- 30 Jul, 2019 20 commits
-
-
Tristram Ha authored
Add DSA tag code for Microchip KSZ8795 switch. The switch is simpler and the tag is only 1 byte, instead of 2 as is the case with KSZ9477. Signed-off-by:
Tristram Ha <Tristram.Ha@microchip.com> Signed-off-by:
Marek Vasut <marex@denx.de> Cc: Andrew Lunn <andrew@lunn.ch> Cc: David S. Miller <davem@davemloft.net> Cc: Florian Fainelli <f.fainelli@gmail.com> Cc: Tristram Ha <Tristram.Ha@microchip.com> Cc: Vivien Didelot <vivien.didelot@gmail.com> Cc: Woojung Huh <woojung.huh@microchip.com> Signed-off-by:
David S. Miller <davem@davemloft.net>
-
Marek Vasut authored
Document Microchip KSZ87xx family switches. These include KSZ8765 - 5 port switch KSZ8794 - 4 port switch KSZ8795 - 5 port switch Signed-off-by:
-
David S. Miller authored
Stefano Garzarella says: ==================== vsock/virtio: optimizations to increase the throughput This series tries to increase the throughput of virtio-vsock with slight changes. While I was testing the v2 of this series I discovered an huge use of memory, so I added patch 1 to mitigate this issue. I put it in this series in order to better track the performance trends. v5: - rebased all patches on net-next - added Stefan's R-b and Michael's A-b v4: https://patchwork.kernel.org/cover/11047717 v3: https://patchwork.kernel.org/cover/10970145 v2: https://patchwork.kernel.org/cover/10938743 v1: https://patchwork.kernel.org/cover/10885431 Below are the benchmarks step by step. I used iperf3 [1] modified with VSOCK support. As Michael suggested in the v1, I booted host and guest with 'nosmap'. A brief description of patches: - Patches 1: limit the memory usage with an extra copy for small packets - Patches 2+3: reduce the number of credit update messages sent to the transmitter - Patches 4+5: allow the host to split packets on multiple buffers and use VIRTIO_VSOCK_MAX_PKT_BUF_SIZE as the max packet size allowed host -> guest [Gbps] pkt_size before opt p 1 p 2+3 p 4+5 32 0.032 0.030 0.048 0.051 64 0.061 0.059 0.108 0.117 128 0.122 0.112 0.227 0.234 256 0.244 0.241 0.418 0.415 512 0.459 0.466 0.847 0.865 1K 0.927 0.919 1.657 1.641 2K 1.884 1.813 3.262 3.269 4K 3.378 3.326 6.044 6.195 8K 5.637 5.676 10.141 11.287 16K 8.250 8.402 15.976 16.736 32K 13.327 13.204 19.013 20.515 64K 21.241 21.341 20.973 21.879 128K 21.851 22.354 21.816 23.203 256K 21.408 21.693 21.846 24.088 512K 21.600 21.899 21.921 24.106 guest -> host [Gbps] pkt_size before opt p 1 p 2+3 p 4+5 32 0.045 0.046 0.057 0.057 64 0.089 0.091 0.103 0.104 128 0.170 0.179 0.192 0.200 256 0.364 0.351 0.361 0.379 512 0.709 0.699 0.731 0.790 1K 1.399 1.407 1.395 1.427 2K 2.670 2.684 2.745 2.835 4K 5.171 5.199 5.305 5.451 8K 8.442 8.500 10.083 9.941 16K 12.305 12.259 13.519 15.385 32K 11.418 11.150 11.988 24.680 64K 10.778 10.659 11.589 35.273 128K 10.421 10.339 10.939 40.338 256K 10.300 9.719 10.508 36.562 512K 9.833 9.808 10.612 35.979 As Stefan suggested in the v1, I measured also the efficiency in this way: efficiency = Mbps / (%CPU_Host + %CPU_Guest) The '%CPU_Guest' is taken inside the VM. I know that it is not the best way, but it's provided for free from iperf3 and could be an indication. host -> guest efficiency [Mbps / (%CPU_Host + %CPU_Guest)] pkt_size before opt p 1 p 2+3 p 4+5 32 0.35 0.45 0.79 1.02 64 0.56 0.80 1.41 1.54 128 1.11 1.52 3.03 3.12 256 2.20 2.16 5.44 5.58 512 4.17 4.18 10.96 11.46 1K 8.30 8.26 20.99 20.89 2K 16.82 16.31 39.76 39.73 4K 30.89 30.79 74.07 75.73 8K 53.74 54.49 124.24 148.91 16K 80.68 83.63 200.21 232.79 32K 132.27 132.52 260.81 357.07 64K 229.82 230.40 300.19 444.18 128K 332.60 329.78 331.51 492.28 256K 331.06 337.22 339.59 511.59 512K 335.58 328.50 331.56 504.56 guest -> host efficiency [Mbps / (%CPU_Host + %CPU_Guest)] pkt_size before opt p 1 p 2+3 p 4+5 32 0.43 0.43 0.53 0.56 64 0.85 0.86 1.04 1.10 128 1.63 1.71 2.07 2.13 256 3.48 3.35 4.02 4.22 512 6.80 6.67 7.97 8.63 1K 13.32 13.31 15.72 15.94 2K 25.79 25.92 30.84 30.98 4K 50.37 50.48 58.79 59.69 8K 95.90 96.15 107.04 110.33 16K 145.80 145.43 143.97 174.70 32K 147.06 144.74 146.02 282.48 64K 145.25 143.99 141.62 406.40 128K 149.34 146.96 147.49 489.34 256K 156.35 149.81 152.21 536.37 512K 151.65 150.74 151.52 519.93 [1] https://github.com/stefano-garzarella/iperf/ ==================== Acked-by:
Michael S. Tsirkin <mst@redhat.com> Signed-off-by:
-
Stefano Garzarella authored
Since now we are able to split packets, we can avoid limiting their sizes to VIRTIO_VSOCK_DEFAULT_RX_BUF_SIZE. Instead, we can use VIRTIO_VSOCK_MAX_PKT_BUF_SIZE as the max packet size. Signed-off-by:
Stefano Garzarella <sgarzare@redhat.com> Reviewed-by:
Stefan Hajnoczi <stefanha@redhat.com> Acked-by:
-
Stefano Garzarella authored
If the packets to sent to the guest are bigger than the buffer available, we can split them, using multiple buffers and fixing the length in the packet header. This is safe since virtio-vsock supports only stream sockets. Signed-off-by:
-
Stefano Garzarella authored
fwd_cnt and last_fwd_cnt are protected by rx_lock, so we should use the same spinlock also if we are in the TX path. Move also buf_alloc under the same lock. Signed-off-by:
-
Stefano Garzarella authored
In order to reduce the number of credit update messages, we send them only when the space available seen by the transmitter is less than VIRTIO_VSOCK_MAX_PKT_BUF_SIZE. Signed-off-by:
-
Stefano Garzarella authored
Since virtio-vsock was introduced, the buffers filled by the host and pushed to the guest using the vring, are directly queued in a per-socket list. These buffers are preallocated by the guest with a fixed size (4 KB). The maximum amount of memory used by each socket should be controlled by the credit mechanism. The default credit available per-socket is 256 KB, but if we use only 1 byte per packet, the guest can queue up to 262144 of 4 KB buffers, using up to 1 GB of memory per-socket. In addition, the guest will continue to fill the vring with new 4 KB free buffers to avoid starvation of other sockets. This patch mitigates this issue copying the payload of small packets (< 128 bytes) into the buffer of last packet queued, in order to avoid wasting memory. Signed-off-by:
-
Stephen Boyd authored
We don't need dev_err() messages when platform_get_irq() fails now that platform_get_irq() prints an error message itself when something goes wrong. Let's remove these prints with a simple semantic patch. // <smpl> @@ expression ret; struct platform_device *E; @@ ret = ( platform_get_irq(E, ...) | platform_get_irq_byname(E, ...) ); if ( \( ret < 0 \| ret <= 0 \) ) { ( -if (ret != -EPROBE_DEFER) -{ ... -dev_err(...); -... } | ... -dev_err(...); ) ... } // </smpl> While we're here, remove braces on if statements that only have one statement (manually). Cc: "David S. Miller" <davem@davemloft.net> Cc: Kalle Valo <kvalo@codeaurora.org> Cc: Saeed Mahameed <saeedm@mellanox.com> Cc: Jeff Kirsher <jeffrey.t.kirsher@intel.com> Cc: Felix Fietkau <nbd@nbd.name> Cc: Lorenzo Bianconi <lorenzo@kernel.org> Cc: netdev@vger.kernel.org Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by:Stephen Boyd <swboyd@chromium.org> Signed-off-by:
-
David S. Miller authored
Jonathan Lemon says: ==================== Finish conversion of skb_frag_t to bio_vec The recent conversion of skb_frag_t to bio_vec did not include skb_frag's page_offset. Add accessor functions for this field, utilize them, and remove the union, restoring the original structure. v2: - rename accessors - follow kdoc conventions ==================== Reviewed-by:
Jakub Kicinski <jakub.kicinski@netronome.com> Reviewed-by:
Jens Axboe <axboe@kernel.dk> Signed-off-by:
-
Jonathan Lemon authored
Now that page_offset is referenced through accessors, remove the union, and use bv_offset. Signed-off-by:
Jonathan Lemon <jonathan.lemon@gmail.com> Signed-off-by:
-
Jonathan Lemon authored
Use accessor functions for skb fragment's page_offset instead of direct references, in preparation for bvec conversion. Signed-off-by:
-
Jonathan Lemon authored
Add skb_frag_off(), skb_frag_off_add(), skb_frag_off_set(), and skb_frag_off_copy() accessors for page_offset. Signed-off-by:
-
David S. Miller authored
Xin Long says: ==================== sctp: clean up __sctp_connect function This patchset is to factor out some common code for sctp_sendmsg_new_asoc() and __sctp_connect() into 2 new functioins. v1->v2: - add the patch 1/5 to avoid a slab-out-of-bounds warning. - add some code comment for the check change in patch 2/5. - remove unused 'addrcnt' as Marcelo noticed in patch 3/5. ==================== Acked-by:
Marcelo Ricardo Leitner <marcelo.leitner@gmail.com> Signed-off-by:
-
Xin Long authored
In this function factored out from sctp_sendmsg_new_asoc() and __sctp_connect(), it adds a peer with the other addr into the asoc after this asoc is created with the 1st addr. Signed-off-by:
Xin Long <lucien.xin@gmail.com> Signed-off-by:
-
Xin Long authored
In this function factored out from sctp_sendmsg_new_asoc() and __sctp_connect(), it creates the asoc and adds a peer with the 1st addr. Signed-off-by:
-
Xin Long authored
__sctp_connect is doing quit similar things as sctp_sendmsg_new_asoc. To factor out common functions, this patch is to clean up their code to make them look more similar: 1. create the asoc and add a peer with the 1st addr. 2. add peers with the other addrs into this asoc one by one. while at it, also remove the unused 'addrcnt'. Signed-off-by:
-
Xin Long authored
Now __sctp_connect() is called by __sctp_setsockopt_connectx() and sctp_inet_connect(), the latter has done addr_size check with size of sa_family_t. In the next patch to clean up __sctp_connect(), we will remove addr_size check with size of sa_family_t from __sctp_connect() for the 1st address. So before doing that, __sctp_setsockopt_connectx() should do this check first, as sctp_inet_connect() does. Signed-off-by:
-
Xin Long authored
'addr' passed to sctp_transport_init is not always a whole size of union sctp_addr, like the path: sctp_sendmsg() -> sctp_sendmsg_new_asoc() -> sctp_assoc_add_peer() -> sctp_transport_new() -> sctp_transport_init() In the next patches, we will also pass the address length of data only to sctp_assoc_add_peer(). So sctp_transport_init() should copy the only available data from addr to peer->ipaddr, instead of 'peer->ipaddr = *addr' which may cause slab-out-of-bounds. Signed-off-by:
-
David Howells authored
rxkad sometimes triggers a warning about oversized stack frames when building with clang for a 32-bit architecture: net/rxrpc/rxkad.c:243:12: error: stack frame size of 1088 bytes in function 'rxkad_secure_packet' [-Werror,-Wframe-larger-than=] net/rxrpc/rxkad.c:501:12: error: stack frame size of 1088 bytes in function 'rxkad_verify_packet' [-Werror,-Wframe-larger-than=] The problem is the combination of SYNC_SKCIPHER_REQUEST_ON_STACK() in rxkad_verify_packet()/rxkad_secure_packet() with the relatively large scatterlist in rxkad_verify_packet_1()/rxkad_secure_packet_encrypt(). The warning does not show up when using gcc, which does not inline the functions as aggressively, but the problem is still the same. Allocate the cipher buffers from the slab instead, caching the allocated packet crypto request memory used for DATA packet crypto in the rxrpc_call struct. Fixes: 17926a79 ("[AF_RXRPC]: Provide secure RxRPC sockets for use by userspace and kernel both") Reported-by:
Arnd Bergmann <arnd@arndb.de> Signed-off-by:
David Howells <dhowells@redhat.com> Acked-by:
-
- 29 Jul, 2019 20 commits
-
-
David S. Miller authored
Michael Chan says: ==================== bnxt_en: Add TPA (GRO_HW and LRO) on 57500 chips. This patchset adds TPA v2 support on the 57500 chips. TPA v2 is different from the legacy TPA scheme on older chips and requires major refactoring and restructuring of the existing TPA logic. The main difference is that the new TPA v2 has on-the-fly aggregation buffer completions before a TPA packet is completed. The larger aggregation ID space also requires a new ID mapping logic to make it more memory efficient. ==================== Signed-off-by: -
Michael Chan authored
Signed-off-by:
Michael Chan <michael.chan@broadcom.com> Signed-off-by:
-
Michael Chan authored
Define the 57508, 57504, and 57502 chip IDs that are all part of the BNXT_CHIP_P5 family of chips. Signed-off-by:
-
Michael Chan authored
With the new TPA feature in the 57500 chips, we need to discover the feature first before setting up the netdev features. Refactor the the firmware probe and init logic more cleanly into 2 functions and and make these calls before setting up the netdev features. Signed-off-by:
-
Michael Chan authored
Support the new expanded TPA v2 counters on 57500 B0 chips for ethtool -S. Signed-off-by:
-
Michael Chan authored
The new TPA implemantation has additional TPA counters that extend the per-ring statistics block. Allocate the proper size accordingly. Signed-off-by:
-
Michael Chan authored
The current code assumes that the per ring statistics counters are fixed. In newer chips that support a newer version of TPA, the TPA counters are also changed. Refactor the code by defining these counter names in arrays so that it is easy to add a new array for a new set of counters supported by the newer chips. Signed-off-by:
-
Michael Chan authored
Add a more optimized hardware GRO function to setup the SKB on 57500 chips. Some workaround code is no longer needed on 57500 chips and the pseudo checksum is also calculated in hardware, so no need to do the software pseudo checksum in the driver. Signed-off-by:
-
Michael Chan authored
The new TPA feature on 57500 supports a larger number of concurrent TPAs (up to 1024) divided among the functions. We need to add some logic to map the hardware TPA ID to a software index that keeps track of each TPA in progress. A 1:1 direct mapping without translation would be too wasteful as we would have to allocate 1024 TPA structures for each RX ring on each PCI function. Signed-off-by:
-
Michael Chan authored
With all the previous refactoring, the TPA fast path can now be modified slightly to support TPA on the new chips. The main difference is that the agg completions are retrieved differently using the bnxt_get_tpa_agg_p5() function on the new chips. Signed-off-by:
-
Michael Chan authored
On 57500 chips, hardware GRO mode cannot be determined from the TPA end, so we need to check bp->flags to determine if we are in hardware GRO mode or not. Modify bnxt_set_features so that the TPA flags in bp->flags don't change until the device is closed. This will ensure that the fast path can safely rely on bp->flags to determine the TPA mode. Signed-off-by:
-
Michael Chan authored
The 2 GRO functions to set up the hardware GRO SKB fields for 2 different hardware chips have practically identical logic for tunneled packets. Refactor the logic into a separate bnxt_gro_tunnel() function that can be used by both functions. Signed-off-by:
-
Michael Chan authored
On the new 57500 chips, these new RX_AGG completions are not coalesced at the TPA_END completion. Handle these by storing them in the array in the bnxt_tpa_info struct, as they are seen when processing the CMPL ring. Signed-off-by:
-
Michael Chan authored
Add an aggregation array to bnxt_tpa_info struct to keep track of the aggregation completions. The aggregation completions are not completed at the TPA_END completion on 57500 chips so we need to keep track of them. The array is only allocated on the new chips when required. An agg_count field is also added to keep track of the number of these completions. The maximum concurrent TPA is now discovered from firmware instead of the hardcoded 64. Add a new bp->max_tpa to keep track of maximum configured TPA. Signed-off-by:
-
Michael Chan authored
Refactor the TPA logic slightly, so that the code can be more easily extended to support TPA on the new 57500 chips. In particular, the logic to get the next aggregation completion is refactored into a new function bnxt_get_agg() so that this operation is made more generalized. This operation will be different on the new chip in TPA mode. The logic to recycle the aggregation buffers has a new start index parameter added for the same purpose. Signed-off-by:
-
Michael Chan authored
The new chips have a slightly modified TPA interface for LRO/GRO_HW. Modify the TPA structures so that the same structures can also be used on the new chips. Signed-off-by:
-
Michael Chan authored
Among the changes are new CoS discard counters and new ctx_hw_stats_ext struct for the latest 5750X B0 chips. Signed-off-by:
-
Oliver Hartkopp authored
Commit 60649d4e ("can: remove obsolete empty ioctl() handler") replaced the almost empty can_ioctl() function with sock_no_ioctl() which always returns -EOPNOTSUPP. Even though we don't have any ioctl() functions on socket/network layer we need to return -ENOIOCTLCMD to be able to forward ioctl commands like SIOCGIFINDEX to the network driver layer. This patch fixes the wrong return codes in the CAN network layer protocols. Reported-by:
kernel test robot <rong.a.chen@intel.com> Fixes: 60649d4e ("can: remove obsolete empty ioctl() handler") Signed-off-by:
Oliver Hartkopp <socketcan@hartkopp.net> Signed-off-by:
-
Rasmus Villemoes authored
We have an ERPS (Ethernet Ring Protection Switching) setup involving mv88e6250 switches which we're in the process of switching to a BSP based on the mainline driver. Breaking any link in the ring works as expected, with the ring reconfiguring itself quickly and traffic continuing with almost no noticable drops. However, when plugging back the cable, we see 5+ second stalls. This has been tracked down to the userspace application in charge of the protocol missing a few CCM messages on the good link (the one that was not unplugged), causing it to broadcast a "signal fail". That message eventually reaches its link partner, which responds by blocking the port. Meanwhile, the first node has continued to block the port with the just plugged-in cable, breaking the network. And the reason for those missing CCM messages has in turn been tracked down to the VTU apparently being too busy servicing load/purge operations that the normal lookups are delayed. Initial state, the link between C and D is blocked in software. _____________________ / \ | | A ----- B ----- C *---- D Unplug the cable between C and D. _____________________ / \ | | A ----- B ----- C * * D Reestablish the link between C and D. _____________________ / \ | | A ----- B ----- C *---- D Somehow, enough VTU/ATU operations happen inside C that prevents the application from receving the CCM messages from B in a timely manner, so a Signal Fail message is sent by C. When B receives that, it responds by blocking its port. _____________________ / \ | | A ----- B *---* C *---- D Very shortly after this, the signal fail condition clears on the BC link (some CCM messages finally make it through), so C unblocks the port. However, a guard timer inside B prevents it from removing the blocking before 5 seconds have elapsed. It is not unlikely that our userspace ERPS implementation could be smarter and/or is simply buggy. However, this patch fixes the symptoms we see, and is a small optimization that should not break anything (knock wood). The idea is simply to avoid doing an VTU load of an entry identical to the one already present. To do that, we need to know whether mv88e6xxx_vtu_get() actually found an existing entry, or has just prepared a struct mv88e6xxx_vtu_entry for us to load. To that end, let vlan->valid be an output parameter. The other two callers of mv88e6xxx_vtu_get() are not affected by this patch since they pass new=false. Signed-off-by:Rasmus Villemoes <rasmus.villemoes@prevas.dk> Signed-off-by:
-
Heiner Kallweit authored
There was a previous attempt to use xmit_more, but the change had to be reverted because under load sometimes a transmit timeout occurred [0]. Maybe this was caused by a missing memory barrier, the new attempt keeps the memory barrier before the call to netif_stop_queue like it is used by the driver as of today. The new attempt also changes the order of some calls as suggested by Eric. [0] https://lkml.org/lkml/2019/2/10/39Signed-off-by:
Heiner Kallweit <hkallweit1@gmail.com> Signed-off-by:
-
