[OLK-5.10] sw64: memset引发的数据错误

【环境信息】
sw3231，英特尔XL710网卡

Linux 4.19.180，commit bc4a1b70

【问题复现步骤】
重启机器时大概率出错，基本3次以内，但不是每次必出。

【预期结果】
机器正常启动

【实际结果】
Unable to handle kernel paging request at virtual address 00000000000006e2
CPU 0 kworker/0:1(25): Oops 0
pc = [<ffffffff814ac69c>]  ra = [<ffffffff814bc228>]  ps = 0000    Not tainted
pc is at i40e_set_num_rings_in_vsi+0x5c/0x180
ra is at i40e_vsi_setup+0x5c8/0x10d0
v0 = 0000000000000000  t0 = 0000000000000020  t1 = 0000000000000000
t2 = 0000000000000000  t3 = 0000000000000007  t4 = 00000000000000a0
t5 = 0000000000000000  t6 = fff0003fe0020a28  t7 = fff0003fe68dc000
s0 = fff0003feb539400  s1 = fff0003feb116000  s2 = fff0003feb116e10
s3 = fff0003fe0020000  s4 = 0000000000000008  s5 = 0000000000000002
s6 = fff0003fe68dfa28
a0 = fff0003fe0020000  a1 = 0000000000000000  a2 = 0000000000000000
a3 = ffffffffffffffff  a4 = 0000000000000000  a5 = 0000000000aaaaaa
t8 = fff0003fe0020a28  t9 = 0000000000000001  t10 = 0000000000000001
t11= 0000000000000299  pv = ffffffff814ac640  at = 0000000000000080
gp = ffffffff81e45d10  sp = (____ptrval____)
Disabling lock debugging due to kernel taint
【解决过程】
出错的文件为drivers/net/ethernet/intel/i40e/i40e_main.c

10080         vsi = kzalloc(sizeof(*vsi), GFP_KERNEL);
10081         if (!vsi) {
10082                 ret = -ENOMEM;
10083                 goto unlock_pf;
10084         }
10085         vsi->type = type;
10086         vsi->back = pf;
10087         set_bit(__I40E_VSI_DOWN, vsi->state);
10088         vsi->flags = 0;
10089         vsi->idx = vsi_idx;
10090         vsi->int_rate_limit = 0;
10091         vsi->rss_table_size = (vsi->type == I40E_VSI_MAIN) ?
10092                                 pf->rss_table_size : 64;
10093         vsi->netdev_registered = false;
10094         vsi->work_limit = I40E_DEFAULT_IRQ_WORK;
10095         hash_init(vsi->mac_filter_hash);
10096         vsi->irqs_ready = false;
10097 
10098         ret = i40e_set_num_rings_in_vsi(vsi);
 9952 static int i40e_set_num_rings_in_vsi(struct i40e_vsi *vsi)
 9953 {
 9954         struct i40e_pf *pf = vsi->back;
 9955 
 9956         switch (vsi->type) {
 9957         case I40E_VSI_MAIN:
 9958                 vsi->alloc_queue_pairs = pf->num_lan_qps;
commit bc4a1b70 ("sw64: optimize simd version of memcpy and memset")把memset里的memb从文件尾部移到了_nc写循环的尾部，来避免不必要的开销。但是位置摆错了，memb并不会被执行，导致_nc写指令有可能被乱序到memset返回后执行。

10080行的kzalloc会调用memset将分配的地址预先清0。如果memset中的_nc写指令被乱序到kzalloc返回后执行，就有可能将10086行的vsi->back清0。那么在进入i40e_set_num_rings_in_vsi()函数后，就会出现访0地址错误。

GVP openEuler/kernel

内容风险标识

评论 (1)

GVPopenEuler/kernel

内容风险标识