The bulk of ATA host state machine is implemented by
ata_sff_hsm_move(). The function is called from either the interrupt
handler or, if polling, a work item. Unlike from the interrupt path,
the polling path calls the function without holding the host lock and
ata_sff_hsm_move() selectively grabs the lock.
This is completely broken. If an IRQ triggers while polling is in
progress, the two can easily race and end up accessing the hardware
and updating state machine state at the same time. This can put the
state machine in an illegal state and lead to a crash like the
following.
kernel BUG at drivers/ata/libata-sff.c:1302!
invalid opcode: 0000 [#1] SMP DEBUG_PAGEALLOC KASAN
Modules linked in:
CPU: 1 PID: 10679 Comm: syz-executor Not tainted 4.5.0-rc1+ #300
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
task: ffff88002bd00000 ti: ffff88002e048000 task.ti: ffff88002e048000
RIP: 0010:[<ffffffff83a83409>] [<ffffffff83a83409>] ata_sff_hsm_move+0x619/0x1c60
...
Call Trace:
<IRQ>
[<ffffffff83a84c31>] __ata_sff_port_intr+0x1e1/0x3a0 drivers/ata/libata-sff.c:1584
[<ffffffff83a85611>] ata_bmdma_port_intr+0x71/0x400 drivers/ata/libata-sff.c:2877
[< inline >] __ata_sff_interrupt drivers/ata/libata-sff.c:1629
[<ffffffff83a85bf3>] ata_bmdma_interrupt+0x253/0x580 drivers/ata/libata-sff.c:2902
[<ffffffff81479f98>] handle_irq_event_percpu+0x108/0x7e0 kernel/irq/handle.c:157
[<ffffffff8147a717>] handle_irq_event+0xa7/0x140 kernel/irq/handle.c:205
[<ffffffff81484573>] handle_edge_irq+0x1e3/0x8d0 kernel/irq/chip.c:623
[< inline >] generic_handle_irq_desc include/linux/irqdesc.h:146
[<ffffffff811a92bc>] handle_irq+0x10c/0x2a0 arch/x86/kernel/irq_64.c:78
[<ffffffff811a7e4d>] do_IRQ+0x7d/0x1a0 arch/x86/kernel/irq.c:240
[<ffffffff86653d4c>] common_interrupt+0x8c/0x8c arch/x86/entry/entry_64.S:520
<EOI>
[< inline >] rcu_lock_acquire include/linux/rcupdate.h:490
[< inline >] rcu_read_lock include/linux/rcupdate.h:874
[<ffffffff8164b4a1>] filemap_map_pages+0x131/0xba0 mm/filemap.c:2145
[< inline >] do_fault_around mm/memory.c:2943
[< inline >] do_read_fault mm/memory.c:2962
[< inline >] do_fault mm/memory.c:3133
[< inline >] handle_pte_fault mm/memory.c:3308
[< inline >] __handle_mm_fault mm/memory.c:3418
[<ffffffff816efb16>] handle_mm_fault+0x2516/0x49a0 mm/memory.c:3447
[<ffffffff8127dc16>] __do_page_fault+0x376/0x960 arch/x86/mm/fault.c:1238
[<ffffffff8127e358>] trace_do_page_fault+0xe8/0x420 arch/x86/mm/fault.c:1331
[<ffffffff8126f514>] do_async_page_fault+0x14/0xd0 arch/x86/kernel/kvm.c:264
[<ffffffff86655578>] async_page_fault+0x28/0x30 arch/x86/entry/entry_64.S:986
Fix it by ensuring that the polling path is holding the host lock
before entering ata_sff_hsm_move() so that all hardware accesses and
state updates are performed under the host lock.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-and-tested-by: Dmitry Vyukov <dvyukov@google.com>
Link: http://lkml.kernel.org/g/CACT4Y+b_JsOxJu2EZyEf+mOXORc_zid5V1-pLZSroJVxyWdSpw@mail.gmail.com
Cc: stable@vger.kernel.org
8eee1d3ed5