* [PATCH 00/12 net-next] cxgb4 update @ 2010-12-14 20:18 Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 01/12] cxgb4: enable PCIe relaxed ordering Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev Here's a set of 12 patches for cxgb4. About half are clean ups, the rest either small non-critical fixes or enhancements. All for net-next. drivers/net/cxgb4/cxgb4.h | 4 +- drivers/net/cxgb4/cxgb4_main.c | 130 +++++++++++++++++++--------------------- drivers/net/cxgb4/sge.c | 25 ++++++-- drivers/net/cxgb4/t4_hw.c | 93 ++++++++++++++-------------- drivers/net/cxgb4/t4fw_api.h | 1 + 5 files changed, 128 insertions(+), 125 deletions(-) ^ permalink raw reply [flat|nested] 15+ messages in thread
* [PATCH 01/12] cxgb4: enable PCIe relaxed ordering 2010-12-14 20:18 [PATCH 00/12 net-next] cxgb4 update Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 02/12] cxgb4: do not read the clock frequency from VPD Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev Enable relaxed ordering for descriptor reads and packet I/O. Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/cxgb4_main.c | 14 ++++++++++++++ drivers/net/cxgb4/sge.c | 5 +++++ 2 files changed, 19 insertions(+), 0 deletions(-) diff --git a/drivers/net/cxgb4/cxgb4_main.c b/drivers/net/cxgb4/cxgb4_main.c index 848f89d..953d62a 100644 --- a/drivers/net/cxgb4/cxgb4_main.c +++ b/drivers/net/cxgb4/cxgb4_main.c @@ -3535,6 +3535,19 @@ static void __devinit print_port_info(struct adapter *adap) } } +static void __devinit enable_pcie_relaxed_ordering(struct pci_dev *dev) +{ + u16 v; + int pos; + + pos = pci_pcie_cap(dev); + if (pos > 0) { + pci_read_config_word(dev, pos + PCI_EXP_DEVCTL, &v); + v |= PCI_EXP_DEVCTL_RELAX_EN; + pci_write_config_word(dev, pos + PCI_EXP_DEVCTL, v); + } +} + /* * Free the following resources: * - memory used for tables @@ -3609,6 +3622,7 @@ static int __devinit init_one(struct pci_dev *pdev, } pci_enable_pcie_error_reporting(pdev); + enable_pcie_relaxed_ordering(pdev); pci_set_master(pdev); pci_save_state(pdev); diff --git a/drivers/net/cxgb4/sge.c b/drivers/net/cxgb4/sge.c index 1702225..cc0b997 100644 --- a/drivers/net/cxgb4/sge.c +++ b/drivers/net/cxgb4/sge.c @@ -2014,6 +2014,8 @@ int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq, flsz = fl->size / 8 + STAT_LEN / sizeof(struct tx_desc); c.iqns_to_fl0congen = htonl(FW_IQ_CMD_FL0PACKEN | + FW_IQ_CMD_FL0FETCHRO(1) | + FW_IQ_CMD_FL0DATARO(1) | FW_IQ_CMD_FL0PADEN); c.fl0dcaen_to_fl0cidxfthresh = htons(FW_IQ_CMD_FL0FBMIN(2) | FW_IQ_CMD_FL0FBMAX(3)); @@ -2106,6 +2108,7 @@ int t4_sge_alloc_eth_txq(struct adapter *adap, struct sge_eth_txq *txq, c.viid_pkd = htonl(FW_EQ_ETH_CMD_VIID(pi->viid)); c.fetchszm_to_iqid = htonl(FW_EQ_ETH_CMD_HOSTFCMODE(2) | FW_EQ_ETH_CMD_PCIECHN(pi->tx_chan) | + FW_EQ_ETH_CMD_FETCHRO(1) | FW_EQ_ETH_CMD_IQID(iqid)); c.dcaen_to_eqsize = htonl(FW_EQ_ETH_CMD_FBMIN(2) | FW_EQ_ETH_CMD_FBMAX(3) | @@ -2158,6 +2161,7 @@ int t4_sge_alloc_ctrl_txq(struct adapter *adap, struct sge_ctrl_txq *txq, c.physeqid_pkd = htonl(0); c.fetchszm_to_iqid = htonl(FW_EQ_CTRL_CMD_HOSTFCMODE(2) | FW_EQ_CTRL_CMD_PCIECHN(pi->tx_chan) | + FW_EQ_CTRL_CMD_FETCHRO | FW_EQ_CTRL_CMD_IQID(iqid)); c.dcaen_to_eqsize = htonl(FW_EQ_CTRL_CMD_FBMIN(2) | FW_EQ_CTRL_CMD_FBMAX(3) | @@ -2207,6 +2211,7 @@ int t4_sge_alloc_ofld_txq(struct adapter *adap, struct sge_ofld_txq *txq, FW_EQ_OFLD_CMD_EQSTART | FW_LEN16(c)); c.fetchszm_to_iqid = htonl(FW_EQ_OFLD_CMD_HOSTFCMODE(2) | FW_EQ_OFLD_CMD_PCIECHN(pi->tx_chan) | + FW_EQ_OFLD_CMD_FETCHRO(1) | FW_EQ_OFLD_CMD_IQID(iqid)); c.dcaen_to_eqsize = htonl(FW_EQ_OFLD_CMD_FBMIN(2) | FW_EQ_OFLD_CMD_FBMAX(3) | -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 02/12] cxgb4: do not read the clock frequency from VPD 2010-12-14 20:18 ` [PATCH 01/12] cxgb4: enable PCIe relaxed ordering Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 03/12] cxgb4: set the number of queues before device registration Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev No need to read the clock frequency from VPD, we already get it a bit later from FW, after any potential adjustments. Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/t4_hw.c | 4 +--- 1 files changed, 1 insertions(+), 3 deletions(-) diff --git a/drivers/net/cxgb4/t4_hw.c b/drivers/net/cxgb4/t4_hw.c index bb813d9..539c49a 100644 --- a/drivers/net/cxgb4/t4_hw.c +++ b/drivers/net/cxgb4/t4_hw.c @@ -370,7 +370,7 @@ int t4_seeprom_wp(struct adapter *adapter, bool enable) static int get_vpd_params(struct adapter *adapter, struct vpd_params *p) { int i, ret; - int ec, sn, v2; + int ec, sn; u8 vpd[VPD_LEN], csum; unsigned int vpdr_len; const struct t4_vpd_hdr *v; @@ -408,10 +408,8 @@ static int get_vpd_params(struct adapter *adapter, struct vpd_params *p) FIND_VPD_KW(ec, "EC"); FIND_VPD_KW(sn, "SN"); - FIND_VPD_KW(v2, "V2"); #undef FIND_VPD_KW - p->cclk = simple_strtoul(vpd + v2, NULL, 10); memcpy(p->id, v->id_data, ID_LEN); strim(p->id); memcpy(p->ec, vpd + ec, EC_LEN); -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 03/12] cxgb4: set the number of queues before device registration 2010-12-14 20:18 ` [PATCH 02/12] cxgb4: do not read the clock frequency from VPD Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 04/12] cxgb4: distinguish between 1-lane KR/KX and 4-lane KR/KX/KX4 ports Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev The number of queues is known early, move the calls to netif_set_real_num_[rt]x_queues before register_netdev. Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/cxgb4_main.c | 10 +++++----- 1 files changed, 5 insertions(+), 5 deletions(-) diff --git a/drivers/net/cxgb4/cxgb4_main.c b/drivers/net/cxgb4/cxgb4_main.c index 953d62a..3012a8a 100644 --- a/drivers/net/cxgb4/cxgb4_main.c +++ b/drivers/net/cxgb4/cxgb4_main.c @@ -2717,10 +2717,6 @@ static int cxgb_open(struct net_device *dev) return err; } - netif_set_real_num_tx_queues(dev, pi->nqsets); - err = netif_set_real_num_rx_queues(dev, pi->nqsets); - if (err) - return err; err = link_start(dev); if (!err) netif_tx_start_all_queues(dev); @@ -3733,6 +3729,10 @@ static int __devinit init_one(struct pci_dev *pdev, * register at least one net device. */ for_each_port(adapter, i) { + pi = adap2pinfo(adapter, i); + netif_set_real_num_tx_queues(adapter->port[i], pi->nqsets); + netif_set_real_num_rx_queues(adapter->port[i], pi->nqsets); + err = register_netdev(adapter->port[i]); if (err) dev_warn(&pdev->dev, @@ -3747,7 +3747,7 @@ static int __devinit init_one(struct pci_dev *pdev, adapter->name = adapter->port[i]->name; __set_bit(i, &adapter->registered_device_map); - adapter->chan_map[adap2pinfo(adapter, i)->tx_chan] = i; + adapter->chan_map[pi->tx_chan] = i; } } if (!adapter->registered_device_map) { -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 04/12] cxgb4: distinguish between 1-lane KR/KX and 4-lane KR/KX/KX4 ports 2010-12-14 20:18 ` [PATCH 03/12] cxgb4: set the number of queues before device registration Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 05/12] cxgb4: print port information after registering each netdev Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev And fix the supported flags ethtool reports for the two cases. Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/cxgb4_main.c | 9 +++++++-- drivers/net/cxgb4/t4fw_api.h | 1 + 2 files changed, 8 insertions(+), 2 deletions(-) diff --git a/drivers/net/cxgb4/cxgb4_main.c b/drivers/net/cxgb4/cxgb4_main.c index 3012a8a..3f33d51 100644 --- a/drivers/net/cxgb4/cxgb4_main.c +++ b/drivers/net/cxgb4/cxgb4_main.c @@ -1375,7 +1375,12 @@ static unsigned int from_fw_linkcaps(unsigned int type, unsigned int caps) } else if (type == FW_PORT_TYPE_KR) v |= SUPPORTED_Backplane | SUPPORTED_10000baseKR_Full; else if (type == FW_PORT_TYPE_BP_AP) - v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC; + v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC | + SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full; + else if (type == FW_PORT_TYPE_BP4_AP) + v |= SUPPORTED_Backplane | SUPPORTED_10000baseR_FEC | + SUPPORTED_10000baseKR_Full | SUPPORTED_1000baseKX_Full | + SUPPORTED_10000baseKX4_Full; else if (type == FW_PORT_TYPE_FIBER_XFI || type == FW_PORT_TYPE_FIBER_XAUI || type == FW_PORT_TYPE_SFP) v |= SUPPORTED_FIBRE; @@ -3489,7 +3494,7 @@ static void __devinit print_port_info(struct adapter *adap) { static const char *base[] = { "R XFI", "R XAUI", "T SGMII", "T XFI", "T XAUI", "KX4", "CX4", - "KX", "KR", "KR SFP+", "KR FEC" + "KX", "KR", "R SFP+", "KR/KX", "KR/KX/KX4" }; int i; diff --git a/drivers/net/cxgb4/t4fw_api.h b/drivers/net/cxgb4/t4fw_api.h index 940584a..edcfd7e 100644 --- a/drivers/net/cxgb4/t4fw_api.h +++ b/drivers/net/cxgb4/t4fw_api.h @@ -1239,6 +1239,7 @@ enum fw_port_type { FW_PORT_TYPE_KR, FW_PORT_TYPE_SFP, FW_PORT_TYPE_BP_AP, + FW_PORT_TYPE_BP4_AP, FW_PORT_TYPE_NONE = FW_PORT_CMD_PTYPE_MASK }; -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 05/12] cxgb4: print port information after registering each netdev 2010-12-14 20:18 ` [PATCH 04/12] cxgb4: distinguish between 1-lane KR/KX and 4-lane KR/KX/KX4 ports Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 06/12] cxgb4: allocate more space for MSI-X interrupt names Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev Print information about each port when its netdev is registered instead of looping separately over the ports at the end. The bulk of this patch is due to indentation change. Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/cxgb4_main.c | 54 ++++++++++++++++----------------------- 1 files changed, 22 insertions(+), 32 deletions(-) diff --git a/drivers/net/cxgb4/cxgb4_main.c b/drivers/net/cxgb4/cxgb4_main.c index 3f33d51..089b753 100644 --- a/drivers/net/cxgb4/cxgb4_main.c +++ b/drivers/net/cxgb4/cxgb4_main.c @@ -3490,50 +3490,41 @@ static int __devinit init_rss(struct adapter *adap) return 0; } -static void __devinit print_port_info(struct adapter *adap) +static void __devinit print_port_info(const struct net_device *dev) { static const char *base[] = { "R XFI", "R XAUI", "T SGMII", "T XFI", "T XAUI", "KX4", "CX4", "KX", "KR", "R SFP+", "KR/KX", "KR/KX/KX4" }; - int i; char buf[80]; + char *bufp = buf; const char *spd = ""; + const struct port_info *pi = netdev_priv(dev); + const struct adapter *adap = pi->adapter; if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_2_5GB) spd = " 2.5 GT/s"; else if (adap->params.pci.speed == PCI_EXP_LNKSTA_CLS_5_0GB) spd = " 5 GT/s"; - for_each_port(adap, i) { - struct net_device *dev = adap->port[i]; - const struct port_info *pi = netdev_priv(dev); - char *bufp = buf; - - if (!test_bit(i, &adap->registered_device_map)) - continue; - - if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M) - bufp += sprintf(bufp, "100/"); - if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G) - bufp += sprintf(bufp, "1000/"); - if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G) - bufp += sprintf(bufp, "10G/"); - if (bufp != buf) - --bufp; - sprintf(bufp, "BASE-%s", base[pi->port_type]); - - netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n", - adap->params.vpd.id, adap->params.rev, - buf, is_offload(adap) ? "R" : "", - adap->params.pci.width, spd, - (adap->flags & USING_MSIX) ? " MSI-X" : - (adap->flags & USING_MSI) ? " MSI" : ""); - if (adap->name == dev->name) - netdev_info(dev, "S/N: %s, E/C: %s\n", - adap->params.vpd.sn, adap->params.vpd.ec); - } + if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_100M) + bufp += sprintf(bufp, "100/"); + if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_1G) + bufp += sprintf(bufp, "1000/"); + if (pi->link_cfg.supported & FW_PORT_CAP_SPEED_10G) + bufp += sprintf(bufp, "10G/"); + if (bufp != buf) + --bufp; + sprintf(bufp, "BASE-%s", base[pi->port_type]); + + netdev_info(dev, "Chelsio %s rev %d %s %sNIC PCIe x%d%s%s\n", + adap->params.vpd.id, adap->params.rev, buf, + is_offload(adap) ? "R" : "", adap->params.pci.width, spd, + (adap->flags & USING_MSIX) ? " MSI-X" : + (adap->flags & USING_MSI) ? " MSI" : ""); + netdev_info(dev, "S/N: %s, E/C: %s\n", + adap->params.vpd.sn, adap->params.vpd.ec); } static void __devinit enable_pcie_relaxed_ordering(struct pci_dev *dev) @@ -3753,6 +3744,7 @@ static int __devinit init_one(struct pci_dev *pdev, __set_bit(i, &adapter->registered_device_map); adapter->chan_map[pi->tx_chan] = i; + print_port_info(adapter->port[i]); } } if (!adapter->registered_device_map) { @@ -3769,8 +3761,6 @@ static int __devinit init_one(struct pci_dev *pdev, if (is_offload(adapter)) attach_ulds(adapter); - print_port_info(adapter); - sriov: #ifdef CONFIG_PCI_IOV if (func < ARRAY_SIZE(num_vf) && num_vf[func] > 0) -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 06/12] cxgb4: allocate more space for MSI-X interrupt names 2010-12-14 20:18 ` [PATCH 05/12] cxgb4: print port information after registering each netdev Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 07/12] cxgb4: correct formatting of " Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev Currently MSI-X names for netdevs with long names are truncated in /proc/interrupts due to insufficient space. Use IFNAMSIZ to size the needed space. Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/cxgb4.h | 2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/drivers/net/cxgb4/cxgb4.h b/drivers/net/cxgb4/cxgb4.h index 3d4253d..f3d9f64 100644 --- a/drivers/net/cxgb4/cxgb4.h +++ b/drivers/net/cxgb4/cxgb4.h @@ -497,7 +497,7 @@ struct adapter { struct { unsigned short vec; - char desc[14]; + char desc[IFNAMSIZ + 10]; } msix_info[MAX_INGQ + 1]; struct sge sge; -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 07/12] cxgb4: correct formatting of MSI-X interrupt names 2010-12-14 20:18 ` [PATCH 06/12] cxgb4: allocate more space for MSI-X interrupt names Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 08/12] cxgb4: remove the name field from the adapter structure Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev The last byte of the buffer for MSI-X names could not be used due to a bogus -1. Also do not explicitly clear the last byte, snprintf will do the right thing. Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/cxgb4_main.c | 17 +++++------------ 1 files changed, 5 insertions(+), 12 deletions(-) diff --git a/drivers/net/cxgb4/cxgb4_main.c b/drivers/net/cxgb4/cxgb4_main.c index 089b753..ba9a67a 100644 --- a/drivers/net/cxgb4/cxgb4_main.c +++ b/drivers/net/cxgb4/cxgb4_main.c @@ -522,39 +522,32 @@ static irqreturn_t t4_nondata_intr(int irq, void *cookie) */ static void name_msix_vecs(struct adapter *adap) { - int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc) - 1; + int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc); /* non-data interrupts */ snprintf(adap->msix_info[0].desc, n, "%s", adap->name); - adap->msix_info[0].desc[n] = 0; /* FW events */ snprintf(adap->msix_info[1].desc, n, "%s-FWeventq", adap->name); - adap->msix_info[1].desc[n] = 0; /* Ethernet queues */ for_each_port(adap, j) { struct net_device *d = adap->port[j]; const struct port_info *pi = netdev_priv(d); - for (i = 0; i < pi->nqsets; i++, msi_idx++) { + for (i = 0; i < pi->nqsets; i++, msi_idx++) snprintf(adap->msix_info[msi_idx].desc, n, "%s-Rx%d", d->name, i); - adap->msix_info[msi_idx].desc[n] = 0; - } } /* offload queues */ - for_each_ofldrxq(&adap->sge, i) { + for_each_ofldrxq(&adap->sge, i) snprintf(adap->msix_info[msi_idx].desc, n, "%s-ofld%d", adap->name, i); - adap->msix_info[msi_idx++].desc[n] = 0; - } - for_each_rdmarxq(&adap->sge, i) { + + for_each_rdmarxq(&adap->sge, i) snprintf(adap->msix_info[msi_idx].desc, n, "%s-rdma%d", adap->name, i); - adap->msix_info[msi_idx++].desc[n] = 0; - } } static int request_msix_queue_irqs(struct adapter *adap) -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 08/12] cxgb4: remove the name field from the adapter structure 2010-12-14 20:18 ` [PATCH 07/12] cxgb4: correct formatting of " Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 09/12] cxgb4: remove a bitmap Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev Remove a field the driver uses to keep track of the name of the first netdev it manages to register. Do this by changing the registration loop to stop the first time it fails so the first registered device is trivial to tell. Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/cxgb4.h | 1 - drivers/net/cxgb4/cxgb4_main.c | 37 +++++++++++++++---------------------- 2 files changed, 15 insertions(+), 23 deletions(-) diff --git a/drivers/net/cxgb4/cxgb4.h b/drivers/net/cxgb4/cxgb4.h index f3d9f64..9caf95f 100644 --- a/drivers/net/cxgb4/cxgb4.h +++ b/drivers/net/cxgb4/cxgb4.h @@ -486,7 +486,6 @@ struct adapter { unsigned int fn; unsigned int flags; - const char *name; int msg_enable; struct adapter_params params; diff --git a/drivers/net/cxgb4/cxgb4_main.c b/drivers/net/cxgb4/cxgb4_main.c index ba9a67a..3db21d1 100644 --- a/drivers/net/cxgb4/cxgb4_main.c +++ b/drivers/net/cxgb4/cxgb4_main.c @@ -525,10 +525,11 @@ static void name_msix_vecs(struct adapter *adap) int i, j, msi_idx = 2, n = sizeof(adap->msix_info[0].desc); /* non-data interrupts */ - snprintf(adap->msix_info[0].desc, n, "%s", adap->name); + snprintf(adap->msix_info[0].desc, n, "%s", adap->port[0]->name); /* FW events */ - snprintf(adap->msix_info[1].desc, n, "%s-FWeventq", adap->name); + snprintf(adap->msix_info[1].desc, n, "%s-FWeventq", + adap->port[0]->name); /* Ethernet queues */ for_each_port(adap, j) { @@ -543,11 +544,11 @@ static void name_msix_vecs(struct adapter *adap) /* offload queues */ for_each_ofldrxq(&adap->sge, i) snprintf(adap->msix_info[msi_idx].desc, n, "%s-ofld%d", - adap->name, i); + adap->port[0]->name, i); for_each_rdmarxq(&adap->sge, i) snprintf(adap->msix_info[msi_idx].desc, n, "%s-rdma%d", - adap->name, i); + adap->port[0]->name, i); } static int request_msix_queue_irqs(struct adapter *adap) @@ -2664,7 +2665,7 @@ static int cxgb_up(struct adapter *adap) } else { err = request_irq(adap->pdev->irq, t4_intr_handler(adap), (adap->flags & USING_MSI) ? 0 : IRQF_SHARED, - adap->name, adap); + adap->port[0]->name, adap); if (err) goto irq_err; } @@ -3627,7 +3628,6 @@ static int __devinit init_one(struct pci_dev *pdev, adapter->pdev = pdev; adapter->pdev_dev = &pdev->dev; adapter->fn = func; - adapter->name = pci_name(pdev); adapter->msg_enable = dflt_msg_enable; memset(adapter->chan_map, 0xff, sizeof(adapter->chan_map)); @@ -3724,26 +3724,19 @@ static int __devinit init_one(struct pci_dev *pdev, err = register_netdev(adapter->port[i]); if (err) - dev_warn(&pdev->dev, - "cannot register net device %s, skipping\n", - adapter->port[i]->name); - else { - /* - * Change the name we use for messages to the name of - * the first successfully registered interface. - */ - if (!adapter->registered_device_map) - adapter->name = adapter->port[i]->name; - - __set_bit(i, &adapter->registered_device_map); - adapter->chan_map[pi->tx_chan] = i; - print_port_info(adapter->port[i]); - } + break; + __set_bit(i, &adapter->registered_device_map); + adapter->chan_map[pi->tx_chan] = i; + print_port_info(adapter->port[i]); } - if (!adapter->registered_device_map) { + if (i == 0) { dev_err(&pdev->dev, "could not register any net devices\n"); goto out_free_dev; } + if (err) { + dev_warn(&pdev->dev, "only %d net devices registered\n", i); + err = 0; + }; if (cxgb4_debugfs_root) { adapter->debugfs_root = debugfs_create_dir(pci_name(pdev), -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 09/12] cxgb4: remove a bitmap 2010-12-14 20:18 ` [PATCH 08/12] cxgb4: remove the name field from the adapter structure Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 10/12] cxgb4: add const to static arrays Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev The driver keeps a bitmap of the netdevs it registered so it knows what to unregister later. Remove that and look at reg_state instead. Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/cxgb4.h | 1 - drivers/net/cxgb4/cxgb4_main.c | 3 +-- 2 files changed, 1 insertions(+), 3 deletions(-) diff --git a/drivers/net/cxgb4/cxgb4.h b/drivers/net/cxgb4/cxgb4.h index 9caf95f..01d49ea 100644 --- a/drivers/net/cxgb4/cxgb4.h +++ b/drivers/net/cxgb4/cxgb4.h @@ -482,7 +482,6 @@ struct adapter { void __iomem *regs; struct pci_dev *pdev; struct device *pdev_dev; - unsigned long registered_device_map; unsigned int fn; unsigned int flags; diff --git a/drivers/net/cxgb4/cxgb4_main.c b/drivers/net/cxgb4/cxgb4_main.c index 3db21d1..d016e10 100644 --- a/drivers/net/cxgb4/cxgb4_main.c +++ b/drivers/net/cxgb4/cxgb4_main.c @@ -3725,7 +3725,6 @@ static int __devinit init_one(struct pci_dev *pdev, err = register_netdev(adapter->port[i]); if (err) break; - __set_bit(i, &adapter->registered_device_map); adapter->chan_map[pi->tx_chan] = i; print_port_info(adapter->port[i]); } @@ -3785,7 +3784,7 @@ static void __devexit remove_one(struct pci_dev *pdev) detach_ulds(adapter); for_each_port(adapter, i) - if (test_bit(i, &adapter->registered_device_map)) + if (adapter->port[i]->reg_state == NETREG_REGISTERED) unregister_netdev(adapter->port[i]); if (adapter->debugfs_root) -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 10/12] cxgb4: add const to static arrays 2010-12-14 20:18 ` [PATCH 09/12] cxgb4: remove a bitmap Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 11/12] cxgb4: extend VPD parsing Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev; +Cc: Joe Perches Patch originally from Joe Perches, unmodified. Signed-off-by: Joe Perches <joe@perches.com> Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/t4_hw.c | 48 ++++++++++++++++++++++---------------------- 1 files changed, 24 insertions(+), 24 deletions(-) diff --git a/drivers/net/cxgb4/t4_hw.c b/drivers/net/cxgb4/t4_hw.c index 6af0452..c7fb549 100644 --- a/drivers/net/cxgb4/t4_hw.c +++ b/drivers/net/cxgb4/t4_hw.c @@ -183,7 +183,7 @@ static void dump_mbox(struct adapter *adap, int mbox, u32 data_reg) int t4_wr_mbox_meat(struct adapter *adap, int mbox, const void *cmd, int size, void *rpl, bool sleep_ok) { - static int delay[] = { + static const int delay[] = { 1, 1, 3, 5, 10, 10, 20, 50, 100, 200 }; @@ -917,7 +917,7 @@ static int t4_handle_intr_status(struct adapter *adapter, unsigned int reg, */ static void pcie_intr_handler(struct adapter *adapter) { - static struct intr_info sysbus_intr_info[] = { + static const struct intr_info sysbus_intr_info[] = { { RNPP, "RXNP array parity error", -1, 1 }, { RPCP, "RXPC array parity error", -1, 1 }, { RCIP, "RXCIF array parity error", -1, 1 }, @@ -925,7 +925,7 @@ static void pcie_intr_handler(struct adapter *adapter) { RFTP, "RXFT array parity error", -1, 1 }, { 0 } }; - static struct intr_info pcie_port_intr_info[] = { + static const struct intr_info pcie_port_intr_info[] = { { TPCP, "TXPC array parity error", -1, 1 }, { TNPP, "TXNP array parity error", -1, 1 }, { TFTP, "TXFT array parity error", -1, 1 }, @@ -937,7 +937,7 @@ static void pcie_intr_handler(struct adapter *adapter) { TDUE, "Tx uncorrectable data error", -1, 1 }, { 0 } }; - static struct intr_info pcie_intr_info[] = { + static const struct intr_info pcie_intr_info[] = { { MSIADDRLPERR, "MSI AddrL parity error", -1, 1 }, { MSIADDRHPERR, "MSI AddrH parity error", -1, 1 }, { MSIDATAPERR, "MSI data parity error", -1, 1 }, @@ -989,7 +989,7 @@ static void pcie_intr_handler(struct adapter *adapter) */ static void tp_intr_handler(struct adapter *adapter) { - static struct intr_info tp_intr_info[] = { + static const struct intr_info tp_intr_info[] = { { 0x3fffffff, "TP parity error", -1, 1 }, { FLMTXFLSTEMPTY, "TP out of Tx pages", -1, 1 }, { 0 } @@ -1006,7 +1006,7 @@ static void sge_intr_handler(struct adapter *adapter) { u64 v; - static struct intr_info sge_intr_info[] = { + static const struct intr_info sge_intr_info[] = { { ERR_CPL_EXCEED_IQE_SIZE, "SGE received CPL exceeding IQE size", -1, 1 }, { ERR_INVALID_CIDX_INC, @@ -1051,7 +1051,7 @@ static void sge_intr_handler(struct adapter *adapter) */ static void cim_intr_handler(struct adapter *adapter) { - static struct intr_info cim_intr_info[] = { + static const struct intr_info cim_intr_info[] = { { PREFDROPINT, "CIM control register prefetch drop", -1, 1 }, { OBQPARERR, "CIM OBQ parity error", -1, 1 }, { IBQPARERR, "CIM IBQ parity error", -1, 1 }, @@ -1061,7 +1061,7 @@ static void cim_intr_handler(struct adapter *adapter) { TIEQOUTPARERRINT, "CIM TIEQ incoming parity error", -1, 1 }, { 0 } }; - static struct intr_info cim_upintr_info[] = { + static const struct intr_info cim_upintr_info[] = { { RSVDSPACEINT, "CIM reserved space access", -1, 1 }, { ILLTRANSINT, "CIM illegal transaction", -1, 1 }, { ILLWRINT, "CIM illegal write", -1, 1 }, @@ -1108,7 +1108,7 @@ static void cim_intr_handler(struct adapter *adapter) */ static void ulprx_intr_handler(struct adapter *adapter) { - static struct intr_info ulprx_intr_info[] = { + static const struct intr_info ulprx_intr_info[] = { { 0x1800000, "ULPRX context error", -1, 1 }, { 0x7fffff, "ULPRX parity error", -1, 1 }, { 0 } @@ -1123,7 +1123,7 @@ static void ulprx_intr_handler(struct adapter *adapter) */ static void ulptx_intr_handler(struct adapter *adapter) { - static struct intr_info ulptx_intr_info[] = { + static const struct intr_info ulptx_intr_info[] = { { PBL_BOUND_ERR_CH3, "ULPTX channel 3 PBL out of bounds", -1, 0 }, { PBL_BOUND_ERR_CH2, "ULPTX channel 2 PBL out of bounds", -1, @@ -1145,7 +1145,7 @@ static void ulptx_intr_handler(struct adapter *adapter) */ static void pmtx_intr_handler(struct adapter *adapter) { - static struct intr_info pmtx_intr_info[] = { + static const struct intr_info pmtx_intr_info[] = { { PCMD_LEN_OVFL0, "PMTX channel 0 pcmd too large", -1, 1 }, { PCMD_LEN_OVFL1, "PMTX channel 1 pcmd too large", -1, 1 }, { PCMD_LEN_OVFL2, "PMTX channel 2 pcmd too large", -1, 1 }, @@ -1167,7 +1167,7 @@ static void pmtx_intr_handler(struct adapter *adapter) */ static void pmrx_intr_handler(struct adapter *adapter) { - static struct intr_info pmrx_intr_info[] = { + static const struct intr_info pmrx_intr_info[] = { { ZERO_E_CMD_ERROR, "PMRX 0-length pcmd", -1, 1 }, { PMRX_FRAMING_ERROR, "PMRX framing error", -1, 1 }, { OCSPI_PAR_ERROR, "PMRX ocspi parity error", -1, 1 }, @@ -1186,7 +1186,7 @@ static void pmrx_intr_handler(struct adapter *adapter) */ static void cplsw_intr_handler(struct adapter *adapter) { - static struct intr_info cplsw_intr_info[] = { + static const struct intr_info cplsw_intr_info[] = { { CIM_OP_MAP_PERR, "CPLSW CIM op_map parity error", -1, 1 }, { CIM_OVFL_ERROR, "CPLSW CIM overflow", -1, 1 }, { TP_FRAMING_ERROR, "CPLSW TP framing error", -1, 1 }, @@ -1205,7 +1205,7 @@ static void cplsw_intr_handler(struct adapter *adapter) */ static void le_intr_handler(struct adapter *adap) { - static struct intr_info le_intr_info[] = { + static const struct intr_info le_intr_info[] = { { LIPMISS, "LE LIP miss", -1, 0 }, { LIP0, "LE 0 LIP error", -1, 0 }, { PARITYERR, "LE parity error", -1, 1 }, @@ -1223,11 +1223,11 @@ static void le_intr_handler(struct adapter *adap) */ static void mps_intr_handler(struct adapter *adapter) { - static struct intr_info mps_rx_intr_info[] = { + static const struct intr_info mps_rx_intr_info[] = { { 0xffffff, "MPS Rx parity error", -1, 1 }, { 0 } }; - static struct intr_info mps_tx_intr_info[] = { + static const struct intr_info mps_tx_intr_info[] = { { TPFIFO, "MPS Tx TP FIFO parity error", -1, 1 }, { NCSIFIFO, "MPS Tx NC-SI FIFO parity error", -1, 1 }, { TXDATAFIFO, "MPS Tx data FIFO parity error", -1, 1 }, @@ -1237,25 +1237,25 @@ static void mps_intr_handler(struct adapter *adapter) { FRMERR, "MPS Tx framing error", -1, 1 }, { 0 } }; - static struct intr_info mps_trc_intr_info[] = { + static const struct intr_info mps_trc_intr_info[] = { { FILTMEM, "MPS TRC filter parity error", -1, 1 }, { PKTFIFO, "MPS TRC packet FIFO parity error", -1, 1 }, { MISCPERR, "MPS TRC misc parity error", -1, 1 }, { 0 } }; - static struct intr_info mps_stat_sram_intr_info[] = { + static const struct intr_info mps_stat_sram_intr_info[] = { { 0x1fffff, "MPS statistics SRAM parity error", -1, 1 }, { 0 } }; - static struct intr_info mps_stat_tx_intr_info[] = { + static const struct intr_info mps_stat_tx_intr_info[] = { { 0xfffff, "MPS statistics Tx FIFO parity error", -1, 1 }, { 0 } }; - static struct intr_info mps_stat_rx_intr_info[] = { + static const struct intr_info mps_stat_rx_intr_info[] = { { 0xffffff, "MPS statistics Rx FIFO parity error", -1, 1 }, { 0 } }; - static struct intr_info mps_cls_intr_info[] = { + static const struct intr_info mps_cls_intr_info[] = { { MATCHSRAM, "MPS match SRAM parity error", -1, 1 }, { MATCHTCAM, "MPS match TCAM parity error", -1, 1 }, { HASHSRAM, "MPS hash SRAM parity error", -1, 1 }, @@ -1354,7 +1354,7 @@ static void ma_intr_handler(struct adapter *adap) */ static void smb_intr_handler(struct adapter *adap) { - static struct intr_info smb_intr_info[] = { + static const struct intr_info smb_intr_info[] = { { MSTTXFIFOPARINT, "SMB master Tx FIFO parity error", -1, 1 }, { MSTRXFIFOPARINT, "SMB master Rx FIFO parity error", -1, 1 }, { SLVFIFOPARINT, "SMB slave FIFO parity error", -1, 1 }, @@ -1370,7 +1370,7 @@ static void smb_intr_handler(struct adapter *adap) */ static void ncsi_intr_handler(struct adapter *adap) { - static struct intr_info ncsi_intr_info[] = { + static const struct intr_info ncsi_intr_info[] = { { CIM_DM_PRTY_ERR, "NC-SI CIM parity error", -1, 1 }, { MPS_DM_PRTY_ERR, "NC-SI MPS parity error", -1, 1 }, { TXFIFO_PRTY_ERR, "NC-SI Tx FIFO parity error", -1, 1 }, @@ -1408,7 +1408,7 @@ static void xgmac_intr_handler(struct adapter *adap, int port) */ static void pl_intr_handler(struct adapter *adap) { - static struct intr_info pl_intr_info[] = { + static const struct intr_info pl_intr_info[] = { { FATALPERR, "T4 fatal parity error", -1, 1 }, { PERRVFID, "PL VFID_MAP parity error", -1, 1 }, { 0 } -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 11/12] cxgb4: extend VPD parsing 2010-12-14 20:18 ` [PATCH 10/12] cxgb4: add const to static arrays Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 12/12] cxgb4: NUMA-aware Tx queue allocations Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev Current code parses the VPD RO section for keywords but makes static assumptions about the location of the section. Remove them and parse the VPD to find it. Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/t4_hw.c | 41 +++++++++++++++++++++-------------------- 1 files changed, 21 insertions(+), 20 deletions(-) diff --git a/drivers/net/cxgb4/t4_hw.c b/drivers/net/cxgb4/t4_hw.c index c7fb549..b9fd8a6 100644 --- a/drivers/net/cxgb4/t4_hw.c +++ b/drivers/net/cxgb4/t4_hw.c @@ -330,18 +330,6 @@ int t4_edc_read(struct adapter *adap, int idx, u32 addr, __be32 *data, u64 *ecc) return 0; } -/* - * Partial EEPROM Vital Product Data structure. Includes only the ID and - * VPD-R header. - */ -struct t4_vpd_hdr { - u8 id_tag; - u8 id_len[2]; - u8 id_data[ID_LEN]; - u8 vpdr_tag; - u8 vpdr_len[2]; -}; - #define EEPROM_STAT_ADDR 0x7bfc #define VPD_BASE 0 #define VPD_LEN 512 @@ -372,23 +360,36 @@ static int get_vpd_params(struct adapter *adapter, struct vpd_params *p) int i, ret; int ec, sn; u8 vpd[VPD_LEN], csum; - unsigned int vpdr_len; - const struct t4_vpd_hdr *v; + unsigned int vpdr_len, kw_offset, id_len; ret = pci_read_vpd(adapter->pdev, VPD_BASE, sizeof(vpd), vpd); if (ret < 0) return ret; - v = (const struct t4_vpd_hdr *)vpd; - vpdr_len = pci_vpd_lrdt_size(&v->vpdr_tag); - if (vpdr_len + sizeof(struct t4_vpd_hdr) > VPD_LEN) { + if (vpd[0] != PCI_VPD_LRDT_ID_STRING) { + dev_err(adapter->pdev_dev, "missing VPD ID string\n"); + return -EINVAL; + } + + id_len = pci_vpd_lrdt_size(vpd); + if (id_len > ID_LEN) + id_len = ID_LEN; + + i = pci_vpd_find_tag(vpd, 0, VPD_LEN, PCI_VPD_LRDT_RO_DATA); + if (i < 0) { + dev_err(adapter->pdev_dev, "missing VPD-R section\n"); + return -EINVAL; + } + + vpdr_len = pci_vpd_lrdt_size(&vpd[i]); + kw_offset = i + PCI_VPD_LRDT_TAG_SIZE; + if (vpdr_len + kw_offset > VPD_LEN) { dev_err(adapter->pdev_dev, "bad VPD-R length %u\n", vpdr_len); return -EINVAL; } #define FIND_VPD_KW(var, name) do { \ - var = pci_vpd_find_info_keyword(&v->id_tag, sizeof(struct t4_vpd_hdr), \ - vpdr_len, name); \ + var = pci_vpd_find_info_keyword(vpd, kw_offset, vpdr_len, name); \ if (var < 0) { \ dev_err(adapter->pdev_dev, "missing VPD keyword " name "\n"); \ return -EINVAL; \ @@ -410,7 +411,7 @@ static int get_vpd_params(struct adapter *adapter, struct vpd_params *p) FIND_VPD_KW(sn, "SN"); #undef FIND_VPD_KW - memcpy(p->id, v->id_data, ID_LEN); + memcpy(p->id, vpd + PCI_VPD_LRDT_TAG_SIZE, id_len); strim(p->id); memcpy(p->ec, vpd + ec, EC_LEN); strim(p->ec); -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* [PATCH 12/12] cxgb4: NUMA-aware Tx queue allocations 2010-12-14 20:18 ` [PATCH 11/12] cxgb4: extend VPD parsing Dimitris Michailidis @ 2010-12-14 20:18 ` Dimitris Michailidis 2010-12-14 21:17 ` Eric Dumazet 0 siblings, 1 reply; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 20:18 UTC (permalink / raw) To: netdev Allocate Tx queue memory on the node indicated by the new netdev_queue_numa_node_read. If that fails we allocate on any node. Signed-off-by: Dimitris Michailidis <dm@chelsio.com> --- drivers/net/cxgb4/sge.c | 20 +++++++++++++------- 1 files changed, 13 insertions(+), 7 deletions(-) diff --git a/drivers/net/cxgb4/sge.c b/drivers/net/cxgb4/sge.c index cc0b997..ed98b8a 100644 --- a/drivers/net/cxgb4/sge.c +++ b/drivers/net/cxgb4/sge.c @@ -579,6 +579,7 @@ static inline void __refill_fl(struct adapter *adap, struct sge_fl *fl) * @phys: the physical address of the allocated ring * @metadata: address of the array holding the SW state for the ring * @stat_size: extra space in HW ring for status information + * @node: preferred node for memory allocations * * Allocates resources for an SGE descriptor ring, such as Tx queues, * free buffer lists, or response queues. Each SGE ring requires @@ -590,7 +591,7 @@ static inline void __refill_fl(struct adapter *adap, struct sge_fl *fl) */ static void *alloc_ring(struct device *dev, size_t nelem, size_t elem_size, size_t sw_size, dma_addr_t *phys, void *metadata, - size_t stat_size) + size_t stat_size, int node) { size_t len = nelem * elem_size + stat_size; void *s = NULL; @@ -599,7 +600,10 @@ static void *alloc_ring(struct device *dev, size_t nelem, size_t elem_size, if (!p) return NULL; if (sw_size) { - s = kcalloc(nelem, sw_size, GFP_KERNEL); + if (node >= 0) + s = kzalloc_node(nelem * sw_size, GFP_KERNEL, node); + if (!s) + s = kcalloc(nelem, sw_size, GFP_KERNEL); if (!s) { dma_free_coherent(dev, len, p, *phys); @@ -1982,7 +1986,7 @@ int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq, iq->size = roundup(iq->size, 16); iq->desc = alloc_ring(adap->pdev_dev, iq->size, iq->iqe_len, 0, - &iq->phys_addr, NULL, 0); + &iq->phys_addr, NULL, 0, NUMA_NO_NODE); if (!iq->desc) return -ENOMEM; @@ -2008,7 +2012,7 @@ int t4_sge_alloc_rxq(struct adapter *adap, struct sge_rspq *iq, bool fwevtq, fl->size = roundup(fl->size, 8); fl->desc = alloc_ring(adap->pdev_dev, fl->size, sizeof(__be64), sizeof(struct rx_sw_desc), &fl->addr, - &fl->sdesc, STAT_LEN); + &fl->sdesc, STAT_LEN, NUMA_NO_NODE); if (!fl->desc) goto fl_nomem; @@ -2095,7 +2099,8 @@ int t4_sge_alloc_eth_txq(struct adapter *adap, struct sge_eth_txq *txq, txq->q.desc = alloc_ring(adap->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, STAT_LEN, + netdev_queue_numa_node_read(netdevq)); if (!txq->q.desc) return -ENOMEM; @@ -2147,7 +2152,7 @@ int t4_sge_alloc_ctrl_txq(struct adapter *adap, struct sge_ctrl_txq *txq, txq->q.desc = alloc_ring(adap->pdev_dev, nentries, sizeof(struct tx_desc), 0, &txq->q.phys_addr, - NULL, 0); + NULL, 0, NUMA_NO_NODE); if (!txq->q.desc) return -ENOMEM; @@ -2198,7 +2203,8 @@ int t4_sge_alloc_ofld_txq(struct adapter *adap, struct sge_ofld_txq *txq, txq->q.desc = alloc_ring(adap->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, STAT_LEN, + NUMA_NO_NODE); if (!txq->q.desc) return -ENOMEM; -- 1.5.4 ^ permalink raw reply related [flat|nested] 15+ messages in thread
* Re: [PATCH 12/12] cxgb4: NUMA-aware Tx queue allocations 2010-12-14 20:18 ` [PATCH 12/12] cxgb4: NUMA-aware Tx queue allocations Dimitris Michailidis @ 2010-12-14 21:17 ` Eric Dumazet 2010-12-14 22:51 ` Dimitris Michailidis 0 siblings, 1 reply; 15+ messages in thread From: Eric Dumazet @ 2010-12-14 21:17 UTC (permalink / raw) To: Dimitris Michailidis; +Cc: netdev Le mardi 14 décembre 2010 à 12:18 -0800, Dimitris Michailidis a écrit : > Allocate Tx queue memory on the node indicated by the new > netdev_queue_numa_node_read. If that fails we allocate on any node. > > Signed-off-by: Dimitris Michailidis <dm@chelsio.com> > --- > drivers/net/cxgb4/sge.c | 20 +++++++++++++------- > 1 files changed, 13 insertions(+), 7 deletions(-) > > diff --git a/drivers/net/cxgb4/sge.c b/drivers/net/cxgb4/sge.c > index cc0b997..ed98b8a 100644 > --- a/drivers/net/cxgb4/sge.c > +++ b/drivers/net/cxgb4/sge.c > @@ -579,6 +579,7 @@ static inline void __refill_fl(struct adapter *adap, struct sge_fl *fl) > * @phys: the physical address of the allocated ring > * @metadata: address of the array holding the SW state for the ring > * @stat_size: extra space in HW ring for status information > + * @node: preferred node for memory allocations > * > * Allocates resources for an SGE descriptor ring, such as Tx queues, > * free buffer lists, or response queues. Each SGE ring requires > @@ -590,7 +591,7 @@ static inline void __refill_fl(struct adapter *adap, struct sge_fl *fl) > */ > static void *alloc_ring(struct device *dev, size_t nelem, size_t elem_size, > size_t sw_size, dma_addr_t *phys, void *metadata, > - size_t stat_size) > + size_t stat_size, int node) > { > size_t len = nelem * elem_size + stat_size; > void *s = NULL; > @@ -599,7 +600,10 @@ static void *alloc_ring(struct device *dev, size_t nelem, size_t elem_size, > if (!p) > return NULL; > if (sw_size) { > - s = kcalloc(nelem, sw_size, GFP_KERNEL); > + if (node >= 0) > + s = kzalloc_node(nelem * sw_size, GFP_KERNEL, node); kzalloc_node() has a fallback, you dont need to retry with kcalloc() > + if (!s) > + s = kcalloc(nelem, sw_size, GFP_KERNEL); > > if (!s) { > dma_free_coherent(dev, len, p, *phys); Also, I am not sure it is going to work, since we can setup XPS only after device being setup ? By the time your driver allocates rings, we probably read -1/NUMA_NO_NODE ^ permalink raw reply [flat|nested] 15+ messages in thread
* Re: [PATCH 12/12] cxgb4: NUMA-aware Tx queue allocations 2010-12-14 21:17 ` Eric Dumazet @ 2010-12-14 22:51 ` Dimitris Michailidis 0 siblings, 0 replies; 15+ messages in thread From: Dimitris Michailidis @ 2010-12-14 22:51 UTC (permalink / raw) To: Eric Dumazet; +Cc: netdev Eric Dumazet wrote: > Le mardi 14 décembre 2010 à 12:18 -0800, Dimitris Michailidis a écrit : >> Allocate Tx queue memory on the node indicated by the new >> netdev_queue_numa_node_read. If that fails we allocate on any node. >> >> Signed-off-by: Dimitris Michailidis <dm@chelsio.com> >> --- >> drivers/net/cxgb4/sge.c | 20 +++++++++++++------- >> 1 files changed, 13 insertions(+), 7 deletions(-) >> >> diff --git a/drivers/net/cxgb4/sge.c b/drivers/net/cxgb4/sge.c >> index cc0b997..ed98b8a 100644 >> --- a/drivers/net/cxgb4/sge.c >> +++ b/drivers/net/cxgb4/sge.c >> @@ -579,6 +579,7 @@ static inline void __refill_fl(struct adapter *adap, struct sge_fl *fl) >> * @phys: the physical address of the allocated ring >> * @metadata: address of the array holding the SW state for the ring >> * @stat_size: extra space in HW ring for status information >> + * @node: preferred node for memory allocations >> * >> * Allocates resources for an SGE descriptor ring, such as Tx queues, >> * free buffer lists, or response queues. Each SGE ring requires >> @@ -590,7 +591,7 @@ static inline void __refill_fl(struct adapter *adap, struct sge_fl *fl) >> */ >> static void *alloc_ring(struct device *dev, size_t nelem, size_t elem_size, >> size_t sw_size, dma_addr_t *phys, void *metadata, >> - size_t stat_size) >> + size_t stat_size, int node) >> { >> size_t len = nelem * elem_size + stat_size; >> void *s = NULL; >> @@ -599,7 +600,10 @@ static void *alloc_ring(struct device *dev, size_t nelem, size_t elem_size, >> if (!p) >> return NULL; >> if (sw_size) { >> - s = kcalloc(nelem, sw_size, GFP_KERNEL); >> + if (node >= 0) >> + s = kzalloc_node(nelem * sw_size, GFP_KERNEL, node); > > kzalloc_node() has a fallback, you dont need to retry with kcalloc() I took this retry part from ixgbe but I can remove it if it's not needed. Luckily it's the last patch in the series. > >> + if (!s) >> + s = kcalloc(nelem, sw_size, GFP_KERNEL); >> >> if (!s) { >> dma_free_coherent(dev, len, p, *phys); > > Also, I am not sure it is going to work, since we can setup XPS only > after device being setup ? > > By the time your driver allocates rings, we probably read > -1/NUMA_NO_NODE XPS is available after registration. The queues are allocated at open time, if one configures XPS prior to that the allocations happen on the right nodes. I've tried this and this is the behavior I see. It is true that setting XPS after open doesn't affect the queue allocations. ^ permalink raw reply [flat|nested] 15+ messages in thread
end of thread, other threads:[~2010-12-14 22:51 UTC | newest] Thread overview: 15+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2010-12-14 20:18 [PATCH 00/12 net-next] cxgb4 update Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 01/12] cxgb4: enable PCIe relaxed ordering Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 02/12] cxgb4: do not read the clock frequency from VPD Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 03/12] cxgb4: set the number of queues before device registration Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 04/12] cxgb4: distinguish between 1-lane KR/KX and 4-lane KR/KX/KX4 ports Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 05/12] cxgb4: print port information after registering each netdev Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 06/12] cxgb4: allocate more space for MSI-X interrupt names Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 07/12] cxgb4: correct formatting of " Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 08/12] cxgb4: remove the name field from the adapter structure Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 09/12] cxgb4: remove a bitmap Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 10/12] cxgb4: add const to static arrays Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 11/12] cxgb4: extend VPD parsing Dimitris Michailidis 2010-12-14 20:18 ` [PATCH 12/12] cxgb4: NUMA-aware Tx queue allocations Dimitris Michailidis 2010-12-14 21:17 ` Eric Dumazet 2010-12-14 22:51 ` Dimitris Michailidis
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).