VMCI: Use threaded irqs instead of tasklets
authorVishnu Dasa <vdasa@vmware.com>
Wed, 30 Nov 2022 07:05:11 +0000 (23:05 -0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 20 Jan 2023 12:24:11 +0000 (13:24 +0100)
commit3daed6345d5880464f46adab871d208e1baa2f3a
tree469ea61771652ff0d31da7de3c2ebd9ef296df2b
parentaaca766c77fcf5aabda846d3372a1d40b0d4735d
VMCI: Use threaded irqs instead of tasklets

The vmci_dispatch_dgs() tasklet function calls vmci_read_data()
which uses wait_event() resulting in invalid sleep in an atomic
context (and therefore potentially in a deadlock).

Use threaded irqs to fix this issue and completely remove usage
of tasklets.

[   20.264639] BUG: sleeping function called from invalid context at drivers/misc/vmw_vmci/vmci_guest.c:145
[   20.264643] in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 762, name: vmtoolsd
[   20.264645] preempt_count: 101, expected: 0
[   20.264646] RCU nest depth: 0, expected: 0
[   20.264647] 1 lock held by vmtoolsd/762:
[   20.264648]  #0: ffff0000874ae440 (sk_lock-AF_VSOCK){+.+.}-{0:0}, at: vsock_connect+0x60/0x330 [vsock]
[   20.264658] Preemption disabled at:
[   20.264659] [<ffff80000151d7d8>] vmci_send_datagram+0x44/0xa0 [vmw_vmci]
[   20.264665] CPU: 0 PID: 762 Comm: vmtoolsd Not tainted 5.19.0-0.rc8.20220727git39c3c396f813.60.fc37.aarch64 #1
[   20.264667] Hardware name: VMware, Inc. VBSA/VBSA, BIOS VEFI 12/31/2020
[   20.264668] Call trace:
[   20.264669]  dump_backtrace+0xc4/0x130
[   20.264672]  show_stack+0x24/0x80
[   20.264673]  dump_stack_lvl+0x88/0xb4
[   20.264676]  dump_stack+0x18/0x34
[   20.264677]  __might_resched+0x1a0/0x280
[   20.264679]  __might_sleep+0x58/0x90
[   20.264681]  vmci_read_data+0x74/0x120 [vmw_vmci]
[   20.264683]  vmci_dispatch_dgs+0x64/0x204 [vmw_vmci]
[   20.264686]  tasklet_action_common.constprop.0+0x13c/0x150
[   20.264688]  tasklet_action+0x40/0x50
[   20.264689]  __do_softirq+0x23c/0x6b4
[   20.264690]  __irq_exit_rcu+0x104/0x214
[   20.264691]  irq_exit_rcu+0x1c/0x50
[   20.264693]  el1_interrupt+0x38/0x6c
[   20.264695]  el1h_64_irq_handler+0x18/0x24
[   20.264696]  el1h_64_irq+0x68/0x6c
[   20.264697]  preempt_count_sub+0xa4/0xe0
[   20.264698]  _raw_spin_unlock_irqrestore+0x64/0xb0
[   20.264701]  vmci_send_datagram+0x7c/0xa0 [vmw_vmci]
[   20.264703]  vmci_datagram_dispatch+0x84/0x100 [vmw_vmci]
[   20.264706]  vmci_datagram_send+0x2c/0x40 [vmw_vmci]
[   20.264709]  vmci_transport_send_control_pkt+0xb8/0x120 [vmw_vsock_vmci_transport]
[   20.264711]  vmci_transport_connect+0x40/0x7c [vmw_vsock_vmci_transport]
[   20.264713]  vsock_connect+0x278/0x330 [vsock]
[   20.264715]  __sys_connect_file+0x8c/0xc0
[   20.264718]  __sys_connect+0x84/0xb4
[   20.264720]  __arm64_sys_connect+0x2c/0x3c
[   20.264721]  invoke_syscall+0x78/0x100
[   20.264723]  el0_svc_common.constprop.0+0x68/0x124
[   20.264724]  do_el0_svc+0x38/0x4c
[   20.264725]  el0_svc+0x60/0x180
[   20.264726]  el0t_64_sync_handler+0x11c/0x150
[   20.264728]  el0t_64_sync+0x190/0x194

Signed-off-by: Vishnu Dasa <vdasa@vmware.com>
Suggested-by: Zack Rusin <zackr@vmware.com>
Reported-by: Nadav Amit <namit@vmware.com>
Reported-by: Nathan Chancellor <nathan@kernel.org>
Tested-by: Nathan Chancellor <nathan@kernel.org>
Fixes: 463713eb6164 ("VMCI: dma dg: add support for DMA datagrams receive")
Cc: <stable@vger.kernel.org> # v5.18+
Cc: VMware PV-Drivers Reviewers <pv-drivers@vmware.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Zack Rusin <zackr@vmware.com>
Link: https://lore.kernel.org/r/20221130070511.46558-1-vdasa@vmware.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/misc/vmw_vmci/vmci_guest.c