- 23 7月, 2011 1 次提交
-
-
由 Mike Marciniszyn 提交于
With ib_qib options: options ib_qib krcvqs=1 pcie_caps=0x51 rcvhdrcnt=4096 singleport=1 ibmtu=4 a run of ib_write_bw -a yields the following: ------------------------------------------------------------------ #bytes #iterations BW peak[MB/sec] BW average[MB/sec] 1048576 5000 2910.64 229.80 ------------------------------------------------------------------ The top cpu use in a profile is: CPU: Intel Architectural Perfmon, speed 2400.15 MHz (estimated) Counted CPU_CLK_UNHALTED events (Clock cycles when not halted) with a unit mask of 0x00 (No unit mask) count 1002300 Counted LLC_MISSES events (Last level cache demand requests from this core that missed the LLC) with a unit mask of 0x41 (No unit mask) count 10000 samples % samples % app name symbol name 15237 29.2642 964 17.1195 ib_qib.ko qib_7322intr 12320 23.6618 1040 18.4692 ib_qib.ko handle_7322_errors 4106 7.8860 0 0 vmlinux vsnprintf Analysis of the stats, profile, the code, and the annotated profile indicate: - All of the overflow interrupts (one per packet overflow) are serviced on CPU0 with no mitigation on the frequency. - All of the receive interrupts are being serviced by CPU0. (That is the way truescale.cmds statically allocates the kctx IRQs to CPU) - The code is spending all of its time servicing QIB_I_C_ERROR RcvEgrFullErr interrupts on CPU0, starving the packet receive processing. - The decode_err routine is very inefficient, using a printf variant to format a "%s" and continues to loop when the errs mask has been cleared. - Both qib_7322intr and handle_7322_errors read pci registers, which is very inefficient. The fix does the following: - Adds a tasklet to service QIB_I_C_ERROR - Replaces the very inefficient scnprintf() with a memcpy(). A field is added to qib_hwerror_msgs to save the sizeof("string") at compile time so that a strlen is not needed during err_decode(). - The most frequent errors (Overflows) are serviced first to exit the loop as early as possible. - The loop now exits as soon as the errs mask is clear rather than fruitlessly looping through the msp array. With this fix the performance changes to: ------------------------------------------------------------------ #bytes #iterations BW peak[MB/sec] BW average[MB/sec] 1048576 5000 2990.64 2941.35 ------------------------------------------------------------------ During testing of the error handling overflow patch, it was determined that some CPU's were slower when servicing both overflow and receive interrupts on CPU0 with different MSI interrupt vectors. This patch adds an option (krcvq01_no_msi) to not use a dedicated MSI interrupt for kctx's < 2 and to service them on the default interrupt. For some CPUs, the cost of the interrupt enter/exit is more costly than then the additional PCI read in the default handler. Signed-off-by: NMike Marciniszyn <mike.marciniszyn@qlogic.com> Signed-off-by: NRoland Dreier <roland@purestorage.com>
-
- 22 7月, 2011 39 次提交
-
-
由 Eric Dumazet 提交于
Some workloads need some headroom (NET_SKB_PAD) to avoid expensive reallocations. Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Don Skidmore 提交于
Private rx_csum flags are now duplicate of netdev->features & NETIF_F_RXCSUM. We remove those duplicates and now use the net_device_ops ndo_set_features. This was based on the original patch submitted by Michal Miroslaw <mirq-linux@rere.qmqm.pl>. I also removed the special case not requiring a reset for X540 hardware. It is needed just as it is in 82599 hardware. Signed-off-by: NDon Skidmore <donald.c.skidmore@intel.com> Cc: Michal Miroslaw <mirq-linux@rere.qmqm.pl> Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com> Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
-
由 Andy Gospodarek 提交于
Martin Wilck <martin.wilck@ts.fujitsu.com> reported that systems using the ixgbe-driver that were capable of WoL were rebooting almost as soon as they were shut down. This is because the default WoL settings enabled magic packet, broadcast, unicast, and multicast. Other Intel devices seem to use the stored eeprom value for initial WoL capabilities. The 82578DM (e1000e) and 82576 (igb) the devices I looked at had only the magic packet enabled in the eeprom, so that seems appropriate on ixgbe-based devices as well. I set the WoL options on my 82578DM to be the same default as the ixgbe devices (umbg) and saw the same as Martin -- almost as soon as my box shutdown, it booted again. This patch changes the default to only be the magic packet. This is the same as the default for most Intel and non-Intel hardware currently upstream. Signed-off-by: NAndy Gospodarek <andy@greyhouse.net> CC: Martin Wilck <martin.wilck@ts.fujitsu.com> Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com> Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
-
由 Emil Tantilov 提交于
Signed-off-by: NEmil Tantilov <emil.s.tantilov@intel.com> Tested-by: NPhil Schmitt <phillip.j.schmitt@intel.com> Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
-
由 Alexander Duyck 提交于
This change is meant to address possible race conditions from the status and error bits on the RX descriptors being re-read by multiple functions in the RX cleanup path. To resolve this I have added code that will pass the staterr value to those functions. Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com> Tested-by: NRoss Brattain <ross.b.brattain@intel.com> Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
-
由 Alexander Duyck 提交于
This change moves work_limit, total_packets, and total_bytes into the ring container struct of the q_vector. The advantage of this is that it should reduce the size of memory used in the event of multiple rings being assigned to a single q_vector. In addition it should help to reduce the total workload for calculating itr since now total_packets and total_bytes will be the total work done of the interrupt instead of for the ring. Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com> Tested-by: NRoss Brattain <ross.b.brattain@intel.com> Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
-
由 Alexander Duyck 提交于
This patch adds support for a ring container structure to be used within the q_vector. The basic idea is to provide a means of separating the RX and TX rings while maintaining a common structure for their containment. The advantage to this is that later we should be able to pass this structure to the update_itr functions without needing to pass individual rings. Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com> Tested-by: NRoss Brattain <ross.b.brattain@intel.com> Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
-
由 Dotan Barak 提交于
Many features were added to this driver, so the driver version should change too. Signed-off-by: NDotan Barak <dotanb@dev.mellanox.co.il> Signed-off-by: NRoland Dreier <roland@purestorage.com>
-
由 Alexander Duyck 提交于
The ixgbe_maybe_stop_tx function is only a few lines long and is called multiple times through the xmit hotpath. In order to streamline things it makes sense to just inline it. Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com> Tested-by: NRoss Brattain <ross.b.brattain@intel.com> Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
-
由 Alexander Duyck 提交于
This change is meant to update ATR so that it will use the recorded RX queue instead of the CPU in the case of routing. This change is meant to help ixgbe default behavior to more closely match that of the kernel. Signed-off-by: NAlexander Duyck <alexander.h.duyck@intel.com> Tested-by: NRoss Brattain <ross.b.brattain@intel.com> Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
-
由 Robert Healy 提交于
On this chipset it is required to configure the MPHY block for loopback tests. If MPHY is not configured then all loopback tests will report failures. Signed-off-by: NRobert Healy <robert.healy@intel.com> Tested-by: NAaron Brown <aaron.f.brown@intel.com> Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
-
由 Nicolas Schichan 提交于
Interrupts about link lost or rx sequence errors are not reported by the ce4100 hardware, leading to transitions from link UP to link DOWN never being reported. Signed-off-by: NNicolas Schichan <nschichan@freebox.fr> Tested-by: NAaron Brown <aaron.f.brown@intel.com> Signed-off-by: NJeff Kirsher <jeffrey.t.kirsher@intel.com>
-
由 Rusty Russell 提交于
We were blatting too much of the register. Linux didn't care, but in theory it might. Reported-by: NJonas Maebe <jonas.maebe@elis.ugent.be> Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
-
由 Rusty Russell 提交于
Also removes a long-unused #define and an extraneous semicolon. Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
-
由 Rusty Russell 提交于
We used to notify the Host every time we updated a device's status. However, it only really needs to know when we're resetting the device, or failed to initialize it, or when we've finished our feature negotiation. In particular, we used to wait for VIRTIO_CONFIG_S_DRIVER_OK in the status byte before starting the device service threads. But this corresponds to the successful finish of device initialization, which might (like virtio_blk's partition scanning) use the device. So we had a hack, if they used the device before we expected we started the threads anyway. Now we hook into the finalize_features hook in the Guest: at that point we tell the Launcher that it can rely on the features we have acked. On the Launcher side, we look at the status at that point, and start servicing the device. Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
-
由 Rusty Russell 提交于
Now we no longer use vmcall, we don't need to rewrite it in the Guest. Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
-
由 Rusty Russell 提交于
The Host used to create some page tables for the Guest to use at the top of Guest memory; it would then tell the Guest where this was. In particular, it created linear mappings for 0 and 0xC0000000 addresses because lguest used to switch to its real page tables quite late in boot. However, since d50d8fe1 Linux initialized boot page tables in head_32.S even before the "are we lguest?" boot jump. So, now we can simplify things: the Host pagetable code assumes 1:1 linear mapping until it first calls the LHCALL_NEW_PGTABLE hypercall, which we now do before we reach C code. This also means that the Host doesn't need to know anything about the Guest's PAGE_OFFSET. (Non-Linux guests might not even have such a thing). Signed-off-by: NRusty Russell <rusty@rustcorp.com.au>
-
由 Amit Kumar Salecha 提交于
o Minimum fw version supported for P3 chip is 4.0.505 o File Fw > 4.0.554 is not supported if flash fw < 4.0.554. o In mn firmware case, file fw older than flash fw is allowed. o Change variable names for readability o Update driver version 4.0.76 Signed-off-by: NAmit Kumar Salecha <amit.salecha@qlogic.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Sathya Perla 提交于
Currently be3-native mode is requested only in probe(). It must be requested, each time the card is reset either after an EEH error or after sleep/hibernation. Also, the be_cmd_check_native_mode() is better named be_cmd_req_native_mode() Signed-off-by: NSathya Perla <sathya.perla@emulex.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Krishna Kumar 提交于
Fix a panic in virtnet_remove. unregister_netdev has already freed up the netdev (and virtnet_info) due to dev->destructor being set, while virtnet_info is still required. Remove virtnet_free altogether, and move the freeing of the per-cpu statistics from virtnet_free to virtnet_remove. Tested patch below. Signed-off-by: NKrishna Kumar <krkumar2@in.ibm.com> Acked-by: NMichael S. Tsirkin <mst@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Eric Dumazet 提交于
IPv6 fragment identification generation is way beyond what we use for IPv4 : It uses a single generator. Its not scalable and allows DOS attacks. Now inetpeer is IPv6 aware, we can use it to provide a more secure and scalable frag ident generator (per destination, instead of system wide) This patch : 1) defines a new secure_ipv6_id() helper 2) extends inet_getid() to provide 32bit results 3) extends ipv6_select_ident() with a new dest parameter Reported-by: NFernando Gont <fernando@gont.com.ar> Signed-off-by: NEric Dumazet <eric.dumazet@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
The function can_get_bittiming is not used anywhere else, so it should be static. Signed-off-by: NThadeu Lima de Souza Cascardo <cascardo@holoscopio.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Shreyas Bhatewara 提交于
NETIF_F_HIGHDMA is being disabled even when dma64 is true. This patch fixes it. CC: Michal Miroslaw <mirq-linux@rere.qmqm.pl> Signed-off-by: NShreyas N Bhatewara <sbhatewara@vmware.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Shreyas Bhatewara 提交于
Parent device for netdev should be set before netdev_info() can be called otherwise there is a NULL pointer dereference and probe() fails. Signed-off-by: NShreyas N Bhatewara <sbhatewara@vmware.com> Signed-off-by: Scott J. Goldman <scottjg@vmware.com>-- Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Marek Vasut 提交于
This device can be found in Acer Iconia TAB W500 tablet dock. Signed-off-by: NMarek Vasut <marek.vasut@gmail.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Giuseppe CAVALLARO 提交于
Prior to this change, most PHY configuration parameters were passed into the STMMAC device as a separate PHY device. As well as being unusual, this made it difficult to make changes to the MAC/PHY relationship. This patch moves all the PHY parameters into the MAC configuration structure, mainly as a separate structure. This allows us to completely ignore the MDIO bus attached to a stmmac if desired, and not create the PHY bus. It also allows the stmmac driver to use a different PHY from the one it is connected to, for example a fixed PHY or bit banging PHY. Also derive the stmmac/PHY connection type (MII/RMII etc) from the mode can be passed into <platf>_configure_ethernet. STLinux kernel at git://git.stlinux.com/stm/linux-sh4-2.6.32.y.git provides several examples how to use this new infrastructure (that actually is easier to maintain and clearer). Signed-off-by: NStuart Menefy <stuart.menefy@st.com> Signed-off-by: NGiuseppe Cavallaro <peppe.cavallaro@st.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Giuseppe CAVALLARO 提交于
The patch removes the following serie of warnings when the driver is compiled as built-in: drivers/net/stmmac/stmmac_main.c: In function stmmac_cmdline_opt: drivers/net/stmmac/stmmac_main.c:1855:12: warning: ignoring return value of kstrtoul, declared with attribute warn_unused_result [snip] Signed-off-by: NGiuseppe Cavallaro <peppe.cavallaro@st.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Giuseppe CAVALLARO 提交于
Signed-off-by: NGiuseppe Cavallaro <peppe.cavallaro@st.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Guo-Fu Tseng 提交于
This patch add the missing dma_unmap(). Which solved the critical issue of system freeze on heavy load. Michal Miroslaw's rejected patch: [PATCH v2 10/46] net: jme: convert to generic DMA API Pointed out the issue also, thank you Michal. But the fix was incorrect. It would unmap needed address when low memory. Got lots of feedback from End user and Gentoo Bugzilla. https://bugs.gentoo.org/show_bug.cgi?id=373109 Thank you all. :) Cc: stable@kernel.org Signed-off-by: NGuo-Fu Tseng <cooldavid@cooldavid.org> Acked-by: NChris Wright <chrisw@sous-sol.org> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiri Pirko 提交于
ndo_vlan_rx_register is no longer in use in any driver so remove it. Signed-off-by: NJiri Pirko <jpirko@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiri Pirko 提交于
Now when all devices are cleaned up, bond can be cleaned up as well - remove bond->vlgrp - remove bond_vlan_rx_register - substitute necessary occurences of vlan_group_get_device Signed-off-by: NJiri Pirko <jpirko@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiri Pirko 提交于
Signed-off-by: NJiri Pirko <jpirko@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiri Pirko 提交于
Signed-off-by: NJiri Pirko <jpirko@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiri Pirko 提交于
- kill priv->vlgrp and stmmac_vlan_rx_register (used for nothing :)) Signed-off-by: NJiri Pirko <jpirko@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiri Pirko 提交于
- unify vlan and nonvlan rx path - kill card->vlangrp and qeth_l3_vlan_rx_register Signed-off-by: NJiri Pirko <jpirko@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiri Pirko 提交于
- unify vlan and nonvlan rx path - kill vdev->vlgrp and vxge_vlan_rx_register Signed-off-by: NJiri Pirko <jpirko@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiri Pirko 提交于
- unify vlan and nonvlan rx path - kill adapter->vlgrp and igb_vlan_rx_register - allow to turn on/off rx/tx vlan accel via ethtool (set_features) Signed-off-by: NJiri Pirko <jpirko@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiri Pirko 提交于
- unify vlan and nonvlan rx path - kill np->vlangrp and nv_vlan_rx_register - allow to turn on/off rx vlan accel via ethtool (set_features) Signed-off-by: NJiri Pirko <jpirko@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-
由 Jiri Pirko 提交于
- unify vlan and nonvlan rx path - kill adapter->vlgrp and e1000_vlan_rx_register - allow to turn on/off rx/tx vlan accel via ethtool (set_features) Signed-off-by: NJiri Pirko <jpirko@redhat.com> Signed-off-by: NDavid S. Miller <davem@davemloft.net>
-