)]}'
{
  "log": [
    {
      "commit": "81c3d5470ecc70564eb9209946730fe2be93ad06",
      "tree": "1efa553c305a6453769dacfaf580bc6ccf146d82",
      "parents": [
        "399de50bbbb2501a6db43daaa8a2dafbc9bcfe0c"
      ],
      "author": {
        "name": "Eric Dumazet",
        "email": "dada1@cosmosbay.com",
        "time": "Mon Oct 03 14:13:38 2005 -0700"
      },
      "committer": {
        "name": "David S. Miller",
        "email": "davem@davemloft.net",
        "time": "Mon Oct 03 14:13:38 2005 -0700"
      },
      "message": "[INET]: speedup inet (tcp/dccp) lookups\n\nArnaldo and I agreed it could be applied now, because I have other\npending patches depending on this one (Thank you Arnaldo)\n\n(The other important patch moves skc_refcnt in a separate cache line,\nso that the SMP/NUMA performance doesnt suffer from cache line ping pongs)\n\n1) First some performance data :\n--------------------------------\n\ntcp_v4_rcv() wastes a *lot* of time in __inet_lookup_established()\n\nThe most time critical code is :\n\nsk_for_each(sk, node, \u0026head-\u003echain) {\n     if (INET_MATCH(sk, acookie, saddr, daddr, ports, dif))\n         goto hit; /* You sunk my battleship! */\n}\n\nThe sk_for_each() does use prefetch() hints but only the begining of\n\"struct sock\" is prefetched.\n\nAs INET_MATCH first comparison uses inet_sk(__sk)-\u003edaddr, wich is far\naway from the begining of \"struct sock\", it has to bring into CPU\ncache cold cache line. Each iteration has to use at least 2 cache\nlines.\n\nThis can be problematic if some chains are very long.\n\n2) The goal\n-----------\n\nThe idea I had is to change things so that INET_MATCH() may return\nFALSE in 99% of cases only using the data already in the CPU cache,\nusing one cache line per iteration.\n\n3) Description of the patch\n---------------------------\n\nAdds a new \u0027unsigned int skc_hash\u0027 field in \u0027struct sock_common\u0027,\nfilling a 32 bits hole on 64 bits platform.\n\nstruct sock_common {\n\tunsigned short\t\tskc_family;\n\tvolatile unsigned char\tskc_state;\n\tunsigned char\t\tskc_reuse;\n\tint\t\t\tskc_bound_dev_if;\n\tstruct hlist_node\tskc_node;\n\tstruct hlist_node\tskc_bind_node;\n\tatomic_t\t\tskc_refcnt;\n+\tunsigned int\t\tskc_hash;\n\tstruct proto\t\t*skc_prot;\n};\n\nStore in this 32 bits field the full hash, not masked by (ehash_size -\n1) Using this full hash as the first comparison done in INET_MATCH\npermits us immediatly skip the element without touching a second cache\nline in case of a miss.\n\nSuppress the sk_hashent/tw_hashent fields since skc_hash (aliased to\nsk_hash and tw_hash) already contains the slot number if we mask with\n(ehash_size - 1)\n\nFile include/net/inet_hashtables.h\n\n64 bits platforms :\n#define INET_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\\\n     (((__sk)-\u003esk_hash \u003d\u003d (__hash))\n     ((*((__u64 *)\u0026(inet_sk(__sk)-\u003edaddr)))\u003d\u003d (__cookie))   \u0026\u0026  \\\n     ((*((__u32 *)\u0026(inet_sk(__sk)-\u003edport))) \u003d\u003d (__ports))   \u0026\u0026  \\\n     (!((__sk)-\u003esk_bound_dev_if) || ((__sk)-\u003esk_bound_dev_if \u003d\u003d (__dif))))\n\n32bits platforms:\n#define TCP_IPV4_MATCH(__sk, __hash, __cookie, __saddr, __daddr, __ports, __dif)\\\n     (((__sk)-\u003esk_hash \u003d\u003d (__hash))                 \u0026\u0026  \\\n     (inet_sk(__sk)-\u003edaddr          \u003d\u003d (__saddr))   \u0026\u0026  \\\n     (inet_sk(__sk)-\u003ercv_saddr      \u003d\u003d (__daddr))   \u0026\u0026  \\\n     (!((__sk)-\u003esk_bound_dev_if) || ((__sk)-\u003esk_bound_dev_if \u003d\u003d (__dif))))\n\n\n- Adds a prefetch(head-\u003echain.first) in \n__inet_lookup_established()/__tcp_v4_check_established() and \n__inet6_lookup_established()/__tcp_v6_check_established() and \n__dccp_v4_check_established() to bring into cache the first element of the \nlist, before the {read|write}_lock(\u0026head-\u003elock);\n\nSigned-off-by: Eric Dumazet \u003cdada1@cosmosbay.com\u003e\nAcked-by: Arnaldo Carvalho de Melo \u003cacme@ghostprotocols.net\u003e\nSigned-off-by: David S. Miller \u003cdavem@davemloft.net\u003e\n"
    },
    {
      "commit": "e6df439b89dbf9eb977c2dc6f7b0644be3704df4",
      "tree": "81ab85c73026e7050f033dac2a6f60a194beda8e",
      "parents": [
        "3874b98c655b9490bea1cf9c7697d5dc5338376f"
      ],
      "author": {
        "name": "Brian Haley",
        "email": "brian.haley@hp.com",
        "time": "Sat Sep 10 00:15:06 2005 -0700"
      },
      "committer": {
        "name": "David S. Miller",
        "email": "davem@davemloft.net",
        "time": "Sat Sep 10 00:15:06 2005 -0700"
      },
      "message": "[IPV6]: Bring Type 0 routing header in-line with rfc3542.\n\nSigned-off-by: Brian Haley \u003cbrian.haley@hp.com\u003e\nSigned-off-by: YOSHIFUJI Hideaki \u003cyoshfuji@linux-ipv6.org\u003e\nSigned-off-by: David S. Miller \u003cdavem@davemloft.net\u003e\n"
    },
    {
      "commit": "41a1f8ea4fbfcdc4232f023732584aae2220de31",
      "tree": "9939686b354ced0b22d81efcbc12f5ed90372b09",
      "parents": [
        "333fad5364d6b457c8d837f7d05802d2aaf8a961"
      ],
      "author": {
        "name": "YOSHIFUJI Hideaki",
        "email": "yoshfuji@linux-ipv6.org",
        "time": "Thu Sep 08 10:19:03 2005 +0900"
      },
      "committer": {
        "name": "YOSHIFUJI Hideaki",
        "email": "yoshfuji@linux-ipv6.org",
        "time": "Thu Sep 08 10:19:03 2005 +0900"
      },
      "message": "[IPV6]: Support IPV6_{RECV,}TCLASS socket options / ancillary data.\n\nBased on patch from David L Stevens \u003cdlstevens@us.ibm.com\u003e\n\nSigned-off-by: David L Stevens \u003cdlstevens@us.ibm.com\u003e\nSigned-off-by: YOSHIFUJI Hideaki \u003cyoshfuji@linux-ipv6.org\u003e\n"
    },
    {
      "commit": "333fad5364d6b457c8d837f7d05802d2aaf8a961",
      "tree": "92149276982a827457e9c556c7ef82ec3b09fb52",
      "parents": [
        "4706df3d3c42af802597d82c8b1542c3d52eab23"
      ],
      "author": {
        "name": "YOSHIFUJI Hideaki",
        "email": "yoshfuji@linux-ipv6.org",
        "time": "Thu Sep 08 09:59:17 2005 +0900"
      },
      "committer": {
        "name": "YOSHIFUJI Hideaki",
        "email": "yoshfuji@linux-ipv6.org",
        "time": "Thu Sep 08 09:59:17 2005 +0900"
      },
      "message": "[IPV6]: Support several new sockopt / ancillary data in Advanced API (RFC3542).\n\nSupport several new socket options / ancillary data:\n  IPV6_RECVPKTINFO, IPV6_PKTINFO,\n  IPV6_RECVHOPOPTS, IPV6_HOPOPTS,\n  IPV6_RECVDSTOPTS, IPV6_DSTOPTS, IPV6_RTHDRDSTOPTS,\n  IPV6_RECVRTHDR, IPV6_RTHDR,\n  IPV6_RECVHOPOPTS, IPV6_HOPOPTS\n\nOld semantics are preserved as IPV6_2292xxxx so that\nwe can maintain backward compatibility.\n\nSigned-off-by: YOSHIFUJI Hideaki \u003cyoshfuji@linux-ipv6.org\u003e\n"
    },
    {
      "commit": "505cbfc577f3fa778005e2800b869eca25727d5f",
      "tree": "a823df4b3e919b06b08d49014ad233c48a5903ff",
      "parents": [
        "b766b305d3f2d8be173e5d9853534ea1afdbabba"
      ],
      "author": {
        "name": "Arnaldo Carvalho de Melo",
        "email": "acme@mandriva.com",
        "time": "Fri Aug 12 09:19:38 2005 -0300"
      },
      "committer": {
        "name": "David S. Miller",
        "email": "davem@sunset.davemloft.net",
        "time": "Mon Aug 29 15:57:24 2005 -0700"
      },
      "message": "[IPV6]: Generalise the tcp_v6_lookup routines\n\nIn the same way as was done with the v4 counterparts, this will be moved\nto inet6_hashtables.c.\n\nSigned-off-by: Arnaldo Carvalho de Melo \u003cacme@mandriva.com\u003e\nSigned-off-by: David S. Miller \u003cdavem@davemloft.net\u003e\n"
    },
    {
      "commit": "3f421baa4720b708022f8bcc52a61e5cd6f10bf8",
      "tree": "e4201b1e2356cea8b7bd8d68dfba06e84002a77d",
      "parents": [
        "463c84b97f24010a67cd871746d6a7e4c925a5f9"
      ],
      "author": {
        "name": "Arnaldo Carvalho de Melo",
        "email": "acme@ghostprotocols.net",
        "time": "Tue Aug 09 20:11:08 2005 -0700"
      },
      "committer": {
        "name": "David S. Miller",
        "email": "davem@sunset.davemloft.net",
        "time": "Mon Aug 29 15:49:14 2005 -0700"
      },
      "message": "[NET]: Just move the inet_connection_sock function from tcp sources\n\nCompleting the previous changeset, this also generalises tcp_v4_synq_add,\nrenaming it to inet_csk_reqsk_queue_hash_add, already geing used in the\nDCCP tree, which I plan to merge RSN.\n\nSigned-off-by: Arnaldo Carvalho de Melo \u003cacme@ghostprotocols.net\u003e\nSigned-off-by: David S. Miller \u003cdavem@davemloft.net\u003e\n"
    },
    {
      "commit": "463c84b97f24010a67cd871746d6a7e4c925a5f9",
      "tree": "48df67ede4ebb5d12b3c0ae55d72531574bd51a6",
      "parents": [
        "87d11ceb9deb7a3f13fdee6e89d9bb6be7d27a71"
      ],
      "author": {
        "name": "Arnaldo Carvalho de Melo",
        "email": "acme@ghostprotocols.net",
        "time": "Tue Aug 09 20:10:42 2005 -0700"
      },
      "committer": {
        "name": "David S. Miller",
        "email": "davem@sunset.davemloft.net",
        "time": "Mon Aug 29 15:43:19 2005 -0700"
      },
      "message": "[NET]: Introduce inet_connection_sock\n\nThis creates struct inet_connection_sock, moving members out of struct\ntcp_sock that are shareable with other INET connection oriented\nprotocols, such as DCCP, that in my private tree already uses most of\nthese members.\n\nThe functions that operate on these members were renamed, using a\ninet_csk_ prefix while not being moved yet to a new file, so as to\nease the review of these changes.\n\nSigned-off-by: Arnaldo Carvalho de Melo \u003cacme@ghostprotocols.net\u003e\nSigned-off-by: David S. Miller \u003cdavem@davemloft.net\u003e\n"
    },
    {
      "commit": "8feaf0c0a5488b3d898a9c207eb6678f44ba3f26",
      "tree": "ddd004afe2f7c8295f6fdb94d34f78a42b5961cb",
      "parents": [
        "33b62231908c58ae04185e4f1063d1e35a7c8576"
      ],
      "author": {
        "name": "Arnaldo Carvalho de Melo",
        "email": "acme@ghostprotocols.net",
        "time": "Tue Aug 09 20:09:30 2005 -0700"
      },
      "committer": {
        "name": "David S. Miller",
        "email": "davem@sunset.davemloft.net",
        "time": "Mon Aug 29 15:42:13 2005 -0700"
      },
      "message": "[INET]: Generalise tcp_tw_bucket, aka TIME_WAIT sockets\n\nThis paves the way to generalise the rest of the sock ID lookup\nroutines and saves some bytes in TCPv4 TIME_WAIT sockets on distro\nkernels (where IPv6 is always built as a module):\n\n[root@qemu ~]# grep tw_sock /proc/slabinfo\ntw_sock_TCPv6  0  0  128  31  1\ntw_sock_TCP    0  0   96  41  1\n[root@qemu ~]#\n\nNow if a protocol wants to use the TIME_WAIT generic infrastructure it\nonly has to set the sk_prot-\u003etwsk_obj_size field with the size of its\ninet_timewait_sock derived sock and proto_register will create\nsk_prot-\u003etwsk_slab, for now its only for INET sockets, but we can\nintroduce timewait_sock later if some non INET transport protocolo\nwants to use this stuff.\n\nNext changesets will take advantage of this new infrastructure to\ngeneralise even more TCP code.\n\n[acme@toy net-2.6.14]$ grep built-in /tmp/before.size /tmp/after.size\n/tmp/before.size: 188646   11764    5068  205478   322a6 net/ipv4/built-in.o\n/tmp/after.size:  188144   11764    5068  204976   320b0 net/ipv4/built-in.o\n[acme@toy net-2.6.14]$\n\nTested with both IPv4 \u0026 IPv6 (::1 (localhost) \u0026 ::ffff:172.20.0.1\n(qemu host)).\n\nSigned-off-by: Arnaldo Carvalho de Melo \u003cacme@ghostprotocols.net\u003e\nSigned-off-by: David S. Miller \u003cdavem@davemloft.net\u003e\n"
    },
    {
      "commit": "60236fdd08b2169045a3bbfc5ffe1576e6c3c17b",
      "tree": "4541c682cc72daf560ec516e2b5868089a88b6ea",
      "parents": [
        "2e6599cb899ba4b133f42cbf9d2b1883d2dc583a"
      ],
      "author": {
        "name": "Arnaldo Carvalho de Melo",
        "email": "acme@ghostprotocols.net",
        "time": "Sat Jun 18 22:47:21 2005 -0700"
      },
      "committer": {
        "name": "David S. Miller",
        "email": "davem@davemloft.net",
        "time": "Sat Jun 18 22:47:21 2005 -0700"
      },
      "message": "[NET] Rename open_request to request_sock\n\nOk, this one just renames some stuff to have a better namespace and to\ndissassociate it from TCP:\n\nstruct open_request  -\u003e struct request_sock\ntcp_openreq_alloc    -\u003e reqsk_alloc\ntcp_openreq_free     -\u003e reqsk_free\ntcp_openreq_fastfree -\u003e __reqsk_free\n\nWith this most of the infrastructure closely resembles a struct\nsock methods subset.\n\nSigned-off-by: Arnaldo Carvalho de Melo \u003cacme@ghostprotocols.net\u003e\nSigned-off-by: David S. Miller \u003cdavem@davemloft.net\u003e\n"
    },
    {
      "commit": "2e6599cb899ba4b133f42cbf9d2b1883d2dc583a",
      "tree": "b5d4fcca4d2a515fc3d3d20cefaaeebd8dbf661f",
      "parents": [
        "1944972d3bb651474a5021c9da8d0166ae19f1eb"
      ],
      "author": {
        "name": "Arnaldo Carvalho de Melo",
        "email": "acme@ghostprotocols.net",
        "time": "Sat Jun 18 22:46:52 2005 -0700"
      },
      "committer": {
        "name": "David S. Miller",
        "email": "davem@davemloft.net",
        "time": "Sat Jun 18 22:46:52 2005 -0700"
      },
      "message": "[NET] Generalise TCP\u0027s struct open_request minisock infrastructure\n\nKept this first changeset minimal, without changing existing names to\nease peer review.\n\nBasicaly tcp_openreq_alloc now receives the or_calltable, that in turn\nhas two new members:\n\n-\u003eslab, that replaces tcp_openreq_cachep\n-\u003eobj_size, to inform the size of the openreq descendant for\n  a specific protocol\n\nThe protocol specific fields in struct open_request were moved to a\nclass hierarchy, with the things that are common to all connection\noriented PF_INET protocols in struct inet_request_sock, the TCP ones\nin tcp_request_sock, that is an inet_request_sock, that is an\nopen_request.\n\nI.e. this uses the same approach used for the struct sock class\nhierarchy, with sk_prot indicating if the protocol wants to use the\nopen_request infrastructure by filling in sk_prot-\u003ersk_prot with an\nor_calltable.\n\nResults? Performance is improved and TCP v4 now uses only 64 bytes per\nopen request minisock, down from 96 without this patch :-)\n\nNext changeset will rename some of the structs, fields and functions\nmentioned above, struct or_calltable is way unclear, better name it\nstruct request_sock_ops, s/struct open_request/struct request_sock/g,\netc.\n\nSigned-off-by: Arnaldo Carvalho de Melo \u003cacme@ghostprotocols.net\u003e\nSigned-off-by: David S. Miller \u003cdavem@davemloft.net\u003e\n"
    },
    {
      "commit": "1da177e4c3f41524e886b7f1b8a0c1fc7321cac2",
      "tree": "0bba044c4ce775e45a88a51686b5d9f90697ea9d",
      "parents": [],
      "author": {
        "name": "Linus Torvalds",
        "email": "torvalds@ppc970.osdl.org",
        "time": "Sat Apr 16 15:20:36 2005 -0700"
      },
      "committer": {
        "name": "Linus Torvalds",
        "email": "torvalds@ppc970.osdl.org",
        "time": "Sat Apr 16 15:20:36 2005 -0700"
      },
      "message": "Linux-2.6.12-rc2\n\nInitial git repository build. I\u0027m not bothering with the full history,\neven though we have it. We can create a separate \"historical\" git\narchive of that later if we want to, and in the meantime it\u0027s about\n3.2GB when imported into git - space that would just make the early\ngit days unnecessarily complicated, when we don\u0027t have a lot of good\ninfrastructure for it.\n\nLet it rip!\n"
    }
  ]
}