【问题实操】银河高级服务器操作系统实例分享,开机之后反复重启

1.服务器环境以及配置

物理机/虚拟机/云/容器

物理机

外网/私有网络/无网络

私有网络

处理器:

 PHYTIUM FT2000PLUS 2200 MHz

内存:

 128 GiB

整机类型/架构:

HIKVISION DS-V

BIOS版本:

HK 601FBE02HK

网卡:

mlx5_core v5.0-0 / fw 14.29.2002 (MT_2420110004)

具体操作系统版本

(7860) Kylin-Server-V10-SP3-General-Release-2303-arm64

内核版本

4.19.90-52.22.v2207.ky10.aarch64

2.问题现象描述

服务器开机之后反复重启,用户现场已经收集了日志,需要分析日志。

3.问题分析

3.1 查看系统环境

cat ./sos_commands/networking/ethtool_-i_enp6s0f1
driver: mlx5_core
version: 5.0-0
firmware-version: 14.29.2002 (MT_2420110004)
expansion-rom-version:
bus-info: 0000:06:00.1
supports-statistics: yes
supports-test: yes
supports-eeprom-access: no
supports-register-dump: no
supports-priv-flags: yes

3.2 vmcore-dmesg分析

分析127.0.0.1-2023-09-19-07:58:37的vmcore-dmesg

[1865270.773139] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865271.773142] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865272.773093] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865273.773195] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865274.773168] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865275.773147] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865276.773151] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865277.773127] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865278.773147] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865279.773174] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865280.773149] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865281.773120] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865282.773112] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865283.773139] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865284.773150] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865285.773097] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865286.773067] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry
[1865287.773095] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1026764): failed to allocate command entry

分析127.0.0.1-2023-10-29-14:17:27的vmcore-dmesg

[3478639.193325] mlx5_core 0000:06:00.0: wait_func:1034:(pid 4068733): MODIFY_CQ(0x403) timeout. Will cause a leak of a command resource
[3478639.193333] mlx5_core 0000:06:00.0: wait_func:1034:(pid 4166044): MODIFY_CQ(0x403) timeout. Will cause a leak of a command resource
[3478639.193334] mlx5_core 0000:06:00.0: wait_func:1034:(pid 2812292): MODIFY_CQ(0x403) timeout. Will cause a leak of a command resource
[3478639.193342] mlx5_core 0000:06:00.0: wait_func:1034:(pid 3401828): MODIFY_CQ(0x403) timeout. Will cause a leak of a command resource
[3478639.193344] mlx5_core 0000:06:00.0: wait_func:1034:(pid 3822606): MODIFY_CQ(0x403) timeout. Will cause a leak of a command resource
[3478639.193354] mlx5_core 0000:06:00.0: wait_func:1034:(pid 188898): MODIFY_CQ(0x403) timeout. Will cause a leak of a command resource
[3478639.193363] mlx5_core 0000:06:00.0: wait_func:1034:(pid 3288787): MODIFY_CQ(0x403) timeout. Will cause a leak of a command resource
[3478639.194469] mlx5_core 0000:06:00.0: wait_func:1038:(pid 192618): MODIFY_CQ(0x403) canceled on out of queue timeout.
[3478639.194487] mlx5_core 0000:06:00.0: wait_func:1038:(pid 192616): MODIFY_CQ(0x403) canceled on out of queue timeout.
[3478639.194495] mlx5_core 0000:06:00.0: wait_func:1038:(pid 192615): MODIFY_CQ(0x403) canceled on out of queue timeout.
[3478639.194505] mlx5_core 0000:06:00.0: wait_func:1038:(pid 192613): MODIFY_CQ(0x403) canceled on out of queue timeout.
[3478639.194515] mlx5_core 0000:06:00.0: wait_func:1038:(pid 192610): MODIFY_CQ(0x403) canceled on out of queue timeout.
[3478640.183135] mlx5_core 0000:06:00.0: cmd_work_handler:906:(pid 1268656): failed to allocate command entry

从日志可见,不断打印failed to allocate command entry,可见该打印来自于drivers/net/ethernet/mellanox/mlx5/core/cmd.c文件的cmd_work_handler函数。

分析异常打印时的函数调用流程如下所示:

-> mlx5_cmd_exec
 -> cmd_exec
  -> mlx5_cmd_invoke
   -> cmd_work_handler
    -> alloc_ent
     -> find_first_bit

static int alloc_ent(struct mlx5_cmd *cmd)
{
    unsigned long flags;
    int ret;

    spin_lock_irqsave(&cmd->alloc_lock, flags);
    ret = find_first_bit(&cmd->bitmask, cmd->max_reg_cmds);
    if (ret < cmd->max_reg_cmds)
        clear_bit(ret, &cmd->bitmask);
    spin_unlock_irqrestore(&cmd->alloc_lock, flags);

    return ret < cmd->max_reg_cmds ? ret : -ENOMEM;
}

/*
 * Find the first set bit in a memory region.
 */
unsigned long find_first_bit(const unsigned long *addr, unsigned long size)
{
    unsigned long idx;

    for (idx = 0; idx * BITS_PER_LONG < size; idx++) {
        if (addr[idx])
            return min(idx * BITS_PER_LONG + __ffs(addr[idx]), size);
    }

    return size;
}

static void free_ent(struct mlx5_cmd *cmd, int idx)
{
    unsigned long flags;

    spin_lock_irqsave(&cmd->alloc_lock, flags);
    set_bit(idx, &cmd->bitmask);
    spin_unlock_irqrestore(&cmd->alloc_lock, flags);
}

从上述代码可见,通过cmd->bitmask进行标记是否有还可以申请,cmd->bitmask的初始化如下,假设获取出来的cmd->log_sz=0x80,则cmd->max_reg_cmds=0xFF,cmd->bitmask=0x1FD。应该是cmd->bitmask初始化的时候有多少bit位为1,就可以alloc_ent多少次。

int mlx5_cmd_init(struct mlx5_core_dev *dev)
{
    int size = sizeof(struct mlx5_cmd_prot_block);
    int align = roundup_pow_of_two(size);
    struct mlx5_cmd *cmd = &dev->cmd;
    u32 cmd_h, cmd_l;
    u16 cmd_if_rev;
    int err;
    int i;

    memset(cmd, 0, sizeof(*cmd));
    cmd_if_rev = cmdif_rev(dev);
    if (cmd_if_rev != CMD_IF_REV) {
        dev_err(&dev->pdev->dev,
            "Driver cmdif rev(%d) differs from firmware's(%d)\n",
            CMD_IF_REV, cmd_if_rev);
        return -EINVAL;
    }

    cmd->pool = dma_pool_create("mlx5_cmd", &dev->pdev->dev, size, align, 0);
    if (!cmd->pool)
        return -ENOMEM;
    err = alloc_cmd_page(dev, cmd);
    if (err)
        goto err_free_pool;

    cmd_l = ioread32be(&dev->iseg->cmdq_addr_l_sz) & 0xff;
    cmd->log_sz = cmd_l >> 4 & 0xf;
    cmd->log_stride = cmd_l & 0xf;
    if (1 << cmd->log_sz > MLX5_MAX_COMMANDS) {
        dev_err(&dev->pdev->dev, "firmware reports too many outstanding commands %d\n",
            1 << cmd->log_sz);
        err = -EINVAL;
        goto err_free_page;
    }

    if (cmd->log_sz + cmd->log_stride > MLX5_ADAPTER_PAGE_SHIFT) {
        dev_err(&dev->pdev->dev, "command queue size overflow\n");
        err = -EINVAL;
        goto err_free_page;
    }

    cmd->checksum_disabled = 1;
    cmd->max_reg_cmds = (1 << cmd->log_sz) - 1;
    cmd->bitmask = (1UL << cmd->max_reg_cmds) - 1;

执行cmd_exec的时候,会去调用alloc_ent,找到cmd->bitmask中最近为1的位,并调用clear_bit清除为0;cmd传到固件执行完以后,要去释放command entry,调用free_ent,将对应位重新标志为1。所以出现问题,有两种可能性:

1、当cmd在传到固件,一直未能执行完,而一直有cmd_exec执行,被积压,导致cmd->bitmask的位使用完,cmd->bitmask所有位被置为了0。alloc_ent无法申请到cmd->bitmask中的标识位。

2、cmd_exec执行完后,由于某种原因,没能调用free_ent,导致cmd->bitmask的所有位被使用完。

3.3 vmcore分析

日志中提示Internal error: SP/PC alignment exception,PC寄存器出现了未对齐的打印。

[3478640.184122] Internal error: SP/PC alignment exception: 8a000000 [#1] SMP
[3478640.184720] Modules linked in: fuse tun bridge stp llc fccore(POE) fchook(OE) bonding ip_set_hash_ip ip_set nfnetlink rfkill ipmi_ssif vfat fat rpcrdma sunrpc rdma_ucm ib_srpt ib_isert iscsi_target_mod target_core_mod ib_iser rdma_cm ib_umad iw_cm ib_ipoib libiscsi scsi_transport_iscsi ib_cm mlx5_ib joydev ib_uverbs ib_core ipmi_si ipmi_devintf ipmi_msghandler sch_fq_codel mlx5_core megaraid_sas igb ast mlxfw dm_mirror dm_region_hash dm_log [last unloaded: ip_tables]
[3478640.188127] Process kworker/u128:0 (pid: 1268656, stack limit = 0x0000000074e0af67)
[3478640.188820] CPU: 15 PID: 1268656 Comm: kworker/u128:0 Kdump: loaded Tainted: P           OE     4.19.90-52.22.v2207.ky10.aarch64 #1
[3478640.189835] Source Version: 52f32c50138f4eec54ab90ce9775f0d75e1d6968
[3478640.190505] Hardware name: HIKVISION DS-V/N/A, BIOS 601FBE02HK 2020/04/20
[3478640.191173] Workqueue: mlx5_cmd_0000:06:00.0 cmd_work_handler [mlx5_core]
[3478640.191790] pstate: 40000005 (nZcv daif -PAN -UAO)
[3478640.192235] pc : 0x42
[3478640.192539] lr : cmd_work_handler+0x434/0x578 [mlx5_core]
[3478640.193029] sp : ffffda857e113d20
[3478640.193348] x29: ffffda857e113d20 x28: ffffda8553b4c400
[3478640.193838] x27: 0000000000000000 x26: ffff5607dffd1000
[3478640.194341] x25: 0000000000000000 x24: ffffda8511e001c8
[3478640.194830] x23: ffffda8511e000e0 x22: ffffda8511e00160
[3478640.195345] x21: ffffda8553b4c548 x20: 0000000000000001
[3478640.195840] x19: ffffda8511e001d8 x18: 0000000000000001
[3478640.196332] x17: 0000fffc58585308 x16: ffff5607def87070
[3478640.196812] x15: ffffffffffffffff x14: 742064656c696166
[3478640.197283] x13: 203a293635363836 x12: 323120646970283a
[3478640.197769] x11: 3630393a72656c64 x10: ffff5607dffd5b70
[3478640.198278] x9 : 5f646d63203a302e x8 : ffff5607deee95d8
[3478640.198769] x7 : ffff5607e07230f8 x6 : 000000000000056b
[3478640.199307] x5 : 0000000000000000 x4 : 0000000000000000
[3478640.199801] x3 : 0000000000000000 x2 : 0000000000000042
[3478640.200358] x1 : 0000000000000000 x0 : 00000000fffffff5
[3478640.200835] Call trace:
[3478640.201084]  0x42
[3478640.201307]  process_one_work+0x1fc/0x4a8
[3478640.201705]  worker_thread+0x50/0x4d8
[3478640.202068]  kthread+0x134/0x138
[3478640.202386]  ret_from_fork+0x10/0x18
[3478640.202781] Code: bad PC value
[3478640.203165] SMP: stopping secondary CPUs
[3478640.205196] Starting crashdump kernel...

crash /usr/lib/debug/lib/modules/4.19.90-52.22.v2207.ky10.aarch64/vmlinux vmcore分析vmcore

LOAD AVERAGE: 102.08, 41.38, 22.56
       TASKS: 7790
    NODENAME: localhost.localdomain
     RELEASE: 4.19.90-52.22.v2207.ky10.aarch64
     VERSION: #1 SMP Tue Mar 14 11:52:45 CST 2023
     MACHINE: aarch64  (unknown Mhz)
      MEMORY: 126.4 GB
       PANIC: ""
         PID: 1268656
     COMMAND: "kworker/u128:0"
        TASK: ffffdc8435076a00  [THREAD_INFO: ffffdc8435076a00]
         CPU: 15
       STATE: TASK_RUNNING (PANIC)

crash> bt
PID: 1268656  TASK: ffffdc8435076a00  CPU: 15  COMMAND: "kworker/u128:0"
 #0 [ffffda857e1138c0] machine_kexec at ffff5607de7b2e8c
 #1 [ffffda857e113920] __crash_kexec at ffff5607de8ca148
 #2 [ffffda857e113a90] crash_kexec at ffff5607de8ca258
 #3 [ffffda857e113ac0] die at ffff5607de79f65c
 #4 [ffffda857e113b00] arm64_notify_die at ffff5607de79f964
 #5 [ffffda857e113b30] do_sp_pc_abort at ffff5607de7914f8
 #6 [ffffda857e113d10] el1_pc at ffff5607de792f94
 #7 [ffffda857e113d20] (null) at 3e
 #8 [ffffda857e113db0] process_one_work at ffff5607de8226f8
 #9 [ffffda857e113e00] worker_thread at ffff5607de8229f4
#10 [ffffda857e113e70] kthread at ffff5607de82a1e0

根据vmcore-dmesg中,pc : 0x42 lr : cmd_work_handler+0x434/0x578 [mlx5_core],查看lr寄存器执行的位置可见执行 blr     x2的时候触发了空指针。从而可知问题发生在drivers/net/ethernet/mellanox/mlx5/core/cmd.c: 908

crash> mod -s mlx5_core
crash> dis -rl cmd_work_handler+0x434 |tail -n 20
0xffff56077bcf6808 <cmd_work_handler+1016>:     mrs     x1, sp_el0
/usr/src/debug/kernel-4.19.90/linux-4.19.90-52.22.v2207.ky10.aarch64/drivers/net/ethernet/mellanox/mlx5/core/cmd.c: 906
0xffff56077bcf680c <cmd_work_handler+1020>:     ldr     x0, [x22,#-128]
0xffff56077bcf6810 <cmd_work_handler+1024>:     adrp    x2, 0xffff56077bd70000
0xffff56077bcf6814 <cmd_work_handler+1028>:     ldr     w4, [x1,#1712]
0xffff56077bcf6818 <cmd_work_handler+1032>:     add     x2, x2, #0xe50
0xffff56077bcf681c <cmd_work_handler+1036>:     add     x2, x2, #0x7c0
0xffff56077bcf6820 <cmd_work_handler+1040>:     mov     w3, #0x38a                      // #906
0xffff56077bcf6824 <cmd_work_handler+1044>:     adrp    x1, 0xffff56077bd7e000
0xffff56077bcf6828 <cmd_work_handler+1048>:     add     x1, x1, #0xcb0
0xffff56077bcf682c <cmd_work_handler+1052>:     bl      0xffff56077bd65e60 <plt:_dev_err>
/usr/src/debug/kernel-4.19.90/linux-4.19.90-52.22.v2207.ky10.aarch64/drivers/net/ethernet/mellanox/mlx5/core/cmd.c: 907
0xffff56077bcf6830 <cmd_work_handler+1056>:     ldr     x2, [x28,#40]
0xffff56077bcf6834 <cmd_work_handler+1060>:     cbz     x2, 0xffff56077bcf6970 <cmd_work_handler+1376>
/usr/src/debug/kernel-4.19.90/linux-4.19.90-52.22.v2207.ky10.aarch64/drivers/net/ethernet/mellanox/mlx5/core/cmd.c: 908
0xffff56077bcf6838 <cmd_work_handler+1064>:     ldr     x1, [x21,#-88]
0xffff56077bcf683c <cmd_work_handler+1068>:     mov     w0, #0xfffffff5                 // #-11
0xffff56077bcf6840 <cmd_work_handler+1072>:     blr     x2
/usr/src/debug/kernel-4.19.90/linux-4.19.90-52.22.v2207.ky10.aarch64/drivers/net/ethernet/mellanox/mlx5/core/cmd.c: 909
0xffff56077bcf6844 <cmd_work_handler+1076>:     ldr     x1, [x28,#16]
crash>

struct mlx5_cmd_work_ent是一个自定义的结构体,它包含了额外的字段和数据,用于处理MLX5驱动程序中的命令工作项。可见callback是一个有问题的值,该值出现问题应是网卡固件导致。

crash> struct mlx5_cmd_work_ent 0xffffda8553b4c400
struct mlx5_cmd_work_ent {
  state = 1,
  in = 0x383ef300000001,
  out = 0x1,
  uout = 0x0,
  uout_size = 0,
  callback = 0x42,
  cb_timeout_work = {
    work = {
      data = {
        counter = 0
      },

4.问题分析结果

通过上述分析,同时查看相关资料https://bugzilla.redhat.com/show_bug.cgi?id=1767678和https://forums.developer.nvidia.com/t/occure-mlx5-core-failed-to-allocate-command-entry-with-mcx4421a-acan-in-ubtuntu-18-04-1/206943,该问题更可能是网卡固件原因。可以尝试升级网卡固件,进行观察。

经查找有新的网卡固件版本为v14.32.1010。

Firmware for ConnectX®-4 Lx EN

具体的OPN号需要联系整机或网卡厂商获取,再找到对应的OPN网卡固件驱动由硬件厂商进行升级

5.参考资料

1767678 – [mlx5_core] OVS offload: mlx5_core 0000:04:00.0: cmd_work_handler:855:(pid 87647): failed to allocate command entry

RHEL8: Mellanox driver [mlx5_core] causing kernel stack overflow. - Red Hat Customer Portal

Occure mlx5_core :" failed to allocate command entry" with MCX4421A-ACAN in Ubtuntu 18.04.1 - Software And Drivers - NVIDIA Developer Forums

https://www.mellanox.com/downloads/MFT/mft-4.21.0-99-x86_64-rpm.tgz

https://www.mellanox.com/downloads/firmware/fw-ConnectX4Lx-rel-14_32_1010-MCX4121A-ACA_Ax-UEFI-14.25.17-FlexBoot-3.6.502.bin.zip

Bug Fixes History - NVIDIA Docs

本文来自互联网用户投稿,该文观点仅代表作者本人,不代表本站立场。本站仅提供信息存储空间服务,不拥有所有权,不承担相关法律责任。如若转载,请注明出处:http://www.mzph.cn/pingmian/11086.shtml

如若内容造成侵权/违法违规/事实不符,请联系多彩编程网进行投诉反馈email:809451989@qq.com,一经查实,立即删除!

相关文章

云原生技术解析

云原生的概念 云原生是一种软件架构和部署方法&#xff0c;旨在利用云计算的优势&#xff0c;以更灵活、可扩展和可靠的方式构建和部署应用程序。它主要关注在容器、微服务、自动化和持续交付等方面。 云原生技术是指以云计算作为基础&#xff0c;以平台和工具为依托&#xff0…

【免费】2024年全新超强版本itvboxfast如意版影视APP源码 TV+手机双端后台PHP源码

首先&#xff0c;让我们了解一下ITVBox如意版影视源码的特点和优势。这一源码基于先进的技术和框架开发&#xff0c;具有稳定、高效的性能&#xff0c;能够满足影视网站的各种需求。与此同时&#xff0c;该源码还提供了丰富的功能和模块&#xff0c;包括影视资源管理、会员系统…

rs6(vmp)瑞某,药某局,商某局,专某局,维某网,cookie + 后缀 的分析解析

文章目录 说在前面rs vmp 特征 介绍解决方法算法补环境运行报错 代码联调补环境框架 补环境导出结果导出cookie导出后缀 效果展示 vx lyj_txd qq 1416279170 # 加我备注来意说在前面 免责声明&#xff1a; 本篇文章只做学习讨论&#xff0c;无商务用途&#xff0c; 未对目标…

线下研讨会 技术沙龙|乐鑫芯片与 ESP RainMaker® 为科技初创企业赋能

众多科技初创企业在智能硬件市场迅猛发展的背景下&#xff0c;对不断变化的需求展现出了高度的敏锐性&#xff0c;期望能够快速将其转化为切实的产品方案。然而&#xff0c;面对复杂繁重的软硬件集成任务&#xff0c;这些企业往往容易陷入研发瓶颈、资金短缺以及效率低下等多重…

C++内存管理new/delete和new[ ]/delete[ ]

1.c/c内存分布 首先看一段代码 int globalVar 1; static int staticGlobalVar 1; void Test() { static int staticVar 1; int localVar 1; int num1[10] { 1, 2, 3, 4 }; char char2[] "abcd";const char* pChar3 "abcd"; //这里不加const会导致…

从 Oracle 到 TiDB,国有大行打造本地生活 APP 新体验

导读 本文介绍了某国有大行推出的本地生活服务类 APP 在数字时代的创新应用实践。该 APP 利用金融科技和互联网平台模式&#xff0c;打造“金融非金融”的线上生态服务平台&#xff0c;满足了用户多样化的生活需求。为应对用户增长和数据量增加带来的挑战&#xff0c;该 APP 决…

【原创】nnUnet V1在win11下的安装与配置

安装之前可以先了解一下论文的主要内容&#xff0c;便于之后网络训练与推理&#xff0c;调试程序。 论文地址&#xff1a;nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation | Nature Methods 也可以从其他博客快速浏览&#xff1a…

计算机网络(网络原理与应用)之高级交换实验------冗余环路与生成树协议

一、实验目的 (1)了解生成树协议的作用&#xff1b; (2)熟悉生成树协议的配置。 二、应用环境 采用生成树协议可以避免环路。 生成树协议的根本目的是将一个存在物理环路的交换网络变成一个没有环路的逻辑树形网络。IEEE802.ID协议通过在交换机上运行一套复杂的算法STA(sp…

JupyterLab OpenCV展示图片

JupyterLab OpenCV展示图片 方式一 注意&#xff1a;此种方式如果在远程服务器上的JupyterLab上运行&#xff0c;可能会出现错误。 import cv2# 读取图片 image cv2.imread(photo/blg.png)# 显示图片 cv2.imshow(image, image)# 等待按键&#xff0c;之后关闭所有窗口 cv2.w…

Deckset for Mac:让演示文稿制作更轻松

还在为繁琐的演示文稿制作而烦恼吗&#xff1f;Deckset for Mac来帮您解决&#xff01;它支持Markdown语言&#xff0c;让您只需专注于内容的创作&#xff0c;无需在排版和设计上耗费过多精力。丰富的主题和布局选项&#xff0c;让您能够轻松打造出专业级的演示文稿。快来体验D…

我的 OpenLiteSpeed 从开始到放弃之经历

昨晚下定决心放弃 OpenLiteSpeed 了&#xff0c;从开始到放弃历时七天。总结下来放弃 OpenLiteSpeed 主要是实在不适合明月当前的需要&#xff0c;用起来不是锦上添花而是个累赘了都&#xff0c;今天明月就给大家总结分享一下这次 OpenLiteSpeed 从开始到放弃的经历。 一、Ngin…

LLM量化

Efficient Finetuning prefix tuning 针对每种任务&#xff0c;学习prefix vector 启发于prompting&#xff0c;调整上下文内容让模型去输出自己想要的内容 核心就是找到一个上下文去引导模型解决NLP生成任务 传统情况下&#xff0c;我们为了得到想要的结果&#xff0c;会…

语言:C#

一、VSCode生成exe 二、

EasyRecovery(易恢复) 使用测试及详细使用方法

你有没有因为数据丢失懊悔不已&#xff0c;EasyRecovery&#xff08;易恢复&#xff09;&#xff0c;来自美国拥有38年数据恢复的软件&#xff0c;只有收费版&#xff0c;重要事情说三遍&#xff0c;EasyRecovery 没有免费版&#xff0c;可以成功找回删除的部分文件&#xff0c…

SQLite .journal 文件

在之前插入大量数据测试的时候&#xff0c;发现在数据库文件同级目录下会产生一个同名.journal的文件&#xff0c;并且不是一直会存在&#xff0c;而是生成一会就会自动删除&#xff0c;然后继续生成继续删除&#xff0c;直到数据插入完成。 初步猜测&#xff0c;应该是类似 re…

增强客户黏性:服装实体店扫码连锁收银软件提供多样的会员营销策略

随着消费者对购物体验的需求不断提升&#xff0c;服装实体店面临着更大的挑战和机遇。为了增强客户黏性&#xff0c;提升销售业绩&#xff0c;越来越多的服装实体店开始引入扫码连锁收银软件&#xff0c;并提供多样化的会员营销策略。这一举措不仅为店铺提供了更加智能化的经营…

C语言——文件相关操作补充

一、文件读取结束的判定 当我们使用例如fgetc、fgets、fscanf、fread等函数来读取文件内容时&#xff0c;我们可能遇到需要判断文件读取的结束&#xff0c;一般情况下都是通过这些函数的返回值来判断文件读取是否结束。 1、fgetc 返回读取的字符的ASCII值&#xff0c;如果读…

Java 守护线程 ( Daemon Thread )详解

在Java中&#xff0c;线程分为两类&#xff1a;用户线程(User Thread)和守护线程(Daemon Thread)。守护线程是后台线程&#xff0c;主要服务于用户线程&#xff0c;当所有的用户线程结束时&#xff0c;守护线程也会自动结束&#xff0c;JVM会随之退出。守护线程的一个典型例子是…

Web前端一套全部清晰 ⑧ day5 CSS.3 选择器、PxCook软件、盒子模型

谁不是一路荆棘而过呢 —— 24.5.12 CSS.3 选择器、PxCook软件、盒子模型 一、选择器 1.结构伪类选择器 1.作用: 根据元素的结构关系查找元素。 选择器 说明 E:first-child 查找第一个 E元素 E:last-child 查找最后一个E元素 E:nth-chil…

【qt】最快的开发界面效率——混合编程

混合编程 一.准备工作1.创建项目2.添加项目资源 二.ui界面设计1.menuBar菜单栏2.action ▲3.toolBar工具栏4.中心组件 三.代码界面设计1.toolBar添加组件2.statusBar状态栏添加组件 四.完成界面的功能1.对action配置信号槽2.对action转到信号槽3.代码添加的组件手动关联槽函数 …