net/mlx5: use indirect call wrappers
Paolo Abeni
maintainercover net/mlx5: use indirect call wrappers
The mlx5_core driver uses several indirect calls in fast-path, some of them are invoked on each ingress packet, even for the XDP-only traffic.
This series leverage the indirect call wrappers infrastructure the avoid the expansive RETPOLINE overhead for 2 indirect calls in fast-path.
Each call is addressed on a different patch, plus we need to introduce a couple of additional helpers to cope with the higher number of possible direct-call alternatives.
Paolo Abeni (3):
net/mlx5e: use indirect calls wrapper for skb allocation indirect call wrappers: add helpers for 3 and 4 ways switch net/mlx5e: use indirect calls wrapper for the rx packet handler
.../net/ethernet/mellanox/mlx5/core/en_rx.c | 26 ++++++++++++++----- include/linux/indirect_call_wrapper.h | 12 +++++++++ 2 files changed, 32 insertions(+), 6 deletions(-)
Signature · 1 line
2.20.1
Paolo Abeni
maintainerPATCH 1/3 net/mlx5e: use indirect calls wrapper for skb allocation
We can avoid an indirect call per packet wrapping the skb creation with the appropriate helper.
--- .../net/ethernet/mellanox/mlx5/core/en_rx.c | 22 ++++++++++++++----- 1 file changed, 17 insertions(+), 5 deletions(-)
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c +17 -5
@@ -34,6 +34,7 @@#include <linux/ip.h> #include <linux/ipv6.h> #include <linux/tcp.h> +#include <linux/indirect_call_wrapper.h> #include <net/ip6_checksum.h> #include <net/page_pool.h> #include <net/inet_ecn.h>
@@ -1092,7 +1093,10 @@ void mlx5e_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe) wi =get_frag (rq, ci); cqe_bcnt =be32_to_cpu (cqe->byte_cnt); - skb = rq->wqe.skb_from_cqe (rq, cqe, wi, cqe_bcnt); + skb =INDIRECT_CALL_2 (rq->wqe.skb_from_cqe, + mlx5e_skb_from_cqe_linear, + mlx5e_skb_from_cqe_nonlinear, + rq, cqe, wi, cqe_bcnt);if (!skb) {/* probably for XDP */ if (__test_and_clear_bit (MLX5E_RQ_FLAG_XDP_XMIT , rq->flags)) {
@@ -1279,8 +1283,10 @@ void mlx5e_handle_rx_cqe_mpwrq(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe) cqe_bcnt =mpwrq_get_cqe_byte_cnt (cqe); - skb = rq->mpwqe.skb_from_cqe_mpwrq (rq, wi, cqe_bcnt, head_offset, - page_idx); + skb =INDIRECT_CALL_2 (rq->mpwqe.skb_from_cqe_mpwrq, + mlx5e_skb_from_cqe_mpwrq_linear, + mlx5e_skb_from_cqe_mpwrq_nonlinear, + rq, wi, cqe_bcnt, head_offset, page_idx);if (!skb) goto mpwrq_cqe_out;
@@ -1437,7 +1443,10 @@ void mlx5i_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe) wi =get_frag (rq, ci); cqe_bcnt =be32_to_cpu (cqe->byte_cnt); - skb = rq->wqe.skb_from_cqe (rq, cqe, wi, cqe_bcnt); + skb =INDIRECT_CALL_2 (rq->wqe.skb_from_cqe, + mlx5e_skb_from_cqe_linear, + mlx5e_skb_from_cqe_nonlinear, + rq, cqe, wi, cqe_bcnt);if (!skb) goto wq_free_wqe;
@@ -1469,7 +1478,10 @@ void mlx5e_ipsec_handle_rx_cqe(struct mlx5e_rq *rq, struct mlx5_cqe64 *cqe) wi =get_frag (rq, ci); cqe_bcnt =be32_to_cpu (cqe->byte_cnt); - skb = rq->wqe.skb_from_cqe (rq, cqe, wi, cqe_bcnt); + skb =INDIRECT_CALL_2 (rq->wqe.skb_from_cqe, + mlx5e_skb_from_cqe_linear, + mlx5e_skb_from_cqe_nonlinear, + rq, cqe, wi, cqe_bcnt);if (unlikely (!skb)) {/* a DROP, save the page-reuse checks */ mlx5e_free_rx_wqe (rq, wi, true);
Signature · 1 line
2.20.1
Paolo Abeni
maintainerPATCH 2/3 indirect call wrappers: add helpers for 3 and 4 ways switch
In reply to Paolo AbeniExperimental results[1] has shown that resorting to several branches and a direct-call is faster than indirect call via retpoline, even when the number of added branches go up 5.
This change adds two additional helpers, to cope with indirect calls with up to 4 available direct call option. We will use them in the next patch.
--- include/linux/indirect_call_wrapper.h | 12 ++++++++++++ 1 file changed, 12 insertions(+)
include/linux/indirect_call_wrapper.h +12 -0
@@ -23,6 +23,16 @@likely (f == f2) ?f2 (__VA_ARGS__) : \INDIRECT_CALL_1 (f, f1, __VA_ARGS__); \ }) +#define INDIRECT_CALL_3 (f, f3, f2, f1, ...) \ + ({ \ + likely(f == f3) ? f3(__VA_ARGS__) : \ + INDIRECT_CALL_2(f, f2, f1, __VA_ARGS__); \ + }) +#define INDIRECT_CALL_4 (f, f4, f3, f2, f1, ...) \ + ({ \ + likely(f == f4) ? f4(__VA_ARGS__) : \ + INDIRECT_CALL_3(f, f3, f2, f1, __VA_ARGS__); \ + })#define INDIRECT_CALLABLE_DECLARE (f) f#define INDIRECT_CALLABLE_SCOPE
@@ -30,6 +40,8 @@#else #define INDIRECT_CALL_1 (f, f1, ...) f(__VA_ARGS__)#define INDIRECT_CALL_2 (f, f2, f1, ...) f(__VA_ARGS__) +#define INDIRECT_CALL_3 (f, f3, f2, f1, ...) f(__VA_ARGS__) +#define INDIRECT_CALL_4 (f, f4, f3, f2, f1, ...) f(__VA_ARGS__)#define INDIRECT_CALLABLE_DECLARE (f)#define INDIRECT_CALLABLE_SCOPE static#endif
Signature · 1 line
2.20.1
Re: [PATCH net-next 2/3] indirect call wrappers: add helpers for 3 and 4 ways switch
Quoted 40 lines · Paolo Abeni
Experimental results[1] has shown that resorting to several branches and a direct-call is faster than indirect call via retpoline, even when the number of added branches go up 5.
This change adds two additional helpers, to cope with indirect calls with up to 4 available direct call option. We will use them in the next patch.
Signed-off-by Paolo Abeni <[email protected]>--- include/linux/indirect_call_wrapper.h | 12 ++++++++++++ 1 file changed, 12 insertions(+)include/linux/indirect_call_wrapper.h +2
-0@@ -23,6 +23,16 @@likely (f == f2) ?f2 (__VA_ARGS__) : \INDIRECT_CALL_1 (f, f1, __VA_ARGS__); \ }) +#define INDIRECT_CALL_3 (f, f3, f2, f1, ...) \ + ({\
include/linux/indirect_call_wrapper.h +1
-0+ likely(f == f3) ? f3(__VA_ARGS__) :\
include/linux/indirect_call_wrapper.h +1
-0+ INDIRECT_CALL_2(f, f2, f1,
__VA_ARGS__); \include/linux/indirect_call_wrapper.h +2
-0+ }) +#define INDIRECT_CALL_4 (f, f4, f3, f2, f1, ...)\include/linux/indirect_call_wrapper.h +1
-0+ ({\
include/linux/indirect_call_wrapper.h
Quoted 1 line
include/linux/indirect_call_wrapper.h +1
-0+ likely(f == f4) ? f4(__VA_ARGS__) :
do we really want "likely" here ? in our cases there is no preference on whuch fN is going to have the top priority, all of them are equally important and statically configured and guranteed to not change on data path ..
include/linux/indirect_call_wrapper.h
Quoted 5 lines
\include/linux/indirect_call_wrapper.h +1
-0+ INDIRECT_CALL_3(f, f3, f2, f1,
__VA_ARGS__); \include/linux/indirect_call_wrapper.h +1
-0+ })
Oh the RETPOLINE!
On which (N) where INDIRECT_CALL_N(f, fN, fN-1, ..., f1,...) , calling the indirection function pointer directly is going to be actually better than this whole INDIRECT_CALL_N wrapper "if else" dance ?
Quoted 11 lines
#define
INDIRECT_CALLABLE_DECLARE(f)f
#defineINDIRECT_CALLABLE_SCOPEinclude/linux/indirect_call_wrapper.h +2
-0@@ -30,6 +40,8 @@#else #define INDIRECT_CALL_1 (f, f1, ...) f(__VA_ARGS__)#define INDIRECT_CALL_2 (f, f2, f1, ...) f(__VA_ARGS__) +#define INDIRECT_CALL_3 (f, f3, f2, f1, ...) f(__VA_ARGS__) +#define INDIRECT_CALL_4 (f, f4, f3, f2, f1, ...) f(__VA_ARGS__)#define INDIRECT_CALLABLE_DECLARE (f)#define INDIRECT_CALLABLE_SCOPE static#endif
Paolo Abeni
maintainerRe: [PATCH net-next 2/3] indirect call wrappers: add helpers for 3 and 4 ways switch
Quoted 39 lines · Saeed Mahameed
Quoted 38 lines · Paolo Abeni
Experimental results[1] has shown that resorting to several branches and a direct-call is faster than indirect call via retpoline, even when the number of added branches go up 5.
This change adds two additional helpers, to cope with indirect calls with up to 4 available direct call option. We will use them in the next patch.
Signed-off-by Paolo Abeni <[email protected]>--- include/linux/indirect_call_wrapper.h | 12 ++++++++++++ 1 file changed, 12 insertions(+)include/linux/indirect_call_wrapper.h +2
-0@@ -23,6 +23,16 @@likely (f == f2) ?f2 (__VA_ARGS__) : \INDIRECT_CALL_1 (f, f1, __VA_ARGS__); \ }) +#define INDIRECT_CALL_3 (f, f3, f2, f1, ...) \ + ({\
include/linux/indirect_call_wrapper.h +1
-0+ likely(f == f3) ? f3(__VA_ARGS__) :\
include/linux/indirect_call_wrapper.h +1
-0+ INDIRECT_CALL_2(f, f2, f1,
__VA_ARGS__); \include/linux/indirect_call_wrapper.h +2
-0+ }) +#define INDIRECT_CALL_4 (f, f4, f3, f2, f1, ...)\
Quoted 8 lines
include/linux/indirect_call_wrapper.h
Quoted 3 lines
include/linux/indirect_call_wrapper.h +1
-0+ ({\
include/linux/indirect_call_wrapper.h +1
-0+ likely(f == f4) ? f4(__VA_ARGS__) :do we really want "likely" here ? in our cases there is no preference on whuch fN is going to have the top priority, all of them are equally important and statically configured and guranteed to not change on data path ..
I was a little undecided about that, too. 'likely()' is there mainly for simmetry with the already existing _1 and _2 variants. In such macros the branch prediction hint represent a real priority of the available choices.
To avoid the branch prediction, a new set of macros should be defined, but that also sounds redundant.
If you have strong opinion against the breanch prediction hint, I could either drop this patch and the next one or resort to custom macros in the mlx code.
Any [alternative] suggestions more than welcome!
\
Quoted 10 lines
include/linux/indirect_call_wrapper.h
Quoted 4 lines
include/linux/indirect_call_wrapper.h +1
-0+ INDIRECT_CALL_3(f, f3, f2, f1,
__VA_ARGS__); \include/linux/indirect_call_wrapper.h +1
-0+ })Oh the RETPOLINE!
On which (N) where
INDIRECT_CALL_N(f, fN, fN-1, ..., f1,...), calling the indirection function pointer directly is going to be actually better than this wholeINDIRECT_CALL_Nwrapper "if else" dance ?
In commit ce02ef06fcf7a399a6276adb83f37373d10cbbe1, it's measured a relevant gain even with more than 5 options. I personally would avoid adding much more options than the above.
Thanks,
Paolo
Re: [PATCH net-next 2/3] indirect call wrappers: add helpers for 3 and 4 ways switch
Quoted 56 lines · Paolo Abeni
Quoted 47 lines · Saeed Mahameed
Quoted 44 lines · Paolo Abeni
Experimental results[1] has shown that resorting to several branches
and a direct-call is faster than indirect call via retpoline, even
when the number of added branches go up 5.This change adds two additional helpers, to cope with indirect calls
with up to 4 available direct call option. We will use them in the next patch.Signed-off-by Paolo Abeni <[email protected]>--- include/linux/indirect_call_wrapper.h | 12 ++++++++++++ 1 file changed, 12 insertions(+)include/linux/indirect_call_wrapper.h +1
-0@@ -23,6 +23,16 @@likely (f == f2) ?f2 (__VA_ARGS__) : \INDIRECT_CALL_1 (f, f1, __VA_ARGS__); \ }) +#define INDIRECT_CALL_3 (f, f3, f2, f1, ...)\
include/linux/indirect_call_wrapper.h +1
-0+ ({\
include/linux/indirect_call_wrapper.h +1
-0+ likely(f == f3) ? f3(__VA_ARGS__) :\
include/linux/indirect_call_wrapper.h +1
-0+ INDIRECT_CALL_2(f, f2, f1,
__VA_ARGS__); \include/linux/indirect_call_wrapper.h +2
-0+ }) +#define INDIRECT_CALL_4 (f, f4, f3, f2, f1, ...)\include/linux/indirect_call_wrapper.h +1
-0+ ({\
include/linux/indirect_call_wrapper.h
Quoted 1 line
include/linux/indirect_call_wrapper.h +1
-0+ likely(f == f4) ? f4(__VA_ARGS__) :Quoted 7 lines
do we really want "likely" here ? in our cases there is no preference
on whuch fN is going to have the top priority, all of them are equally
important and statically configured and guranteed to not change on data
path ..
Quoted 5 lines
I was a little undecided about that, too. '
likely()' is there mainly for simmetry with the already existing _1 and _2 variants. In such macros the branch prediction hint represent a real priority of the available choices.
For macro _1 it make sense to have the likely keyword but for _2 it doesn't, by looking at most of the usecases of INDIRECT_CALL_2, they seem to be all around improving tcp/udp related indirection calls in the protocol stack, and they seem to prefer tcp over udp. But IMHO at least for the above usecase I think the likely keyword is being misused here and should be remove from all INDIRECT_CALL_N where N > 1;
Eric, what do you think ?
Quoted 11 lines
To avoid the branch prediction, a new set of macros should be defined,
but that also sounds redundant.If you have strong opinion against the breanch prediction hint, I could
either drop this patch and the next one or resort to custom macros in the mlx code.Any [alternative] suggestions more than welcome!
\
custom macros can work, but in case you don't want to introduce such macros in a vendor specific driver, then i think your patches are still an improvement after all..
In any case, just make sure to use the order i suggested in next patch
with: MLX5_RX_INDIRECT_CALL_LIST
Quoted 21 lines
Quoted 11 lines
include/linux/indirect_call_wrapper.h
Quoted 4 lines
include/linux/indirect_call_wrapper.h +1
-0+ INDIRECT_CALL_3(f, f3, f2, f1,
__VA_ARGS__); \include/linux/indirect_call_wrapper.h +1
-0+ })Oh the RETPOLINE!
On which (N) where
INDIRECT_CALL_N(f, fN, fN-1, ..., f1,...), calling
the indirection function pointer directly is going to be actually better than this wholeINDIRECT_CALL_Nwrapper "if else" dance ?In commit
ce02ef06fcf7a399a6276adb83f37373d10cbbe1, it's measured a relevant gain even with more than 5 options. I personally would avoid adding much more options than the above.Thanks,
Paolo
Paolo Abeni
maintainerRe: [PATCH net-next 2/3] indirect call wrappers: add helpers for 3 and 4 ways switch
Quoted 63 lines · Saeed Mahameed
Quoted 56 lines · Paolo Abeni
Quoted 47 lines · Saeed Mahameed
Quoted 44 lines · Paolo Abeni
Experimental results[1] has shown that resorting to several branches
and a direct-call is faster than indirect call via retpoline, even
when the number of added branches go up 5.This change adds two additional helpers, to cope with indirect calls
with up to 4 available direct call option. We will use them in the next patch.Signed-off-by Paolo Abeni <[email protected]>--- include/linux/indirect_call_wrapper.h | 12 ++++++++++++ 1 file changed, 12 insertions(+)include/linux/indirect_call_wrapper.h +1
-0@@ -23,6 +23,16 @@likely (f == f2) ?f2 (__VA_ARGS__) : \INDIRECT_CALL_1 (f, f1, __VA_ARGS__); \ }) +#define INDIRECT_CALL_3 (f, f3, f2, f1, ...)\
include/linux/indirect_call_wrapper.h +1
-0+ ({\
include/linux/indirect_call_wrapper.h +1
-0+ likely(f == f3) ? f3(__VA_ARGS__) :\
include/linux/indirect_call_wrapper.h +1
-0+ INDIRECT_CALL_2(f, f2, f1,
__VA_ARGS__); \include/linux/indirect_call_wrapper.h +2
-0+ }) +#define INDIRECT_CALL_4 (f, f4, f3, f2, f1, ...)\include/linux/indirect_call_wrapper.h +1
-0+ ({\
include/linux/indirect_call_wrapper.h
Quoted 1 line
include/linux/indirect_call_wrapper.h +1
-0+ likely(f == f4) ? f4(__VA_ARGS__) :Quoted 7 lines
do we really want "likely" here ? in our cases there is no preference
on whuch fN is going to have the top priority, all of them are equally
important and statically configured and guranteed to not change on data
path ..Quoted 5 lines
I was a little undecided about that, too. '
likely()' is there mainly for simmetry with the already existing _1 and _2 variants. In such macros the branch prediction hint represent a real priority of the available choices.
Quoted 6 lines
For macro _1 it make sense to have the likely keyword but for _2 it doesn't, by looking at most of the usecases of
INDIRECT_CALL_2, they seem to be all around improving tcp/udp related indirection calls in the protocol stack, and they seem to prefer tcp over udp. But IMHO at least for the above usecase I think the likely keyword is being misused here and should be remove from allINDIRECT_CALL_Nwhere N > 1;
I experimented a bit with gcc 8.3.1 and some BP hint variations:
* with current macros we have single test for fN and an incresing number of conditional jumps and tests for the following functions, as the generated code looks somehow alike:
cmp f4, function_ptr
jne test_f3
call f4post_call:
// ... // ...
test_f3:
cmp f3, function_ptr
jne test_f2
call f3
jmp post_call
test_f2:
cmp f2, function_ptr //...
* keeping 'likely' only on INDIRECT_CALL_1 we get a conditinal jump for fN and the number of conditional jumps and tests grows for the next functions, as the generated code looks somehow alike:
cmp f4, function_ptr
je call_f4
cmp f3, function_ptr
je call_f3
//...
cmp f1, function_ptr
jne indirect_call
call f1post_call:
// ... // ...
call_f4:
call f4
jmp post_callcall_f3:
call f3 jmp post_call // ...
* without any BP hints, is quite alike the above, except for the last test, the indirect call don't need an additional jump:
cmp f4, function_ptr
je call_f4
cmp f3, function_ptr
je call_f3
//...
cmp f1, function_ptr
je call_f1
call retpoline_helper
I think the first option should be overall better then the 2nd. The 3rd one is the worse.
Quoted 2 lines
In any case, just make sure to use the order i suggested in next patch with:
MLX5_RX_INDIRECT_CALL_LIST
Sure! will do in next iteration, as soon as the above topic is settled.
Thanks!
Paolo
Paolo Abeni
maintainerPATCH 3/3 net/mlx5e: use indirect calls wrapper for the rx packet handler
In reply to Paolo AbeniWe can avoid another indirect call per packet wrapping the rx handler call with the proper helper.
To ensure that even the last listed direct call experience measurable gain, despite the additional conditionals we must traverse before reaching it, I tested reversing the order of the listed options, with performance differences below noise level.
Together with the previous indirect call patch, this gives ~6% performance improvement in raw UDP tput.
--- drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c +3 -1
@@ -1333,7 +1333,9 @@ int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget)mlx5_cqwq_pop (cqwq); - rq->handle_rx_cqe (rq, cqe); +INDIRECT_CALL_4 (rq->handle_rx_cqe, mlx5e_handle_rx_cqe_mpwrq, + mlx5e_handle_rx_cqe, mlx5e_handle_rx_cqe_rep, + mlx5e_ipsec_handle_rx_cqe, rq, cqe); }while ((++work_done < budget) && (cqe =mlx5_cqwq_get_cqe (cqwq))); out:
Signature · 1 line
2.20.1
Re: [PATCH net-next 3/3] net/mlx5e: use indirect calls wrapper for the rx packet handler
Quoted 11 lines · Paolo Abeni
We can avoid another indirect call per packet wrapping the rx handler call with the proper helper.
To ensure that even the last listed direct call experience measurable gain, despite the additional conditionals we must traverse before reaching it, I tested reversing the order of the listed options, with performance differences below noise level.
Together with the previous indirect call patch, this gives ~6% performance improvement in raw UDP tput.
Nice ! I like it.
Quoted 11 lines
Signed-off-by Paolo Abeni <[email protected]>--- drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)drivers/net/ethernet/mellanox/mlx5/core/en_rx.c +0
-0
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int · line 1341
Quoted 10 lines
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c +3
-1@@ -1333,7 +1333,9 @@ int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget)mlx5_cqwq_pop (cqwq ); - rq->handle_rx_cqe (rq, cqe); +INDIRECT_CALL_4 (rq->handle_rx_cqe, mlx5e_handle_rx_cqe_mpwrq, + mlx5e_handle_rx_cqe, mlx5e_handle_rx_cqe_rep, + mlx5e_ipsec_handle_rx_cqe, rq, cqe);
you missed mlx5i_handle_rx_cqe, anyway don't add INDIRECT_CALL_5 :D
just replace mlx5e_handle_rx_cqe_rep with mlx5i_handle_rx_cqe, mlx5e_handle_rx_cqe_rep is actually a slow path of switchdev mode.
Maybe define the list somewhere in en.h where it is visible for every one:
#define MLX5_RX_INDIRECT_CALL_LIST \mlx5e_handle_rx_cqe_mpwrq, mlx5e_handle_rx_cqe, mlx5i_handle_rx_cqe, mlx5e_ipsec_handle_rx_cqe
and here:INDIRECT_CALL_4(rq->handle_rx_cqe, MLX5_RX_INDIRECT_CALL_LIST, rq, cqe);
Quoted 4 lines
} while ((++work_done < budget) && (cqe =
mlx5_cqwq_get_cqe(cqwq)));out:
Paolo Abeni
maintainerRe: [PATCH net-next 3/3] net/mlx5e: use indirect calls wrapper for the rx packet handler
Quoted 39 lines · Saeed Mahameed
Quoted 11 lines · Paolo Abeni
We can avoid another indirect call per packet wrapping the rx handler call with the proper helper.
To ensure that even the last listed direct call experience measurable gain, despite the additional conditionals we must traverse before reaching it, I tested reversing the order of the listed options, with performance differences below noise level.
Together with the previous indirect call patch, this gives ~6% performance improvement in raw UDP tput.
Nice ! I like it.
Quoted 11 lines
Signed-off-by Paolo Abeni <[email protected]>--- drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)drivers/net/ethernet/mellanox/mlx5/core/en_rx.c +0
-0drivers/net/ethernet/mellanox/mlx5/core/en_rx.c int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int · line 1341
Quoted 10 lines
drivers/net/ethernet/mellanox/mlx5/core/en_rx.c +3
-1@@ -1333,7 +1333,9 @@ int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget)mlx5_cqwq_pop (cqwq ); - rq->handle_rx_cqe (rq, cqe); +INDIRECT_CALL_4 (rq->handle_rx_cqe, mlx5e_handle_rx_cqe_mpwrq, + mlx5e_handle_rx_cqe, mlx5e_handle_rx_cqe_rep, + mlx5e_ipsec_handle_rx_cqe, rq, cqe);you missed
mlx5i_handle_rx_cqe, anyway don't addINDIRECT_CALL_5:D
Quoted 2 lines
just replace
mlx5e_handle_rx_cqe_repwithmlx5i_handle_rx_cqe,mlx5e_handle_rx_cqe_repis actually a slow path of switchdev mode.
Thank you! This is exactly the kind of feedback I was looking for! I hoped some of the options was less relevant than the other, but I do not know the driver well enough to guess. Also I missed completely mlx5i_handle_rx_cqe, as you noded.
Quoted 10 lines
Maybe define the list somewhere in
en.hwhere it is visible for every one:#define
MLX5_RX_INDIRECT_CALL_LIST\mlx5e_handle_rx_cqe_mpwrq,mlx5e_handle_rx_cqe,mlx5i_handle_rx_cqe,mlx5e_ipsec_handle_rx_cqeand here:
INDIRECT_CALL_4(rq->handle_rx_cqe, MLX5_RX_INDIRECT_CALL_LIST, rq, cqe);
Will do in v2, unless this patch will be dropped, please see my reply to patch 2/3.
Thanks,
Paolo