#include "../cxgb4/t4fw_api.h"
#include "../cxgb4/t4_msg.h"
-/*
- * Decoded Adapter Parameters.
- */
-static u32 FL_PG_ORDER; /* large page allocation size */
-static u32 STAT_LEN; /* length of status page at ring end */
-static u32 PKTSHIFT; /* padding between CPL and packet data */
-static u32 FL_ALIGN; /* response queue message alignment */
-
/*
* Constants ...
*/
TX_QCHECK_PERIOD = (HZ / 2),
MAX_TIMER_TX_RECLAIM = 100,
- /*
- * An FL with <= FL_STARVE_THRES buffers is starving and a periodic
- * timer will attempt to refill it.
- */
- FL_STARVE_THRES = 4,
-
/*
* Suspend an Ethernet TX queue with fewer available descriptors than
* this. We always want to have room for a maximum sized packet:
* we can specify for immediate data in the firmware Ethernet TX
* Work Request.
*/
- MAX_IMM_TX_PKT_LEN = FW_WR_IMMDLEN_MASK,
+ MAX_IMM_TX_PKT_LEN = FW_WR_IMMDLEN_M,
/*
* Max size of a WR sent through a control TX queue.
/**
* fl_starving - return whether a Free List is starving.
+ * @adapter: pointer to the adapter
* @fl: the Free List
*
* Tests specified Free List to see whether the number of buffers
* available to the hardware has falled below our "starvation"
* threshold.
*/
-static inline bool fl_starving(const struct sge_fl *fl)
+static inline bool fl_starving(const struct adapter *adapter,
+ const struct sge_fl *fl)
{
- return fl->avail - fl->pend_cred <= FL_STARVE_THRES;
+ const struct sge *s = &adapter->sge;
+
+ return fl->avail - fl->pend_cred <= s->fl_starve_thres;
}
/**
/**
* get_buf_size - return the size of an RX Free List buffer.
+ * @adapter: pointer to the associated adapter
* @sdesc: pointer to the software buffer descriptor
*/
-static inline int get_buf_size(const struct rx_sw_desc *sdesc)
+static inline int get_buf_size(const struct adapter *adapter,
+ const struct rx_sw_desc *sdesc)
{
- return FL_PG_ORDER > 0 && (sdesc->dma_addr & RX_LARGE_BUF)
- ? (PAGE_SIZE << FL_PG_ORDER)
- : PAGE_SIZE;
+ const struct sge *s = &adapter->sge;
+
+ return (s->fl_pg_order > 0 && (sdesc->dma_addr & RX_LARGE_BUF)
+ ? (PAGE_SIZE << s->fl_pg_order) : PAGE_SIZE);
}
/**
if (is_buf_mapped(sdesc))
dma_unmap_page(adapter->pdev_dev, get_buf_addr(sdesc),
- get_buf_size(sdesc), PCI_DMA_FROMDEVICE);
+ get_buf_size(adapter, sdesc),
+ PCI_DMA_FROMDEVICE);
put_page(sdesc->page);
sdesc->page = NULL;
if (++fl->cidx == fl->size)
if (is_buf_mapped(sdesc))
dma_unmap_page(adapter->pdev_dev, get_buf_addr(sdesc),
- get_buf_size(sdesc), PCI_DMA_FROMDEVICE);
+ get_buf_size(adapter, sdesc),
+ PCI_DMA_FROMDEVICE);
sdesc->page = NULL;
if (++fl->cidx == fl->size)
fl->cidx = 0;
static unsigned int refill_fl(struct adapter *adapter, struct sge_fl *fl,
int n, gfp_t gfp)
{
+ struct sge *s = &adapter->sge;
struct page *page;
dma_addr_t dma_addr;
unsigned int cred = fl->avail;
*/
BUG_ON(fl->avail + n > fl->size - FL_PER_EQ_UNIT);
+ gfp |= __GFP_NOWARN;
+
/*
* If we support large pages, prefer large buffers and fail over to
* small pages if we can't allocate large pages to satisfy the refill.
* If we don't support large pages, drop directly into the small page
* allocation code.
*/
- if (FL_PG_ORDER == 0)
+ if (s->fl_pg_order == 0)
goto alloc_small_pages;
while (n) {
- page = alloc_pages(gfp | __GFP_COMP | __GFP_NOWARN,
- FL_PG_ORDER);
+ page = __dev_alloc_pages(gfp, s->fl_pg_order);
if (unlikely(!page)) {
/*
* We've failed inour attempt to allocate a "large
fl->large_alloc_failed++;
break;
}
- poison_buf(page, PAGE_SIZE << FL_PG_ORDER);
+ poison_buf(page, PAGE_SIZE << s->fl_pg_order);
dma_addr = dma_map_page(adapter->pdev_dev, page, 0,
- PAGE_SIZE << FL_PG_ORDER,
+ PAGE_SIZE << s->fl_pg_order,
PCI_DMA_FROMDEVICE);
if (unlikely(dma_mapping_error(adapter->pdev_dev, dma_addr))) {
/*
* because DMA mapping resources are typically
* critical resources once they become scarse.
*/
- __free_pages(page, FL_PG_ORDER);
+ __free_pages(page, s->fl_pg_order);
goto out;
}
dma_addr |= RX_LARGE_BUF;
alloc_small_pages:
while (n--) {
- page = __skb_alloc_page(gfp | __GFP_NOWARN, NULL);
+ page = __dev_alloc_page(gfp);
if (unlikely(!page)) {
fl->alloc_failed++;
break;
fl->pend_cred += cred;
ring_fl_db(adapter, fl);
- if (unlikely(fl_starving(fl))) {
+ if (unlikely(fl_starving(adapter, fl))) {
smp_wmb();
set_bit(fl->cntxt_id, adapter->sge.starving_fl);
}
sgl->addr0 = cpu_to_be64(addr[1]);
}
- sgl->cmd_nsge = htonl(ULPTX_CMD(ULP_TX_SC_DSGL) |
+ sgl->cmd_nsge = htonl(ULPTX_CMD_V(ULP_TX_SC_DSGL) |
ULPTX_NSGE(nfrags));
if (likely(--nfrags == 0))
return;
goto out_free;
}
- wr_mid = FW_WR_LEN16(DIV_ROUND_UP(flits, 2));
+ wr_mid = FW_WR_LEN16_V(DIV_ROUND_UP(flits, 2));
if (unlikely(credits < ETHTXQ_STOP_THRES)) {
/*
* After we're done injecting the Work Request for this
* has opened up.
*/
txq_stop(txq);
- wr_mid |= FW_WR_EQUEQ | FW_WR_EQUIQ;
+ wr_mid |= FW_WR_EQUEQ_F | FW_WR_EQUIQ_F;
}
/*
int eth_xtra_len = skb_network_offset(skb) - ETH_HLEN;
wr->op_immdlen =
- cpu_to_be32(FW_WR_OP(FW_ETH_TX_PKT_VM_WR) |
- FW_WR_IMMDLEN(sizeof(*lso) +
- sizeof(*cpl)));
+ cpu_to_be32(FW_WR_OP_V(FW_ETH_TX_PKT_VM_WR) |
+ FW_WR_IMMDLEN_V(sizeof(*lso) +
+ sizeof(*cpl)));
/*
* Fill in the LSO CPL message.
*/
len = is_eth_imm(skb) ? skb->len + sizeof(*cpl) : sizeof(*cpl);
wr->op_immdlen =
- cpu_to_be32(FW_WR_OP(FW_ETH_TX_PKT_VM_WR) |
- FW_WR_IMMDLEN(len));
+ cpu_to_be32(FW_WR_OP_V(FW_ETH_TX_PKT_VM_WR) |
+ FW_WR_IMMDLEN_V(len));
/*
* Set up TX Packet CPL pointer, control word and perform
static void do_gro(struct sge_eth_rxq *rxq, const struct pkt_gl *gl,
const struct cpl_rx_pkt *pkt)
{
+ struct adapter *adapter = rxq->rspq.adapter;
+ struct sge *s = &adapter->sge;
int ret;
struct sk_buff *skb;
return;
}
- copy_frags(skb, gl, PKTSHIFT);
- skb->len = gl->tot_len - PKTSHIFT;
+ copy_frags(skb, gl, s->pktshift);
+ skb->len = gl->tot_len - s->pktshift;
skb->data_len = skb->len;
skb->truesize += skb->data_len;
skb->ip_summed = CHECKSUM_UNNECESSARY;
bool csum_ok = pkt->csum_calc && !pkt->err_vec &&
(rspq->netdev->features & NETIF_F_RXCSUM);
struct sge_eth_rxq *rxq = container_of(rspq, struct sge_eth_rxq, rspq);
+ struct adapter *adapter = rspq->adapter;
+ struct sge *s = &adapter->sge;
/*
* If this is a good TCP packet and we have Generic Receive Offload
rxq->stats.rx_drops++;
return 0;
}
- __skb_pull(skb, PKTSHIFT);
+ __skb_pull(skb, s->pktshift);
skb->protocol = eth_type_trans(skb, rspq->netdev);
skb_record_rx_queue(skb, rspq->idx);
rxq->stats.pkts++;
static int process_responses(struct sge_rspq *rspq, int budget)
{
struct sge_eth_rxq *rxq = container_of(rspq, struct sge_eth_rxq, rspq);
+ struct adapter *adapter = rspq->adapter;
+ struct sge *s = &adapter->sge;
int budget_left = budget;
while (likely(budget_left)) {
BUG_ON(frag >= MAX_SKB_FRAGS);
BUG_ON(rxq->fl.avail == 0);
sdesc = &rxq->fl.sdesc[rxq->fl.cidx];
- bufsz = get_buf_size(sdesc);
+ bufsz = get_buf_size(adapter, sdesc);
fp->page = sdesc->page;
fp->offset = rspq->offset;
fp->size = min(bufsz, len);
*/
ret = rspq->handler(rspq, rspq->cur_desc, &gl);
if (likely(ret == 0))
- rspq->offset += ALIGN(fp->size, FL_ALIGN);
+ rspq->offset += ALIGN(fp->size, s->fl_align);
else
restore_rx_bufs(&gl, &rxq->fl, frag);
} else if (likely(rsp_type == RSP_TYPE_CPL)) {
* schedule napi but the FL is no longer starving.
* No biggie.
*/
- if (fl_starving(fl)) {
+ if (fl_starving(adapter, fl)) {
struct sge_eth_rxq *rxq;
rxq = container_of(fl, struct sge_eth_rxq, fl);
int intr_dest,
struct sge_fl *fl, rspq_handler_t hnd)
{
+ struct sge *s = &adapter->sge;
struct port_info *pi = netdev_priv(dev);
struct fw_iq_cmd cmd, rpl;
int ret, iqandst, flsz = 0;
* into OS-independent common code ...
*/
memset(&cmd, 0, sizeof(cmd));
- cmd.op_to_vfn = cpu_to_be32(FW_CMD_OP(FW_IQ_CMD) |
- FW_CMD_REQUEST |
- FW_CMD_WRITE |
- FW_CMD_EXEC);
+ cmd.op_to_vfn = cpu_to_be32(FW_CMD_OP_V(FW_IQ_CMD) |
+ FW_CMD_REQUEST_F |
+ FW_CMD_WRITE_F |
+ FW_CMD_EXEC_F);
cmd.alloc_to_len16 = cpu_to_be32(FW_IQ_CMD_ALLOC |
FW_IQ_CMD_IQSTART(1) |
FW_LEN16(cmd));
fl->size = roundup(fl->size, FL_PER_EQ_UNIT);
fl->desc = alloc_ring(adapter->pdev_dev, fl->size,
sizeof(__be64), sizeof(struct rx_sw_desc),
- &fl->addr, &fl->sdesc, STAT_LEN);
+ &fl->addr, &fl->sdesc, s->stat_len);
if (!fl->desc) {
ret = -ENOMEM;
goto err;
* free list ring) in Egress Queue Units.
*/
flsz = (fl->size / FL_PER_EQ_UNIT +
- STAT_LEN / EQ_UNIT);
+ s->stat_len / EQ_UNIT);
/*
* Fill in all the relevant firmware Ingress Queue Command
struct net_device *dev, struct netdev_queue *devq,
unsigned int iqid)
{
+ struct sge *s = &adapter->sge;
int ret, nentries;
struct fw_eq_eth_cmd cmd, rpl;
struct port_info *pi = netdev_priv(dev);
* Calculate the size of the hardware TX Queue (including the Status
* Page on the end of the TX Queue) in units of TX Descriptors.
*/
- nentries = txq->q.size + STAT_LEN / sizeof(struct tx_desc);
+ nentries = txq->q.size + s->stat_len / sizeof(struct tx_desc);
/*
* Allocate the hardware ring for the TX ring (with space for its
txq->q.desc = alloc_ring(adapter->pdev_dev, txq->q.size,
sizeof(struct tx_desc),
sizeof(struct tx_sw_desc),
- &txq->q.phys_addr, &txq->q.sdesc, STAT_LEN);
+ &txq->q.phys_addr, &txq->q.sdesc, s->stat_len);
if (!txq->q.desc)
return -ENOMEM;
* into the common code ...
*/
memset(&cmd, 0, sizeof(cmd));
- cmd.op_to_vfn = cpu_to_be32(FW_CMD_OP(FW_EQ_ETH_CMD) |
- FW_CMD_REQUEST |
- FW_CMD_WRITE |
- FW_CMD_EXEC);
+ cmd.op_to_vfn = cpu_to_be32(FW_CMD_OP_V(FW_EQ_ETH_CMD) |
+ FW_CMD_REQUEST_F |
+ FW_CMD_WRITE_F |
+ FW_CMD_EXEC_F);
cmd.alloc_to_len16 = cpu_to_be32(FW_EQ_ETH_CMD_ALLOC |
FW_EQ_ETH_CMD_EQSTART |
FW_LEN16(cmd));
*/
static void free_txq(struct adapter *adapter, struct sge_txq *tq)
{
+ struct sge *s = &adapter->sge;
+
dma_free_coherent(adapter->pdev_dev,
- tq->size * sizeof(*tq->desc) + STAT_LEN,
+ tq->size * sizeof(*tq->desc) + s->stat_len,
tq->desc, tq->phys_addr);
tq->cntxt_id = 0;
tq->sdesc = NULL;
static void free_rspq_fl(struct adapter *adapter, struct sge_rspq *rspq,
struct sge_fl *fl)
{
+ struct sge *s = &adapter->sge;
unsigned int flid = fl ? fl->cntxt_id : 0xffff;
t4vf_iq_free(adapter, FW_IQ_TYPE_FL_INT_CAP,
if (fl) {
free_rx_bufs(adapter, fl, fl->avail);
dma_free_coherent(adapter->pdev_dev,
- fl->size * sizeof(*fl->desc) + STAT_LEN,
+ fl->size * sizeof(*fl->desc) + s->stat_len,
fl->desc, fl->addr);
kfree(fl->sdesc);
fl->sdesc = NULL;
u32 fl0 = sge_params->sge_fl_buffer_size[0];
u32 fl1 = sge_params->sge_fl_buffer_size[1];
struct sge *s = &adapter->sge;
+ unsigned int ingpadboundary, ingpackboundary;
/*
* Start by vetting the basic SGE parameters which have been set up by
* Now translate the adapter parameters into our internal forms.
*/
if (fl1)
- FL_PG_ORDER = ilog2(fl1) - PAGE_SHIFT;
- STAT_LEN = ((sge_params->sge_control & EGRSTATUSPAGESIZE_MASK)
- ? 128 : 64);
- PKTSHIFT = PKTSHIFT_GET(sge_params->sge_control);
- FL_ALIGN = 1 << (INGPADBOUNDARY_GET(sge_params->sge_control) +
- SGE_INGPADBOUNDARY_SHIFT);
+ s->fl_pg_order = ilog2(fl1) - PAGE_SHIFT;
+ s->stat_len = ((sge_params->sge_control & EGRSTATUSPAGESIZE_MASK)
+ ? 128 : 64);
+ s->pktshift = PKTSHIFT_GET(sge_params->sge_control);
+
+ /* T4 uses a single control field to specify both the PCIe Padding and
+ * Packing Boundary. T5 introduced the ability to specify these
+ * separately. The actual Ingress Packet Data alignment boundary
+ * within Packed Buffer Mode is the maximum of these two
+ * specifications. (Note that it makes no real practical sense to
+ * have the Pading Boudary be larger than the Packing Boundary but you
+ * could set the chip up that way and, in fact, legacy T4 code would
+ * end doing this because it would initialize the Padding Boundary and
+ * leave the Packing Boundary initialized to 0 (16 bytes).)
+ */
+ ingpadboundary = 1 << (INGPADBOUNDARY_GET(sge_params->sge_control) +
+ X_INGPADBOUNDARY_SHIFT);
+ if (is_t4(adapter->params.chip)) {
+ s->fl_align = ingpadboundary;
+ } else {
+ /* T5 has a different interpretation of one of the PCIe Packing
+ * Boundary values.
+ */
+ ingpackboundary = INGPACKBOUNDARY_G(sge_params->sge_control2);
+ if (ingpackboundary == INGPACKBOUNDARY_16B_X)
+ ingpackboundary = 16;
+ else
+ ingpackboundary = 1 << (ingpackboundary +
+ INGPACKBOUNDARY_SHIFT_X);
+
+ s->fl_align = max(ingpadboundary, ingpackboundary);
+ }
+
+ /* A FL with <= fl_starve_thres buffers is starving and a periodic
+ * timer will attempt to refill it. This needs to be larger than the
+ * SGE's Egress Congestion Threshold. If it isn't, then we can get
+ * stuck waiting for new packets while the SGE is waiting for us to
+ * give it more Free List entries. (Note that the SGE's Egress
+ * Congestion Threshold is in units of 2 Free List pointers.)
+ */
+ s->fl_starve_thres
+ = EGRTHRESHOLD_GET(sge_params->sge_congestion_control)*2 + 1;
/*
* Set up tasklet timers.