Thread (7 messages) 7 messages, 3 authors, 2018-01-26

Re: [dpdk-stable] [PATCH v2] app/procinfo: Fix memory leak by rte_service_init

From: Van Haaren, Harry <hidden>
Date: 2018-01-26 17:15:17

From: Thomas Monjalon [mailto:thomas@monjalon.net]
Sent: Friday, January 26, 2018 5:00 PM
To: Varghese, Vipin <redacted>; Van Haaren, Harry
[off-list ref]
Cc: stable@dpdk.org; dev@dpdk.org
Subject: Re: [dpdk-stable] [PATCH v2] app/procinfo: Fix memory leak by
rte_service_init

11/01/2018 20:47, Vipin Varghese:
quoted
When procinfo is run multiple times against primary application, it
consumes huge page memory by rte_service_init. Which is not released
at exit of application.

Invoking rte_service_finalize to real memory and prevent memory leak.
I don't think it is correct to call rte_service_finalize in applications,
while rte_service_init is called in EAL.

Maybe we need a new function in EAL.
Yes correct - we need a rte_eal_deinit(), cleanup() or finalize() or something. This ties in with splitting EAL to be more modular on startup, and DPDK in general behaving more like a library and less like a single-monolith.

For the 18.02 timeframe, the simplest solution to solve the secondary process mem-leak issue than to merge into these applications, unfortunately. 

The only other option I see is to add an rte_eal_finalize() function, and hide this call behind it, however it is quite late to add such a function, and what do we do with cases like rte_panic(), rte_exit(), or system signals like SIGINT, SIGHUP etc? It seems too complicated to add "quickly" to me.

If there is technically a better solution viable in the given timeframe, I'm open to suggestions?
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help