From: Joey Zhong <[email protected]>
ulp_ha_mgr_timer_cb() acquires the global ULP context spinlock via
bnxt_ulp_cntxt_entry_acquire(). When myclient_cnt is transiently 0,
the function did a bare return without calling
bnxt_ulp_cntxt_entry_release(), leaking the lock and deadlocking all
subsequent flow operations on the device.
Replace the bare return with goto cb_restart, matching every other
error path in the callback: it releases the lock and restarts the
timer. Since a zero client count is transient, restarting the timer
is the correct behavior, so the misleading "kill timer" log message
is updated accordingly.
Fixes: 1993b267dbcb ("net/bnxt: cleanup ULP parser and mapper")
Cc: [email protected]
Signed-off-by: Joey Zhong <[email protected]>
Signed-off-by: Manish Kurup <[email protected]>
---
drivers/net/bnxt/tf_ulp/ulp_ha_mgr.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/net/bnxt/tf_ulp/ulp_ha_mgr.c
b/drivers/net/bnxt/tf_ulp/ulp_ha_mgr.c
index 7b7c10d360..d558f78c8c 100644
--- a/drivers/net/bnxt/tf_ulp/ulp_ha_mgr.c
+++ b/drivers/net/bnxt/tf_ulp/ulp_ha_mgr.c
@@ -256,8 +256,8 @@ ulp_ha_mgr_timer_cb(void *arg)
myclient_cnt = bnxt_ulp_cntxt_num_shared_clients_get(ulp_ctx);
if (myclient_cnt == 0) {
BNXT_DRV_DBG(ERR,
- "PANIC Client Count is zero kill timer\n.");
- return;
+ "Client Count is zero, skip and restart timer.\n");
+ goto cb_restart;
}
tfp = bnxt_ulp_cntxt_tfp_get(ulp_ctx, BNXT_ULP_SESSION_TYPE_SHARED_WC);
--
2.31.1
