| Message ID | 20150818121439.3E5621A241A@localhost.localdomain |
|---|---|
| State | Rejected, archived |
| Delegated to: | David Miller |
| Headers | show |
On Tue, 2015-08-18 at 14:14 +0200, Christophe Leroy wrote: > ip_send_check() is a function that does nothing but calling > ip_fast_csum() after having zeroed iph->check > > We get (on PPC) a quite long function for just that, as it requires > setting the stack frame, saving volatile regs, restoring them after > the call, etc ... > > c02d3e3c <ip_send_check>: > c02d3e3c: 7c 08 02 a6 mflr r0 > c02d3e40: 94 21 ff f0 stwu r1,-16(r1) > c02d3e44: 39 20 00 00 li r9,0 > c02d3e48: 93 e1 00 0c stw r31,12(r1) > c02d3e4c: 90 01 00 14 stw r0,20(r1) > c02d3e50: b1 23 00 0a sth r9,10(r3) > c02d3e54: 88 83 00 00 lbz r4,0(r3) > c02d3e58: 7c 7f 1b 78 mr r31,r3 > c02d3e5c: 54 84 07 3e clrlwi r4,r4,28 > c02d3e60: 4b d3 ab 7d bl c000e9dc <ip_fast_csum> > c02d3e64: b0 7f 00 0a sth r3,10(r31) > c02d3e68: 80 01 00 14 lwz r0,20(r1) > c02d3e6c: 83 e1 00 0c lwz r31,12(r1) > c02d3e70: 7c 08 03 a6 mtlr r0 > c02d3e74: 38 21 00 10 addi r1,r1,16 > c02d3e78: 4e 80 00 20 blr That's because you should inline ip_fast_csum() instead. We do not want to inline ip_send_check() on eg x86, as it is clearly too long. ffffffff816d72e0 <ip_send_check>: ffffffff816d72e0: e8 9b e3 0c 00 callq ffffffff817a5680 <__fentry__> ffffffff816d72e5: 0f b6 17 movzbl (%rdi),%edx ffffffff816d72e8: 55 push %rbp ffffffff816d72e9: 31 c0 xor %eax,%eax ffffffff816d72eb: 66 89 47 0a mov %ax,0xa(%rdi) ffffffff816d72ef: 48 89 f8 mov %rdi,%rax ffffffff816d72f2: 48 89 e5 mov %rsp,%rbp ffffffff816d72f5: 83 e2 0f and $0xf,%edx ffffffff816d72f8: 8b 08 mov (%rax),%ecx ffffffff816d72fa: 83 ea 04 sub $0x4,%edx ffffffff816d72fd: 76 24 jbe ffffffff816d7323 <ip_send_check+0x43> ffffffff816d72ff: 03 48 04 add 0x4(%rax),%ecx ffffffff816d7302: 13 48 08 adc 0x8(%rax),%ecx ffffffff816d7305: 13 48 0c adc 0xc(%rax),%ecx ffffffff816d7308: 13 48 10 adc 0x10(%rax),%ecx ffffffff816d730b: 48 8d 40 04 lea 0x4(%rax),%rax ffffffff816d730f: ff ca dec %edx ffffffff816d7311: 75 f5 jne ffffffff816d7308 <ip_send_check+0x28> ffffffff816d7313: 83 d1 00 adc $0x0,%ecx ffffffff816d7316: 89 ca mov %ecx,%edx ffffffff816d7318: c1 e9 10 shr $0x10,%ecx ffffffff816d731b: 66 01 d1 add %dx,%cx ffffffff816d731e: 83 d1 00 adc $0x0,%ecx ffffffff816d7321: f7 d1 not %ecx ffffffff816d7323: 66 89 4f 0a mov %cx,0xa(%rdi) ffffffff816d7327: 5d pop %rbp ffffffff816d7328: c3 retq -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/include/net/ip.h b/include/net/ip.h index c0c26c3..b88d592 100644 --- a/include/net/ip.h +++ b/include/net/ip.h @@ -32,6 +32,7 @@ #include <net/snmp.h> #include <net/flow.h> #include <net/flow_keys.h> +#include <net/checksum.h> struct sock; @@ -110,7 +111,13 @@ int ip_output(struct sock *sk, struct sk_buff *skb); int ip_mc_output(struct sock *sk, struct sk_buff *skb); int ip_fragment(struct sk_buff *skb, int (*output)(struct sk_buff *)); int ip_do_nat(struct sk_buff *skb); -void ip_send_check(struct iphdr *ip); +/* Generate a checksum for an outgoing IP datagram. */ +static inline void ip_send_check(struct iphdr *iph) +{ + iph->check = 0; + iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); +} + int __ip_local_out(struct sk_buff *skb); int ip_local_out_sk(struct sock *sk, struct sk_buff *skb); static inline int ip_local_out(struct sk_buff *skb) diff --git a/net/ipv4/ip_output.c b/net/ipv4/ip_output.c index 357c2a9..fa054cf 100644 --- a/net/ipv4/ip_output.c +++ b/net/ipv4/ip_output.c @@ -83,14 +83,6 @@ int sysctl_ip_default_ttl __read_mostly = IPDEFTTL; EXPORT_SYMBOL(sysctl_ip_default_ttl); -/* Generate a checksum for an outgoing IP datagram. */ -void ip_send_check(struct iphdr *iph) -{ - iph->check = 0; - iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl); -} -EXPORT_SYMBOL(ip_send_check); - int __ip_local_out(struct sk_buff *skb) { struct iphdr *iph = ip_hdr(skb);
ip_send_check() is a function that does nothing but calling ip_fast_csum() after having zeroed iph->check We get (on PPC) a quite long function for just that, as it requires setting the stack frame, saving volatile regs, restoring them after the call, etc ... c02d3e3c <ip_send_check>: c02d3e3c: 7c 08 02 a6 mflr r0 c02d3e40: 94 21 ff f0 stwu r1,-16(r1) c02d3e44: 39 20 00 00 li r9,0 c02d3e48: 93 e1 00 0c stw r31,12(r1) c02d3e4c: 90 01 00 14 stw r0,20(r1) c02d3e50: b1 23 00 0a sth r9,10(r3) c02d3e54: 88 83 00 00 lbz r4,0(r3) c02d3e58: 7c 7f 1b 78 mr r31,r3 c02d3e5c: 54 84 07 3e clrlwi r4,r4,28 c02d3e60: 4b d3 ab 7d bl c000e9dc <ip_fast_csum> c02d3e64: b0 7f 00 0a sth r3,10(r31) c02d3e68: 80 01 00 14 lwz r0,20(r1) c02d3e6c: 83 e1 00 0c lwz r31,12(r1) c02d3e70: 7c 08 03 a6 mtlr r0 c02d3e74: 38 21 00 10 addi r1,r1,16 c02d3e78: 4e 80 00 20 blr Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr> --- include/net/ip.h | 9 ++++++++- net/ipv4/ip_output.c | 8 -------- 2 files changed, 8 insertions(+), 9 deletions(-)