crypto: BUG: spinlock recursion when doing iperf over ipsec with crypto hardware device

[Corentin Labbe]

Le Fri, Jan 14, 2022 at 03:23:26PM +1100, Herbert Xu a écrit :
> On Fri, Jan 14, 2022 at 03:14:38PM +1100, Herbert Xu wrote:
> > On Tue, Jan 11, 2022 at 10:47:12AM +0100, Corentin Labbe wrote:
> > >
> > > [   44.646050] [<c0100afc>] (__irq_svc) from [<c080b9d4>] (xfrm_replay_advance+0x11c/0x3dc)
> > > [   44.654143] [<c080b9d4>] (xfrm_replay_advance) from [<c0809388>] (xfrm_input+0x4d0/0x1304)
> > > [   44.662408] [<c0809388>] (xfrm_input) from [<c03a3d88>] (crypto_finalize_request+0x5c/0xc4)
> > > [   44.670766] [<c03a3d88>] (crypto_finalize_request) from [<c06a0888>] (sun8i_ce_cipher_run+0x34/0x3c)
> > > [   44.679900] [<c06a0888>] (sun8i_ce_cipher_run) from [<c03a4264>] (crypto_pump_work+0x1a8/0x330)
> > 
> > So did sun8i_ce_cipher_run ensure that BH is disabled before
> > invoking xfrm_input? If not then this explains the dead-lock.
> 
> The issue appears to be with crypto_engine.  It needs to ensure
> that completion functions are called with BH disabled, not IRQ
> disabled and definitely not BH enabled.
> 

Hello

This minimal patch fix my issue, does it is the rigth way ?

Thanks for your help
Regards

diff --git a/crypto/crypto_engine.c b/crypto/crypto_engine.c
index fb07da9920ee..b3844f6d98a3 100644
--- a/crypto/crypto_engine.c
+++ b/crypto/crypto_engine.c
@@ -7,6 +7,7 @@
  * Author: Baolin Wang <baolin.wang at linaro.org>
  */
 
+#include <linux/bottom_half.h>
 #include <linux/err.h>
 #include <linux/delay.h>
 #include <linux/device.h>
@@ -53,7 +54,9 @@ static void crypto_finalize_request(struct crypto_engine *engine,
                                dev_err(engine->dev, "failed to unprepare request\n");
                }
        }
+       local_bh_disable();
        req->complete(req, err);
+       local_bh_enable();
 
        kthread_queue_work(engine->kworker, &engine->pump_requests);
 }