On Mon, Jun 17, 2024 at 11:51 PM Dragos Tatulea <dtatulea@xxxxxxxxxx> wrote: > > Hi, > > After commit ba168b52bf8e "mm: use rwsem assertion macros for > mmap_lock") was submitted, we started getting a lot of the > following warnings about a missing mmap write lock during VM boot: > > ------------[ cut here ]------------ > WARNING: CPU: 1 PID: 58633 at include/linux/rwsem.h:85 > track_pfn_remap+0x12b/0x130 > Modules linked in: act_mirred act_skbedit vhost_vdpa cls_matchall > nfnetlink_cttimeout act_gact cls_flower sch_ingress mlx5_vdpa vringh vdpa > openvswitch nsh vhost_net vhost vhost_iotlb tap ip6table_mangle ip6table_nat > iptable_mangle nf_tables ip6table_filter ip6_tables xt_conntrack xt_MASQUERADE > nf_conntrack_netlink nfnetlink xt_addrtype iptable_nat nf_nat br_netfilter > rpcsec_gss_krb5 auth_rpcgss oid_registry overlay rpcrdma rdma_ucm ib_iser > libiscsi ib_umad scsi_transport_iscsi ib_ipoib rdma_cm iw_cm ib_cm mlx5_ib > ib_uverbs ib_core fuse mlx5_core > CPU: 1 PID: 58633 Comm: CPU 0/KVM Tainted: G W > 6.10.0-rc1_for_upstream_min_debug_2024_05_29_17_06 #1 > Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS > rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014 > RIP: 0010:track_pfn_remap+0x12b/0x130 > Code: 48 83 c4 08 b8 ea ff ff ff 5b 5d 41 5c 41 5d c3 48 83 c4 08 48 89 ef 48 > 89 f2 5b 31 c9 4c 89 c6 5d 41 5c 41 5d e9 f5 fb ff ff <0f> 0b eb 9b 90 0f 1f 44 > 00 00 80 3d ac 59 96 01 00 74 01 c3 48 89 > RSP: 0018:ffff888350f8b8e0 EFLAGS: 00010246 > RAX: 0000000000000000 RBX: 0000000000001000 RCX: 0000000000000000 > RDX: ffff8881080ca300 RSI: 0000000000001000 RDI: 0000000544003000 > RBP: 0000000544003000 R08: ffff888106730a60 R09: 0000000000000000 > R10: ffff888116eeff60 R11: 0000000000000000 R12: ffff888350f8b918 > R13: ffff888149f99da8 R14: 0000000000001000 R15: 0000000000001000 > FS: 00007f678d800700(0000) GS:ffff88852c880000(0000) knlGS:0000000000000000 > CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > CR2: 00000000004e54f8 CR3: 0000000112290004 CR4: 0000000000372eb0 > Call Trace: > <TASK> > ? __warn+0x78/0x110 > ? track_pfn_remap+0x12b/0x130 > ? report_bug+0x16d/0x180 > ? handle_bug+0x3c/0x60 > ? exc_invalid_op+0x14/0x70 > ? asm_exc_invalid_op+0x16/0x20 > ? track_pfn_remap+0x12b/0x130 > remap_pfn_range+0x41/0xa0 > vhost_vdpa_fault+0x6c/0xa0 [vhost_vdpa] > __do_fault+0x2f/0xb0 > __handle_mm_fault+0x13d3/0x2210 > handle_mm_fault+0xb0/0x260 > fixup_user_fault+0x77/0x170 > hva_to_pfn+0x2c5/0x4b0 > kvm_faultin_pfn+0xd7/0x510 > kvm_tdp_page_fault+0x111/0x190 > kvm_mmu_do_page_fault+0x105/0x230 > kvm_mmu_page_fault+0x7d/0x620 > ? vmx_deliver_interrupt+0x110/0x190 > ? __apic_accept_irq+0x16c/0x270 > ? vmx_vmexit+0x8d/0xc0 > vmx_handle_exit+0x110/0x640 > kvm_arch_vcpu_ioctl_run+0xdb0/0x1c20 > kvm_vcpu_ioctl+0x263/0x6a0 > ? futex_wake+0x81/0x180 > __x64_sys_ioctl+0x4a7/0x9d0 > ? __x64_sys_futex+0x73/0x1c0 > ? kvm_on_user_return+0x86/0x90 > do_syscall_64+0x4c/0x100 > entry_SYSCALL_64_after_hwframe+0x4b/0x53 > RIP: 0033:0x7f679186a17b > Code: 0f 1e fa 48 8b 05 1d ad 0c 00 64 c7 00 26 00 00 00 48 c7 c0 ff ff ff ff > c3 66 0f 1f 44 00 00 f3 0f 1e fa b8 10 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 > c3 48 8b 0d ed ac 0c 00 f7 d8 64 89 01 48 > RSP: 002b:00007f678d7ff788 EFLAGS: 00000246 ORIG_RAX: 0000000000000010 > RAX: ffffffffffffffda RBX: 000000000000ae80 RCX: 00007f679186a17b > RDX: 0000000000000000 RSI: 000000000000ae80 RDI: 0000000000000059 > RBP: 000055da5ee22050 R08: 000055da44b28160 R09: 0000000000000000 > R10: 0000000000000001 R11: 0000000000000246 R12: 0000000000000000 > R13: 000055da452b05e0 R14: 0000000000000001 R15: 0000000000000000 > </TASK> > ---[ end trace 0000000000000000 ]--- > > The warnings show up only when the vdpa page-per-vq option is used (doorbell > mapping to guest). > > The issue seems to have existed before, but was visible only with CONFIG_LOCKDEP > enabled. I tried finding if this was introduced in more recent kernels, but > stopped after going as far back as 6.5: the issue was still visible there. > > The warning is triggered for the following call chain: > vhost_vdpa_fault() > -> remap_pfn_range() > -> remap_pfn_range_notrack() > -> vm_flags_set() > -> vma_start_write() > -> __is_vma_write_locked() > -> mmap_assert_write_locked() > > > I've been trying to follow how the mm write lock is dropped in the above call > chain or not taken at all. But I couldn't make much sense of it... I've also had a glance at vfio_pci_mmap_fault, it seems to do something similar. > Any ideas of what could have gone wrong here? Adding Peter for more thought here. Thanks > > Thanks, > Dragos
