I am currently seeing an issue when using SDK 2.0 on an LS1043ARDB reference board when running a fairly standard build (fsl-image-full + fsl-image-kernelitb). When I issue the rpm -aq command I see a lot of page faults being issued by the kernel sometimes ending in an oops. The following is the kind of kernel stack trace I see:
[ 96.377478] BUG: Bad page map in process rpm pte:a481a481a481a481 pmd:e00000e1a00fd3
[ 96.385321] addr:0000ffffb1600000 vm_flags:00100071 anon_vma:ffff800072aab9c8 mapping: (null) index:ffffb1600
[ 96.396019] file: (null) fault: (null) mmap: (null) readpage: (null)
[ 96.405497] CPU: 1 PID: 1796 Comm: rpm Tainted: G S 4.1.8-rt8+gbd51baf #1
[ 96.413409] Hardware name: LS1043A RDB Board (DT)
[ 96.418101] Call trace:
[ 96.420543] [<ffff800000089808>] dump_backtrace+0x0/0x11c
[ 96.425939] [<ffff800000089934>] show_stack+0x10/0x1c
[ 96.430981] [<ffff8000007eaff8>] dump_stack+0x7c/0x98
[ 96.436030] [<ffff800000160870>] print_bad_pte+0x164/0x200
[ 96.441506] [<ffff800000161f18>] vm_normal_page+0x70/0xa0
[ 96.446901] [<ffff800000162270>] unmap_single_vma+0x328/0x710
[ 96.452807] [<ffff800000162e90>] unmap_vmas+0x54/0xc4
[ 96.459258] [<ffff8000001680c0>] unmap_region+0x8c/0x140
[ 96.465930] [<ffff80000016a204>] do_munmap+0x240/0x3b8
[ 96.472427] [<ffff80000016a3bc>] vm_munmap+0x40/0x64
[ 96.478773] [<ffff80000016b228>] SyS_munmap+0x20/0x34
[ 96.485198] Disabling lock debugging due to kernel taint
[ 96.418101] Call trace:
I only ever see SyS_munmap in the stack dump.
This smells like a H/W configuration issue (running too fast maybe) but I have not changed anything in this area and the boards I have encountered this fault on were perfectly happy with the LS1043A 0.5 SDK.
Note that RPM is doing a lot of mmap/mprotect/munmap system calls so it is likely to be stressing the memory sub-system more than the other applications I have tried.
Any tips on where to look or solutions gratefully received.
- FCW unchanged since delivery of the board
- Standard kernel with systemd enabled (following systemd guidance regarding kernel config options) and changed openvswitch to be a loadable module
- Standard fsl-image-full file system but with systemd enabled instead of sysvinit (plus some additional packages added such as openvswitch, networkmanager, openvpn and some other odds and sods)
- Kernel booting off SD card using SD card variant of U-Boot (run from SD card) along with firmware version fsl_fman_ucode_t2080_r1.1_108_4_5.bin (loaded from SD-card)
- Had to remove RAM FS from the FIT image (kernel.itb) due to the issue reported here: SDK 2.0 inittab entries not found booting from /dev/mmcblk0p2
- Have also tried with NFS mounted file system with no change in behaviour
Note that I have performed a full rebuild of all binaries direct from source.