Re: [Qemu-devel] [PATCH v2 5/5] libqos: Add virtio MMIO support

[Marc Marí]

El Mon, 24 Nov 2014 11:59:19 +0000
Stefan Hajnoczi <address@hidden> escribió:
> On Sun, Nov 23, 2014 at 12:41:25PM +0100, Marc Marí wrote:
> > El Mon, 17 Nov 2014 15:48:09 +0000
> > Stefan Hajnoczi <address@hidden> escribió:
> > > On Sat, Nov 01, 2014 at 06:02:30PM +0100, Marc Marí wrote:
> > > 
> > > > +static void mmio_basic(void)
> > > > +{
> > > > +    QVirtioMMIODevice *dev;
> > > > +    QVirtQueue *vq;
> > > > +    QGuestAllocator *alloc;
> > > > +    QVirtioBlkReq req;
> > > > +    int n_size = TEST_IMAGE_SIZE / 2;
> > > > +    uint64_t req_addr;
> > > > +    uint64_t capacity;
> > > > +    uint32_t features;
> > > > +    uint32_t free_head;
> > > > +    uint8_t status;
> > > > +    char *data;
> > > > +
> > > > +    arm_test_start();
> > > > +
> > > > +    dev = qvirtio_mmio_init_device(MMIO_DEV_BASE_ADDR,
> > > > MMIO_PAGE_SIZE);
> > > > +    g_assert(dev != NULL);
> > > > +    g_assert_cmphex(dev->vdev.device_type, ==,
> > > > QVIRTIO_BLK_DEVICE_ID); +
> > > > +    qvirtio_reset(&qvirtio_mmio, &dev->vdev);
> > > > +    qvirtio_set_acknowledge(&qvirtio_mmio, &dev->vdev);
> > > > +    qvirtio_set_driver(&qvirtio_mmio, &dev->vdev);
> > > > +
> > > > +    capacity = qvirtio_config_readq(&qvirtio_mmio, &dev->vdev,
> > > > +
> > > > QVIRTIO_MMIO_DEVICE_SPECIFIC);
> > > > +    g_assert_cmpint(capacity, ==, TEST_IMAGE_SIZE / 512);
> > > > +
> > > > +    features = qvirtio_get_features(&qvirtio_mmio, &dev->vdev);
> > > > +    features = features & ~(QVIRTIO_F_RING_INDIRECT_DESC |
> > > > +                                QVIRTIO_F_RING_EVENT_IDX |
> > > > QVIRTIO_BLK_F_SCSI);
> > > > +    qvirtio_set_features(&qvirtio_mmio, &dev->vdev, features);
> > > > +
> > > > +    alloc = generic_alloc_init(MMIO_RAM_ADDR, MMIO_RAM_SIZE,
> > > > MMIO_PAGE_SIZE);
> > > > +    vq = qvirtqueue_setup(&qvirtio_mmio, &dev->vdev, alloc, 0);
> > > > +
> > > > +    qvirtio_set_driver_ok(&qvirtio_mmio, &dev->vdev);
> > > > +
> > > > +    qmp("{ 'execute': 'block_resize', 'arguments': { 'device':
> > > > 'drive0', "
> > > > +                                                    " 'size':
> > > > %d } }", n_size); +
> > > > +    qvirtio_wait_queue_isr(&qvirtio_mmio, &dev->vdev, vq,
> > > > +                           QVIRTIO_BLK_TIMEOUT_US);
> > > > +
> > > > +    capacity = qvirtio_config_readq(&qvirtio_mmio, &dev->vdev,
> > > > +
> > > > QVIRTIO_MMIO_DEVICE_SPECIFIC);
> > > > +    g_assert_cmpint(capacity, ==, n_size / 512);
> > > > +
> > > > +    /* Write request */
> > > > +    req.type = QVIRTIO_BLK_T_OUT;
> > > > +    req.ioprio = 1;
> > > > +    req.sector = 0;
> > > > +    req.data = g_malloc0(512);
> > > > +    strcpy(req.data, "TEST");
> > > > +
> > > > +    req_addr = virtio_blk_request(alloc, &req, 512);
> > > > +
> > > > +    g_free(req.data);
> > > > +
> > > > +    free_head = qvirtqueue_add(vq, req_addr, 528, false, true);
> > > > +    qvirtqueue_add(vq, req_addr + 528, 1, true, false);
> > > > +    qvirtqueue_kick(&qvirtio_mmio, &dev->vdev, vq, free_head);
> > > > +
> > > > +    qvirtio_wait_queue_isr(&qvirtio_mmio, &dev->vdev, vq,
> > > > +                           QVIRTIO_BLK_TIMEOUT_US);
> > > > +    status = readb(req_addr + 528);
> > > > +    g_assert_cmpint(status, ==, 0);
> > > > +
> > > > +    guest_free(alloc, req_addr);
> > > > +
> > > > +    /* Read request */
> > > > +    req.type = QVIRTIO_BLK_T_IN;
> > > > +    req.ioprio = 1;
> > > > +    req.sector = 0;
> > > > +    req.data = g_malloc0(512);
> > > > +
> > > > +    req_addr = virtio_blk_request(alloc, &req, 512);
> > > > +
> > > > +    g_free(req.data);
> > > > +
> > > > +    free_head = qvirtqueue_add(vq, req_addr, 16, false, true);
> > > > +    qvirtqueue_add(vq, req_addr + 16, 513, true, false);
> > > > +
> > > > +    qvirtqueue_kick(&qvirtio_mmio, &dev->vdev, vq, free_head);
> > > > +
> > > > +    qvirtio_wait_queue_isr(&qvirtio_mmio, &dev->vdev, vq,
> > > > +                           QVIRTIO_BLK_TIMEOUT_US);
> > > > +    status = readb(req_addr + 528);
> > > > +    g_assert_cmpint(status, ==, 0);
> > > > +
> > > > +    data = g_malloc0(512);
> > > > +    memread(req_addr + 16, data, 512);
> > > > +    g_assert_cmpstr(data, ==, "TEST");
> > > > +    g_free(data);
> > > 
> > > There is a lot of code duplication here.  Can the test logic but
> > > shared between PCI and MMIO?
> > 
> > The code duplication that can be easily extracted and shared is
> > performing a simple write - read operation on the block device
> > (which is used in various test cases). Other places (for example,
> > checking the image size) use arch specific offsets. This could be
> > abstracted, but I think is a bit too complicated for a test case.
> 
> Virtio config space is not transport-specific.
> 
> The Linux virtio_blk driver does:
> virtio_cread(vdev, struct virtio_blk_config, capacity, &capacity);
> 
> Why does this patch use:
> capacity = qvirtio_config_readq(&qvirtio_mmio, &dev->vdev,
> QVIRTIO_MMIO_DEVICE_SPECIFIC); ?
> 
> I don't have the virtio spec open right now, but I guess
> qvirtio_config_readq() should always start at virtio configuration
> byte 0 and not rely on transport-specific offsets.
> 
> Stefan

As usual, I arrived to wrong conclusions because I didn't look at it
carefully enough. Sorry.

What is variable is the position of the device-specific configuration
(in PCI depends of MSIX enabled or disabled and it can be 0x14 or 0x18,
and in MMIO is 0x100). But the capacity is always in the position 0 of
the device-specific configuration. With the base position, as well as an
allocator, a device and a bus, the basic tests can be performed (and
expanded later).

Thanks
Marc