sge.c

来自「linux 内核源代码」· C语言 代码 · 共 2,293 行 · 第 1/5 页

/*
 * Copyright (c) 2005-2007 Chelsio, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */
#include <linux/skbuff.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/if_vlan.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/dma-mapping.h>
#include "common.h"
#include "regs.h"
#include "sge_defs.h"
#include "t3_cpl.h"
#include "firmware_exports.h"

#define USE_GTS 0

#define SGE_RX_SM_BUF_SIZE 1536

#define SGE_RX_COPY_THRES  256
#define SGE_RX_PULL_LEN    128

/*
 * Page chunk size for FL0 buffers if FL0 is to be populated with page chunks.
 * It must be a divisor of PAGE_SIZE.  If set to 0 FL0 will use sk_buffs
 * directly.
 */
#define FL0_PG_CHUNK_SIZE  2048

#define SGE_RX_DROP_THRES 16

/*
 * Period of the Tx buffer reclaim timer.  This timer does not need to run
 * frequently as Tx buffers are usually reclaimed by new Tx packets.
 */
#define TX_RECLAIM_PERIOD (HZ / 4)

/* WR size in bytes */
#define WR_LEN (WR_FLITS * 8)

/*
 * Types of Tx queues in each queue set.  Order here matters, do not change.
 */
enum { TXQ_ETH, TXQ_OFLD, TXQ_CTRL };

/* Values for sge_txq.flags */
enum {
	TXQ_RUNNING = 1 << 0,	/* fetch engine is running */
	TXQ_LAST_PKT_DB = 1 << 1,	/* last packet rang the doorbell */
};

struct tx_desc {
	__be64 flit[TX_DESC_FLITS];
};

struct rx_desc {
	__be32 addr_lo;
	__be32 len_gen;
	__be32 gen2;
	__be32 addr_hi;
};

struct tx_sw_desc {		/* SW state per Tx descriptor */
	struct sk_buff *skb;
};

struct rx_sw_desc {                /* SW state per Rx descriptor */
	union {
		struct sk_buff *skb;
		struct fl_pg_chunk pg_chunk;
	};
	DECLARE_PCI_UNMAP_ADDR(dma_addr);
};

struct rsp_desc {		/* response queue descriptor */
	struct rss_header rss_hdr;
	__be32 flags;
	__be32 len_cq;
	u8 imm_data[47];
	u8 intr_gen;
};

struct unmap_info {		/* packet unmapping info, overlays skb->cb */
	int sflit;		/* start flit of first SGL entry in Tx descriptor */
	u16 fragidx;		/* first page fragment in current Tx descriptor */
	u16 addr_idx;		/* buffer index of first SGL entry in descriptor */
	u32 len;		/* mapped length of skb main body */
};

/*
 * Holds unmapping information for Tx packets that need deferred unmapping.
 * This structure lives at skb->head and must be allocated by callers.
 */
struct deferred_unmap_info {
	struct pci_dev *pdev;
	dma_addr_t addr[MAX_SKB_FRAGS + 1];
};

/*
 * Maps a number of flits to the number of Tx descriptors that can hold them.
 * The formula is
 *
 * desc = 1 + (flits - 2) / (WR_FLITS - 1).
 *
 * HW allows up to 4 descriptors to be combined into a WR.
 */
static u8 flit_desc_map[] = {
	0,
#if SGE_NUM_GENBITS == 1
	1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
	2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
	3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
	4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4
#elif SGE_NUM_GENBITS == 2
	1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
	2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2, 2,
	3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3,
	4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4,
#else
# error "SGE_NUM_GENBITS must be 1 or 2"
#endif
};

static inline struct sge_qset *fl_to_qset(const struct sge_fl *q, int qidx)
{
	return container_of(q, struct sge_qset, fl[qidx]);
}

static inline struct sge_qset *rspq_to_qset(const struct sge_rspq *q)
{
	return container_of(q, struct sge_qset, rspq);
}

static inline struct sge_qset *txq_to_qset(const struct sge_txq *q, int qidx)
{
	return container_of(q, struct sge_qset, txq[qidx]);
}

/**
 *	refill_rspq - replenish an SGE response queue
 *	@adapter: the adapter
 *	@q: the response queue to replenish
 *	@credits: how many new responses to make available
 *
 *	Replenishes a response queue by making the supplied number of responses
 *	available to HW.
 */
static inline void refill_rspq(struct adapter *adapter,
			       const struct sge_rspq *q, unsigned int credits)
{
	t3_write_reg(adapter, A_SG_RSPQ_CREDIT_RETURN,
		     V_RSPQ(q->cntxt_id) | V_CREDITS(credits));
}

/**
 *	need_skb_unmap - does the platform need unmapping of sk_buffs?
 *
 *	Returns true if the platfrom needs sk_buff unmapping.  The compiler
 *	optimizes away unecessary code if this returns true.
 */
static inline int need_skb_unmap(void)
{
	/*
	 * This structure is used to tell if the platfrom needs buffer
	 * unmapping by checking if DECLARE_PCI_UNMAP_ADDR defines anything.
	 */
	struct dummy {
		DECLARE_PCI_UNMAP_ADDR(addr);
	};

	return sizeof(struct dummy) != 0;
}

/**
 *	unmap_skb - unmap a packet main body and its page fragments
 *	@skb: the packet
 *	@q: the Tx queue containing Tx descriptors for the packet
 *	@cidx: index of Tx descriptor
 *	@pdev: the PCI device
 *
 *	Unmap the main body of an sk_buff and its page fragments, if any.
 *	Because of the fairly complicated structure of our SGLs and the desire
 *	to conserve space for metadata, we keep the information necessary to
 *	unmap an sk_buff partly in the sk_buff itself (in its cb), and partly
 *	in the Tx descriptors (the physical addresses of the various data
 *	buffers).  The send functions initialize the state in skb->cb so we
 *	can unmap the buffers held in the first Tx descriptor here, and we
 *	have enough information at this point to update the state for the next
 *	Tx descriptor.
 */
static inline void unmap_skb(struct sk_buff *skb, struct sge_txq *q,
			     unsigned int cidx, struct pci_dev *pdev)
{
	const struct sg_ent *sgp;
	struct unmap_info *ui = (struct unmap_info *)skb->cb;
	int nfrags, frag_idx, curflit, j = ui->addr_idx;

	sgp = (struct sg_ent *)&q->desc[cidx].flit[ui->sflit];

	if (ui->len) {
		pci_unmap_single(pdev, be64_to_cpu(sgp->addr[0]), ui->len,
				 PCI_DMA_TODEVICE);
		ui->len = 0;	/* so we know for next descriptor for this skb */
		j = 1;
	}

	frag_idx = ui->fragidx;
	curflit = ui->sflit + 1 + j;
	nfrags = skb_shinfo(skb)->nr_frags;

	while (frag_idx < nfrags && curflit < WR_FLITS) {
		pci_unmap_page(pdev, be64_to_cpu(sgp->addr[j]),
			       skb_shinfo(skb)->frags[frag_idx].size,
			       PCI_DMA_TODEVICE);
		j ^= 1;
		if (j == 0) {
			sgp++;
			curflit++;
		}
		curflit++;
		frag_idx++;
	}

	if (frag_idx < nfrags) {	/* SGL continues into next Tx descriptor */
		ui->fragidx = frag_idx;
		ui->addr_idx = j;
		ui->sflit = curflit - WR_FLITS - j;	/* sflit can be -1 */
	}
}

/**
 *	free_tx_desc - reclaims Tx descriptors and their buffers
 *	@adapter: the adapter
 *	@q: the Tx queue to reclaim descriptors from
 *	@n: the number of descriptors to reclaim
 *
 *	Reclaims Tx descriptors from an SGE Tx queue and frees the associated
 *	Tx buffers.  Called with the Tx queue lock held.
 */
static void free_tx_desc(struct adapter *adapter, struct sge_txq *q,
			 unsigned int n)
{
	struct tx_sw_desc *d;
	struct pci_dev *pdev = adapter->pdev;
	unsigned int cidx = q->cidx;

	const int need_unmap = need_skb_unmap() &&
			       q->cntxt_id >= FW_TUNNEL_SGEEC_START;

	d = &q->sdesc[cidx];
	while (n--) {
		if (d->skb) {	/* an SGL is present */
			if (need_unmap)
				unmap_skb(d->skb, q, cidx, pdev);
			if (d->skb->priority == cidx)
				kfree_skb(d->skb);
		}
		++d;
		if (++cidx == q->size) {
			cidx = 0;
			d = q->sdesc;
		}
	}
	q->cidx = cidx;
}

/**
 *	reclaim_completed_tx - reclaims completed Tx descriptors
 *	@adapter: the adapter
 *	@q: the Tx queue to reclaim completed descriptors from
 *
 *	Reclaims Tx descriptors that the SGE has indicated it has processed,
 *	and frees the associated buffers if possible.  Called with the Tx
 *	queue's lock held.
 */
static inline void reclaim_completed_tx(struct adapter *adapter,
					struct sge_txq *q)
{
	unsigned int reclaim = q->processed - q->cleaned;

	if (reclaim) {
		free_tx_desc(adapter, q, reclaim);
		q->cleaned += reclaim;
		q->in_use -= reclaim;
	}
}

/**
 *	should_restart_tx - are there enough resources to restart a Tx queue?
 *	@q: the Tx queue
 *
 *	Checks if there are enough descriptors to restart a suspended Tx queue.
 */
static inline int should_restart_tx(const struct sge_txq *q)
{
	unsigned int r = q->processed - q->cleaned;

	return q->in_use - r < (q->size >> 1);
}

/**
 *	free_rx_bufs - free the Rx buffers on an SGE free list
 *	@pdev: the PCI device associated with the adapter
 *	@rxq: the SGE free list to clean up
 *
 *	Release the buffers on an SGE free-buffer Rx queue.  HW fetching from
 *	this queue should be stopped before calling this function.
 */
static void free_rx_bufs(struct pci_dev *pdev, struct sge_fl *q)
{
	unsigned int cidx = q->cidx;

	while (q->credits--) {
		struct rx_sw_desc *d = &q->sdesc[cidx];

		pci_unmap_single(pdev, pci_unmap_addr(d, dma_addr),
				 q->buf_size, PCI_DMA_FROMDEVICE);
		if (q->use_pages) {
			put_page(d->pg_chunk.page);
			d->pg_chunk.page = NULL;
		} else {
			kfree_skb(d->skb);
			d->skb = NULL;
		}
		if (++cidx == q->size)
			cidx = 0;
	}

	if (q->pg_chunk.page) {
		__free_page(q->pg_chunk.page);
		q->pg_chunk.page = NULL;
	}
}

/**
 *	add_one_rx_buf - add a packet buffer to a free-buffer list
 *	@va:  buffer start VA
 *	@len: the buffer length
 *	@d: the HW Rx descriptor to write
 *	@sd: the SW Rx descriptor to write
 *	@gen: the generation bit value
 *	@pdev: the PCI device associated with the adapter
 *
 *	Add a buffer of the given length to the supplied HW and SW Rx
 *	descriptors.
 */
static inline void add_one_rx_buf(void *va, unsigned int len,
				  struct rx_desc *d, struct rx_sw_desc *sd,
				  unsigned int gen, struct pci_dev *pdev)
{
	dma_addr_t mapping;

	mapping = pci_map_single(pdev, va, len, PCI_DMA_FROMDEVICE);
	pci_unmap_addr_set(sd, dma_addr, mapping);

	d->addr_lo = cpu_to_be32(mapping);
	d->addr_hi = cpu_to_be32((u64) mapping >> 32);
	wmb();
	d->len_gen = cpu_to_be32(V_FLD_GEN1(gen));
	d->gen2 = cpu_to_be32(V_FLD_GEN2(gen));
}

static int alloc_pg_chunk(struct sge_fl *q, struct rx_sw_desc *sd, gfp_t gfp)
{
	if (!q->pg_chunk.page) {
		q->pg_chunk.page = alloc_page(gfp);
		if (unlikely(!q->pg_chunk.page))
			return -ENOMEM;
		q->pg_chunk.va = page_address(q->pg_chunk.page);
		q->pg_chunk.offset = 0;
	}
	sd->pg_chunk = q->pg_chunk;

	q->pg_chunk.offset += q->buf_size;
	if (q->pg_chunk.offset == PAGE_SIZE)
		q->pg_chunk.page = NULL;
	else {
		q->pg_chunk.va += q->buf_size;
		get_page(q->pg_chunk.page);
	}
	return 0;
}

/**
 *	refill_fl - refill an SGE free-buffer list
 *	@adapter: the adapter
 *	@q: the free-list to refill
 *	@n: the number of new buffers to allocate
 *	@gfp: the gfp flags for allocating new buffers
 *
 *	(Re)populate an SGE free-buffer list with up to @n new packet buffers,
 *	allocated with the supplied gfp flags.  The caller must assure that
 *	@n does not exceed the queue's capacity.
 */
static void refill_fl(struct adapter *adap, struct sge_fl *q, int n, gfp_t gfp)
{
	void *buf_start;
	struct rx_sw_desc *sd = &q->sdesc[q->pidx];
	struct rx_desc *d = &q->desc[q->pidx];

	while (n--) {
		if (q->use_pages) {
			if (unlikely(alloc_pg_chunk(q, sd, gfp))) {
nomem:				q->alloc_failed++;
				break;
			}
			buf_start = sd->pg_chunk.va;
		} else {
			struct sk_buff *skb = alloc_skb(q->buf_size, gfp);

			if (!skb)
				goto nomem;

			sd->skb = skb;
			buf_start = skb->data;
		}

		add_one_rx_buf(buf_start, q->buf_size, d, sd, q->gen,
			       adap->pdev);
		d++;
		sd++;
		if (++q->pidx == q->size) {
			q->pidx = 0;
			q->gen ^= 1;
			sd = q->sdesc;
			d = q->desc;
		}
		q->credits++;
	}