From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga04.intel.com (mga04.intel.com [192.55.52.120]) by dpdk.org (Postfix) with ESMTP id D6D822583 for ; Wed, 25 Apr 2018 11:48:17 +0200 (CEST) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga006.fm.intel.com ([10.253.24.20]) by fmsmga104.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 25 Apr 2018 02:48:16 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.49,325,1520924400"; d="scan'208";a="223238418" Received: from fyigit-mobl.ger.corp.intel.com (HELO [10.237.221.42]) ([10.237.221.42]) by fmsmga006.fm.intel.com with ESMTP; 25 Apr 2018 02:48:14 -0700 To: Ophir Munk , Pascal Mazon Cc: "dev@dpdk.org" , Mordechay Haimovsky , Olga Shern , Thomas Monjalon , Raslan Darawsheh , Shahaf Shuler References: <20180423093846.81133-1-ferruh.yigit@intel.com> <20180424175408.42099-1-ferruh.yigit@intel.com> From: Ferruh Yigit Openpgp: preference=signencrypt Autocrypt: addr=ferruh.yigit@intel.com; prefer-encrypt=mutual; keydata= xsFNBFXZCFABEADCujshBOAaqPZpwShdkzkyGpJ15lmxiSr3jVMqOtQS/sB3FYLT0/d3+bvy qbL9YnlbPyRvZfnP3pXiKwkRoR1RJwEo2BOf6hxdzTmLRtGtwWzI9MwrUPj6n/ldiD58VAGQ +iR1I/z9UBUN/ZMksElA2D7Jgg7vZ78iKwNnd+vLBD6I61kVrZ45Vjo3r+pPOByUBXOUlxp9 GWEKKIrJ4eogqkVNSixN16VYK7xR+5OUkBYUO+sE6etSxCr7BahMPKxH+XPlZZjKrxciaWQb +dElz3Ab4Opl+ZT/bK2huX+W+NJBEBVzjTkhjSTjcyRdxvS1gwWRuXqAml/sh+KQjPV1PPHF YK5LcqLkle+OKTCa82OvUb7cr+ALxATIZXQkgmn+zFT8UzSS3aiBBohg3BtbTIWy51jNlYdy ezUZ4UxKSsFuUTPt+JjHQBvF7WKbmNGS3fCid5Iag4tWOfZoqiCNzxApkVugltxoc6rG2TyX CmI2rP0mQ0GOsGXA3+3c1MCdQFzdIn/5tLBZyKy4F54UFo35eOX8/g7OaE+xrgY/4bZjpxC1 1pd66AAtKb3aNXpHvIfkVV6NYloo52H+FUE5ZDPNCGD0/btFGPWmWRmkPybzColTy7fmPaGz cBcEEqHK4T0aY4UJmE7Ylvg255Kz7s6wGZe6IR3N0cKNv++O7QARAQABzSVGZXJydWggWWln aXQgPGZlcnJ1aC55aWdpdEBpbnRlbC5jb20+wsF+BBMBAgAoAhsDBgsJCAcDAgYVCAIJCgsE FgIDAQIeAQIXgAUCWZR3VQUJB33WBQAKCRD5M+tD3xNhH6DWEACVhEb8q1epPwZrUDoxzu7E TS1b8tmabOmnjXZRs6+EXgUVHkp2xxkCfDmL3pa5bC0G/74aJnWjNsdvE05V1cb4YK4kRQ62 FwDQ+hlrFrwFB3PtDZk1tpkzCRHvJgnIil+0MuEh32Y57ig6hy8yO8ql7Lohyrnpfk/nNpm4 jQGEF5qEeHcEFe1AZQlPHN/STno8NZSz2nl0b2cw+cujN1krmvB52Ah/2KugQ6pprVyrGrzB c34ZQO9OsmSjJlETCZk6EZzuhfe16iqBFbOSadi9sPcJRwaUQBid+xdFWl7GQ8qC3zNPibSF HmU43yBZUqJDZlhIcl6/cFpOSjv2sDWdtjEXTDn5y/0FsuY0mFE78ItC4kCTIVk17VZoywcd fmbbnwOSWzDq7hiUYuQGkIudJw5k/A1CMsyLkoUEGN3sLfsw6KASgS4XrrmPO4UVr3mH5bP1 yC7i1OVNpzvOxtahmzm481ID8sk72GC2RktTOHb0cX+qdoiMMfYgo3wRRDYCBt6YoGYUxF1p msjocXyqToKhhnFbXLaZlVfnQ9i2i8jsj9SKig+ewC2p3lkPj6ncye9q95bzhmUeJO6sFhJg Hiz6syOMg8yCcq60j07airybAuHIDNFWk0gaWAmtHZxLObZx2PVn2nv9kLYGohFekw0AOsIW ta++5m48dnCoAc7BTQRX1ky+ARAApzQNvXvE2q1LAS+Z+ni2R13Bb1cDS1ZYq1jgpR13+OKN ipzd8MPngRJilXxBaPTErhgzR0vGcNTYhjGMSyFIHVOoBq1VbP1a0Fi/NqWzJOowo/fDfgVy K4vuitc/gCJs+2se4hdZA4EQJxVlNM51lgYDNpjPGIA43MX15OLAip73+ho6NPBMuc5qse3X pAClNhBKfENRCWN428pi3WVkT+ABRTE0taxjJNP7bb+9TQYNRqGwnGzX5/XISv44asWIQCaq vOkXSUJLd//cdVNTqtL1wreCVVR5pMXj7VIrlk07fmmJVALCmGbFr53BMb8O+8dgK2A5mitM n44d+8KdJWOwziRxcaMk/LclmZS3Iv1TERtiWt98Y9AjeAtcgYPkA3ld0BcUKONogP8pHVz1 Ed3s5rDQ91yr1S0wuAzW91fxGUO4wY+uPmxCtFVuBgd9VT9NAKTUL0qHM7CDgCnZPe0TW6Zj 8OqtdCCyAfvU9cW5xWM7Icxhde6AtPxhDSBwE8fL2ZmrDmaA4jmUKXp3i4JxRPSX84S08b+s DWXHPxy10UFU5A7EK/BEbZAKBwn9ROfm+WK+6X5xOGLoRE++OqNuUudxC1GDyLOPaqCbBCS9 +P6HsTHzxsjyJa27n4jcrcuY3P9TEcFJYSZSeSDh8mVGvugi0exnSJrrBZDyVCcAEQEAAcLB ZQQYAQIADwIbDAUCWZR1ZwUJA59cIQAKCRD5M+tD3xNhH5b+D/9XG44Ci6STdcA5RO/ur05J EE3Ux1DCHZ5V7vNAtX/8Wg4l4GZfweauXwuJ1w7Sp7fklwcNC6wsceI+EmNjGMqfIaukGetG +jBGqsQ7moOZodfXUoCK98gblKgt/BPYMVidzlGC8Q/+lZg1+o29sPnwImW+MXt/Z5az/Z17 Qc265g+p5cqJHzq6bpQdnF7Fu6btKU/kv6wJghENvgMXBuyThqsyFReJWFh2wfaKyuix3Zyj ccq7/blkhzIKmtFWgDcgaSc2UAuJU+x9nuYjihW6WobpKP/nlUDu3BIsbIq09UEke+uE/QK+ FJ8PTJkAsXOf1Bc2C0XbW4Y2hf103+YY6L8weUCBsWC5VH5VtVmeuh26ENURclwfeXhWQ9Og 77yzpTXWr5g1Z0oLpYpWPv745J4bE7pv+dzxOrFdM1xNkzY2pvXph/A8OjxZNQklDkHQ7PIB Lki5L2F4XkEOddUUQchJwzMqTPsggPDmGjgLZrqgO+s4ECZK5+nLD3HEpAbPa3JLDaScy+90 Nu1lAqPUHSnP3vYZVw85ZYm6UCxHE4VLMnnJsN09ZhsOSVR+GyP5Nyw9rT1V3lcsuH7M5Naa 2Xobn9m7l9bRCD/Ji8kG15eV1WTxx1HXVQGjdUYDI7UwegBNbwMLh17XDy+3sn/6SgcqtECA Q6pZKA2mTQxEKMLBZQQYAQIADwIbDAUCWZR3hQUJA59eRwAKCRD5M+tD3xNhH4a/D/4jLAZu UhvU1swWcNEVVCELZ0D3LOV14XcY2MXa3QOpeZ9Bgq7YYJ4S5YXK+SBQS0FkRZdjGNvlGZoG ZdpU+NsQmQFhqHGwX0IT9MeTFM8uvKgxNKGwMVcV9g0IOqwBhGHne+BFboRA9362fgGW5AYQ zT0mzzRKEoOh4r3AQvbM6kLISxo0k1ujdYiI5nj/5WoKDqxTwwfuN1uDUHsWo3tzenRmpMyU NyW3Dc+1ajvXLyo09sRRq7BnM99Rix1EGL8Qhwy+j0YAv+FuspWxUX9FxXYho5PvGLHLsHfK FYQ7x/RRbpMjkJWVfIe/xVnfvn4kz+MTA5yhvsuNi678fLwY9hBP0y4lO8Ob2IhEPdfnTuIs tFVxXuelJ9xAe5TyqP0f+fQjf1ixsBZkqOohsBXDfje0iaUpYa/OQ/BBeej0dUdg2JEu4jAC x41HpVCnP9ipLpD0fYz1d/dX0F/VY2ovW6Eba/y/ngOSAR6C+u881m7oH2l0G47MTwkaQCBA bLGXPj4TCdX3lftqt4bcBPBJ+rFAnJmRHtUuyyaewBnZ81ZU2YAptqFM1kTh+aSvMvGhfVsQ qZL2rk2OPN1hg+KXhErlbTZ6oPtLCFhSHQmuxQ4oc4U147wBTUuOdwNjtnNatUhRCp8POc+3 XphVR5G70mnca1E2vzC77z+XSlTyRA== Message-ID: Date: Wed, 25 Apr 2018 10:48:13 +0100 User-Agent: Mozilla/5.0 (Windows NT 10.0; WOW64; rv:52.0) Gecko/20100101 Thunderbird/52.7.0 MIME-Version: 1.0 In-Reply-To: Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: 8bit Subject: Re: [dpdk-dev] [PATCH v3] net/tap: remove queue specific offload support X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 25 Apr 2018 09:48:18 -0000 On 4/25/2018 10:18 AM, Ophir Munk wrote: > Hi Ferruh, > > I should have mentioned earlier that TAP does support queue specific capabilities. > Please look in tap_queue_setup() and note that each TAP queue is created with a distinct file descriptor (fd). > Then supporting an offload capability is just implementing it in SW (e.g. calculating IP checksum). > > If the main assumption of this patch was that TAP does not support queue specific offloads - then please consider this patch again. Yes that was the initial question, is tap supports queue specific offloads or not. Thanks for the answer. > > On the other hand there is no port specific capability supported by TAP. If so verify functions are wrong, that was the error I got. It seems copy/paste of mlx one but the port_supp_offloads has different meaning there. > However, in order to support legacy applications, port capabilities are usually reported as the OR operation between queue & port capabilities. > TAP currently clones the queue capabilities to port capabilities. We could optimize this cloning by always return queue capabilities when queried about queues or ports. In this case - tap_rx_offload_get_port_capa() and tap_tx_offload_get_port_capa() could be removed. Instead of removing the functions I think you can keep them but return correct values, in this case return empty, this will make the exiting validation functions correct. Can you send a fix for that? If no fix sent, I suggest going with this patch to remove queue level offload support until it is fixed. > > Please find more comments inline. > >> -----Original Message----- >> From: Ferruh Yigit [mailto:ferruh.yigit@intel.com] >> Sent: Tuesday, April 24, 2018 8:54 PM >> To: Pascal Mazon >> Cc: dev@dpdk.org; Ferruh Yigit ; Mordechay >> Haimovsky ; Ophir Munk >> Subject: [PATCH v3] net/tap: remove queue specific offload support >> >> It is not clear if tap PMD supports queue specific offloads, removing the >> related code. >> >> Fixes: 95ae196ae10b ("net/tap: use new Rx offloads API") >> Fixes: 818fe14a9891 ("net/tap: use new Tx offloads API") >> Cc: motih@mellanox.com >> >> Signed-off-by: Ferruh Yigit >> --- >> Cc: Ophir Munk >> >> v2: >> * rebased >> >> v3: >> * txq->csum restored, >> - ETH_TXQ_FLAGS_IGNORE check removed since ethdev layer takes care of >> it >> - tx_conf != NULL check removed, this is internal api who calls this is >> ethdev and it doesn't pass null tx_conf >> --- >> drivers/net/tap/rte_eth_tap.c | 102 +++++------------------------------------- >> 1 file changed, 10 insertions(+), 92 deletions(-) >> >> diff --git a/drivers/net/tap/rte_eth_tap.c b/drivers/net/tap/rte_eth_tap.c >> index ef33aace9..61b4b5df3 100644 >> --- a/drivers/net/tap/rte_eth_tap.c >> +++ b/drivers/net/tap/rte_eth_tap.c >> @@ -278,31 +278,6 @@ tap_rx_offload_get_port_capa(void) >> DEV_RX_OFFLOAD_CRC_STRIP; >> } >> >> -static uint64_t >> -tap_rx_offload_get_queue_capa(void) >> -{ >> - return DEV_RX_OFFLOAD_SCATTER | >> - DEV_RX_OFFLOAD_IPV4_CKSUM | >> - DEV_RX_OFFLOAD_UDP_CKSUM | >> - DEV_RX_OFFLOAD_TCP_CKSUM | >> - DEV_RX_OFFLOAD_CRC_STRIP; >> -} >> - > > TAP PMD supports all of these RX queue specific offloads. I suggest to leave this function in place. > >> -static bool >> -tap_rxq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads) -{ >> - uint64_t port_offloads = dev->data->dev_conf.rxmode.offloads; >> - uint64_t queue_supp_offloads = tap_rx_offload_get_queue_capa(); >> - uint64_t port_supp_offloads = tap_rx_offload_get_port_capa(); >> - >> - if ((offloads & (queue_supp_offloads | port_supp_offloads)) != >> - offloads) >> - return false; >> - if ((port_offloads ^ offloads) & port_supp_offloads) >> - return false; >> - return true; >> -} >> - > > Putting aside the fact that queue offloads equals port offloads (so could ignore "port_supp_offload" variable) - this function is essential to validate that the configured Rx offloads are supported by TAP. I suggest to leave this function in place. > Without it - testpmd falsely confirms non supported offloads. > For example before this patch: offloading "hw-vlan-filter" will fail as expected: > > testpmd> port config all > testpmd> port config all hw-vlan-filter on > testpmd> port start all > Configuring Port 0 (socket 0) > PMD: net_tap0: 0x1209fc0: TX configured queues number: 1 > PMD: net_tap0: 0x1209fc0: RX configured queues number: 1 > PMD: 0x1209fc0: Rx queue offloads 0x120e don't match port offloads 0x120e or supported offloads 0x300e > Fail to configure port 0 rx queues > > However, with this patch this configuration is falsely accepted. > >> /* Callback to handle the rx burst of packets to the correct interface and >> * file descriptor(s) in a multi-queue setup. >> */ >> @@ -411,31 +386,6 @@ tap_tx_offload_get_port_capa(void) >> DEV_TX_OFFLOAD_TCP_CKSUM; >> } >> >> -static uint64_t >> -tap_tx_offload_get_queue_capa(void) >> -{ >> - return DEV_TX_OFFLOAD_MULTI_SEGS | >> - DEV_TX_OFFLOAD_IPV4_CKSUM | >> - DEV_TX_OFFLOAD_UDP_CKSUM | >> - DEV_TX_OFFLOAD_TCP_CKSUM; >> -} >> - > > TAP PMD supports all of these TX queue specific offloads. I suggest to leave this function in place. > >> -static bool >> -tap_txq_are_offloads_valid(struct rte_eth_dev *dev, uint64_t offloads) -{ >> - uint64_t port_offloads = dev->data->dev_conf.txmode.offloads; >> - uint64_t queue_supp_offloads = tap_tx_offload_get_queue_capa(); >> - uint64_t port_supp_offloads = tap_tx_offload_get_port_capa(); >> - >> - if ((offloads & (queue_supp_offloads | port_supp_offloads)) != >> - offloads) >> - return false; >> - /* Verify we have no conflict with port offloads */ >> - if ((port_offloads ^ offloads) & port_supp_offloads) >> - return false; >> - return true; >> -} >> - > > This function is essential to validate that the configured Tx offloads are supported by TAP. > I suggest to leave this function in place. > >> static void >> tap_tx_offload(char *packet, uint64_t ol_flags, unsigned int l2_len, >> unsigned int l3_len) >> @@ -763,12 +713,10 @@ tap_dev_info(struct rte_eth_dev *dev, struct >> rte_eth_dev_info *dev_info) >> dev_info->max_tx_queues = RTE_PMD_TAP_MAX_QUEUES; >> dev_info->min_rx_bufsize = 0; >> dev_info->speed_capa = tap_dev_speed_capa(); >> - dev_info->rx_queue_offload_capa = >> tap_rx_offload_get_queue_capa(); >> - dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() | >> - dev_info->rx_queue_offload_capa; >> - dev_info->tx_queue_offload_capa = >> tap_tx_offload_get_queue_capa(); >> - dev_info->tx_offload_capa = tap_tx_offload_get_port_capa() | >> - dev_info->tx_queue_offload_capa; >> + dev_info->rx_offload_capa = tap_rx_offload_get_port_capa(); >> + dev_info->tx_offload_capa = tap_tx_offload_get_port_capa(); >> + dev_info->rx_queue_offload_capa = 0; >> + dev_info->tx_queue_offload_capa = 0; >> } >> > > Rx_queue_offloads_capa should be reported as before: > dev_info->tx_queue_offload_capa = tap_tx_offload_get_queue_capa(); > Same for TX offloads. > > Port capabilities could return queue capabilities: > > Instead of: > > dev_info->rx_offload_capa = tap_rx_offload_get_port_capa() | > dev_info->rx_queue_offload_capa; > > We could return: > > dev_info->rx_offload_capa = dev_info->rx_queue_offload_capa; > > The same argument is valid for Tx as well. > >> static int >> @@ -1094,19 +1042,6 @@ tap_rx_queue_setup(struct rte_eth_dev *dev, >> return -1; >> } >> >> - /* Verify application offloads are valid for our port and queue. */ >> - if (!tap_rxq_are_offloads_valid(dev, rx_conf->offloads)) { >> - rte_errno = ENOTSUP; >> - RTE_LOG(ERR, PMD, >> - "%p: Rx queue offloads 0x%" PRIx64 >> - " don't match port offloads 0x%" PRIx64 >> - " or supported offloads 0x%" PRIx64 "\n", >> - (void *)dev, rx_conf->offloads, >> - dev->data->dev_conf.rxmode.offloads, >> - (tap_rx_offload_get_port_capa() | >> - tap_rx_offload_get_queue_capa())); >> - return -rte_errno; >> - } > > The tap_rxq_are_offloads_valid() call is essential. I suggest to leave it in place. > The RTE_LOG could drop port references to become: > > RTE_LOG(ERR, PMD, > "%p: Rx queue offloads 0x%" PRIx64 > " don't match" > " supported offloads 0x%" PRIx64 "\n", > (void *)dev, rx_conf->offloads, > tap_rx_offload_get_queue_capa())); > > >> rxq->mp = mp; >> rxq->trigger_seen = 1; /* force initial burst */ >> rxq->in_port = dev->data->port_id; >> @@ -1175,29 +1110,12 @@ tap_tx_queue_setup(struct rte_eth_dev *dev, >> return -1; >> dev->data->tx_queues[tx_queue_id] = &internals->txq[tx_queue_id]; >> txq = dev->data->tx_queues[tx_queue_id]; >> - /* >> - * Don't verify port offloads for application which >> - * use the old API. >> - */ >> - if (tx_conf != NULL && >> - !!(tx_conf->txq_flags & ETH_TXQ_FLAGS_IGNORE)) { >> - if (tap_txq_are_offloads_valid(dev, tx_conf->offloads)) { >> - txq->csum = !!(tx_conf->offloads & >> - (DEV_TX_OFFLOAD_IPV4_CKSUM | >> - DEV_TX_OFFLOAD_UDP_CKSUM | >> - DEV_TX_OFFLOAD_TCP_CKSUM)); >> - } else { >> - rte_errno = ENOTSUP; >> - RTE_LOG(ERR, PMD, >> - "%p: Tx queue offloads 0x%" PRIx64 >> - " don't match port offloads 0x%" PRIx64 >> - " or supported offloads 0x%" PRIx64, >> - (void *)dev, tx_conf->offloads, >> - dev->data->dev_conf.txmode.offloads, >> - tap_tx_offload_get_port_capa()); >> - return -rte_errno; >> - } >> - } >> + > > The tap_txq_are_offloads_valid() call is essential. I suggest to leave it in place. > The RTE_LOG message could drop comparison between queue and port capabilities: > > RTE_LOG(ERR, PMD, > "%p: Tx queue offloads 0x%" PRIx64 > " don't match" > " supported offloads 0x%" PRIx64, > (void *)dev, tx_conf->offloads, > tap_tx_offload_get_queue_capa()); > >> + txq->csum = !!(tx_conf->offloads & >> + (DEV_TX_OFFLOAD_IPV4_CKSUM | >> + DEV_TX_OFFLOAD_UDP_CKSUM | >> + DEV_TX_OFFLOAD_TCP_CKSUM)); >> + >> ret = tap_setup_queue(dev, internals, tx_queue_id, 0); >> if (ret == -1) >> return -1; >> -- >> 2.14.3 >