Add a long-running DMA memcpy operation to vfio_pci_liveupdate_kexec_test so that the device attempts to perform DMAs continuously during the Live Update.
At this point iommufd preservation is not supported and bus mastering is not kept enabled on the device during across the kexec, so most of these DMAs will be dropped. However this test ensures that the current device preservation support does not lead to system instability or crashes if the device is active. And once iommufd and bus mastering are preserved, this test can be relaxed to check that the DMA operations completed successfully.
Signed-off-by: David Matlack dmatlack@google.com --- .../vfio/vfio_pci_liveupdate_kexec_test.c | 135 ++++++++++++++++++ 1 file changed, 135 insertions(+)
diff --git a/tools/testing/selftests/vfio/vfio_pci_liveupdate_kexec_test.c b/tools/testing/selftests/vfio/vfio_pci_liveupdate_kexec_test.c index 925c5fc30d56..3af50110a65d 100644 --- a/tools/testing/selftests/vfio/vfio_pci_liveupdate_kexec_test.c +++ b/tools/testing/selftests/vfio/vfio_pci_liveupdate_kexec_test.c @@ -1,8 +1,16 @@ // SPDX-License-Identifier: GPL-2.0-only
+#include <linux/sizes.h> +#include <sys/mman.h> + #include <libliveupdate.h> #include <libvfio.h>
+#define MEMCPY_SIZE SZ_1G +#define DRIVER_SIZE SZ_1M +#define MEMFD_SIZE (MEMCPY_SIZE + DRIVER_SIZE) + +static struct dma_region memcpy_region; static const char *device_bdf;
#define STATE_SESSION "567e1b8d-daf1-4a4a-ac13-35d6d275f646" @@ -11,8 +19,100 @@ static const char *device_bdf; enum { STATE_TOKEN, DEVICE_TOKEN, + MEMFD_TOKEN, };
+static void dma_memcpy_one(struct vfio_pci_device *device) +{ + void *src = memcpy_region.vaddr, *dst; + u64 size; + + size = min_t(u64, memcpy_region.size / 2, device->driver.max_memcpy_size); + dst = src + size; + + memset(src, 1, size); + memset(dst, 0, size); + + printf("Kicking off 1 DMA memcpy operations of size 0x%lx...\n", size); + vfio_pci_driver_memcpy(device, + to_iova(device, src), + to_iova(device, dst), + size); + + VFIO_ASSERT_EQ(memcmp(src, dst, size), 0); +} + +static void dma_memcpy_start(struct vfio_pci_device *device) +{ + void *src = memcpy_region.vaddr, *dst; + u64 count, size; + + size = min_t(u64, memcpy_region.size / 2, device->driver.max_memcpy_size); + dst = src + size; + + /* + * Rough Math: If we assume the device will perform memcpy at a rate of + * 30GB/s then 7200GB of transfers will run for about 4 minutes. + */ + count = (u64)7200 * SZ_1G / size; + count = min_t(u64, count, device->driver.max_memcpy_count); + + memset(src, 1, size / 2); + memset(dst, 0, size / 2); + + printf("Kicking off %lu DMA memcpy operations of size 0x%lx...\n", count, size); + vfio_pci_driver_memcpy_start(device, + to_iova(device, src), + to_iova(device, dst), + size, count); +} + +static void dma_memfd_map(struct vfio_pci_device *device, int fd) +{ + void *vaddr; + + vaddr = mmap(NULL, MEMFD_SIZE, PROT_WRITE, MAP_SHARED, fd, 0); + VFIO_ASSERT_NE(vaddr, MAP_FAILED); + + memcpy_region.iova = SZ_4G; + memcpy_region.size = MEMCPY_SIZE; + memcpy_region.vaddr = vaddr; + iommu_map(device->iommu, &memcpy_region); + + device->driver.region.iova = memcpy_region.iova + memcpy_region.size; + device->driver.region.size = DRIVER_SIZE; + device->driver.region.vaddr = vaddr + memcpy_region.size; + iommu_map(device->iommu, &device->driver.region); +} + +static void dma_memfd_setup(struct vfio_pci_device *device, int session_fd) +{ + int fd, ret; + + fd = memfd_create("dma-buffer", 0); + VFIO_ASSERT_GE(fd, 0); + + ret = fallocate(fd, 0, 0, MEMFD_SIZE); + VFIO_ASSERT_EQ(ret, 0); + + printf("Preserving memfd of size 0x%x in session\n", MEMFD_SIZE); + ret = luo_session_preserve_fd(session_fd, fd, MEMFD_TOKEN); + VFIO_ASSERT_EQ(ret, 0); + + dma_memfd_map(device, fd); +} + +static void dma_memfd_restore(struct vfio_pci_device *device, int session_fd) +{ + int fd; + + printf("Retrieving memfd from LUO\n"); + fd = luo_session_retrieve_fd(session_fd, MEMFD_TOKEN); + VFIO_ASSERT_GE(fd, 0); + + dma_memfd_map(device, fd); +} + static void before_kexec(int luo_fd) { struct vfio_pci_device *device; @@ -32,6 +132,27 @@ static void before_kexec(int luo_fd) ret = luo_session_preserve_fd(session_fd, device->fd, DEVICE_TOKEN); VFIO_ASSERT_EQ(ret, 0);
+ dma_memfd_setup(device, session_fd); + + /* + * If the device has a selftests driver, kick off a long-running DMA + * operation to exercise the device trying to DMA during a Live Update. + * Since iommufd preservation is not supported yet, these DMAs should be + * dropped. So this is just looking to verify that the system does not + * fall over and crash as a result of a busy device being preserved. + */ + if (device->driver.ops) { + vfio_pci_driver_init(device); + dma_memcpy_start(device); + + /* + * Disable interrupts on the device or freeze() will fail. + * Unfortunately there isn't a way to easily have a test for + * that here since the check happens during shutdown. + */ + vfio_pci_msix_disable(device); + } + close(luo_fd); daemonize_and_wait(); } @@ -106,9 +227,23 @@ static void after_kexec(int luo_fd, int state_session_fd) */ device = __vfio_pci_device_init(device_bdf, iommu, device_fd);
+ dma_memfd_restore(device, session_fd); + printf("Finishing the session\n"); VFIO_ASSERT_EQ(luo_session_finish(session_fd), 0);
+ /* + * Once iommufd preservation is supported and the device is kept fully + * running across the Live Update, this should wait for the long- + * running DMA memcpy operation kicked off in before_kexec() to + * complete. But for now we expect the device to be reset so just + * trigger a single memcpy to make sure it's still functional. + */ + if (device->driver.ops) { + vfio_pci_driver_init(device); + dma_memcpy_one(device); + } + vfio_pci_device_cleanup(device); iommu_cleanup(iommu); }