i.mx6Q board sabreSD with SDK_4.1.0, slab allocator failed, kernel panic. "Unable to handle kernel paging request at virtual address"

[wallyyeh]

good day, every one:

   since we using realtek NIC rtl8111g instead FEC, system will hang for iperf testing about 32 hours, and I got a kernel panic in attach file.

the LR register's value is "8043cf88", kernel panic message is "Unable to handle kernel paging request at virtual address 00100104".

from the kernel dump:

static inline int bit_spin_trylock(int bitnum, unsigned long *addr)

{

        preempt_disable();

8043cf70:       e5993004        ldr     r3, [r9, #4]

8043cf74:       e2833001        add     r3, r3, #1

8043cf78:       e5893004        str     r3, [r9, #4]

#if defined(CONFIG_SMP) || defined(CONFIG_DEBUG_SPINLOCK)

        if (unlikely(test_and_set_bit_lock(bitnum, addr))) {

8043cf7c:       e3a00000        mov     r0, #0

8043cf80:       e1a01004        mov     r1, r4

8043cf84:       ebf6a891        bl      801e71d0 <_test_and_set_bit>

8043cf88:       e3500000        cmp     r0, #0

8043cf8c:       0a000007        beq     8043cfb0 <__slab_alloc.isra.53.constprop.57+0x1a8>

                preempt_enable();

......

according kernel panic message, I find out the entire kernel hang in the following steps:

    rtl8168_rx_interrupt() -> rtl8168_try_rx_copy() -> dev_alloc_skb() -> alloc_skb() -> kmalloc_track_caller() -> slab_alloc() -> slab_lock() ->lock_and_freeze_slab() -> bit_spin_lock() -> test_and_set_bit_lock() -> test_and_set_bit() <-- failed

the reason seems test_and_set_bit() operate the invalid memory address, and break the kernel.

I wonder some one inside the slab allocator functions gives the invalid memory address.

does anybody else using SDK_4.1.0's kernel and have a solution about this problem?

Original Attachment has been moved to: kernel_panic.zip

[wallyyeh · Accepted solution]

After applied patches from this post, the panic issue seems resolved.

freescale android kernel have scheduling problem !!!