- 13 May, 2020 9 commits
-
-
Florian Fainelli authored
The GENET controller on the Raspberry Pi 4 (2711) is typically interfaced with an external Broadcom PHY via a RGMII electrical interface. To make sure that delays are properly configured at the PHY side, ensure that we the dedicated Broadcom PHY driver (CONFIG_BROADCOM_PHY) is enabled for this to happen. Fixes: 402482a6 ("net: bcmgenet: Clear ID_MODE_DIS in EXT_RGMII_OOB_CTRL when not needed") Reported-by:
Marek Szyprowski <m.szyprowski@samsung.com> Signed-off-by:
Florian Fainelli <f.fainelli@gmail.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
David S. Miller authored
Tuong Lien says: ==================== tipc: add some patches This series adds patches to fix some issues in TIPC streaming & service subscription. ==================== Signed-off-by: -
Tuong Lien authored
When a service subscription is expired or canceled by user, it needs to be deleted from the subscription list, so that new subscriptions can be registered (max = 65535 per net). However, there are two issues in code that can cause such an unused subscription to persist: 1) The 'tipc_conn_delete_sub()' has a loop on the subscription list but it makes a break shortly when the 1st subscription differs from the one specified, so the subscription will not be deleted. 2) In case a subscription is canceled, the code to remove the 'TIPC_SUB_CANCEL' flag from the subscription filter does not work if it is a local subscription (i.e. the little endian isn't involved). So, it will be no matches when looking for the subscription to delete later. The subscription(s) will be removed eventually when the user terminates its topology connection but that could be a long time later. Meanwhile, the number of available subscriptions may be exhausted. This commit fixes the two issues above, so as needed a subscription can be deleted correctly. Acked-by:
Ying Xue <ying.xue@windriver.com> Acked-by:
Jon Maloy <jmaloy@redhat.com> Signed-off-by:
Tuong Lien <tuong.t.lien@dektech.com.au> Signed-off-by:
-
Tuong Lien authored
Upon receipt of a service subscription request from user via a topology connection, one 'sub' object will be allocated in kernel, so it will be able to send an event of the service if any to the user correspondingly then. Also, in case of any failure, the connection will be shutdown and all the pertaining 'sub' objects will be freed. However, there is a race condition as follows resulting in memory leak: receive-work connection send-work | | | sub-1 |<------//-------| | sub-2 |<------//-------| | | |<---------------| evt for sub-x sub-3 |<------//-------| | : : : : : : | /--------| | | | * peer closed | | | | | | | |<-------X-------| evt for sub-y | | |<===============| sub-n |<------/ X shutdown | -> orphan | | That is, the 'receive-work' may get the last subscription request while the 'send-work' is shutting down the connection due to peer close. We had a 'lock' on the connection, so the two actions cannot be carried out simultaneously. If the last subscription is allocated e.g. 'sub-n', before the 'send-work' closes the connection, there will be no issue at all, the 'sub' objects will be freed. In contrast the last subscription will become orphan since the connection was closed, and we released all references. This commit fixes the issue by simply adding one test if the connection remains in 'connected' state right after we obtain the connection lock, then a subscription object can be created as usual, otherwise we ignore it. Acked-by:Thang Ngo <thang.h.ngo@dektech.com.au> Signed-off-by:
-
Tuong Lien authored
Currently when a connection is in Nagle mode, we set the 'ack_required' bit in the last sending buffer and wait for the corresponding ACK prior to pushing more data. However, on the receiving side, the ACK is issued only when application really reads the whole data. Even if part of the last buffer is received, we will not do the ACK as required. This might cause an unnecessary delay since the receiver does not always fetch the message as fast as the sender, resulting in a large latency in the user message sending, which is: [one RTT + the receiver processing time]. The commit makes Nagle ACK as soon as possible i.e. when a message with the 'ack_required' arrives in the receiving side's stack even before it is processed or put in the socket receive queue... This way, we can limit the streaming latency to one RTT as committed in Nagle mode. Acked-by:
-
David S. Miller authored
Ursula Braun says: ==================== s390/net: updates 2020-05-13 please apply the fix from Wei Yongjun to netdev's net tree and add Karsten Graul as co-maintainer for drivers/s390/net. ==================== Signed-off-by: -
Ursula Braun authored
Add Karsten as additional maintainer for drivers/s390/net . One of his focal points is the ism driver. Cc: Julian Wiedmann <jwi@linux.ibm.com> Acked-by:
Julian Wiedmann <jwi@linux.ibm.com> Signed-off-by:
Ursula Braun <ubraun@linux.ibm.com> Signed-off-by:
-
Wei Yongjun authored
Fix to return negative error code -ENOMEM from the smcd_alloc_dev() error handling case instead of 0, as done elsewhere in this function. Fixes: 684b89bc ("s390/ism: add device driver for internal shared memory") Reported-by:
Hulk Robot <hulkci@huawei.com> Signed-off-by:
Wei Yongjun <weiyongjun1@huawei.com> Signed-off-by:
-
Paolo Abeni authored
The cipso and calipso code can set the MLS_CAT attribute on successful parsing, even if the corresponding catmap has not been allocated, as per current configuration and external input. Later, selinux code tries to access the catmap if the MLS_CAT flag is present via netlbl_catmap_getlong(). That may cause null ptr dereference while processing incoming network traffic. Address the issue setting the MLS_CAT flag only if the catmap is really allocated. Additionally let netlbl_catmap_getlong() cope with NULL catmap. Reported-by:
Matthew Sheets <matthew.sheets@gd-ms.com> Fixes: 4b8feff2 ("netlabel: fix the horribly broken catmap functions") Fixes: ceba1832 ("calipso: Set the calipso socket label to match the secattr.") Signed-off-by:
Paolo Abeni <pabeni@redhat.com> Acked-by:
Paul Moore <paul@paul-moore.com> Signed-off-by:
-
- 12 May, 2020 7 commits
-
-
Eric Dumazet authored
We autotune rcvbuf whenever SO_RCVLOWAT is set to account for 100% overhead in tcp_set_rcvlowat() This works well when skb->len/skb->truesize ratio is bigger than 0.5 But if we receive packets with small MSS, we can end up in a situation where not enough bytes are available in the receive queue to satisfy RCVLOWAT setting. As our sk_rcvbuf limit is hit, we send zero windows in ACK packets, preventing remote peer from sending more data. Even autotuning does not help, because it only triggers at the time user process drains the queue. If no EPOLLIN is generated, this can not happen. Note poll() has a similar issue, after commit c7004482 ("tcp: Respect SO_RCVLOWAT in tcp_poll().") Fixes: 03f45c88 ("tcp: avoid extra wakeups for SO_RCVLOWAT users") Signed-off-by:
Eric Dumazet <edumazet@google.com> Acked-by:
Soheil Hassas Yeganeh <soheil@google.com> Signed-off-by:
-
Clay McClure authored
My recent commit b6d49cab ("net: Make PTP-specific drivers depend on PTP_1588_CLOCK") exposes a missing dependency in defconfigs that select TI_CPTS without selecting PTP_1588_CLOCK, leading to linker errors of the form: drivers/net/ethernet/ti/cpsw.o: in function `cpsw_ndo_stop': cpsw.c:(.text+0x680): undefined reference to `cpts_unregister' ... That's because TI_CPTS_MOD (which is the symbol gating the _compilation_ of cpts.c) now depends on PTP_1588_CLOCK, and so is not enabled in these configurations, but TI_CPTS (which is the symbol gating _calls_ to the cpts functions) _is_ enabled. So we end up compiling calls to functions that don't exist, resulting in the linker errors. This patch fixes build errors and restores previous behavior by: - ensure PTP_1588_CLOCK=y in TI specific configs and CPTS will be built - remove TI_CPTS_MOD and, instead, add dependencies from CPTS in TI_CPSW/TI_KEYSTONE_NETCP/TI_CPSW_SWITCHDEV as below: config TI_CPSW_SWITCHDEV ... depends on TI_CPTS || !TI_CPTS which will ensure proper dependencies PTP_1588_CLOCK -> TI_CPTS -> TI_CPSW/TI_KEYSTONE_NETCP/TI_CPSW_SWITCHDEV and build type selection. Note. For NFS boot + CPTS all of above configs have to be built-in. Cc: Arnd Bergmann <arnd@arndb.de> Cc: Dan Murphy <dmurphy@ti.com> Cc: Tony Lindgren <tony@atomide.com> Fixes: b6d49cab ("net: Make PTP-specific drivers depend on PTP_1588_CLOCK") Reported-by:
kbuild test robot <lkp@intel.com> Signed-off-by:
Clay McClure <clay@daemons.net> [grygorii.strashko@ti.com: rewording, add deps cpsw/netcp from cpts, drop IS_REACHABLE] Signed-off-by:
Grygorii Strashko <grygorii.strashko@ti.com> Reviewed-by:
Arnd Bergmann <arnd@arndb.de> Tested-by:
Tony Lindgren <tony@atomide.com> Signed-off-by:
-
David S. Miller authored
Shannon Nelson says: ==================== ionic fixes These are a couple more fixes after more fw-upgrade testing. ==================== Signed-off-by: -
Shannon Nelson authored
Since the fw has been re-inited, we need to refresh the port information dma address so we can see fresh port information. Let's call ionic_port_init again, and tweak it to allow for a call to simply refresh the existing dma address. Fixes: c672412f ("ionic: remove lifs on fw reset") Signed-off-by:
Shannon Nelson <snelson@pensando.io> Signed-off-by:
-
Shannon Nelson authored
When running in a bond setup, or some other potential configurations, the netdev mac may have been changed from the default device mac. Since the userland doesn't know about the changes going on under the covers in a fw-upgrade it doesn't know the re-push the mac filter. The driver needs to leave the netdev mac filter alone when rebuilding after the fw-upgrade. Fixes: c672412f ("ionic: remove lifs on fw reset") Signed-off-by:
-
Jacob Keller authored
The do_aux_work callback had documentation in the structure comment which referred to it as "do_work". Signed-off-by:
Jacob Keller <jacob.e.keller@intel.com> Cc: Richard Cochran <richardcochran@gmail.com> Acked-by:
Richard Cochran <richardcochran@gmail.com> Signed-off-by:
-
Christoph Paasch authored
When the other MPTCP-peer uses 32-bit data-sequence numbers, we rely on map_seq to indicate how to expand to a 64-bit data-sequence number in expand_seq() when receiving data. For new subflows, this field is not initialized, thus results in an "invalid" mapping being discarded. Fix this by initializing map_seq upon subflow establishment time. Fixes: f296234c ("mptcp: Add handling of incoming MP_JOIN requests") Signed-off-by:
Christoph Paasch <cpaasch@apple.com> Reviewed-by:
-
- 11 May, 2020 4 commits
-
-
David S. Miller authored
Alex Elder says: ==================== net: ipa: fix cleanup after modem crash The first patch in this series fixes a bug where the size of a data transfer request was never set, meaning it was 0. The consequence of this was that such a transfer request would never complete if attempted, and led to a hung task timeout. This data transfer is required for cleaning up IPA hardware state when recovering from a modem crash. The code to implement this cleanup is already present, but its use was commented out because it hit the bug described above. So the second patch in this series enables the use of that "tag process" cleanup code. ==================== Signed-off-by: -
Alex Elder authored
One part of recovering from a modem crash is performing a "tag sequence" of several IPA immediate commands, to clear the hardware pipeline. The sequence ends with a data transfer request on the command endpoint (which is not otherwise done). Unfortunately, attempting to do the data transfer led to a hang, so that request plus two other commands were commented out. The previous commit fixes the bug that was causing that hang. And with that bug fixed we can properly issue the tag sequence when the modem crashes, to return the hardware to a known state. Signed-off-by:
Alex Elder <elder@linaro.org> Signed-off-by:
-
Alex Elder authored
When a command gets added to a transaction for the AP->command channel we set the DMA address of its scatterlist entry, but not its DMA length. Fix this bug. Signed-off-by:
-
Luo bin authored
if some function in ndo_stop interface returns failure because of hardware fault, must go on excuting rest steps rather than return failure directly, otherwise will cause memory leak.And bump the timeout for SET_FUNC_STATE to ensure that cmd won't return failure when hw is busy. Otherwise hw may stomp host memory if we free memory regardless of the return value of SET_FUNC_STATE. Fixes: 51ba902a ("net-next/hinic: Initialize hw interface") Signed-off-by:
Luo bin <luobin9@huawei.com> Signed-off-by:
Jakub Kicinski <kuba@kernel.org>
-
- 10 May, 2020 2 commits
-
-
Florian Fainelli authored
There is a soft dependency against dsa_loop_bdinfo.ko which sets up the MDIO device registration, since there are no symbols referenced by dsa_loop.ko, there is no automatic loading of dsa_loop_bdinfo.ko which is needed. Fixes: 98cd1552 ("net: dsa: Mock-up driver") Signed-off-by:
-
Zefan Li authored
If systemd is configured to use hybrid mode which enables the use of both cgroup v1 and v2, systemd will create new cgroup on both the default root (v2) and netprio_cgroup hierarchy (v1) for a new session and attach task to the two cgroups. If the task does some network thing then the v2 cgroup can never be freed after the session exited. One of our machines ran into OOM due to this memory leak. In the scenario described above when sk_alloc() is called cgroup_sk_alloc() thought it's in v2 mode, so it stores the cgroup pointer in sk->sk_cgrp_data and increments the cgroup refcnt, but then sock_update_netprioidx() thought it's in v1 mode, so it stores netprioidx value in sk->sk_cgrp_data, so the cgroup refcnt will never be freed. Currently we do the mode switch when someone writes to the ifpriomap cgroup control file. The easiest fix is to also do the switch when a task is attached to a new cgroup. Fixes: bd1060a1 ("sock, cgroup: add sock->sk_cgroup") Reported-by:
Yang Yingliang <yangyingliang@huawei.com> Tested-by:
Zefan Li <lizefan@huawei.com> Acked-by:
Tejun Heo <tj@kernel.org> Signed-off-by:
-
- 09 May, 2020 8 commits
-
-
Arnd Bergmann authored
I ran into a randconfig build failure with CONFIG_FIXED_PHY=m and CONFIG_GIANFAR=y: x86_64-linux-ld: drivers/net/ethernet/freescale/gianfar.o:(.rodata+0x418): undefined reference to `fixed_phy_change_carrier' It seems the same thing can happen with dpaa and ucc_geth, so change all three to do an explicit 'select FIXED_PHY'. The fixed-phy driver actually has an alternative stub function that theoretically allows building network drivers when fixed-phy is disabled, but I don't see how that would help here, as the drivers presumably would not work then. Signed-off-by:
-
Paolo Abeni authored
In commit b406472b ("net: ipv4: avoid mixed n_redirects and rate_tokens usage") I missed the fact that a 0 'rate_tokens' will bypass the backoff algorithm. Since rate_tokens is cleared after a redirect silence, and never incremented on redirects, if the host keeps receiving packets requiring redirect it will reply ignoring the backoff. Additionally, the 'rate_last' field will be updated with the cadence of the ingress packet requiring redirect. If that rate is high enough, that will prevent the host from generating any other kind of ICMP messages The check for a zero 'rate_tokens' value was likely a shortcut to avoid the more complex backoff algorithm after a redirect silence period. Address the issue checking for 'n_redirects' instead, which is incremented on successful redirect, and does not interfere with other ICMP replies. Fixes: b406472b ("net: ipv4: avoid mixed n_redirects and rate_tokens usage") Reported-and-tested-by:
Colin Walters <walters@redhat.com> Signed-off-by:
-
Wei Yongjun authored
Fix to return negative error code -ENOMEM from the alloc failed error handling case instead of 0, as done elsewhere in this function. Fixes: 3184fb5b ("octeontx2-vf: Virtual function driver support") Reported-by:
-
Vincent Minet authored
If a UMH process created by fork_usermode_blob() fails to execute, a pair of struct file allocated by umh_pipe_setup() will leak. Under normal conditions, the caller (like bpfilter) needs to manage the lifetime of the UMH and its two pipes. But when fork_usermode_blob() fails, the caller doesn't really have a way to know what needs to be done. It seems better to do the cleanup ourselves in this case. Fixes: 449325b5 ("umh: introduce fork_usermode_blob() helper") Signed-off-by:
Vincent Minet <v.minet@criteo.com> Signed-off-by:
-
git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpfJakub Kicinski authored
Daniel Borkmann says: ==================== pull-request: bpf 2020-05-09 The following pull-request contains BPF updates for your *net* tree. We've added 4 non-merge commits during the last 9 day(s) which contain a total of 4 files changed, 11 insertions(+), 6 deletions(-). The main changes are: 1) Fix msg_pop_data() helper incorrectly setting an sge length in some cases as well as fixing bpf_tcp_ingress() wrongly accounting bytes in sg.size, from John Fastabend. 2) Fix to return an -EFAULT error when copy_to_user() of the value fails in map_lookup_and_delete_elem(), from Wei Yongjun. 3) Fix sk_psock refcnt leak in tcp_bpf_recvmsg(), from Xiyu Yang. ==================== Signed-off-by: -
Dan Carpenter authored
The "location" is controlled by the user via the ethtool_set_rxnfc() function. This update_cls_rule() function checks for array overflows but it doesn't check if the value is negative. I have changed the type to unsigned to prevent array underflows. Fixes: afb90dbb ("dpaa2-eth: Add ethtool support for flow classification") Signed-off-by:
Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by:
-
Wei Yongjun authored
Fix to return negative error code -ENOMEM from the kzalloc() error handling case instead of 0, as done elsewhere in this function. Fixes: 174ab544 ("nfp: abm: add cls_u32 offload for simple band classification") Reported-by:
-
Heiner Kallweit authored
r8169 never had native support for the RTL8401, however it reportedly worked with the fallback to RTL8101e [0]. Therefore let's add this as an explicit assignment. [0] https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=956868 Fixes: b4cc2dcc ("r8169: remove default chip versions") Reported-by:
Camaleón <noelamac@gmail.com> Signed-off-by:
Heiner Kallweit <hkallweit1@gmail.com> Signed-off-by:
-
- 08 May, 2020 6 commits
-
-
Kelly Littlepage authored
The stated intent of the original commit is to is to "return the timestamp corresponding to the highest sequence number data returned." The current implementation returns the timestamp for the last byte of the last fully read skb, which is not necessarily the last byte in the recv buffer. This patch converts behavior to the original definition, and to the behavior of the previous draft versions of commit 98aaa913 ("tcp: Extend SOF_TIMESTAMPING_RX_SOFTWARE to TCP recvmsg") which also match this behavior. Fixes: 98aaa913 ("tcp: Extend SOF_TIMESTAMPING_RX_SOFTWARE to TCP recvmsg") Co-developed-by:
Iris Liu <iris@onechronos.com> Signed-off-by:
Kelly Littlepage <kelly@onechronos.com> Signed-off-by:
Willem de Bruijn <willemb@google.com> Signed-off-by:
-
Cong Wang authored
syzbot managed to trigger a recursive NETDEV_FEAT_CHANGE event between bonding master and slave. I managed to find a reproducer for this: ip li set bond0 up ifenslave bond0 eth0 brctl addbr br0 ethtool -K eth0 lro off brctl addif br0 bond0 ip li set br0 up When a NETDEV_FEAT_CHANGE event is triggered on a bonding slave, it captures this and calls bond_compute_features() to fixup its master's and other slaves' features. However, when syncing with its lower devices by netdev_sync_lower_features() this event is triggered again on slaves when the LRO feature fails to change, so it goes back and forth recursively until the kernel stack is exhausted. Commit 17b85d29 intentionally lets __netdev_update_features() return -1 for such a failure case, so we have to just rely on the existing check inside netdev_sync_lower_features() and skip NETDEV_FEAT_CHANGE event only for this specific failure case. Fixes: fd867d51 ("net/core: generic support for disabling netdev features down stack") Reported-by: syzbot+e73ceacfd8560cc8a3ca@syzkaller.appspotmail.com Reported-by: syzbot+c2fb6f9ddcea95ba49b5@syzkaller.appspotmail.com Cc: Jarod Wilson <jarod@redhat.com> Cc: Nikolay Aleksandrov <nikolay@cumulusnetworks.com> Cc: Josh Poimboeuf <jpoimboe@redhat.com> Cc: Jann Horn <jannh@google.com> Reviewed-by:
Jay Vosburgh <jay.vosburgh@canonical.com> Signed-off-by:
Cong Wang <xiyou.wangcong@gmail.com> Acked-by:
Nikolay Aleksandrov <nikolay@cumulusnetworks.com> Signed-off-by:
-
Paolo Abeni authored
When a subflow is created via mptcp_subflow_create_socket(), a new 'struct socket' is allocated, with a new i_ino value. When inspecting TCP sockets via the procfs and or the diag interface, the above ones are not related to the process owning the MPTCP master socket, even if they are a logical part of it ('ss -p' shows an empty process field) Additionally, subflows created by the path manager get the uid/gid from the running workqueue. Subflows are part of the owning MPTCP master socket, let's adjust the vfs info to reflect this. After this patch, 'ss' correctly displays subflows as belonging to the msk socket creator. Fixes: 2303f994 ("mptcp: Associate MPTCP context with TCP socket") Signed-off-by: -
Chuhong Yuan authored
This driver calls kthread_run() in probe, but forgets to call kthread_stop() in probe failure and remove. Add the missed kthread_stop() to fix it. Signed-off-by:
Chuhong Yuan <hslester96@gmail.com> Signed-off-by:
-
Maciej Żenczykowski authored
This reverts commit 19bda36c: | ipv6: add mtu lock check in __ip6_rt_update_pmtu | | Prior to this patch, ipv6 didn't do mtu lock check in ip6_update_pmtu. | It leaded to that mtu lock doesn't really work when receiving the pkt | of ICMPV6_PKT_TOOBIG. | | This patch is to add mtu lock check in __ip6_rt_update_pmtu just as ipv4 | did in __ip_rt_update_pmtu. The above reasoning is incorrect. IPv6 *requires* icmp based pmtu to work. There's already a comment to this effect elsewhere in the kernel: $ git grep -p -B1 -A3 'RTAX_MTU lock' net/ipv6/route.c=4813= static int rt6_mtu_change_route(struct fib6_info *f6i, void *p_arg) ... /* In IPv6 pmtu discovery is not optional, so that RTAX_MTU lock cannot disable it. We still use this lock to block changes caused by addrconf/ndisc. */ This reverts to the pre-4.9 behaviour. Cc: Eric Dumazet <edumazet@google.com> Cc: Willem de Bruijn <willemb@google.com> Cc: Xin Long <lucien.xin@gmail.com> Cc: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by:
Maciej Żenczykowski <maze@google.com> Fixes: 19bda36c ("ipv6: add mtu lock check in __ip6_rt_update_pmtu") Signed-off-by:
-
Arnd Bergmann authored
clang points out that building without IPv6 would lead to returning an uninitialized variable if a packet with family!=AF_INET is passed into bareudp_udp_encap_recv(): drivers/net/bareudp.c:139:6: error: variable 'err' is used uninitialized whenever 'if' condition is false [-Werror,-Wsometimes-uninitialized] if (family == AF_INET) ^~~~~~~~~~~~~~~~~ drivers/net/bareudp.c:146:15: note: uninitialized use occurs here if (unlikely(err)) { ^~~ include/linux/compiler.h:78:42: note: expanded from macro 'unlikely' # define unlikely(x) __builtin_expect(!!(x), 0) ^ drivers/net/bareudp.c:139:2: note: remove the 'if' if its condition is always true if (family == AF_INET) ^~~~~~~~~~~~~~~~~~~~~~ This cannot happen in practice, so change the condition in a way that gcc sees the IPv4 case as unconditionally true here. For consistency, change all the similar constructs in this file the same way, using "if(IS_ENABLED())" instead of #if IS_ENABLED()". Fixes: 571912c6 ("net: UDP tunnel encapsulation module for tunnelling different protocols like MPLS, IP, NSH etc.") Signed-off-by:Nathan Chancellor <natechancellor@gmail.com> Signed-off-by:
-
- 07 May, 2020 4 commits
-
-
Geert Uytterhoeven authored
The HNS config symbol enables the framework support for the Hisilicon Network Subsystem. It is already selected by all of its users, so there is no reason to make it visible. Signed-off-by:
Geert Uytterhoeven <geert+renesas@glider.be> Signed-off-by:
-
Oliver Neukum authored
If you do not find the OUT endpoint, you should say so, rather than copy the error message for the IN endpoint. Presumably a copy and paste error. Signed-off-by:
Oliver Neukum <oneukum@suse.com> Signed-off-by:
-
Maciej Żenczykowski authored
it doesn't actually exist... Test: builds and 'git grep tcp_default_init_rwnd' comes up empty Signed-off-by:
-
Michael S. Tsirkin authored
When we fill up a receive VQ, try_fill_recv currently tries to count kicks using a 64 bit stats counter. Turns out, on a 32 bit kernel that uses a seqcount. sequence counts are "lock" constructs where you need to make sure that writers are serialized. In turn, this means that we mustn't run two try_fill_recv concurrently. Which of course we don't. We do run try_fill_recv sometimes from a softirq napi context, and sometimes from a fully preemptible context, but the later always runs with napi disabled. However, when it comes to the seqcount, lockdep is trying to enforce the rule that the same lock isn't accessed from preemptible and softirq context - it doesn't know about napi being enabled/disabled. This causes a false-positive warning: WARNING: inconsistent lock state ... inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage. As a work around, shut down the warning by switching to u64_stats_update_begin_irqsave - that works by disabling interrupts on 32 bit only, is a NOP on 64 bit. Reported-by:Thomas Gleixner <tglx@linutronix.de> Suggested-by:
Eric Dumazet <eric.dumazet@gmail.com> Signed-off-by:
Michael S. Tsirkin <mst@redhat.com> Signed-off-by:
-
