From patchwork Thu Aug 11 12:02:09 2016 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Bert Kenward X-Patchwork-Id: 658198 X-Patchwork-Delegate: davem@davemloft.net Return-Path: X-Original-To: patchwork-incoming@ozlabs.org Delivered-To: patchwork-incoming@ozlabs.org Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by ozlabs.org (Postfix) with ESMTP id 3s96gd2cxLz9t0m for ; Thu, 11 Aug 2016 22:26:05 +1000 (AEST) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932346AbcHKM0A (ORCPT ); Thu, 11 Aug 2016 08:26:00 -0400 Received: from nbfkord-smmo03.seg.att.com ([209.65.160.84]:3652 "EHLO nbfkord-smmo03.seg.att.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932388AbcHKMZw (ORCPT ); Thu, 11 Aug 2016 08:25:52 -0400 Received: from unknown [193.34.186.16] (EHLO nbfkord-smmo03.seg.att.com) by nbfkord-smmo03.seg.att.com(mxl_mta-7.2.4-7) with ESMTP id 0de6ca75.2ba0e2469940.1608458.00-2492.4233714.nbfkord-smmo03.seg.att.com (envelope-from ); Thu, 11 Aug 2016 12:25:52 +0000 (UTC) X-MXL-Hash: 57ac6ed03c3aca48-a2a3404a23f24364d243f96a9f20353aa45eb1f0 Received: from unknown [193.34.186.16] by nbfkord-smmo03.seg.att.com(mxl_mta-7.2.4-7) with SMTP id 2da6ca75.0.1607168.00-2318.4230697.nbfkord-smmo03.seg.att.com (envelope-from ); Thu, 11 Aug 2016 12:08:52 +0000 (UTC) X-MXL-Hash: 57ac6ad40cf32903-4c2feefdeb40b6fcf81ed7abb6dcc49b190a2e8d Received: from hgk-desktop.uk.solarflarecom.com (10.17.20.82) by ukex01.SolarFlarecom.com (10.17.10.4) with Microsoft SMTP Server (TLS) id 15.0.1044.25; Thu, 11 Aug 2016 13:02:13 +0100 From: Bert Kenward Subject: [PATCH net-next 5/6] sfc: set interrupt moderation via MCDI References: <78a8505b-8c63-b879-2cd2-d78d9890aa6c@solarflare.com> To: Dave Miller CC: , linux-net-drivers Message-ID: <201b64d1-5946-c4fb-52b2-052986f4cd83@solarflare.com> Date: Thu, 11 Aug 2016 13:02:09 +0100 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:45.0) Gecko/20100101 Thunderbird/45.2.0 MIME-Version: 1.0 In-Reply-To: <78a8505b-8c63-b879-2cd2-d78d9890aa6c@solarflare.com> X-Originating-IP: [10.17.20.82] X-ClientProxiedBy: ocex03.SolarFlarecom.com (10.20.40.36) To ukex01.SolarFlarecom.com (10.17.10.4) X-TM-AS-Product-Ver: SMEX-11.0.0.1191-8.000.1202-22506.003 X-TM-AS-Result: No--17.277000-8.000000-31 X-TM-AS-User-Approved-Sender: No X-TM-AS-User-Blocked-Sender: No X-AnalysisOut: [v=2.1 cv=OcWee0nY c=1 sm=1 tr=0 a=8P+NB+fYZDP74ap4g4d9Kw==] X-AnalysisOut: [:17 a=OjXlt12Agc0A:10 a=N659UExz7-8A:10 a=7z1cN_iqozsA:10 ] X-AnalysisOut: [a=zRKbQ67AAAAA:8 a=1zt4NYxvTnF3imsyvRQA:9 a=93U0qDcwgPFIoS] X-AnalysisOut: [OD:21 a=8NMmk8gRN-9rZSic:21 a=pILNOxqGKmIA:10 a=PA03WX8tBz] X-AnalysisOut: [eizutn5_OT:22] X-Spam: [F=0.2000000000; CM=0.500; S=0.200(2015072901)] X-MAIL-FROM: X-SOURCE-IP: [193.34.186.16] Sender: netdev-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: netdev@vger.kernel.org SFN8000-series NICs require a new method of setting interrupt moderation, via MCDI. This is indicated by a workaround flag. This new MCDI command takes an explicit time value rather than a number of ticks. It therefore makes sense to also store the moderation values in terms of time, since that is what the ethtool interface is interested in. Signed-off-by: Bert Kenward --- drivers/net/ethernet/sfc/ef10.c | 38 +++++++++++---- drivers/net/ethernet/sfc/efx.c | 89 ++++++++++++++++++---------------- drivers/net/ethernet/sfc/efx.h | 2 + drivers/net/ethernet/sfc/falcon.c | 7 ++- drivers/net/ethernet/sfc/net_driver.h | 10 ++-- drivers/net/ethernet/sfc/nic.h | 2 + drivers/net/ethernet/sfc/ptp.c | 2 +- drivers/net/ethernet/sfc/siena.c | 11 +++-- drivers/net/ethernet/sfc/workarounds.h | 4 ++ 9 files changed, 103 insertions(+), 62 deletions(-) diff --git a/drivers/net/ethernet/sfc/ef10.c b/drivers/net/ethernet/sfc/ef10.c index e57fa2e..5f6c4fa 100644 --- a/drivers/net/ethernet/sfc/ef10.c +++ b/drivers/net/ethernet/sfc/ef10.c @@ -1749,27 +1749,43 @@ static size_t efx_ef10_update_stats_vf(struct efx_nic *efx, u64 *full_stats, static void efx_ef10_push_irq_moderation(struct efx_channel *channel) { struct efx_nic *efx = channel->efx; - unsigned int mode, value; + unsigned int mode, usecs; efx_dword_t timer_cmd; - if (channel->irq_moderation) { + if (channel->irq_moderation_us) { mode = 3; - value = channel->irq_moderation - 1; + usecs = channel->irq_moderation_us; } else { mode = 0; - value = 0; + usecs = 0; } - if (EFX_EF10_WORKAROUND_35388(efx)) { + if (EFX_EF10_WORKAROUND_61265(efx)) { + MCDI_DECLARE_BUF(inbuf, MC_CMD_SET_EVQ_TMR_IN_LEN); + unsigned int ns = usecs * 1000; + + MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_INSTANCE, + channel->channel); + MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_LOAD_REQ_NS, ns); + MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_RELOAD_REQ_NS, ns); + MCDI_SET_DWORD(inbuf, SET_EVQ_TMR_IN_TMR_MODE, mode); + + efx_mcdi_rpc_async(efx, MC_CMD_SET_EVQ_TMR, + inbuf, sizeof(inbuf), 0, NULL, 0); + } else if (EFX_EF10_WORKAROUND_35388(efx)) { + unsigned int ticks = efx_usecs_to_ticks(efx, usecs); + EFX_POPULATE_DWORD_3(timer_cmd, ERF_DD_EVQ_IND_TIMER_FLAGS, EFE_DD_EVQ_IND_TIMER_FLAGS, ERF_DD_EVQ_IND_TIMER_MODE, mode, - ERF_DD_EVQ_IND_TIMER_VAL, value); + ERF_DD_EVQ_IND_TIMER_VAL, ticks); efx_writed_page(efx, &timer_cmd, ER_DD_EVQ_INDIRECT, channel->channel); } else { + unsigned int ticks = efx_usecs_to_ticks(efx, usecs); + EFX_POPULATE_DWORD_2(timer_cmd, ERF_DZ_TC_TIMER_MODE, mode, - ERF_DZ_TC_TIMER_VAL, value); + ERF_DZ_TC_TIMER_VAL, ticks); efx_writed_page(efx, &timer_cmd, ER_DZ_EVQ_TMR, channel->channel); } @@ -2615,10 +2631,11 @@ static int efx_ef10_ev_init(struct efx_channel *channel) /* Successfully created event queue on channel 0 */ rc = efx_mcdi_get_workarounds(efx, &implemented, &enabled); if (rc == -ENOSYS) { - /* GET_WORKAROUNDS was implemented before the bug26807 - * workaround, thus the latter must be unavailable in this fw + /* GET_WORKAROUNDS was implemented before these workarounds, + * thus they must be unavailable in this firmware. */ nic_data->workaround_26807 = false; + nic_data->workaround_61265 = false; rc = 0; } else if (rc) { goto fail; @@ -2658,6 +2675,9 @@ static int efx_ef10_ev_init(struct efx_channel *channel) rc = 0; } } + + nic_data->workaround_61265 = + !!(implemented & MC_CMD_GET_WORKAROUNDS_OUT_BUG61265); } if (!rc) diff --git a/drivers/net/ethernet/sfc/efx.c b/drivers/net/ethernet/sfc/efx.c index 14b821b..b26e1f9 100644 --- a/drivers/net/ethernet/sfc/efx.c +++ b/drivers/net/ethernet/sfc/efx.c @@ -281,6 +281,27 @@ static int efx_process_channel(struct efx_channel *channel, int budget) * NAPI guarantees serialisation of polls of the same device, which * provides the guarantee required by efx_process_channel(). */ +static void efx_update_irq_mod(struct efx_nic *efx, struct efx_channel *channel) +{ + int step = efx->irq_mod_step_us; + + if (channel->irq_mod_score < irq_adapt_low_thresh) { + if (channel->irq_moderation_us > step) { + channel->irq_moderation_us -= step; + efx->type->push_irq_moderation(channel); + } + } else if (channel->irq_mod_score > irq_adapt_high_thresh) { + if (channel->irq_moderation_us < + efx->irq_rx_moderation_us) { + channel->irq_moderation_us += step; + efx->type->push_irq_moderation(channel); + } + } + + channel->irq_count = 0; + channel->irq_mod_score = 0; +} + static int efx_poll(struct napi_struct *napi, int budget) { struct efx_channel *channel = @@ -301,22 +322,7 @@ static int efx_poll(struct napi_struct *napi, int budget) if (efx_channel_has_rx_queue(channel) && efx->irq_rx_adaptive && unlikely(++channel->irq_count == 1000)) { - if (unlikely(channel->irq_mod_score < - irq_adapt_low_thresh)) { - if (channel->irq_moderation > 1) { - channel->irq_moderation -= 1; - efx->type->push_irq_moderation(channel); - } - } else if (unlikely(channel->irq_mod_score > - irq_adapt_high_thresh)) { - if (channel->irq_moderation < - efx->irq_rx_moderation) { - channel->irq_moderation += 1; - efx->type->push_irq_moderation(channel); - } - } - channel->irq_count = 0; - channel->irq_mod_score = 0; + efx_update_irq_mod(efx, channel); } efx_filter_rfs_expire(channel); @@ -1703,6 +1709,7 @@ static int efx_probe_nic(struct efx_nic *efx) netif_set_real_num_rx_queues(efx->net_dev, efx->n_rx_channels); /* Initialise the interrupt moderation settings */ + efx->irq_mod_step_us = DIV_ROUND_UP(efx->timer_quantum_ns, 1000); efx_init_irq_moderation(efx, tx_irq_mod_usec, rx_irq_mod_usec, true, true); @@ -1949,14 +1956,21 @@ static void efx_remove_all(struct efx_nic *efx) * Interrupt moderation * **************************************************************************/ - -static unsigned int irq_mod_ticks(unsigned int usecs, unsigned int quantum_ns) +unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs) { if (usecs == 0) return 0; - if (usecs * 1000 < quantum_ns) + if (usecs * 1000 < efx->timer_quantum_ns) return 1; /* never round down to 0 */ - return usecs * 1000 / quantum_ns; + return usecs * 1000 / efx->timer_quantum_ns; +} + +unsigned int efx_ticks_to_usecs(struct efx_nic *efx, unsigned int ticks) +{ + /* We must round up when converting ticks to microseconds + * because we round down when converting the other way. + */ + return DIV_ROUND_UP(ticks * efx->timer_quantum_ns, 1000); } /* Set interrupt moderation parameters */ @@ -1968,18 +1982,12 @@ int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs, unsigned int irq_mod_max = DIV_ROUND_UP(efx->type->timer_period_max * efx->timer_quantum_ns, 1000); - unsigned int tx_ticks; - unsigned int rx_ticks; - EFX_ASSERT_RESET_SERIALISED(efx); if (tx_usecs > irq_mod_max || rx_usecs > irq_mod_max) return -EINVAL; - tx_ticks = irq_mod_ticks(tx_usecs, efx->timer_quantum_ns); - rx_ticks = irq_mod_ticks(rx_usecs, efx->timer_quantum_ns); - - if (tx_ticks != rx_ticks && efx->tx_channel_offset == 0 && + if (tx_usecs != rx_usecs && efx->tx_channel_offset == 0 && !rx_may_override_tx) { netif_err(efx, drv, efx->net_dev, "Channels are shared. " "RX and TX IRQ moderation must be equal\n"); @@ -1987,12 +1995,12 @@ int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs, } efx->irq_rx_adaptive = rx_adaptive; - efx->irq_rx_moderation = rx_ticks; + efx->irq_rx_moderation_us = rx_usecs; efx_for_each_channel(channel, efx) { if (efx_channel_has_rx_queue(channel)) - channel->irq_moderation = rx_ticks; + channel->irq_moderation_us = rx_usecs; else if (efx_channel_has_tx_queues(channel)) - channel->irq_moderation = tx_ticks; + channel->irq_moderation_us = tx_usecs; } return 0; @@ -2001,26 +2009,21 @@ int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs, void efx_get_irq_moderation(struct efx_nic *efx, unsigned int *tx_usecs, unsigned int *rx_usecs, bool *rx_adaptive) { - /* We must round up when converting ticks to microseconds - * because we round down when converting the other way. - */ - *rx_adaptive = efx->irq_rx_adaptive; - *rx_usecs = DIV_ROUND_UP(efx->irq_rx_moderation * - efx->timer_quantum_ns, - 1000); + *rx_usecs = efx->irq_rx_moderation_us; /* If channels are shared between RX and TX, so is IRQ * moderation. Otherwise, IRQ moderation is the same for all * TX channels and is not adaptive. */ - if (efx->tx_channel_offset == 0) + if (efx->tx_channel_offset == 0) { *tx_usecs = *rx_usecs; - else - *tx_usecs = DIV_ROUND_UP( - efx->channel[efx->tx_channel_offset]->irq_moderation * - efx->timer_quantum_ns, - 1000); + } else { + struct efx_channel *tx_channel; + + tx_channel = efx->channel[efx->tx_channel_offset]; + *tx_usecs = tx_channel->irq_moderation_us; + } } /************************************************************************** diff --git a/drivers/net/ethernet/sfc/efx.h b/drivers/net/ethernet/sfc/efx.h index c3ae739..342ae16 100644 --- a/drivers/net/ethernet/sfc/efx.h +++ b/drivers/net/ethernet/sfc/efx.h @@ -204,6 +204,8 @@ int efx_try_recovery(struct efx_nic *efx); /* Global */ void efx_schedule_reset(struct efx_nic *efx, enum reset_type type); +unsigned int efx_usecs_to_ticks(struct efx_nic *efx, unsigned int usecs); +unsigned int efx_ticks_to_usecs(struct efx_nic *efx, unsigned int ticks); int efx_init_irq_moderation(struct efx_nic *efx, unsigned int tx_usecs, unsigned int rx_usecs, bool rx_adaptive, bool rx_may_override_tx); diff --git a/drivers/net/ethernet/sfc/falcon.c b/drivers/net/ethernet/sfc/falcon.c index d790cb8..d3d0bcc 100644 --- a/drivers/net/ethernet/sfc/falcon.c +++ b/drivers/net/ethernet/sfc/falcon.c @@ -378,12 +378,15 @@ static void falcon_push_irq_moderation(struct efx_channel *channel) struct efx_nic *efx = channel->efx; /* Set timer register */ - if (channel->irq_moderation) { + if (channel->irq_moderation_us) { + unsigned int ticks; + + ticks = efx_usecs_to_ticks(efx, channel->irq_moderation_us); EFX_POPULATE_DWORD_2(timer_cmd, FRF_AB_TC_TIMER_MODE, FFE_BB_TIMER_MODE_INT_HLDOFF, FRF_AB_TC_TIMER_VAL, - channel->irq_moderation - 1); + ticks - 1); } else { EFX_POPULATE_DWORD_2(timer_cmd, FRF_AB_TC_TIMER_MODE, diff --git a/drivers/net/ethernet/sfc/net_driver.h b/drivers/net/ethernet/sfc/net_driver.h index 9ff062a..3e8c6fb 100644 --- a/drivers/net/ethernet/sfc/net_driver.h +++ b/drivers/net/ethernet/sfc/net_driver.h @@ -392,7 +392,7 @@ enum efx_sync_events_state { * @eventq_init: Event queue initialised flag * @enabled: Channel enabled indicator * @irq: IRQ number (MSI and MSI-X only) - * @irq_moderation: IRQ moderation value (in hardware ticks) + * @irq_moderation_us: IRQ moderation value (in microseconds) * @napi_dev: Net device used with NAPI * @napi_str: NAPI control structure * @state: state for NAPI vs busy polling @@ -433,7 +433,7 @@ struct efx_channel { bool eventq_init; bool enabled; int irq; - unsigned int irq_moderation; + unsigned int irq_moderation_us; struct net_device *napi_dev; struct napi_struct napi_str; #ifdef CONFIG_NET_RX_BUSY_POLL @@ -811,7 +811,8 @@ struct vfdi_status; * @interrupt_mode: Interrupt mode * @timer_quantum_ns: Interrupt timer quantum, in nanoseconds * @irq_rx_adaptive: Adaptive IRQ moderation enabled for RX event queues - * @irq_rx_moderation: IRQ moderation time for RX event queues + * @irq_rx_mod_step_us: Step size for IRQ moderation for RX event queues + * @irq_rx_moderation_us: IRQ moderation time for RX event queues * @msg_enable: Log message enable flags * @state: Device state number (%STATE_*). Serialised by the rtnl_lock. * @reset_pending: Bitmask for pending resets @@ -941,7 +942,8 @@ struct efx_nic { enum efx_int_mode interrupt_mode; unsigned int timer_quantum_ns; bool irq_rx_adaptive; - unsigned int irq_rx_moderation; + unsigned int irq_mod_step_us; + unsigned int irq_rx_moderation_us; u32 msg_enable; enum nic_state state; diff --git a/drivers/net/ethernet/sfc/nic.h b/drivers/net/ethernet/sfc/nic.h index 531c5a8..d8b1694 100644 --- a/drivers/net/ethernet/sfc/nic.h +++ b/drivers/net/ethernet/sfc/nic.h @@ -507,6 +507,7 @@ enum { * @stats: Hardware statistics * @workaround_35388: Flag: firmware supports workaround for bug 35388 * @workaround_26807: Flag: firmware supports workaround for bug 26807 + * @workaround_61265: Flag: firmware supports workaround for bug 61265 * @must_check_datapath_caps: Flag: @datapath_caps needs to be revalidated * after MC reboot * @datapath_caps: Capabilities of datapath firmware (FLAGS1 field of @@ -542,6 +543,7 @@ struct efx_ef10_nic_data { u64 stats[EF10_STAT_COUNT]; bool workaround_35388; bool workaround_26807; + bool workaround_61265; bool must_check_datapath_caps; u32 datapath_caps; u32 datapath_caps2; diff --git a/drivers/net/ethernet/sfc/ptp.c b/drivers/net/ethernet/sfc/ptp.c index c771e0a..dd204d9 100644 --- a/drivers/net/ethernet/sfc/ptp.c +++ b/drivers/net/ethernet/sfc/ptp.c @@ -1306,7 +1306,7 @@ static int efx_ptp_probe_channel(struct efx_channel *channel) { struct efx_nic *efx = channel->efx; - channel->irq_moderation = 0; + channel->irq_moderation_us = 0; channel->rx_queue.core_index = 0; return efx_ptp_probe(efx, channel); diff --git a/drivers/net/ethernet/sfc/siena.c b/drivers/net/ethernet/sfc/siena.c index 2219b54..994e33f 100644 --- a/drivers/net/ethernet/sfc/siena.c +++ b/drivers/net/ethernet/sfc/siena.c @@ -34,19 +34,24 @@ static void siena_init_wol(struct efx_nic *efx); static void siena_push_irq_moderation(struct efx_channel *channel) { + struct efx_nic *efx = channel->efx; efx_dword_t timer_cmd; - if (channel->irq_moderation) + if (channel->irq_moderation_us) { + unsigned int ticks; + + ticks = efx_usecs_to_ticks(efx, channel->irq_moderation_us); EFX_POPULATE_DWORD_2(timer_cmd, FRF_CZ_TC_TIMER_MODE, FFE_CZ_TIMER_MODE_INT_HLDOFF, FRF_CZ_TC_TIMER_VAL, - channel->irq_moderation - 1); - else + ticks - 1); + } else { EFX_POPULATE_DWORD_2(timer_cmd, FRF_CZ_TC_TIMER_MODE, FFE_CZ_TIMER_MODE_DIS, FRF_CZ_TC_TIMER_VAL, 0); + } efx_writed_page_locked(channel->efx, &timer_cmd, FR_BZ_TIMER_COMMAND_P0, channel->channel); } diff --git a/drivers/net/ethernet/sfc/workarounds.h b/drivers/net/ethernet/sfc/workarounds.h index 2310b75..351cd14 100644 --- a/drivers/net/ethernet/sfc/workarounds.h +++ b/drivers/net/ethernet/sfc/workarounds.h @@ -50,4 +50,8 @@ #define EFX_WORKAROUND_35388(efx) \ (efx_nic_rev(efx) == EFX_REV_HUNT_A0 && EFX_EF10_WORKAROUND_35388(efx)) +/* Moderation timer access must go through MCDI */ +#define EFX_EF10_WORKAROUND_61265(efx) \ + (((struct efx_ef10_nic_data *)efx->nic_data)->workaround_61265) + #endif /* EFX_WORKAROUNDS_H */