linux - mmap messing up - causing kernel to hang -

i building application embedded board using microblaze processor. have linux cross-compiled this. board has 512 mb of memory kernel knows 256 mb of (the lower half of ddr3).

this memory map of system:

0x4000_0000 0x5fff_ffff -> physical range of ddr3 (verified cat /proc/meminfo) 0x4000_0000 0x4fff_ffff -> known linux , used kernel 0x5000_0000 0x5fff_ffff -> mmaped using "/dev/mem" application 

i have have direct access region in ddr, 2nd half of ddr. wrote application mmap's physical memory this:

typedef struct  {   int fd;                      // file descriptor   unsigned long *hw_addr;      // hardware base address   unsigned long *vaddr;        // virtual address pointer   unsigned long memsize;       // size of physical region   off_t page_offset;   off_t page_address;   off_t page_size; } mem_region_t ;  mem->vaddr = (unsigned long *) mmap(0,                    mem->memsize,                    prot_read|prot_write,                    map_shared,                    mem->fd,                    mem->page_address); 

to test working, doing simple memory test (memtest) on requested region.

here of problems noticing:

1. i can allocate amount of memory , mmap it, no problems. when use small region 0x100 bytes memtest, test runs successfully.

2. when allocate larger regions (0x1000, 0x1_0000, etc), kernel not segfault, hangs !! behavior normal ?

can suggest ways work around problem ? mmap'ing correctly.

my suspicion smaller page of memory works , more screws badly. need enable feature in kernel ?

thanks attention. let me know if need more info.

rrs

ok... solved problem.

i saw kernel didnt have enough pages work. created no-go zone using "memmap" (see kernel parameters). allowed me use mmap mentioned above.

it seems work large allocation of contiguous physical memory.