On Thu, Aug 07, 2008 at 06:07:50PM -0500, Woodruff, Richard wrote: > If I write a series of control register commands to device A, then > write a go operation to device B, I would hope all of A's writes > had completed before B gets the go. SO gives you this. DEVICE may > not with out barriers. This, I think, is where the problem lies. Device regions of the same type *are* ordered with respect to each other. So, shared device accesses occur in program order. Unshared device accesses occur in program order. However, shared device accesses may occur out of order with unshared device accesses or memory accesses. Unshared device accesses may occur out of order with shared device accesses or memory accesses. So, if both device A and device B are mapped as shared devices, then accesses to both occur in program order. If device A is mapped as a shared device and device B as an unshared device, then you have to use read backs and possibly barriers to ensure ordering. Remember, a barrier only affects up to the CPU. It doesn't affect write posting downstream of the CPU. -- To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
