From patchwork Thu Sep 5 20:13:09 2013 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Michal Schmidt X-Patchwork-Id: 272944 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 DE6892C00AC for ; Fri, 6 Sep 2013 06:13:17 +1000 (EST) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755583Ab3IEUNO (ORCPT ); Thu, 5 Sep 2013 16:13:14 -0400 Received: from mx1.redhat.com ([209.132.183.28]:1566 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753914Ab3IEUNM (ORCPT ); Thu, 5 Sep 2013 16:13:12 -0400 Received: from int-mx02.intmail.prod.int.phx2.redhat.com (int-mx02.intmail.prod.int.phx2.redhat.com [10.5.11.12]) by mx1.redhat.com (8.14.4/8.14.4) with ESMTP id r85KDBpr018870 (version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=OK); Thu, 5 Sep 2013 16:13:11 -0400 Received: from hp.brq.redhat.com (dhcp-27-114.brq.redhat.com [10.34.27.114]) by int-mx02.intmail.prod.int.phx2.redhat.com (8.13.8/8.13.8) with ESMTP id r85KD92L022380; Thu, 5 Sep 2013 16:13:10 -0400 From: Michal Schmidt To: netdev@vger.kernel.org Cc: Ariel Elior , Eilon Greenstein , David Miller Subject: [PATCH] bnx2x: avoid atomic allocations during initialization Date: Thu, 5 Sep 2013 22:13:09 +0200 Message-Id: <1378411989-19775-1-git-send-email-mschmidt@redhat.com> X-Scanned-By: MIMEDefang 2.67 on 10.5.11.12 Sender: netdev-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: netdev@vger.kernel.org During initialization bnx2x allocates significant amounts of memory (for rx data, rx SGEs, TPA pool) using atomic allocations. I received a report where bnx2x failed to allocate SGEs and it had to fall back to TPA-less operation. Let's use GFP_KERNEL allocations during initialization, which runs in process context. Add gfp_t parameters to functions that are used both in initialization and in the receive path. Use an unlikely branch in bnx2x_frag_alloc() to avoid atomic allocation by netdev_alloc_frag(). The branch is taken several thousands of times during initialization, but then never more. Note that fp->rx_frag_size is never greater than PAGE_SIZE, so __get_free_page() can be used here. Signed-off-by: Michal Schmidt --- drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c | 38 +++++++++++++++---------- 1 file changed, 23 insertions(+), 15 deletions(-) diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c index 8d726f6..884e8ad 100644 --- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c +++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c @@ -484,10 +484,10 @@ static void bnx2x_set_gro_params(struct sk_buff *skb, u16 parsing_flags, NAPI_GRO_CB(skb)->count = num_of_coalesced_segs; } -static int bnx2x_alloc_rx_sge(struct bnx2x *bp, - struct bnx2x_fastpath *fp, u16 index) +static int bnx2x_alloc_rx_sge(struct bnx2x *bp, struct bnx2x_fastpath *fp, + u16 index, gfp_t gfp_mask) { - struct page *page = alloc_pages(GFP_ATOMIC, PAGES_PER_SGE_SHIFT); + struct page *page = alloc_pages(gfp_mask, PAGES_PER_SGE_SHIFT); struct sw_rx_page *sw_buf = &fp->rx_page_ring[index]; struct eth_rx_sge *sge = &fp->rx_sge_ring[index]; dma_addr_t mapping; @@ -566,7 +566,7 @@ static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp, /* If we fail to allocate a substitute page, we simply stop where we are and drop the whole packet */ - err = bnx2x_alloc_rx_sge(bp, fp, sge_idx); + err = bnx2x_alloc_rx_sge(bp, fp, sge_idx, GFP_ATOMIC); if (unlikely(err)) { bnx2x_fp_qstats(bp, fp)->rx_skb_alloc_failed++; return err; @@ -610,12 +610,17 @@ static void bnx2x_frag_free(const struct bnx2x_fastpath *fp, void *data) kfree(data); } -static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp) +static void *bnx2x_frag_alloc(const struct bnx2x_fastpath *fp, gfp_t gfp_mask) { - if (fp->rx_frag_size) + if (fp->rx_frag_size) { + /* GFP_KERNEL allocations are used only during initialization */ + if (unlikely(gfp_mask & __GFP_WAIT)) + return (void *)__get_free_page(gfp_mask); + return netdev_alloc_frag(fp->rx_frag_size); + } - return kmalloc(fp->rx_buf_size + NET_SKB_PAD, GFP_ATOMIC); + return kmalloc(fp->rx_buf_size + NET_SKB_PAD, gfp_mask); } #ifdef CONFIG_INET @@ -695,7 +700,7 @@ static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp, goto drop; /* Try to allocate the new data */ - new_data = bnx2x_frag_alloc(fp); + new_data = bnx2x_frag_alloc(fp, GFP_ATOMIC); /* Unmap skb in the pool anyway, as we are going to change pool entry status to BNX2X_TPA_STOP even if new skb allocation fails. */ @@ -746,15 +751,15 @@ drop: bnx2x_fp_stats(bp, fp)->eth_q_stats.rx_skb_alloc_failed++; } -static int bnx2x_alloc_rx_data(struct bnx2x *bp, - struct bnx2x_fastpath *fp, u16 index) +static int bnx2x_alloc_rx_data(struct bnx2x *bp, struct bnx2x_fastpath *fp, + u16 index, gfp_t gfp_mask) { u8 *data; struct sw_rx_bd *rx_buf = &fp->rx_buf_ring[index]; struct eth_rx_bd *rx_bd = &fp->rx_desc_ring[index]; dma_addr_t mapping; - data = bnx2x_frag_alloc(fp); + data = bnx2x_frag_alloc(fp, gfp_mask); if (unlikely(data == NULL)) return -ENOMEM; @@ -947,7 +952,8 @@ int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget) memcpy(skb->data, data + pad, len); bnx2x_reuse_rx_data(fp, bd_cons, bd_prod); } else { - if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod) == 0)) { + if (likely(bnx2x_alloc_rx_data(bp, fp, bd_prod, + GFP_ATOMIC) == 0)) { dma_unmap_single(&bp->pdev->dev, dma_unmap_addr(rx_buf, mapping), fp->rx_buf_size, @@ -1307,7 +1313,8 @@ void bnx2x_init_rx_rings(struct bnx2x *bp) struct sw_rx_bd *first_buf = &tpa_info->first_buf; - first_buf->data = bnx2x_frag_alloc(fp); + first_buf->data = + bnx2x_frag_alloc(fp, GFP_KERNEL); if (!first_buf->data) { BNX2X_ERR("Failed to allocate TPA skb pool for queue[%d] - disabling TPA on this queue!\n", j); @@ -1329,7 +1336,8 @@ void bnx2x_init_rx_rings(struct bnx2x *bp) for (i = 0, ring_prod = 0; i < MAX_RX_SGE_CNT*NUM_RX_SGE_PAGES; i++) { - if (bnx2x_alloc_rx_sge(bp, fp, ring_prod) < 0) { + if (bnx2x_alloc_rx_sge(bp, fp, ring_prod, + GFP_KERNEL) < 0) { BNX2X_ERR("was only able to allocate %d rx sges\n", i); BNX2X_ERR("disabling TPA for queue[%d]\n", @@ -4214,7 +4222,7 @@ static int bnx2x_alloc_rx_bds(struct bnx2x_fastpath *fp, * fp->eth_q_stats.rx_skb_alloc_failed = 0 */ for (i = 0; i < rx_ring_size; i++) { - if (bnx2x_alloc_rx_data(bp, fp, ring_prod) < 0) { + if (bnx2x_alloc_rx_data(bp, fp, ring_prod, GFP_KERNEL) < 0) { failure_cnt++; continue; }