On 10/6/21 18:49, Philippe Mathieu-Daudé wrote:
> nvme_create_queue_pair() allocates resources with qemu_vfio_dma_map(),
> but we never release them. Do it in nvme_free_queue() which is called
> from nvme_free_queue_pair().
>
> Reported by valgrind:
>
> ==252858== 520,192 bytes in 1 blocks are still reachable in loss record
> 8,293 of 8,302
BTW the "8302" number is kinda depressing...
Good news, with this patch I'm now at 8301.
> ==252858== at 0x4846803: memalign (vg_replace_malloc.c:1265)
> ==252858== by 0x484691F: posix_memalign (vg_replace_malloc.c:1429)
> ==252858== by 0xB8AFE4: qemu_try_memalign (oslib-posix.c:210)
> ==252858== by 0xA9E315: nvme_create_queue_pair (nvme.c:229)
> ==252858== by 0xAA0125: nvme_init (nvme.c:799)
> ==252858== by 0xAA081C: nvme_file_open (nvme.c:953)
> ==252858== by 0xA23DDD: bdrv_open_driver (block.c:1550)
> ==252858== by 0xA24806: bdrv_open_common (block.c:1827)
> ==252858== by 0xA2889B: bdrv_open_inherit (block.c:3747)
> ==252858== by 0xA28DE4: bdrv_open (block.c:3840)
> ==252858== by 0x9E0F8E: bds_tree_init (blockdev.c:675)
> ==252858== by 0x9E7C74: qmp_blockdev_add (blockdev.c:3551)
>
> Fixes: bdd6a90a9e5 ("block: Add VFIO based NVMe driver")
> Signed-off-by: Philippe Mathieu-Daudé <[email protected]>
> ---
> block/nvme.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/block/nvme.c b/block/nvme.c
> index 6e476f54b9f..903c8ffa060 100644
> --- a/block/nvme.c
> +++ b/block/nvme.c
> @@ -185,6 +185,7 @@ static bool nvme_init_queue(BDRVNVMeState *s, NVMeQueue
> *q,
>
> static void nvme_free_queue(BDRVNVMeState *s, NVMeQueue *q)
> {
> + qemu_vfio_dma_unmap(s->vfio, q->queue);
> qemu_vfree(q->queue);
> }
>
>
