On 12-Jul-18 11:36 PM, Thomas Monjalon wrote:
26/06/2018 12:53, Anatoly Burakov:
From: Jianfeng Tan <jianfeng....@intel.com>
Next commit will make asynchronous IPC requests rely on alarm API,
which in turn relies on interrupts to work. Therefore, move the EAL
interrupt initialization before IPC initialization to avoid breaking
IPC in the next commit.
Signed-off-by: Jianfeng Tan <jianfeng....@intel.com>
Signed-off-by: Anatoly Burakov <anatoly.bura...@intel.com>
---
--- a/lib/librte_eal/linuxapp/eal/eal.c
+++ b/lib/librte_eal/linuxapp/eal/eal.c
@@ -839,6 +839,11 @@ rte_eal_init(int argc, char **argv)
rte_config_init();
+ if (rte_eal_intr_init() < 0) {
+ rte_eal_init_alert("Cannot init interrupt-handling thread\n");
+ return -1;
+ }
+
/* Put mp channel init before bus scan so that we can init the vdev
* bus through mp channel in the secondary process before the bus scan.
*/
@@ -968,11 +973,6 @@ rte_eal_init(int argc, char **argv)
rte_config.master_lcore, (int)thread_id, cpuset,
ret == 0 ? "" : "...");
- if (rte_eal_intr_init() < 0) {
- rte_eal_init_alert("Cannot init interrupt-handling thread\n");
- return -1;
- }
-
RTE_LCORE_FOREACH_SLAVE(i) {
I am almost sure it will bring regressions.
Please think again about the consequences of initializing interrupt thread
before affinity setting, memory init, device init.
Opinions / ideas from anyone?
Since interrupt thread is no longer relying on rte_malloc, i do not
expect there to be any consequences for memory.
I also do not expect any consequences due to moving intr init before
setting CPU thread affinity, because first of all interrupt thread is
*already* run before setting thread affinity of the lcores (but after
setting thread affinity of the master), and it picks its own affinity
based on parsed coremask anyway, which is already parsed at a point
where i'm moving it to. Affinities will be set similarly to how they
were set before, because lcore information is already parsed.
As for device init, that is debatable. The only consequence i can think
of is if device interrupts happen right after enabling the intr handler
for that device and this causes some kind of issue or a race. But, we
already support device hotplug which essentially causes the same
situation to happen (interrupt handler initialized before bus
scan/probe), so i'm not really convinced this could have any negative
consequences either.
--
Thanks,
Anatoly
