Re: [Ocf-linux-users] ocf-linux-20071215's ixp4xx freeze as cryptotest with LinuxLink 2.6.16.11 and NPE v2.4

[David M. <Dav...@se...>]

Jivin JUN J lays it down ...
> Hi, I did tests by Intel's crypto codelets stuff and have the result in the 
> attachment crypto_codelets_4096_Mar19th.log
> 
> And I turned the debug on of the NPE driver and added some
> ixCryptoAccShow() functions to see the status of write queue and read
> queue, > It seems like the write queue(29) is full while no one is
> really draining it, 
> which leads to the "cannot allocate memory" error.
> see attachment cryptotest.log

Sounds like your access library code is a little mixed up.
Are you loading the corrent NPE image with crypto ?

If the codelets are having problems then I would guess something is
wrong in there.

Cheers,
Davidm

> David McCullough <Dav...@se...> wrote: 
> Jivin JUN J lays it down ...
> > 
> > I just subscribed and not to follow the old discussions, sorry.
> 
> No problems.  See below.
> 
> > Platform: Intel IXP425, LinuxLink 2.6.16.11, Intel Access Library With Crypto v2.4.
> > ocf-linux-20071215 software.
> > 
> > I did:
> > insmod ixp400.ko
> > cat IxNpeMicrocode.dat > /dev/ixNpe
> > insmod ixp400_eth.ko
> > 
> > (ethernet works fine)
> > 
> > Then I did:
> > # insmod /lib/modules/2.6.16.11/kernel/crypto/ocf/ocf.ko crypto_debug=1
> > 
> > crypto_init(0xbf1acbc8)
> > crypto/ocf/crypto.c,1227:Q_LOCK()
> > crypto_proc-sleeping(qe=1 qb=0 kqe=1 kqb=0)
> > crypto/ocf/crypto.c,1374:Q_UNLOCK()
> > crypto/ocf/crypto.c,1415:RETQ_LOCK
> > crypto_ret_proc-sleeping
> > crypto/ocf/crypto.c,1446: RETQ_UNLOCK
> > 
> > 
> > # insmod /lib/modules/2.6.16.11/kernel/crypto/ocf/cryptodev.ko cryptodev_debug=1
> > 
> > cryptodev_init(bf1b7000)
> > 
> > # insmod /lib/modules/2.6.16.11/kernel/crypto/ocf/ixp4xx/ixp4xx.ko ixp_debug=1
> > 
> > ixp_init(bf1b8270)
> > crypto/ocf/crypto.c,469:DRIVER_LOCK()
> > crypto/ocf/crypto.c,514:DRIVER_UNLOCK()
> > crypto_register(id=0x0, alg=1, maxoplen=0, flags=0x0)
> > crypto/ocf/crypto.c,619:DRIVER_LOCK()
> > crypto/ocf/crypto.c,645:DRIVER_UNLOCK()
> > crypto_register(id=0x0, alg=2, maxoplen=0, flags=0x0)
> > crypto/ocf/crypto.c,619:DRIVER_LOCK()
> > crypto/ocf/crypto.c,645:DRIVER_UNLOCK()
> > crypto_register(id=0x0, alg=11, maxoplen=0, flags=0x0)
> > crypto/ocf/crypto.c,619:DRIVER_LOCK()
> > crypto/ocf/crypto.c,645:DRIVER_UNLOCK()
> > crypto_register(id=0x0, alg=13, maxoplen=0, flags=0x0)
> > crypto/ocf/crypto.c,619:DRIVER_LOCK()
> > crypto/ocf/crypto.c,645:DRIVER_UNLOCK()
> > crypto_register(id=0x0, alg=14, maxoplen=0, flags=0x0)
> > crypto/ocf/crypto.c,619:DRIVER_LOCK()
> > crypto/ocf/crypto.c,645:DRIVER_UNLOCK()
> > crypto_register(id=0x0, alg=6, maxoplen=0, flags=0x0)
> > crypto/ocf/crypto.c,619:DRIVER_LOCK()
> > crypto/ocf/crypto.c,645:DRIVER_UNLOCK()
> > crypto_register(id=0x0, alg=7, maxoplen=0, flags=0x0)
> > crypto/ocf/crypto.c,619:DRIVER_LOCK()
> > crypto/ocf/crypto.c,645: DRIVER_UNLOCK()
> > 
> > 
> > 
> > At last I did:
> > 
> > #cryptotest
> > 
> > cryptodev_open()
> > cryptodev_ioctl(cmd=c0046364 arg=bef84af4)
> > cryptodev_ioctl(CRIOGET)
> > cryptodev_ioctl(cmd=c030636a arg=bef84b3c)
> > cryptodev_ioctl(CIOCGSESSION2)
> > cryptodev_ioctl(CIOCGSESSION2) - no mac
> > crypto/ocf/crypto.c,355: DRIVER_LOCK()
> > crypto/ocf/crypto.c,379: DRIVER_UNLOCK()
> > ixp_newsession():alg 2
> > crypto/ocf/crypto.c,381: DRIVER_LOCK()
> > crypto/ocf/crypto.c,391: DRIVER_UNLOCK()
> > csecreate()
> > cseadd()
> > cryptodev_ioctl(cmd=c01c6367 arg=bef84b20)
> > cryptodev_ioctl(CIOCCRYPT)
> > csefind()
> > cryptodev_op()
> > crypto_dispatch()
> > crypto_invoke()
> > ixp_process()
> > cryptodev_op about to WAIT
> > ixp_register_cb(999, 00000000, 0)
> > ixp_process_pending(c2885000)
> > ixp_q_process(c2e5a9e0)
> > 
> > (freezed here)
> > 
> > ___________________________________________________________________
> > 
> > I tracked down the source a little and found that it was waiting for the crp_flags
> > to be set to CRYPTO_F_DONE around line 310 in crypto/ocf/cryptodev.c.
> > And I found there's only one function:
> > void crypto_done(struct cryptop *crp);
> > to set crp_flags to CRYPTO_F_DONE.
> > (Ctrl-c also sends signal to break out the loop but it's been disabled intently.) 
> > 
> > 
> > >From the debug output of ixp4xx.ko I followed the last function ixp_q_process() 
> > and saw it returned successfully with IX_CRYPTO_ACC_STATUS_SUCCESS == status
> > (line 550 in crypto/ocf/ixp4xx/ixp4xx.c) ... 
> > which means crypto_done() has never been called since the request was sent to
> > NPE stuff regardless of the ixCryptoAcc.... function returned successfully.
> > 
> > I don't understand. Actually I met this situation when I was tring to integrate 
> > MontaVista4.01 plus NPE v2.3 and ocf-linux ...
> > 
> > Any suggestion or hints would be appreciated, thank you very much.
> 
> Thanks for all the detail,  makes it easy for see what is happening ;-)
> 
> It looks fairly simple,  try the attached "untested" patch and see how you go.
> 
> Somehow you have generated a request that does not call back through the
> ixp_perform_cb or ixp_hash_perform_cb routines.  Obviously we hadn't hit
> that one before.  Of course all we had to do was run cryptotest without
> args, but that is obviously too hard :-) :-)
> 
> Cheers,
> Davidm
> 
> 
> -- 
> David McCullough,  dav...@se...,   Ph:+61 734352815
> Secure Computing - SnapGear  http://www.uCdot.org http://www.cyberguard.com
> Index: ocf/ixp4xx/ixp4xx.c
> ===================================================================
> RCS file: ocf/ixp4xx/ixp4xx.c,v
> retrieving revision 1.46
> diff -u -r1.46 ixp4xx.c
> --- ocf/ixp4xx/ixp4xx.c 12 Oct 2007 01:10:31 -0000 1.46
> +++ ocf/ixp4xx/ixp4xx.c 18 Mar 2008 04:48:30 -0000
> @@ -547,16 +547,15 @@
>     q->ixp_q_iv);
>   }
>  
> - if (IX_CRYPTO_ACC_STATUS_SUCCESS == status)
> -  return;
> -
>   if (IX_CRYPTO_ACC_STATUS_QUEUE_FULL == status) {
>    q->ixp_q_crp->crp_etype = ENOMEM;
>    goto done;
>   }
>  
> - printk("ixp: %s failed %u\n", crypt_func, status);
> - q->ixp_q_crp->crp_etype = EINVAL;
> + if (IX_CRYPTO_ACC_STATUS_SUCCESS != status) {
> +  printk("ixp: %s failed %u\n", crypt_func, status);
> +  q->ixp_q_crp->crp_etype = EINVAL;
> + }
>  
>  done:
>   crypto_done(q->ixp_q_crp);
> 
> 
>        
> ---------------------------------
> Be a better friend, newshound, and know-it-all with Yahoo! Mobile.  Try it now.



-- 
David McCullough,  dav...@se...,   Ph:+61 734352815
Secure Computing - SnapGear  http://www.uCdot.org http://www.cyberguard.com