HI3515的Unable to handle kernel paging request错误

Divingbear

最近在用海思的HI3515设计一款工控机，第一版做完后，开始开发第二版，主要改动是增删了部分硬件功能，DDR颗粒从原来的HY5PS1G1631CFP-Y5C换成了H5PS1G63EFRY5C。
) {! J" r% r& k$ A/ W 结果发现第二版的程序在引导过程中偶尔会出现Unable to handle kernel paging request的错误，且后边的地址经常会变化，并不是固定的错误。错误的引导信息如下：
Unable to handle kernel NULL pointer dereference at virtual address 0000003c
pgd = c1de4000
6 }) Y* b1 i, \& R1 p- A$ ^) x [0000003c] *pgd=c1de3031, *pte=00000000, *ppte=00000000
Internal error: Oops: 817 [#1]
Modules linked in:
! z, P% H, H: t8 _& l3 n CPU: 0    Not tainted  (2.6.24-rt1-hi3515v100 #24)
PC is at generic_file_aio_read+0x20/0x1a4
+ O5 [8 W4 z2 d& } LR is at do_sync_read+0xc8/0x114
( b8 M  E$ C3 k( l3 L pc : [<c0060904>]    lr : [<c007cc1c>]    psr: a0000013
sp : c1dd9d18  ip : 00000000  fp : c1dd9d60
r10: 00000000  r9 : 00000000  r8 : c1dd9e50
3 @# c5 E' ]/ V r7 : c1dd9d6c  r6 : c1dbb660  r5 : c1dd9dbc  r4 : c1dd9d74
r3 : 00000000  r2 : 00000001  r1 : 0000003c  r0 : c1dd9d74
Flags: NzCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment kernel
Control: 0005317f  Table: c1de4000  DAC: 00000017
  x9 B* Z- D0 y) }7 e* v  I/ ^7 Y Process rcS (pid: 219, stack limit = 0xc1dd8258)
Stack: (0xc1dd9d18 to 0xc1dda000)
* e9 e* [0 y- w$ U3 S 9d00:                                                       c0093cc4 c00f95cc
9d20: 00000000 c1dd9e28 c1dd9da4 c1dd9d38 c0085be8 c0093cb4 c1dd9d74 c1dd9dbc
3 a% ]- w1 @' A9 F4 W% {; w" _) ^ 9d40: c1dbb660 c1dd9d6c c1dd9e50 00000000 00000000 c1dd9e24 c1dd9d68 c007cc1c
9d60: c00608f8 00000000 00000000 c1d6bdc0 00000080 00000000 c1dd9fb0 00000000
9d80: 00000001 ffffffff c1dbb660 00000000 00000000 00000000 00000000 c1c5c0e0
9da0: 00000000 00000000 c1dd9de0 c1c5c0e0 c0049d48 c1dd9db4 c1dd9db4 00000000
9dc0: 00000000 c0044454 c0044004 c1dd9df8 c1dbb660 c1dd9e08 00000080 c007bde0
  p7 V6 T. h% ~ 9de0: c0065b7c c1dbb660 c1dd9e28 c1dd9fb0 c1dd8000 c0294ac8 c1dd9fb0 00000080
9e00: c1dbb660 c1d6bdc0 c1dd9e50 c007cb54 00000000 c1dd9fb0 c1dd9e4c c1dd9e28
9e20: c007cd14 c007cb64 c1801b40 bf000000 00000000 c1dd9e50 c1dd8000 c0294ac8
9e40: c1dd9e70 c1dd9e50 c00814b8 c007cc78 00000000 00000000 00000080 c1d6bdc0
9e60: 00000000 c1dd9e8c c1dd9e74 c0081d24 c0081474 c1dd9e98 c1d6bdc0 c1dd9fb0
9e80: c1dd9f30 c1dd9e90 c00ad608 c0081c98 c1d6bdc2 00000000 6e69622f 0068732f
0 S* Y. h% Z# C; ~6 y. L4 g 9ea0: c1db2320 beffff7b c1dd9ee8 c1dd9eb8 c006e3e4 00000000 00000020 00000000
9ec0: c1dd8000 00000001 00000000 c1d6bdc0 00000000 00000000 00000000 c1dd9f1c
9ee0: 00000000 00000000 00000000 c1dd9fb0 00000000 c1dd9efc 00000000 00000f7b
% X: Y& t; k) M$ j! B% H 9f00: c1dd9f1c c1dd9f10 c0294e54 c1d6bdc0 fffffffe c1dd8000 c0294e34 c1d6bdc0
1 B' R8 t% A: V: Z) A 9f20: fffffff8 c1dd9f5c c1dd9f34 c0081fc0 c00ad3fc c1d6bdc0 beffff92 00000000
9f40: 001d7bcc 001d7bc4 c1dd8000 c1dd9fb0 c1dd9f84 c1dd9f60 c0082248 c0081f2c
- w# L( q4 K" i, W) @ 9f60: c1c99000 001d7bcc c1dd9fb0 c1c99000 c001ffe4 001c71fc c1dd9fa4 c1dd9f88
, }+ y: U6 x$ b, [- z' H 9f80: c00237c4 c0082124 001d7bc4 001d7ba4 00000001 0000000b 00000000 c1dd9fa8
9fa0: c001fe40 c0023798 001d7bc4 001d7ba4 001d7ba4 001d7bc4 001d7bcc 001d7ba4
, B: k4 r- W% q3 A 9fc0: 001d7bc4 001d7ba4 00000001 001d7bcc 001d7bcc 00000001 001c71fc 00000000
9fe0: 400833c4 be813a44 00069d44 400833cc 20000010 001d7ba4 00000000 00000000
7 G+ i2 `  F* z  |& i! w Backtrace:
/ g" u- M5 @8 N: P [<c00608e8>] (generic_file_aio_read+0x4/0x1a4) from [<c007cc1c>] (do_sync_read+0xc8/0x114)
[<c007cb54>] (do_sync_read+0x0/0x114) from [<c007cd14>] (vfs_read+0xac/0x144)
. o9 t) a) {+ A& R& ~ [<c007cc68>] (vfs_read+0x0/0x144) from [<c00814b8>] (kernel_read+0x54/0x84)
  r8:c0294ac8 r7:c1dd8000 r6:c1dd9e50 r5:00000000 r4:bf000000
[<c0081464>] (kernel_read+0x0/0x84) from [<c0081d24>] (prepare_binprm+0x9c/0x108)
  r6:00000000 r5:c1d6bdc0 r4:00000080
! G2 i, M7 z6 w6 i; L* @' Y+ \. Z [<c0081c88>] (prepare_binprm+0x0/0x108) from [<c00ad608>] (load_script+0x21c/0x240)
3 l& W: L, F# p& _- H0 J  r6:c1dd9fb0 r5:c1d6bdc0 r4:c1dd9e98
5 |0 m! ^+ c$ A, ^+ E9 ^ [<c00ad3ec>] (load_script+0x0/0x240) from [<c0081fc0>] (search_binary_handler+0xa4/0x1f8)
" m% z7 \/ l$ H* ?$ x8 x  r6:fffffff8 r5:c1d6bdc0 r4:c0294e34
4 H) R9 |* |5 {2 q; G [<c0081f1c>] (search_binary_handler+0x0/0x1f8) from [<c0082248>] (do_execve+0x134/0x184)
[<c0082114>] (do_execve+0x0/0x184) from [<c00237c4>] (sys_execve+0x3c/0x5c)
[<c0023788>] (sys_execve+0x0/0x5c) from [<c001fe40>] (ret_fast_syscall+0x0/0x2c)
* b+ m6 w0 w& X7 E/ a2 n5 f! A  r7:0000000b r6:00000001 r5:001d7ba4 r4:001d7bc4
, }; ?8 K3 @' e, j/ f( q1 {# Q Code: e24dd020 e3a0c000 e58b3004 e1a04000 (e1a0a001)
* l8 K( r( Y7 B( A ---[ end trace e292f16b7bf51848 ]---
Segmentation fault
我先是把第一版和第二版上的NAND FLASH颗粒进行互换，发现新旧板上的颗粒在第一版的板子上都没有报错，而两个颗粒在第二版的板子上都会出现报错，初步确定问题是随着板子走的，而不是颗粒。
  j/ k0 Z7 _' I 然后将第二版上的硬件改动还原，尽量还原到和第一版一样，然后启动还是会出现报错信息，说明外围硬件的改动没有影响到引导驱动程序出错。
: ~$ c4 \7 {# G& M6 F4 _' d 现在在把新板子寄回工厂，准备把DDR颗粒换成和第一版一样，然后再尝试一下，这个需要一些时间。
然后比对了前后两种DDR颗粒的spec，未见明显的参数差别。


PS:
& n4 Q( T4 E. H7 |这几天集中研究了一下这个问题，汇报一下：
* e3 Z0 o$ E* P$ ~ 1.启动UBOOT后，用网络下载后边的文件到内存中的方式来启动，问题依旧会出现，排除掉NAND flash的问题。
2.抓了1.8V，VREF_DDR的电压波形,没有发现drop和ripple.排除电压的问题
3.把老板子上HY5PS1G1631CLFP-Y5C和新板子上的H5PS1G63EFR-Y5C互换了一下。旧版运行到现在一次都没出错
& f- C+ v2 `; {0 {2 f  c 而新版运行到现在报过一次错。
4.买了几颗HY5PS1G1631CFP-Y5C（与原来板子上DDR比较少了一个L,原来的DDR颗粒停产了，现在这个颗粒IDD6电流略高，其他参数一样）换到板上，暂时还没看到报错，还需多跑跑看。
) d8 c& q+ F9 S4 U 5.基于以上的实验，感觉是layout的问题，查了新旧版的走线，是完全一样的，etch length和manhattan length一样。
6.准备再次比对堆叠厚度，因为前后两版PCB是不同的厂子压合的，板子堆叠厚度是根据我提供的匹配阻抗由板厂自己计算的。

wenqing89

牛人，不过3515 支持的DDR2,最大200MHz吧，对布线要求这么高嘛？