On Thu, Aug 06, 2015 at 06:15:57PM +0200, Laszlo Ersek wrote:
On 08/06/15 14:35, Stefan Hajnoczi wrote:
On Thu, Aug 6, 2015 at 12:02 PM, Marc Marí markmb@redhat.com wrote:
- if (qemu_cfg_dma_enabled()) {
QemuCfgDmaAccess access;access.address = (u64)(u32)buf;access.length = len;access.control = QEMU_CFG_DMA_CTL_READ;/** The out is done before the write of the variables on memory. This* causes misread on the QEMU side.*/barrier();outl((u32)&access, PORT_QEMU_CFG_DMA_ADDR);I thought PORT_QEMU_CFG_DMA_ADDR is a 64-bit register according to the spec you posted?
while(access.length != 0 && !(access.control & QEMU_CFG_DMA_CTL_ERROR));Either the field accesses need to be marked volatile, or a barrier is needed to force the compiler to reload these register from memory each iteration of the loop.
Better yet, I would recommend declaring (and here, also defining) "access" as "volatile".
volatile QemuCfgDmaAccess access;
I'd prefer to avoid volatile for the reasons described at:
https://www.kernel.org/doc/Documentation/volatile-considered-harmful.txt
I think barrier() and yield() are what we want - in particular, we really do want to yield() while busy so that interrupts and other code can continue to run.
-Kevin