| debuginfo for kernel [message #25865] |
Wed, 09 January 2008 22:47  |
amorsen
Messages: 10
Registered: June 2007
|
Junior Member |
|
|
I am trying to nail down a performance problem with routing.
CPU: AMD64 processors, speed 2194.5 MHz (estimated)
Counted CPU_CLK_UNHALTED events (Cycles outside of halt state) with a unit mask of 0x00 (No unit mask) count 100000
vma samples % linenr info symbol name
ffffffff811e83a2 586492 25.2725 (no location information) dev_hard_start_xmit
ffffffff811e83a2 68 0.0116 (no location information)
ffffffff811e83a4 63 0.0107 (no location information)
ffffffff811e83aa 61 0.0104 (no location information)
ffffffff811e83b1 92 0.0157 (no location information)
ffffffff811e83c5 55 0.0094 (no location information)
ffffffff811e83cb 3 5.1e-04 (no location information)
ffffffff811e83d2 61 0.0104 (no location information)
ffffffff811e83d4 1 1.7e-04 (no location information)
ffffffff811e83dd 75 0.0128 (no location information)
ffffffff811e83ee 2 3.4e-04 (no location information)
ffffffff811e83f2 492753 84.0170 (no location information)
[..]
So a lot of the kernel time is spent in the dev_hard_start_xmit function, and in particular on the address ffffffff811e83f2. The big question is -- what is that instruction doing?
I think I need the debuginfo package to find out, but I can't find the debuginfo package.
/Benny
|
|
|
|
|
|
| Re: debuginfo for kernel [message #25911 is a reply to message #25872] |
Thu, 10 January 2008 18:39  |
amorsen
Messages: 10
Registered: June 2007
|
Junior Member |
|
|
I tried with gdb, but it is a bit difficult to figure out which bit in the assembly corresponds to which bit of the source code.
(gdb) disassemble dev_hard_start_xmit
Dump of assembler code for function dev_hard_start_xmit:
0xffffffff811e83a2 <dev_hard_start_xmit+0>: push %r13
0xffffffff811e83a4 <dev_hard_start_xmit+2>: push %r12
0xffffffff811e83a6 <dev_hard_start_xmit+4>: push %rbp
0xffffffff811e83a7 <dev_hard_start_xmit+5>: mov %rdi,%rbp
0xffffffff811e83aa <dev_hard_start_xmit+8>: push %rbx
0xffffffff811e83ab <dev_hard_start_xmit+9>: push %r13
0xffffffff811e83ad <dev_hard_start_xmit+11>: cmpq $0x0,(%rdi)
0xffffffff811e83b1 <dev_hard_start_xmit+15>: mov %rsi,%r13
0xffffffff811e83b4 <dev_hard_start_xmit+18>: jne 0xffffffff811e85ad <dev_hard_start_xmit+523>
0xffffffff811e83ba <dev_hard_start_xmit+24>: cmpq $0xffffffff8169b160,4926875(%rip) # 0xffffffff8169b160 <ptype_all>
0xffffffff811e83c5 <dev_hard_start_xmit+35>: je 0xffffffff811e84ed <dev_hard_start_xmit+331>
0xffffffff811e83cb <dev_hard_start_xmit+41>: cmpl $0x0,6474374(%rip) # 0xffffffff81814e58 <netstamp_needed>
0xffffffff811e83d2 <dev_hard_start_xmit+48>: je 0xffffffff811e83df <dev_hard_start_xmit+61>
0xffffffff811e83d4 <dev_hard_start_xmit+50>: callq 0xffffffff81045b0c <ktime_get_real>
0xffffffff811e83d9 <dev_hard_start_xmit+55>: mov %rax,0x18(%rbp)
0xffffffff811e83dd <dev_hard_start_xmit+59>: jmp 0xffffffff811e83e7 <dev_hard_start_xmit+69>
0xffffffff811e83df <dev_hard_start_xmit+61>: movq $0x0,0x18(%rdi)
0xffffffff811e83e7 <dev_hard_start_xmit+69>: mov 4926834(%rip),%rdx # 0xffffffff8169b160 <ptype_all>
0xffffffff811e83ee <dev_hard_start_xmit+76>: lea 0xffffffffffffffd0(%rdx),%r12
0xffffffff811e83f2 <dev_hard_start_xmit+80>: mov 0x30(%r12),%rax
0xffffffff811e83f7 <dev_hard_start_xmit+85>: prefetcht0 (%rax)
0xffffffff811e83fa <dev_hard_start_xmit+88>: cmp $0xffffffff8169b160,%rdx
0xffffffff811e8401 <dev_hard_start_xmit+95>: je 0xffffffff811e84ed <dev_hard_start_xmit+331>
[..]
I guess I will have to build the kernel myself. The problematic instruction is at dev_hard_start_xmit+80.
The performance problem is dropped packets when routing more than about 8000 packets/s.
|
|
|
|
| Re: debuginfo for kernel [message #25969 is a reply to message #25872] |
Sat, 12 January 2008 23:32 |
amorsen
Messages: 10
Registered: June 2007
|
Junior Member |
|
|
I tried rebuilding the kernel rpm to get debuginfo-packages. It didn't work.
I set builddebuginfo to 1, but no debuginfo packages were generated from rpmbuild -ba kernel-ovz.spec. I'm building for x86_64.
|
|
|
|
OpenVZ Forum
Members
Search
Help
Register
Login
Home
