Valgrind does not run well with VPP for a various of reasons (too slow, VPP
doing too many nasty tricks and using its own memory allocators etc.).
- for profiling, I'd recommend using Linux perf
- for memory errors there is a support for Address Sanitizer (I need to spend
some time to update it but it should work apart from running make test):
https://fd.io/docs/vpp/master/troubleshooting/sanitizer.html
- for memory leaks you can use VPP memory traces:
https://fd.io/docs/vpp/master/troubleshooting/mem.html
Best
ben
> -----Original Message-----
> From: vpp-dev@lists.fd.io <vpp-dev@lists.fd.io> On Behalf Of bhishma via
> lists.fd.io
> Sent: vendredi 7 mai 2021 11:27
> To: vpp-dev@lists.fd.io
> Subject: [vpp-dev] Running VPP in valgrind #vpp-dev
>
> Hi All,
>
>
>
>
> We are using VPP as our x86 based forwarder. For our use case, we have
> extended VPP by adding our own specific nodes including process nodes.
>
>
>
>
> The startup.conf as shown below:
>
>
>
> unix {
>
> nodaemon
>
> log /var/log/vpp/vpp.log
>
> full-coredump
>
> cli-listen localhost:5003
>
> runtime-dir /shm/run/vpp/
>
> }
>
>
> api-trace {
>
> on
>
> }
>
>
> cpu {
>
> main-core 2
>
> workers 2
>
> }
>
>
> heapsize 500M
>
>
> ip {
>
> heap-size 32M
>
> }
>
>
> ip6 {
>
> heap-size 20M
>
> hash-buckets 131072
>
> }
>
>
> statseg {
>
> size 150M
>
> }
>
>
> plugins {
>
> plugin dpdk_plugin.so { disable }
>
> }
>
>
>
>
> We would like to run valgrind/callgrind for profiling the non-vpp portions
> of code running in the same process.
>
>
>
>
>
>
> When we start vpp with valgrind, we are seeing the below issue:
>
>
>
> unix_config:492: couldn't open log '/var/log/vpp/vpp.log'
>
> /usr/local/bin/vpp[1741]: clib_elf_parse_file: open `/usr/local/bin/vpp':
> No such file or directory
>
> /usr/local/bin/vpp[1741]: clib_sysfs_prealloc_hugepages:261: pre-
> allocating 19 additional 2048K hugepages on numa node 0
>
> /usr/local/bin/vpp[1741]: buffer: vlib_physmem_shared_map_create:
> pmalloc_map_pages: failed to mmap 19 pages at 0x0 fd 15 numa 0 flags 0x11:
> Operation not permitted
>
>
> /usr/local/bin/vpp[1741]: buffer: falling back to non-hugepage backed
> buffer pool
>
> ==1741== Warning: unimplemented fcntl command: 1033
>
> /usr/local/bin/vpp[1741]: vlib_physmem_shared_map_create:
> clib_mem_create_fd: fcntl (F_ADD_SEALS): Invalid argument
>
> /usr/local/bin/vpp[1741]: vlib_buffer_main_init: failed to allocate buffer
> pool(s)
>
> 2021-05-06T16:46:01.987175+0000 loop_poll.c:368 [trace] grown poll array
> to 10 for FD 6
>
>
>
>
> Is there any specific options or anything that needs to be tweaked in the
> startup conf file to run valgrind/callgrind successfully ?
>
>
>
>
> To reiterate we would just like valgrind/callgrind for profiling the non-
> VPP portions, but of course VPP needs to run first as it is the top of our
> event loop.
>
>
> Thanks,
> Bhishma
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#19360): https://lists.fd.io/g/vpp-dev/message/19360
Mute This Topic: https://lists.fd.io/mt/82651125/21656
Mute #vpp-dev:https://lists.fd.io/g/vpp-dev/mutehashtag/vpp-dev
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-
