there was always page_alloc/slowpath on top in perf

Quoted 2 lines

Currently testing RX4096/TX256 - this helps with bandwidth like +10% more bandwidth with less interrupts...

The problem is if you are going for less interrupts you are setting yourself up for buffer bloat. Basically you are going to use much more cache and much more memory then you actually need and if things are properly configured NAPI should take care of the interrupts anyway since under maximum load you shouldn't stop polling normally.

One issue I have seen is people delay interrupts for as long as possible which isn't really a good thing since most network controllers will use NAPI which will disable the interrupts and leave them disabled whenever the system is under heavy stress so you should be able to get the maximum performance by configuring an adapter with small ring sizes and for high interrupt rates.

It is easiest to think of it this way. Your total packet rate is equal to your interrupt rate times the number of buffers you will store in the ring. So if you have some fixed rate "X" for packets and an interrupt rate of "i" then your optimal ring size should be "X/i". So if you lower the interrupt rate you end up hurting the throughput unless you increase the buffer size. However at a certain point the buffer size starts becoming an issue. For example with UDP flows I often see massive packet drops if you tune the interrupt rate too low and then put the system under heavy stress.

- Alex

W dniu 12.11.2018 o 00:05, Alexander Duyck pisze:

Quoted 115 lines · Paweł Staszewski

Quoted 114 lines

W dniu 05.11.2018 o 16:44, Alexander Duyck pisze:

Quoted 91 lines · Aaron Lu

Quoted 73 lines

page_frag_free() calls __free_pages_ok() to free the page back to Buddy. This is OK for high order page, but for order-0 pages, it misses the optimization opportunity of using Per-Cpu-Pages and can cause zone lock contention when called frequently.

Paweł Staszewski recently shared his result of 'how Linux kernel handles normal traffic'[1] and from perf data, Jesper Dangaard Brouer found the lock contention comes from page allocator:

    mlx5e_poll_tx_cq
    |
     --16.34%--napi_consume_skb
               |
               |--12.65%--__free_pages_ok
               |          |
               |           --11.86%--free_one_page
               |                     |
               |                     |--10.10%--queued_spin_lock_slowpath
               |                     |
               |                      --0.65%--_raw_spin_lock
               |
               |--1.55%--page_frag_free
               |
                --1.44%--skb_release_data

Jesper explained how it happened: mlx5 driver RX-page recycle mechanism is not effective in this workload and pages have to go through the page allocator. The lock contention happens during mlx5 DMA TX completion cycle. And the page allocator cannot keep up at these speeds.[2]

I thought that __free_pages_ok() are mostly freeing high order pages and thought this is an lock contention for high order pages but Jesper explained in detail that __free_pages_ok() here are actually freeing order-0 pages because mlx5 is using order-0 pages to satisfy its page pool allocation request.[3]

The free path as pointed out by Jesper is:
skb_free_head()

    -> skb_free_frag()
      -> skb_free_frag()
        -> page_frag_free()

And the pages being freed on this path are order-0 pages.

Fix this by doing similar things as in __page_frag_cache_drain() - send the being freed page to PCP if it's an order-0 page, or directly to Buddy if it is a high order page.

With this change, Paweł hasn't noticed lock contention yet in his workload and Jesper has noticed a 7% performance improvement using a micro benchmark and lock contention is gone.

[1]: https://www.spinics.net/lists/netdev/msg531362.html
[2]: https://www.spinics.net/lists/netdev/msg531421.html
[3]: https://www.spinics.net/lists/netdev/msg531556.html

Reported-by Paweł Staszewski <[email protected]>

Analysed-by Jesper Dangaard Brouer <[email protected]>

Signed-off-by Aaron Lu <[email protected]>

---
   mm/page_alloc.c | 10 ++++++++--
   1 file changed, 8 insertions(+), 2 deletions(-)

mm/page_alloc.c +1 -2

index ae31839874b8..91a9a6af41a2 100644
@@ -4555,8 +4555,14 @@ void page_frag_free(void *addr)
   {
          struct page *page = virt_to_head_page(addr);
 
-       if (unlikely(put_page_testzero(page)))
-               __free_pages_ok(page, compound_order(page));
+       if (unlikely(put_page_testzero(page))) {

mm/page_alloc.c void page_frag_free(void *addr) · line 4568

Quoted 10 lines

mm/page_alloc.c +7 -0

+               unsigned int order = compound_order(page);
+
+               if (order == 0)
+                       free_unref_page(page);
+               else
+                       __free_pages_ok(page, order);
+       }
   }
   EXPORT_SYMBOL(page_frag_free);
 

One thing I would suggest for Pawel to try would be to reduce the Tx qdisc size on his transmitting interfaces, Reduce the Tx ring size, and possibly increase the Tx interrupt rate. Ideally we shouldn't have too many packets in-flight and I suspect that is the issue that Pawel is seeing that is leading to the page pool allocator freeing up the memory. I know we like to try to batch things but the issue is processing too many Tx buffers in one batch leads to us eating up too

Quoted 8 lines

much memory and causing evictions from the cache. Ideally the Rx and Tx rings and queues should be sized as small as possible while still allowing us to process up to our NAPI budget. Usually I run things with a 128 Rx / 128 Tx setup and then reduce the Tx queue length so we don't have more buffers stored there than we can place in the Tx ring. Then we can avoid the extra thrash of having to pull/push memory into and out of the freelists. Essentially the issue here ends up being another form of buffer bloat.

Thanks Aleksandar - yes it can be - but in my scenario setting RX buffer <4096 producing more interface rx drops - and no_rx_buffer on network controller that is receiving more packets
So i need to stick with 3000-4000 on RX - and yes i was trying to lower the TX buff on connectx4 - but that changed nothing before Aaron patch

After Aaron patch - decreasing TX buffer influencing total bandwidth that can be handled by the router/server
Dono why before this patch there was no difference there no matter what i set there there was always page_alloc/slowpath on top in perf

Quoted 7 lines

Quoted 2 lines

Currently testing RX4096/TX256 - this helps with bandwidth like +10% more bandwidth with less interrupts...

The problem is if you are going for less interrupts you are setting yourself up for buffer bloat. Basically you are going to use much more cache and much more memory then you actually need and if things are properly configured NAPI should take care of the interrupts anyway since under maximum load you shouldn't stop polling normally.

Im trying to balance here - there is problem cause server is forwarding all kingd of protocols packets/different size etc

The problem is im trying to go in high interrupt rate - but

Setting coalescence to adaptative for rx killing cpu's at 22Gbit/s RX and 22Gbit with rly high interrupt rate

So adding a little more latency i can turn off adaptative rx and setup rx-usecs from range 16-64 - and this gives me more or less interrupts - but the problem is - always same bandwidth as maximum

Quoted 7 lines

One issue I have seen is people delay interrupts for as long as possible which isn't really a good thing since most network controllers will use NAPI which will disable the interrupts and leave them disabled whenever the system is under heavy stress so you should be able to get the maximum performance by configuring an adapter with small ring sizes and for high interrupt rates.

Sure this is bad to setup rx-usec for high values - cause at some point this will add high latency for packet traversing both sides - and start to hurt buffers

But my problem is a little different now i have no problems with RX side
- cause i can setup anything like:

coalescence from 16 to 64

rx ring from 3000 to max 8192

And it does not change my max bw - only produces less or more interrupts.

So I start to change params for TX side - and for now i know that the best for me is

coalescence adaptative on

TX buffer 128

This helps with max BW that for now is close to 70Gbit/s RX and 70Gbit TX but after this change i have increasing DROPS on TX side for vlan interfaces.

And only 50% cpu (max was 50% for 70Gbit/s)

Quoted 9 lines

It is easiest to think of it this way. Your total packet rate is equal to your interrupt rate times the number of buffers you will store in the ring. So if you have some fixed rate "X" for packets and an interrupt rate of "i" then your optimal ring size should be "X/i". So if you lower the interrupt rate you end up hurting the throughput unless you increase the buffer size. However at a certain point the buffer size starts becoming an issue. For example with UDP flows I often see massive packet drops if you tune the interrupt rate too low and then put the system under heavy stress.

Yes - in normal life traffic - most of ddos'es are like this many pps with small frames.

It sounds to me like XDP would probably be your best bet. With that you could probably get away with smaller ring sizes, higher interrupt rates, and get the advantage of it batching the Tx without having to drop packets.

Quoted 5 lines

It sounds to me like XDP would probably be your best bet. With that you could probably get away with smaller ring sizes, higher interrupt rates, and get the advantage of it batching the Tx without having to drop packets.

Add to this that with XDP (or anything lowering per packet processing costs) you can reduce number of cpus/queues, get better latencies, and bigger TX batches.

Quoted 123 lines

W dniu 12.11.2018 o 00:05, Alexander Duyck pisze:

Quoted 114 lines · Paweł Staszewski

Quoted 113 lines

W dniu 05.11.2018 o 16:44, Alexander Duyck pisze:

Quoted 91 lines · Aaron Lu

Quoted 73 lines

page_frag_free() calls __free_pages_ok() to free the page back to Buddy. This is OK for high order page, but for order-0 pages, it misses the optimization opportunity of using Per-Cpu-Pages and can cause zone lock contention when called frequently.

Paweł Staszewski recently shared his result of 'how Linux kernel handles normal traffic'[1] and from perf data, Jesper Dangaard Brouer found the lock contention comes from page allocator:

     mlx5e_poll_tx_cq
     |
      --16.34%--napi_consume_skb
                |
                |--12.65%--__free_pages_ok
                |          |
                |           --11.86%--free_one_page
                |                     |
                |                     |--10.10%--queued_spin_lock_slowpath
                |                     |
                |                      --0.65%--_raw_spin_lock
                |
                |--1.55%--page_frag_free
                |
                 --1.44%--skb_release_data

Jesper explained how it happened: mlx5 driver RX-page recycle mechanism is not effective in this workload and pages have to go through the page allocator. The lock contention happens during mlx5 DMA TX completion cycle. And the page allocator cannot keep up at these speeds.[2]

I thought that __free_pages_ok() are mostly freeing high order pages and thought this is an lock contention for high order pages but Jesper explained in detail that __free_pages_ok() here are actually freeing order-0 pages because mlx5 is using order-0 pages to satisfy its page pool allocation request.[3]

The free path as pointed out by Jesper is:
skb_free_head()

     -> skb_free_frag()
       -> skb_free_frag()
         -> page_frag_free()

And the pages being freed on this path are order-0 pages.

Fix this by doing similar things as in __page_frag_cache_drain() - send the being freed page to PCP if it's an order-0 page, or directly to Buddy if it is a high order page.

With this change, Paweł hasn't noticed lock contention yet in his workload and Jesper has noticed a 7% performance improvement using a micro benchmark and lock contention is gone.

[1]: https://www.spinics.net/lists/netdev/msg531362.html
[2]: https://www.spinics.net/lists/netdev/msg531421.html
[3]: https://www.spinics.net/lists/netdev/msg531556.html

Reported-by Paweł Staszewski <[email protected]>

Analysed-by Jesper Dangaard Brouer <[email protected]>

Signed-off-by Aaron Lu <[email protected]>

---
    mm/page_alloc.c | 10 ++++++++--
    1 file changed, 8 insertions(+), 2 deletions(-)

mm/page_alloc.c +1 -2

index ae31839874b8..91a9a6af41a2 100644
@@ -4555,8 +4555,14 @@ void page_frag_free(void *addr)
    {
           struct page *page = virt_to_head_page(addr);
 
-       if (unlikely(put_page_testzero(page)))
-               __free_pages_ok(page, compound_order(page));
+       if (unlikely(put_page_testzero(page))) {

mm/page_alloc.c void page_frag_free(void *addr) · line 4568

Quoted 10 lines

mm/page_alloc.c +7 -0

+               unsigned int order = compound_order(page);
+
+               if (order == 0)
+                       free_unref_page(page);
+               else
+                       __free_pages_ok(page, order);
+       }
    }
    EXPORT_SYMBOL(page_frag_free);
 

One thing I would suggest for Pawel to try would be to reduce the Tx qdisc size on his transmitting interfaces, Reduce the Tx ring size, and possibly increase the Tx interrupt rate. Ideally we shouldn't have too many packets in-flight and I suspect that is the issue that Pawel is seeing that is leading to the page pool allocator freeing up the memory. I know we like to try to batch things but the issue is processing too many Tx buffers in one batch leads to us eating up too

Quoted 8 lines

much memory and causing evictions from the cache. Ideally the Rx and Tx rings and queues should be sized as small as possible while still allowing us to process up to our NAPI budget. Usually I run things with a 128 Rx / 128 Tx setup and then reduce the Tx queue length so we don't have more buffers stored there than we can place in the Tx ring. Then we can avoid the extra thrash of having to pull/push memory into and out of the freelists. Essentially the issue here ends up being another form of buffer bloat.

Thanks Aleksandar - yes it can be - but in my scenario setting RX buffer <4096 producing more interface rx drops - and no_rx_buffer on network controller that is receiving more packets
So i need to stick with 3000-4000 on RX - and yes i was trying to lower the TX buff on connectx4 - but that changed nothing before Aaron patch

After Aaron patch - decreasing TX buffer influencing total bandwidth that can be handled by the router/server
Dono why before this patch there was no difference there no matter what i set there there was always page_alloc/slowpath on top in perf

Quoted 7 lines

Quoted 2 lines

Currently testing RX4096/TX256 - this helps with bandwidth like +10% more bandwidth with less interrupts...

The problem is if you are going for less interrupts you are setting yourself up for buffer bloat. Basically you are going to use much more cache and much more memory then you actually need and if things are properly configured NAPI should take care of the interrupts anyway since under maximum load you shouldn't stop polling normally.

Quoted 7 lines

Im trying to balance here - there is problem cause server is forwarding all kingd of protocols packets/different size etc

The problem is im trying to go in high interrupt rate - but

Setting coalescence to adaptative for rx killing cpu's at 22Gbit/s RX and 22Gbit with rly high interrupt rate

I wouldn't recommend adaptive just because the behavior would be hard to predict.

Quoted 3 lines

So adding a little more latency i can turn off adaptative rx and setup rx-usecs from range 16-64 - and this gives me more or less interrupts - but the problem is - always same bandwidth as maximum

What about the tx-usecs, is that a functional thing for the adapter you are using?

The Rx side logic should be pretty easy to figure out. Essentially you want to keep the Rx ring size as small as possible while at the same time avoiding storming the system with interrupts. I know for 10Gb/s I have used a value of 25us in the past. What you want to watch for is if you are dropping packets on the Rx side or not. Ideally you want enough buffers that you can capture any burst while you wait for the interrupt routine to catch up.

Quoted 10 lines

Quoted 6 lines

One issue I have seen is people delay interrupts for as long as possible which isn't really a good thing since most network controllers will use NAPI which will disable the interrupts and leave them disabled whenever the system is under heavy stress so you should be able to get the maximum performance by configuring an adapter with small ring sizes and for high interrupt rates.

Sure this is bad to setup rx-usec for high values - cause at some point this will add high latency for packet traversing both sides - and start to hurt buffers

Quoted 8 lines

But my problem is a little different now i have no problems with RX side
- cause i can setup anything like:

coalescence from 16 to 64

rx ring from 3000 to max 8192

And it does not change my max bw - only produces less or more interrupts.

Right so the issue itself isn't Rx, you aren't throttled there. We are probably looking at an issue of PCIe bandwidth or Tx slowing things down. The fact that you are still filing interrupts is a bit surprising though. Are the Tx and Rx interrupts linked for the device you are using or are they firing them seperately? Normally Rx traffic won't generate many interrupts under a stress test as NAPI will leave the interrupts disabled unless it can keep up. Anyway, my suggestion would be to look at tuning things for as small a ring size as possible.

Quoted 10 lines

So I start to change params for TX side - and for now i know that the best for me is

coalescence adaptative on

TX buffer 128

This helps with max BW that for now is close to 70Gbit/s RX and 70Gbit TX but after this change i have increasing DROPS on TX side for vlan interfaces.

So this sounds like you are likely bottlenecked due to either PCIe bandwidth or latency. When you start putting back-pressure on the Tx like you have described it starts pushing packets onto the Qdisc layer. One thing that happens when packets are on the qdisc layer is that they can start to perform a bulk dequeue. The side effect of this is that you write multiple packets to the descriptor ring and then update the hardware doorbell only once for the entire group of packets instead of once per packet.

Quoted 12 lines

And only 50% cpu (max was 50% for 70Gbit/s)

Quoted 9 lines

It is easiest to think of it this way. Your total packet rate is equal to your interrupt rate times the number of buffers you will store in the ring. So if you have some fixed rate "X" for packets and an interrupt rate of "i" then your optimal ring size should be "X/i". So if you lower the interrupt rate you end up hurting the throughput unless you increase the buffer size. However at a certain point the buffer size starts becoming an issue. For example with UDP flows I often see massive packet drops if you tune the interrupt rate too low and then put the system under heavy stress.

Quoted 2 lines

Yes - in normal life traffic - most of ddos'es are like this many pps with small frames.

It sounds to me like XDP would probably be your best bet. With that you could probably get away with smaller ring sizes, higher interrupt rates, and get the advantage of it batching the Tx without having to drop packets.

page_frag_free() calls __free_pages_ok() to free the page back to Buddy. This is OK for high order page, but for order-0 pages, it misses the optimization opportunity of using Per-Cpu-Pages and can cause zone lock contention when called frequently.

Paweł Staszewski recently shared his result of 'how Linux kernel handles normal traffic'[1] and from perf data, Jesper Dangaard Brouer found the lock contention comes from page allocator:

   mlx5e_poll_tx_cq
   |
    --16.34%--napi_consume_skb
              |
              |--12.65%--__free_pages_ok
              |          |
              |           --11.86%--free_one_page
              |                     |
              |                     |--10.10%--queued_spin_lock_slowpath
              |                     |
              |                      --0.65%--_raw_spin_lock
              |
              |--1.55%--page_frag_free
              |
               --1.44%--skb_release_data

Jesper explained how it happened: mlx5 driver RX-page recycle mechanism is not effective in this workload and pages have to go through the page allocator. The lock contention happens during mlx5 DMA TX completion cycle. And the page allocator cannot keep up at these speeds.[2]

I thought that __free_pages_ok() are mostly freeing high order pages and thought this is an lock contention for high order pages but Jesper explained in detail that __free_pages_ok() here are actually freeing order-0 pages because mlx5 is using order-0 pages to satisfy its page pool allocation request.[3]

The free path as pointed out by Jesper is:
skb_free_head()
-> skb_free_frag()

     -> page_frag_free()

And the pages being freed on this path are order-0 pages.

Fix this by doing similar things as in __page_frag_cache_drain() - send the being freed page to PCP if it's an order-0 page, or directly to Buddy if it is a high order page.

With this change, Paweł hasn't noticed lock contention yet in his workload and Jesper has noticed a 7% performance improvement using a micro benchmark and lock contention is gone. Ilias' test on a 'low' speed 1Gbit interface on an cortex-a53 shows ~11% performance boost testing with 64byte packets and __free_pages_ok() disappeared from perf top.

[1]: https://www.spinics.net/lists/netdev/msg531362.html
[2]: https://www.spinics.net/lists/netdev/msg531421.html
[3]: https://www.spinics.net/lists/netdev/msg531556.html

Reported-by Paweł Staszewski <[email protected]>

Analysed-by Jesper Dangaard Brouer <[email protected]>

Acked-by Vlastimil Babka <[email protected]>

Acked-by Mel Gorman <[email protected]>

Acked-by Jesper Dangaard Brouer <[email protected]>

Acked-by Ilias Apalodimas <[email protected]>

Tested-by Ilias Apalodimas <[email protected]>

Acked-by Alexander Duyck <[email protected]>

Signed-off-by Aaron Lu <[email protected]>

---
v2: only changelog changes:
- remove the duplicated skb_free_frag() as pointed by Jesper;
- add Ilias' test result;
- add people's ack/test tag.

  mm/page_alloc.c | 10 ++++++++--
  1 file changed, 8 insertions(+), 2 deletions(-)

mm/page_alloc.c +8 -2

index ae31839874b8..91a9a6af41a2 100644
@@ -4555,8 +4555,14 @@ void page_frag_free(void *addr)
  {
  	struct page *page = virt_to_head_page(addr);
 
-	if (unlikely(put_page_testzero(page)))
-		__free_pages_ok(page, compound_order(page));
+	if (unlikely(put_page_testzero(page))) {
+		unsigned int order = compound_order(page);
+
+		if (order == 0)
+			free_unref_page(page);
+		else
+			__free_pages_ok(page, order);
+	}
  }
  EXPORT_SYMBOL(page_frag_free);