VIRTIO_BLK_F_SEG_MAX incompatible with qcow2 and vhdx? #6876
Description
Related issue: #5164
I tried to turn on the VIRTIO_BLK_F_SEG_MAX feature in the cloud-hypervisor
I refer to the implementation in qemu and simply set seg_max to queue_size - 2
diff --git a/virtio-devices/src/block.rs b/virtio-devices/src/block.rs
index 793242e2..69087719 100644
--- a/virtio-devices/src/block.rs
+++ b/virtio-devices/src/block.rs
@@ -626,6 +626,7 @@ impl Block {
| (1u64 << VIRTIO_BLK_F_BLK_SIZE)
| (1u64 << VIRTIO_BLK_F_TOPOLOGY)
| (1u64 << VIRTIO_RING_F_EVENT_IDX)
+ | (1u64 << VIRTIO_BLK_F_SEG_MAX)
| (1u64 << VIRTIO_RING_F_INDIRECT_DESC);
if iommu {
avail_features |= 1u64 << VIRTIO_F_IOMMU_PLATFORM;
@@ -660,6 +661,7 @@ impl Block {
physical_block_exp,
min_io_size: (topology.minimum_io_size / logical_block_size) as u16,
opt_io_size: (topology.optimal_io_size / logical_block_size) as u32,
+ seg_max: std::cmp::max(queue_size as u32, 3) - 2,
..Default::default()
};The default queue_size of virtio-blk is 128, so seg_max is set to 128 - 2 = 126 by default
When I tested this change, I found that the raw format system image works well,
But when using qcow2 or vhdx format system image, the system startup failure (x86 and arm64 neither).
Guest:
[ 1.292606] Run /init as init process
[ 1.676795] virtio_net virtio2 ens3: renamed from eth0
[ 1.690603] virtio_blk virtio1: [vda] 4612096 512-byte logical blocks (2.36 GB/2.20 GiB)
[ 1.709903] vda: vda1 vda15
[ 3.107821] raid6: neonx8 gen() 6862 MB/s
[ 3.179819] raid6: neonx8 xor() 4986 MB/s
[ 3.247818] raid6: neonx4 gen() 6969 MB/s
[ 3.315820] raid6: neonx4 xor() 5161 MB/s
[ 3.383821] raid6: neonx2 gen() 6147 MB/s
[ 3.451819] raid6: neonx2 xor() 4753 MB/s
[ 3.519817] raid6: neonx1 gen() 4759 MB/s
[ 3.587818] raid6: neonx1 xor() 3885 MB/s
[ 3.655819] raid6: int64x8 gen() 3640 MB/s
[ 3.723822] raid6: int64x8 xor() 2201 MB/s
[ 3.791820] raid6: int64x4 gen() 4068 MB/s
[ 3.859819] raid6: int64x4 xor() 2312 MB/s
[ 3.927820] raid6: int64x2 gen() 3819 MB/s
[ 3.995820] raid6: int64x2 xor() 2055 MB/s
[ 4.063819] raid6: int64x1 gen() 2932 MB/s
[ 4.131824] raid6: int64x1 xor() 1530 MB/s
[ 4.132578] raid6: using algorithm neonx4 gen() 6969 MB/s
[ 4.133525] raid6: .... xor() 5161 MB/s, rmw enabled
[ 4.134397] raid6: using neon recovery algorithm
[ 4.140248] xor: measuring software checksum speed
[ 4.143138] 8regs : 5196 MB/sec
[ 4.145804] 32regs : 5600 MB/sec
[ 4.148562] arm64_neon : 5209 MB/sec
[ 4.149393] xor: using function: 32regs (5600 MB/sec)
[ 4.151979] async_tx: api initialized (async)
[ 4.355293] Btrfs loaded, crc32c=crc32c-generic, zoned=yes, fsverity=yes
[ 4.432271] EXT4-fs (vda1): mounted filesystem with ordered data mode. Opts: (null). Quota mode: none.
[ 4.434355] EXT4-fs error (device vda1): ext4_lookup:1785: inode #37: comm sh: iget: checksum invalid
[ 4.486987] EXT4-fs error (device vda1): ext4_lookup:1785: inode #37: comm sh: iget: checksum invalid
[ 4.488823] EXT4-fs error (device vda1): ext4_lookup:1785: inode #37: comm sh: iget: checksum invalid
[ 4.505144] EXT4-fs error (device vda1): ext4_lookup:1785: inode #22: comm mount: iget: checksum invalid
[ 4.508933] EXT4-fs error (device vda1): ext4_lookup:1785: inode #22: comm mount: iget: checksum invalid
[ 4.519358] EXT4-fs error (device vda1): ext4_lookup:1785: inode #1501: comm sh: iget: checksum invalid
[ 4.522159] EXT4-fs error (device vda1): ext4_lookup:1785: inode #1501: comm mkdir: iget: checksum invalid
[ 4.541621] EXT4-fs error (device vda1): ext4_lookup:1785: inode #1509: comm mount: iget: checksum invalid
[ 4.545476] EXT4-fs error (device vda1): ext4_lookup:1785: inode #1513: comm run-init: iget: checksum invalid
[ 4.549299] EXT4-fs error (device vda1): ext4_lookup:1785: inode #1513: comm run-init: iget: checksum invalid
I didn't fix this, and I didn't see qemu doing anything special for the seg_max feature.
Does anyone have any suggestions? I'm happy to contribute after solving the problem