[dpdk-users] Crash in FM10k - fm10k_rxtx_vec.c

[Van Haaren, Harry]

> -----Original Message-----
> From: users [mailto:users-bounces at dpdk.org] On Behalf Of Anthony Hart
> Sent: Thursday, February 9, 2017 11:43 AM
> To: users at dpdk.org
> Subject: [dpdk-users] Crash in FM10k - fm10k_rxtx_vec.c

Hi Anthony,

> Found a crash in the FM10k vector driver when it tries to send a packet with a VLAN header
> priority (PCP) field value >= 4.  This results in the FM10k returning the following error.
> 
> testpmd> PMD: fm10k_dev_interrupt_handler_pf(): INT: find fault!
> PMD: fm10k_dev_handle_fault(): THI_MAL_DIS_Q_FAULT: PF(0) Addr:0x0 Spec: 0x0
> 
> The reason is the (pkt->vlan_tci << 16) value gets sign extended and causes illegal values to
> be written into the TX descriptor.   The following is a patch to fix the issue.

I'm not familiar with the fm10k pmd, but this sounds like something that should be fixed.


> I don’t know
> what the procedure is for getting this into the fix stream.

There is documentation on how to create patches, and submit them to the mailing list:
http://dpdk.org/doc/guides/contributing/patches.html#creating-patches

When sending the patch, CC the fm10k maintainer: Jing Chen <jing.d.chen at intel.com>

Hope that helps, -Harry


> --- ./drivers/net/fm10k/fm10k_rxtx_vec.c	2016-11-13 09:28:12.000000000 -0500
> +++ ./drivers/net/fm10k/fm10k_rxtx_vec.c.new	2017-02-09 06:37:00.362960064 -0500
> @@ -718,7 +718,7 @@ vtx1(volatile struct fm10k_tx_desc *txdp
>  		struct rte_mbuf *pkt, uint64_t flags)
>  {
>  	__m128i descriptor = _mm_set_epi64x(flags << 56 |
> -			pkt->vlan_tci << 16 | pkt->data_len,
> +                        (uint64_t)(pkt->vlan_tci) << 16 | (uint64_t)pkt->data_len,
>  			MBUF_DMA_ADDR(pkt));
>  	_mm_store_si128((__m128i *)txdp, descriptor);
>  }