Hi, Nicolin
在 2025/11/8 01:52, Nicolin Chen 写道:
On Fri, Nov 07, 2025 at 12:49:32PM -0400, Jason Gunthorpe wrote:
This is on github: https://github.com/jgunthorpe/linux/commits/iommufd_dmabuf
v2:
- Rebase on Leon's v7
- Fix mislocking in an iopt_fill_domain() error path
I have verified this v2 using the branch above, by drafting a QEMU patch for dmabuf on top of Shameer's vSMMU v5 series: https://github.com/nicolinc/qemu/commits/wip/iommufd_dmabuf/
With that, I see GPU BAR memory be correctly fetched in the QEMU: vfio_region_dmabuf Device 0009:01:00.0, region "0009:01:00.0 BAR 0", offset: 0x0, size: 0x1000000 vfio_region_dmabuf Device 0009:01:00.0, region "0009:01:00.0 BAR 2", offset: 0x0, size: 0x44f00000 vfio_region_dmabuf Device 0009:01:00.0, region "0009:01:00.0 BAR 4", offset: 0x0, size: 0x17a0000000
Tested-by: Nicolin Chen nicolinc@nvidia.com
I tested with your QEMU branch and this v2 branch, and observed the following:
**QEMU successfully maps the BAR regions via dmabuf:**
vfio_region_mmap Region 0008:01:00.0 BAR 0 mmaps[0] [0x0 - 0x3ffffff] vfio_region_dmabuf Device 0008:01:00.0, region "0008:01:00.0 BAR 0", offset: 0x0, size: 0x4000000 vfio_region_mmap Region 0008:01:00.0 BAR 2 mmaps[0] [0x0 - 0x3fffffffff] vfio_region_dmabuf Device 0008:01:00.0, region "0008:01:00.0 BAR 2", offset: 0x0, size: 0x4000000000 vfio_region_mmap Region 0008:01:00.0 BAR 4 mmaps[0] [0x0 - 0x2e41efffff] vfio_region_dmabuf Device 0008:01:00.0, region "0008:01:00.0 BAR 4", offset: 0x0, size: 0x2e41f00000
**IOMMUFD backend successfully maps the DMA regions:**
iommufd_backend_map_file_dma iommufd=303 ioas=2 iova=0x30000000000 size=0xb90000 fd=306 start=0 readonly=0 (0) vfio_listener_region_add_ram region_add [ram] 0x30000b91000 - 0x30003ffffff [0xffcae8b91000] iommufd_backend_map_file_dma iommufd=303 ioas=2 iova=0x30000b91000 size=0x346f000 fd=306 start=12128256 readonly=0 (0) vfio_listener_region_add_ram region_add [ram] 0x2c000000000 - 0x2ffffffffff [0xff8ac0000000] iommufd_backend_map_file_dma iommufd=303 ioas=2 iova=0x2c000000000 size=0x4000000000 fd=307 start=0 readonly=0 (0) vfio_listener_region_add_ram region_add [ram] 0x28000000000 - 0x2ae41efffff [0xff5c7e000000] iommufd_backend_map_file_dma iommufd=303 ioas=2 iova=0x28000000000 size=0x2e41f00000 fd=308 start=0 readonly=0 (0)
**However, the guest kernel crashes during PCI initialization:**
[ 1.422280] acpi PNP0A08:01: ECAM area [mem 0x4010100000-0x40102fffff] reserved by PNP0C02:00^M [ 1.422907] acpi PNP0A08:01: ECAM at [mem 0x4010100000-0x40102fffff] for [bus 01-02]^M [ 1.423541] ------------[ cut here ]------------^M [ 1.423905] kernel BUG at mm/vmalloc.c:109!^M [ 1.424240] Internal error: Oops - BUG: 00000000f2000800 [#1] SMP^M [ 1.424724] Modules linked in:^M [ 1.424972] CPU: 19 PID: 1 Comm: swapper/0 Not tainted 6.6.102-5.100_rc1.1.al8.aarch64+64k #1^M [ 1.425647] Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015^M [ 1.426201] pstate: 01401005 (nzcv daif +PAN -UAO -TCO +DIT +SSBS BTYPE=--)^M [ 1.426751] pc : vmap_pte_range+0x158/0x2b8^M [ 1.427101] lr : vmap_pte_range+0x178/0x2b8^M [ 1.427455] sp : ffff800083fcf6d0^M [ 1.427758] x29: ffff800083fcf6d0 x28: 0068000040820713 x27: 0000000000004083^M [ 1.428328] x26: ffff800083fcf7d4 x25: 0068000040820713 x24: 000000000000003f^M [ 1.428895] x23: 0068000000000713 x22: 0010000000000001 x21: ffffffbffe801000^M [ 1.429460] x20: ffffffc000040000 x19: ffff0018f9010020 x18: ffffffffffffffff^M [ 1.430034] x17: 75625b20726f6620 x16: 5d66666666663230 x15: 0000000000000002^M [ 1.430599] x14: 0000000000000000 x13: ffff0000c96ccf36 x12: 75625b20726f6620^M [ 1.431189] x11: 000000000000002e x10: 000003bffe800000 x9 : ffff800080341544^M [ 1.431802] x8 : ffff0000c96b8b80 x7 : ffff0018f9000000 x6 : ffff800083fcf7d4^M [ 1.432376] x5 : 000000000000003f x4 : 0068000000000713 x3 : 0000000000000000^M [ 1.432936] x2 : fffffffffe7d1000 x1 : ffff0031fff81380 x0 : 0000000000000008^M [ 1.433495] Call trace:^M [ 1.433689] vmap_pte_range+0x158/0x2b8^M [ 1.433996] vmap_range_noflush+0x19c/0x260^M [ 1.434323] ioremap_page_range+0x20/0x30^M [ 1.434636] pci_remap_iospace+0x78/0x90^M [ 1.434960] acpi_pci_probe_root_resources+0x124/0x260^M [ 1.435378] pci_acpi_root_prepare_resources+0x20/0xd0^M [ 1.435814] acpi_pci_root_create+0x94/0x2e8^M [ 1.436154] pci_acpi_scan_root+0x98/0x130^M [ 1.436474] acpi_pci_root_add+0x15c/0x2f8^M [ 1.436795] acpi_bus_attach+0x1bc/0x338^M [ 1.437104] acpi_dev_for_one_check+0x38/0x48^M [ 1.437444] device_for_each_child+0x60/0xb8^M [ 1.437799] acpi_dev_for_each_child+0x40/0x70^M [ 1.438148] acpi_bus_attach+0x234/0x338^M [ 1.438454] acpi_dev_for_one_check+0x38/0x48^M [ 1.438796] device_for_each_child+0x60/0xb8^M [ 1.439148] acpi_dev_for_each_child+0x40/0x70^M [ 1.439531] acpi_bus_attach+0x234/0x338^M [ 1.439854] acpi_bus_scan+0x68/0x208^M [ 1.440142] acpi_scan_init+0xd0/0x2b8^M [ 1.440445] acpi_init+0x80/0x140^M [ 1.440704] do_one_initcall+0x48/0x348^M [ 1.441010] do_initcalls+0x104/0x200^M [ 1.441305] kernel_init_freeable+0x128/0x2a0^M [ 1.441645] kernel_init+0x28/0x1e8^M [ 1.441943] ret_from_fork+0x10/0x20^M [ 1.442231] Code: 97f3ddf6 aa0003e1 b4fffa41 d503201f (d4210000) ^M
Could you share more details for testing?, e.g. QEMU cmdline, Guest kernel version.
I'd like to compare with my setup to identify the difference.
Thanks. Shuai