eal/linux: prevent deadlocks on rte init and cleanup
Checks
Commit Message
Resolves a deadlock that can occur when multiple secondary
processes are starting and stopping. A deadlock can occur because
eal_memalloc_init() is taking a second unnecessary read lock.
If another DPDK process that is terminating enters rte_eal_memory_detach()
and acquires a write lock wait state before the starting process can
acquire it's second read lock then no process will be able to proceed.
Cc: stable@dpdk.org
Signed-off-by: Jonathan Erb <jonathan.erb@threatblockr.com>
---
.mailmap | 2 +-
lib/eal/linux/eal_memalloc.c | 4 ++--
2 files changed, 3 insertions(+), 3 deletions(-)
Comments
Hello Jonathan,
On Thu, Jul 20, 2023 at 7:19 AM Jonathan Erb
<jonathan.erb@threatblockr.com> wrote:
>
> Resolves a deadlock that can occur when multiple secondary
> processes are starting and stopping. A deadlock can occur because
> eal_memalloc_init() is taking a second unnecessary read lock.
> If another DPDK process that is terminating enters rte_eal_memory_detach()
> and acquires a write lock wait state before the starting process can
> acquire it's second read lock then no process will be able to proceed.
>
> Cc: stable@dpdk.org
>
> Signed-off-by: Jonathan Erb <jonathan.erb@threatblockr.com>
Thanks for the report and fix and sorry for the late reply.
Artemy came with a similar report and a more complete fix.
Could you confirm it works for you?
https://patchwork.dpdk.org/project/dpdk/list/?series=29463&state=%2A&archive=both
Thanks,
David,
Thanks for the follow-up. The proposed patch below from Artemy will work
for me.
Jonathan
On 9/29/23 10:32, David Marchand wrote:
> Hello Jonathan,
>
> On Thu, Jul 20, 2023 at 7:19 AM Jonathan Erb
> <jonathan.erb@threatblockr.com> wrote:
>> Resolves a deadlock that can occur when multiple secondary
>> processes are starting and stopping. A deadlock can occur because
>> eal_memalloc_init() is taking a second unnecessary read lock.
>> If another DPDK process that is terminating enters rte_eal_memory_detach()
>> and acquires a write lock wait state before the starting process can
>> acquire it's second read lock then no process will be able to proceed.
>>
>> Cc:stable@dpdk.org
>>
>> Signed-off-by: Jonathan Erb<jonathan.erb@threatblockr.com>
> Thanks for the report and fix and sorry for the late reply.
>
> Artemy came with a similar report and a more complete fix.
> Could you confirm it works for you?
>
> https://patchwork.dpdk.org/project/dpdk/list/?series=29463&state=%2A&archive=both
>
>
> Thanks,
>
@@ -663,7 +663,7 @@ John OLoughlin <john.oloughlin@intel.com>
John Ousterhout <ouster@cs.stanford.edu>
John W. Linville <linville@tuxdriver.com>
Jonas Pfefferle <jpf@zurich.ibm.com> <pepperjo@japf.ch>
-Jonathan Erb <jonathan.erb@banduracyber.com>
+Jonathan Erb <jonathan.erb@threatblockr.com>
Jon DeVree <nuxi@vault24.org>
Jon Loeliger <jdl@netgate.com>
Joongi Kim <joongi@an.kaist.ac.kr>
@@ -1740,7 +1740,7 @@ eal_memalloc_init(void)
eal_get_internal_configuration();
if (rte_eal_process_type() == RTE_PROC_SECONDARY)
- if (rte_memseg_list_walk(secondary_msl_create_walk, NULL) < 0)
+ if (rte_memseg_list_walk_thread_unsafe(secondary_msl_create_walk, NULL) < 0)
return -1;
if (rte_eal_process_type() == RTE_PROC_PRIMARY &&
internal_conf->in_memory) {
@@ -1778,7 +1778,7 @@ eal_memalloc_init(void)
}
/* initialize all of the fd lists */
- if (rte_memseg_list_walk(fd_list_create_walk, NULL))
+ if (rte_memseg_list_walk_thread_unsafe(fd_list_create_walk, NULL))
return -1;
return 0;
}