- 03 Apr, 2010 21 commits
-
-
James Chapman authored
Reader/write locks are discouraged because they are slower than spin locks. So this patch converts the rwlocks used in the per_net structs to rcu. Signed-off-by:
James Chapman <jchapman@katalix.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
James Chapman authored
In L2TPv3, we need to create/delete/modify/query L2TP tunnel and session contexts. The number of parameters is significant. So let's use netlink. Userspace uses this API to control L2TP tunnel/session contexts in the kernel. The previous pppol2tp driver was managed using [gs]etsockopt(). This API is retained for backwards compatibility. Unlike L2TPv2 which carries only PPP frames, L2TPv3 can carry raw ethernet frames or other frame types and these do not always have an associated socket family. Therefore, we need a way to use L2TP sessions that doesn't require a socket type for each supported frame type. Hence netlink is used. Signed-off-by:
Randy Dunlap <randy.dunlap@oracle.com> Signed-off-by:
-
James Chapman authored
This lets kernel modules which use genl netlink APIs serialize netlink processing. Signed-off-by:
-
James Chapman authored
This patch adds a new L2TPIP socket family and modifies the core to handle the case where there is no UDP header in the L2TP packet. L2TP/IP uses IP protocol 115. Since L2TP/UDP and L2TP/IP packets differ in layout, the datapath packet handling code needs changes too. Userspace uses an L2TPIP socket instead of a UDP socket when IP encapsulation is required. We can't use raw sockets for this because the semantics of raw sockets don't lend themselves to the socket-per-tunnel model - we need to Signed-off-by: -
James Chapman authored
This patch makes changes to the L2TP PPP code for L2TPv3. The existing code has some assumptions about the L2TP header which are broken by L2TPv3. Also the sockaddr_pppol2tp structure of the original code is too small to support the increased size of the L2TPv3 tunnel and session id, so a new sockaddr_pppol2tpv3 structure is needed. In the socket calls, the size of this structure is used to tell if the operation is for L2TPv2 or L2TPv3. Signed-off-by:
-
James Chapman authored
The L2TPv3 protocol changes the layout of the L2TP packet header. Tunnel and session ids change from 16-bit to 32-bit values, data sequence numbers change from 16-bit to 24-bit values and PPP-specific fields are moved into protocol-specific subheaders. Although this patch introduces L2TPv3 protocol support, there are no userspace interfaces to create L2TPv3 sessions yet. Signed-off-by:
-
James Chapman authored
When dumping L2TP PPP sessions using /proc/net/pppol2tp, get the assigned PPP device name from PPP using ppp_dev_name(). Signed-off-by:
-
James Chapman authored
ppp_dev_name() gives PPP users visibility of a ppp channel's device name. This can be used by L2TP drivers to dump the assigned PPP interface name. Signed-off-by:
-
James Chapman authored
This patch splits the pppol2tp driver into separate L2TP and PPP parts to prepare for L2TPv3 support. In L2TPv3, protocols other than PPP can be carried, so this split creates a common L2TP core that will handle the common L2TP bits which protocol support modules such as PPP will use. Note that the existing pppol2tp module is split into l2tp_core and l2tp_ppp by this change. There are no feature changes here. Internally, however, there are significant changes, mostly to handle the separation of PPP-specific data from the L2TP session and to provide hooks in the core for modules like PPP to access. Signed-off-by:
-
James Chapman authored
This patch moves the existing pppol2tp driver from drivers/net into a new net/l2tp directory, which is where the upcoming L2TPv3 code will live. The existing CONFIG_PPPOL2TP config option is left in its current place to avoid "make oldconfig" issues when an existing pppol2tp user takes this change. (This is the same approach used for the pppoatm driver, which moved to net/atm.) There are no code changes. The existing drivers/net/pppol2tp.c is simply moved to net/l2tp. Signed-off-by:
-
Jiri Pirko authored
Converts the list and the core manipulating with it to be the same as uc_list. +uses two functions for adding/removing mc address (normal and "global" variant) instead of a function parameter. +removes dev_mcast.c completely. +exposes netdev_hw_addr_list_* macros along with __hw_addr_* functions for manipulation with lists on a sandbox (used in bonding and 80211 drivers) Signed-off-by:
Jiri Pirko <jpirko@redhat.com> Signed-off-by:
-
Jiri Pirko authored
+little renaming of unicast functions to be smooth with multicast ones Signed-off-by:
-
Amit Kumar Salecha authored
Signed-off-by:
Amit Kumar Salecha <amit.salecha@qlogic.com> Signed-off-by:
-
Amit Kumar Salecha authored
cpu_to_le32 was missing and used improperly. Signed-off-by:
-
Amit Kumar Salecha authored
Interface should be visible even if resource allocation fails. netif_device_attach should be called for every netif_device_detach. Signed-off-by:
-
Amit Kumar Salecha authored
Add debug print in driver, can be tuned by ethtool msg level callback. Signed-off-by:
-
Sucheta Chakraborty authored
o USE/Read IDC defined timeout value from ROM. o While resetting chip, don't wait for other pci-func to respond, more than reset_ack_timeo seconds, Signed-off-by:
Sucheta Chakraborty <sucheta.chakraborty@qlogic.com> Signed-off-by:
-
Dhananjay Phadke authored
Fix incorrect offset calculation and remove unnecessary remap of the region in bar 0 to access onchip memory. This was leading to read incorrect values by debug tools. Signed-off-by:
Dhananjay Phadke <dhananjay.phadke@qlogic.com> Signed-off-by:
-
Dhananjay Phadke authored
All QLogic converged NICs have 128-bit 128MB on card memory. Fix the limit check from 64MB to 128MB and remove unnecessary 64-bit read/write checks. Signed-off-by:
-
Dhananjay Phadke authored
Check the access by tools for hardware queue engine and handle it separately than other block registers, otherwise incorrect data is returned. Signed-off-by:
-
Amit Kumar Salecha authored
Rarely: Fw file size can be unaligned to 8. Signed-off-by:
-
- 02 Apr, 2010 19 commits
-
-
David Woodhouse authored
Signed-off-by:
David Woodhouse <David.Woodhouse@intel.com> Signed-off-by:
-
David Woodhouse authored
We don't use the normal hotplug mechanism because it doesn't work. It will load the module some time after the device appears, but that's not good enough for us -- we need the driver loaded _immediately_ because otherwise the NIC driver may just abort and then the phy 'device' goes away. [bwh: s/phy/mdio/ in module alias, kerneldoc for struct mdio_device_id] Signed-off-by:
Ben Hutchings <ben@decadent.org.uk> Acked-by:
Andy Fleming <afleming@freescale.com> Signed-off-by:
-
Eric Dumazet authored
Followup to commit 5acbbd42 (net: change illegal_highdma to use dma_mask) If dev->dev.parent is NULL, we should not try to dereference it. Dont force inline illegal_highdma() as its pretty big now. Signed-off-by:
Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by:
-
Sarveshwar Bandi authored
- Patch adds support to enable PCI SRIOV in the driver and changes to handle initialization of PCI virtual functions. - Function handler to change mac addresses for VF from its corresponding PF. Signed-off-by:
Sarveshwar Bandi <sarveshwarb@serverengines.com> Signed-off-by:
-
Alexander Kurz authored
Signed-off-by:
Alexander Kurz <linux@kbdbabel.org> Signed-off-by:
-
Bryan Wu authored
BugLink: http://bugs.launchpad.net/bugs/457878 v2: - remove duplicated phy_speed caculation - fix the phy_speed caculation according to the DataSheet v1: - removed old MII phy control code - add phylib supporting - add ethtool interface to make user space NetworkManager works Tested on Freescale i.MX51 Babbage board. This patch is based on a patch from Frederic Rodo <fred.rodo@gmail.com> Cc: Frederic Rodo <fred.rodo@gmail.com> Signed-off-by:
Bryan Wu <bryan.wu@canonical.com> Acked-by:
Amit Kucheria <amit.kucheria@canonical.com> Acked-by:
Sascha Hauer <s.hauer@pengutronix.de> Signed-off-by:
-
FUJITA Tomonori authored
Robert Hancock pointed out two problems about NETIF_F_HIGHDMA: -Many drivers only set the flag when they detect they can use 64-bit DMA, since otherwise they could receive DMA addresses that they can't handle (which on platforms without IOMMU/SWIOTLB support is fatal). This means that if 64-bit support isn't available, even buffers located below 4GB will get copied unnecessarily. -Some drivers set the flag even though they can't actually handle 64-bit DMA, which would mean that on platforms without IOMMU/SWIOTLB they would get a DMA mapping error if the memory they received happened to be located above 4GB. http://lkml.org/lkml/2010/3/3/530 We can use the dma_mask if we need bouncing or not here. Then we can safely fix drivers that misuse NETIF_F_HIGHDMA. Signed-off-by:
FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp> Signed-off-by:
-
Dmitry Kravkov authored
Adding GRO support on top of the HW LRO (TPA) support – there is no measurable performance drawback of adding GRO on top of it, and it allows better performance when LRO (TPA) is turned off for virtualization or bridging. Signed-off-by:
Dmitry Kravkov <dmitry@broadcom.com> Signed-off-by:
Eilon Greenstein <eilong@broadcom.com> Signed-off-by:
-
Timo Teräs authored
Group all per-cpu data to one structure instead of having many globals. Also prepare the internals so that we can have multiple instances of the flow cache if needed. Only the kmem_cache is left as a global as all flow caches share the same element size, and benefit from using a common cache. Signed-off-by:
Timo Teras <timo.teras@iki.fi> Acked-by:
Herbert Xu <herbert@gondor.apana.org.au> Signed-off-by:
-
Timo Teräs authored
All of the code considers ->dead as a hint that the cached policy needs to get refreshed. The read side can just drop the read lock without any side effects. The write side needs to make sure that it's written only exactly once. Only possible race is at xfrm_policy_kill(). This is fixed by checking result of __xfrm_policy_unlink() when needed. It will always succeed if the policy object is looked up from the hash list (so some checks are removed), but it needs to be checked if we are trying to unlink policy via a reference (appropriate checks added). Since policy->walk.dead is written exactly once, it no longer needs to be protected with a write lock. Signed-off-by:
-
Timo Teräs authored
Add missing check for policy direction verification. This is especially important since without this xfrm_user may end up deleting per-socket policy which is not allowed. Signed-off-by:
-
Herbert Xu authored
The xfrm state genid only needs to be matched against the copy saved in xfrm_dst. So we don't need a global genid at all. In fact, we don't even need to initialise it. Based on observation by Timo Teräs. Signed-off-by:
-
YOSHIFUJI Hideaki / 吉藤英明 authored
The head element of rt6_info{} is dst_entry{}, and IPv6 specific elements follow. Because elements at the end of dst_entry{} are frequently updated, it is not good to put frequently-used static elements, such as rt6i_idev, rt6i_dst or rt6i_flags in the same cache line. On the other hand, fib6_table, rt6i_node or rt6i_gateway are rarely used, so it is okay to stay in the same cache line. Let's rearrange rt6_info{}. Signed-off-by:YOSHIFUJI Hideaki <yoshfuji@linux-ipv6.org> Signed-off-by:
-
Changli Gao authored
keep the old behavior on SMP without rps RPS introduces a lock operation to per cpu variable input_pkt_queue on SMP whenever rps is enabled or not. On SMP without RPS, this lock isn't needed at all. Signed-off-by:
Changli Gao <xiaosuo@gmail.com> ---- net/core/dev.c | 42 ++++++++++++++++++++++++++++-------------- 1 file changed, 28 insertions(+), 14 deletions(-) Signed-off-by:
-
Eric Dumazet authored
In case a reset is performed, rtl8169_rx_interrupt() is called from process context instead of softirq context. Special care must be taken to call appropriate network core services (netif_rx() instead of netif_receive_skb()). VLAN handling also corrected. Reported-by:
Sergey Senozhatsky <sergey.senozhatsky@gmail.com> Tested-by:
Oleg Nesterov <oleg@redhat.com> Signed-off-by:
-
Eric Dumazet authored
One of my test machine got a deadlock during "tc" sessions, adding/deleting classes & filters, using traffic estimators. After some analysis, I believe we have a potential use after free case in est_timer() : spin_lock(e->stats_lock); << HERE >> read_lock(&est_lock); if (e->bstats == NULL) << TEST >> goto skip; Test is done a bit late, because after estimator is killed, and before rcu grace period elapsed, we might already have freed/reuse memory where e->stats_locks points to (some qdisc->q.lock) A possible fix is to respect a rcu grace period at Qdisc dismantle time. On 64bit, sizeof(struct Qdisc) is exactly 192 bytes. Adding 16 bytes to it (for struct rcu_head) is a problem because it might change performance, given QDISC_ALIGNTO is 32 bytes. This is why I also change QDISC_ALIGNTO to 64 bytes, to satisfy most current alignment requirements. Signed-off-by:
-
Hagen Paul Pfeifer authored
The check if error signaling is wanted (inet->recverr != 0) is done by the caller: raw.c:raw_err() and udp.c:__udp4_lib_err(), so there is no need to check this condition again. Signed-off-by:
Hagen Paul Pfeifer <hagen@jauu.net> Signed-off-by:
-
FUJITA Tomonori authored
The DMA API is preferred. Signed-off-by:
-
FUJITA Tomonori authored
acenic wrongly assumes that zero is an invalid dma address (calls dma_unmap_page for only non zero dma addresses). Zero is a valid dma address on some architectures. The dma length can be used here. Signed-off-by:
-
