From patchwork Tue Aug 23 18:26:17 2011 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Leann Ogasawara X-Patchwork-Id: 111167 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Received: from chlorine.canonical.com (chlorine.canonical.com [91.189.94.204]) by ozlabs.org (Postfix) with ESMTP id D52C2B6F68 for ; Wed, 24 Aug 2011 04:26:41 +1000 (EST) Received: from localhost ([127.0.0.1] helo=chlorine.canonical.com) by chlorine.canonical.com with esmtp (Exim 4.71) (envelope-from ) id 1Qvvg6-0002o5-Uc; Tue, 23 Aug 2011 18:26:23 +0000 Received: from youngberry.canonical.com ([91.189.89.112]) by chlorine.canonical.com with esmtp (Exim 4.71) (envelope-from ) id 1Qvvg3-0002nr-R7 for kernel-team@lists.ubuntu.com; Tue, 23 Aug 2011 18:26:19 +0000 Received: from c-24-21-156-70.hsd1.or.comcast.net ([24.21.156.70] helo=[192.168.1.3]) by youngberry.canonical.com with esmtpsa (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1Qvvg3-0008Tt-Av for kernel-team@lists.ubuntu.com; Tue, 23 Aug 2011 18:26:19 +0000 Subject: [Natty] [PATCH 1/1] ath9k: use split rx buffers to get rid of order-1 skb allocations From: Leann Ogasawara To: kernel-team Date: Tue, 23 Aug 2011 11:26:17 -0700 X-Mailer: Evolution 3.1.5- Message-ID: <1314123979.2020.9.camel@adamo> Mime-Version: 1.0 X-BeenThere: kernel-team@lists.ubuntu.com X-Mailman-Version: 2.1.13 Precedence: list List-Id: Kernel team discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: kernel-team-bounces@lists.ubuntu.com Errors-To: kernel-team-bounces@lists.ubuntu.com BugLink: http://bugs.launchpad.net/bugs/728835 SRU Justification: Impact: Without the fix, users will experience a page allocation failure with the ath9k driver Fix: Upstream commit 0d95521ea74735826cb2e28bebf6a07392c75bfa User has tested and confirmed the upstream patch resolves this issue. It is a clean cherry-pick from upstream. Please consider for Natty SRU. Thanks, Leann From 920ef7cfde1ffe0bf85ed27d608f4e38630f911d Mon Sep 17 00:00:00 2001 From: Felix Fietkau Date: Wed, 26 Jan 2011 18:23:27 +0100 Subject: [PATCH] ath9k: use split rx buffers to get rid of order-1 skb allocations BugLink: http://bugs.launchpad.net/bugs/728835 With this change, less CPU time is spent trying to look for consecutive pages for rx skbs. This also reduces the socket memory required for IP/UDP reassembly. Only two buffers per frame are supported. Frames spanning more buffers will be dropped, but the buffer size is enough to handle the required AMSDU size. Signed-off-by: Jouni Malinen Signed-off-by: Felix Fietkau Signed-off-by: John W. Linville (cherry picked from commit 0d95521ea74735826cb2e28bebf6a07392c75bfa) Signed-off-by: Leann Ogasawara --- drivers/net/wireless/ath/ath9k/ath9k.h | 2 + drivers/net/wireless/ath/ath9k/main.c | 5 ++ drivers/net/wireless/ath/ath9k/recv.c | 88 +++++++++++++++++++++++--------- 3 files changed, 71 insertions(+), 24 deletions(-) diff --git a/drivers/net/wireless/ath/ath9k/ath9k.h b/drivers/net/wireless/ath/ath9k/ath9k.h index 1a7fa6e..33e4431 100644 --- a/drivers/net/wireless/ath/ath9k/ath9k.h +++ b/drivers/net/wireless/ath/ath9k/ath9k.h @@ -301,6 +301,8 @@ struct ath_rx { struct ath_descdma rxdma; struct ath_buf *rx_bufptr; struct ath_rx_edma rx_edma[ATH9K_RX_QUEUE_MAX]; + + struct sk_buff *frag; }; int ath_startrecv(struct ath_softc *sc); diff --git a/drivers/net/wireless/ath/ath9k/main.c b/drivers/net/wireless/ath/ath9k/main.c index 0848e09..97dbc8f 100644 --- a/drivers/net/wireless/ath/ath9k/main.c +++ b/drivers/net/wireless/ath/ath9k/main.c @@ -1323,6 +1323,11 @@ static void ath9k_stop(struct ieee80211_hw *hw) } else sc->rx.rxlink = NULL; + if (sc->rx.frag) { + dev_kfree_skb_any(sc->rx.frag); + sc->rx.frag = NULL; + } + /* disable HAL and put h/w to sleep */ ath9k_hw_disable(ah); ath9k_hw_configpcipowersave(ah, 1, 1); diff --git a/drivers/net/wireless/ath/ath9k/recv.c b/drivers/net/wireless/ath/ath9k/recv.c index 1e0c1e3..1b074f2 100644 --- a/drivers/net/wireless/ath/ath9k/recv.c +++ b/drivers/net/wireless/ath/ath9k/recv.c @@ -230,11 +230,6 @@ static int ath_rx_edma_init(struct ath_softc *sc, int nbufs) int error = 0, i; u32 size; - - common->rx_bufsize = roundup(IEEE80211_MAX_MPDU_LEN + - ah->caps.rx_status_len, - min(common->cachelsz, (u16)64)); - ath9k_hw_set_rx_bufsize(ah, common->rx_bufsize - ah->caps.rx_status_len); @@ -321,12 +316,12 @@ int ath_rx_init(struct ath_softc *sc, int nbufs) sc->sc_flags &= ~SC_OP_RXFLUSH; spin_lock_init(&sc->rx.rxbuflock); + common->rx_bufsize = IEEE80211_MAX_MPDU_LEN / 2 + + sc->sc_ah->caps.rx_status_len; + if (sc->sc_ah->caps.hw_caps & ATH9K_HW_CAP_EDMA) { return ath_rx_edma_init(sc, nbufs); } else { - common->rx_bufsize = roundup(IEEE80211_MAX_MPDU_LEN, - min(common->cachelsz, (u16)64)); - ath_dbg(common, ATH_DBG_CONFIG, "cachelsz %u rxbufsize %u\n", common->cachelsz, common->rx_bufsize); @@ -860,15 +855,9 @@ static bool ath9k_rx_accept(struct ath_common *common, if (rx_stats->rs_datalen > (common->rx_bufsize - rx_status_len)) return false; - /* - * rs_more indicates chained descriptors which can be used - * to link buffers together for a sort of scatter-gather - * operation. - * reject the frame, we don't support scatter-gather yet and - * the frame is probably corrupt anyway - */ + /* Only use error bits from the last fragment */ if (rx_stats->rs_more) - return false; + return true; /* * The rx_stats->rs_status will not be set until the end of the @@ -1020,6 +1009,10 @@ static int ath9k_rx_skb_preprocess(struct ath_common *common, if (!ath9k_rx_accept(common, hdr, rx_status, rx_stats, decrypt_error)) return -EINVAL; + /* Only use status info from the last fragment */ + if (rx_stats->rs_more) + return 0; + ath9k_process_rssi(common, hw, hdr, rx_stats); if (ath9k_process_rate(common, hw, rx_stats, rx_status)) @@ -1621,7 +1614,7 @@ div_comb_done: int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp) { struct ath_buf *bf; - struct sk_buff *skb = NULL, *requeue_skb; + struct sk_buff *skb = NULL, *requeue_skb, *hdr_skb; struct ieee80211_rx_status *rxs; struct ath_hw *ah = sc->sc_ah; struct ath_common *common = ath9k_hw_common(ah); @@ -1672,8 +1665,17 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp) if (!skb) continue; - hdr = (struct ieee80211_hdr *) (skb->data + rx_status_len); - rxs = IEEE80211_SKB_RXCB(skb); + /* + * Take frame header from the first fragment and RX status from + * the last one. + */ + if (sc->rx.frag) + hdr_skb = sc->rx.frag; + else + hdr_skb = skb; + + hdr = (struct ieee80211_hdr *) (hdr_skb->data + rx_status_len); + rxs = IEEE80211_SKB_RXCB(hdr_skb); hw = ath_get_virt_hw(sc, hdr); @@ -1684,12 +1686,12 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp) * chain it back at the queue without processing it. */ if (flush) - goto requeue; + goto requeue_drop_frag; retval = ath9k_rx_skb_preprocess(common, hw, hdr, &rs, rxs, &decrypt_error); if (retval) - goto requeue; + goto requeue_drop_frag; rxs->mactime = (tsf & ~0xffffffffULL) | rs.rs_tstamp; if (rs.rs_tstamp > tsf_lower && @@ -1709,7 +1711,7 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp) * skb and put it at the tail of the sc->rx.rxbuf list for * processing. */ if (!requeue_skb) - goto requeue; + goto requeue_drop_frag; /* Unmap the frame */ dma_unmap_single(sc->dev, bf->bf_buf_addr, @@ -1720,8 +1722,9 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp) if (ah->caps.rx_status_len) skb_pull(skb, ah->caps.rx_status_len); - ath9k_rx_skb_postprocess(common, skb, &rs, - rxs, decrypt_error); + if (!rs.rs_more) + ath9k_rx_skb_postprocess(common, hdr_skb, &rs, + rxs, decrypt_error); /* We will now give hardware our shiny new allocated skb */ bf->bf_mpdu = requeue_skb; @@ -1738,6 +1741,38 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp) break; } + if (rs.rs_more) { + /* + * rs_more indicates chained descriptors which can be + * used to link buffers together for a sort of + * scatter-gather operation. + */ + if (sc->rx.frag) { + /* too many fragments - cannot handle frame */ + dev_kfree_skb_any(sc->rx.frag); + dev_kfree_skb_any(skb); + skb = NULL; + } + sc->rx.frag = skb; + goto requeue; + } + + if (sc->rx.frag) { + int space = skb->len - skb_tailroom(hdr_skb); + + sc->rx.frag = NULL; + + if (pskb_expand_head(hdr_skb, 0, space, GFP_ATOMIC) < 0) { + dev_kfree_skb(skb); + goto requeue_drop_frag; + } + + skb_copy_from_linear_data(skb, skb_put(hdr_skb, skb->len), + skb->len); + dev_kfree_skb_any(skb); + skb = hdr_skb; + } + /* * change the default rx antenna if rx diversity chooses the * other antenna 3 times in a row. @@ -1763,6 +1798,11 @@ int ath_rx_tasklet(struct ath_softc *sc, int flush, bool hp) ath_rx_send_to_mac80211(hw, sc, skb); +requeue_drop_frag: + if (sc->rx.frag) { + dev_kfree_skb_any(sc->rx.frag); + sc->rx.frag = NULL; + } requeue: if (edma) { list_add_tail(&bf->list, &sc->rx.rxbuf);