Actions
Bug #7378
open
LS
LS
dpdk: having too few hugepages can lead to segfault on startup
Bug #7378:
dpdk: having too few hugepages can lead to segfault on startup
Affected Versions:
Effort:
Difficulty:
Label:
Description
Starting Suri with the usual params and mempool size set to ~262143.
Just one 2MB Hugepage allocated.
The output:
... Config: detect: No rules loaded from /dev/null [SigLoadSignatures:detect-engine-loader.c:351] Info: detect: No signatures supplied. [SigLoadSignatures:detect-engine-loader.c:361] RING: Cannot reserve memory HASH: memory allocation failed ice_fdir_init_filter_list(): Failed to create fdir hash table! ice_fdir_setup(): Failed to init FDIR filter list. ice_flow_init(): Failed to initialize engine 1 ice_dev_init(): Failed to initialize flow Segmentation fault
This is up for a discussion if Suricata should still be responsible for this as the driver crashes Suricata. It is possible however to preestimate the size of hugepages by considering - mempool size, (potentially number of mempools - mempool per interface/queue), mbuf size (mtu + extra).
OT Updated by OISF Ticketbot over 1 year ago
- Subtask #7379 added
OT Updated by OISF Ticketbot over 1 year ago
- Label deleted (
Needs backport to 7.0)
LS Updated by Lukas Sismis about 1 year ago
- Status changed from New to Assigned
VJ Updated by Victor Julien about 1 year ago
An important part of this is to understand where the actual crash happens. I suspect it may be in Suricata's handling of the DPDK error, see also https://github.com/OISF/suricata/pull/12654#issuecomment-2678077622
VJ Updated by Victor Julien 12 months ago
- Target version changed from 8.0.0-beta1 to 8.0.0-rc1
VJ Updated by Victor Julien 10 months ago
- Target version changed from 8.0.0-rc1 to 9.0.0-beta1
Actions