[dpdk-dev] [PATCH] crypto/qat: fix second seg fault in QAT PMD

[Iremonger, Bernard]

Hi Fiona,

<snip>

> > > Subject: [PATCH] crypto/qat: fix second seg fault in QAT PMD
> > >
> > > While running the IPsec unit tests the following segmentation fault
> > > is
> > > occurring:
> > >
> > > Thread 1 "test" received signal SIGSEGV, Segmentation fault.
> > > 0x0000000000bf3b50 in qat_sym_process_response (
> > > op=0x5355280 <unittest_params+512>, resp=0x100455100 "") at
> > > /root/dpdk/drivers/crypto/qat/qat_sym.h:161
> > > 161	if (sess->bpi_ctx)
> > >
> > > Fixes: 91614c73b67c ("crypto/qat: make response process function
> > > inline")
> > > Signed-off-by: Bernard Iremonger <bernard.iremonger at intel.com>
> > > ---
> > >  drivers/crypto/qat/qat_sym.h | 8 +++++---
> > >  1 file changed, 5 insertions(+), 3 deletions(-)
> > >
> > > diff --git a/drivers/crypto/qat/qat_sym.h
> > > b/drivers/crypto/qat/qat_sym.h index bc6426c..685c6c0 100644
> > > --- a/drivers/crypto/qat/qat_sym.h
> > > +++ b/drivers/crypto/qat/qat_sym.h
> > > @@ -157,10 +157,12 @@ qat_sym_process_response(void **op, uint8_t
> > *resp)
> > >  						rx_op->sym->session,
> > >  						cryptodev_qat_driver_id);
> > >
> > > -
> > > -		if (sess->bpi_ctx)
> > > +		if (sess == NULL)
> > > +			rx_op->status =
> > RTE_CRYPTO_OP_STATUS_AUTH_FAILED;
> > > +		else if (sess->bpi_ctx) {
> > >  			qat_bpicipher_postprocess(sess, rx_op);
> > > -		rx_op->status = RTE_CRYPTO_OP_STATUS_SUCCESS;
> > > +			rx_op->status = RTE_CRYPTO_OP_STATUS_SUCCESS;
> > > +		}
> > >  	}
> > >  	*op = (void *)rx_op;
> > >  }
> > > --
> > > 2.7.4
> > [Fiona] As for previous seg-fault- this is on data-path and likely due
> > to the application sending in an invalid session - or a session
> > created for a symmetric PMD, but being sent to a QAT asymmetric PMD -
> > whose driver_id would not appear in the session array.
> > Can you make sure to check for SYMMETRIC support in capabilities.
> >
> I have the asym PMD disabled in the common/config_base  file.
> I will investigate further.

Self NAK of this patch as causing  test failures.
 
 Regards,
 
 Bernard.