On 07/16/2012 06:07 PM, Adam Hraska wrote:
> We could add one more barrier to pt_mapping_make_global()
> in order to be extra safe:
>
> void pt_mapping_make_global(uintptr_t base, size_t size)
> {
> // snip
>
> for (addr = ALIGN_DOWN(base, ptl0_step); addr - 1 < base + size - 1;
> addr += ptl0_step) {
> uintptr_t l1;
>
> l1 = (uintptr_t) frame_alloc(order, FRAME_KA | FRAME_LOWMEM);
> memsetb((void *) l1, FRAME_SIZE << order, 0);
> SET_PTL1_ADDRESS(ptl0, PTL0_INDEX(addr), KA2PA(l1));
>
> /* Make memsetb() visible before setting the PAGE_PRESENT flag.
> */
> write_barrier(); // <---- THIS IS NEW
>
> SET_PTL1_FLAGS(ptl0, PTL0_INDEX(addr),
> PAGE_PRESENT | PAGE_USER | PAGE_CACHEABLE |
> PAGE_EXEC | PAGE_WRITE | PAGE_READ);
> }
> }
This case is purely theoretical because this piece of code always runs
single-threaded before the other CPUs are even detected.
Jakub
_______________________________________________
HelenOS-devel mailing list
[email protected]
http://lists.modry.cz/cgi-bin/listinfo/helenos-devel
