[Qemu-devel] [Bug 1783422] Re: qemu_clock_get_ns does not take into account icount_time_shift

[SilverOnemi]

Actually implementing this fix causes a linux boot to crash. So there
may be something more to this. I will investigate more.

[  892.294463] BUG: soft lockup - CPU#0 stuck for 347s! [swapper/0:1]
[  892.490416] Modules linked in:
[  892.635725] 
[  892.790816] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.16.0-6-armmp-lpae #1 
Debian 3.16.56-1+deb8u1
[  893.098786] task: ee059640 ti: ee05a000 task.ti: ee05a000
[  893.322395] PC is at vprintk_emit+0x2c8/0x650
[  893.525852] LR is at console_unlock+0x234/0x4dc
[  893.749395] pc : [<c00bdd18>]    lr : [<c00bd7a8>]    psr: 60000153
[  893.749395] sp : ee05bdc8  ip : ee05bd70  fp : ee05be2c
[  894.117789] r10: 00000001  r9 : 00000000  r8 : 00000006
[  894.329454] r7 : 00000000  r6 : 00000028  r5 : c0a3dd98  r4 : c0a3dd98
[  894.571266] r3 : c09b44cc  r2 : 60000153  r1 : 60000153  r0 : c0a3dd98
[  894.807621] Flags: nZCv  IRQs on  FIQs off  Mode SVC_32  ISA ARM  Segment 
kernel
[  895.089213] Control: 30c7387d  Table: 40003000  DAC: fffffffd
[  895.329468] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 3.16.0-6-armmp-lpae #1 
Debian 3.16.56-1+deb8u1
[  895.720505] [<c002cf60>] (unwind_backtrace) from [<c002764c>] 
(show_stack+0x20/0x24)
[  896.065798] [<c002764c>] (show_stack) from [<c05ed770>] 
(dump_stack+0x9c/0xb0)
[  896.393691] [<c05ed770>] (dump_stack) from [<c0024370>] (show_regs+0x30/0x34)
[  896.713637] [<c0024370>] (show_regs) from [<c0103084>] 
(watchdog_timer_fn+0x18c/0x1e4)
[  897.062977] [<c0103084>] (watchdog_timer_fn) from [<c008d9c0>] 
(__run_hrtimer+0x88/0x2a8)
[  897.412317] [<c008d9c0>] (__run_hrtimer) from [<c008ecf0>] 
(hrtimer_run_queues+0xc0/0x1a4)
[  897.785790] [<c008ecf0>] (hrtimer_run_queues) from [<c0076ef0>] 
(update_process_times+0x3c/0x70)
[  898.169815] [<c0076ef0>] (update_process_times) from [<c00d5ed4>] 
(tick_periodic+0x4c/0xcc)
[  898.532622] [<c00d5ed4>] (tick_periodic) from [<c00d61b0>] 
(tick_handle_periodic+0x94/0x98)
[  898.895806] [<c00d61b0>] (tick_handle_periodic) from [<c04ad8fc>] 
(arch_timer_handler_virt+0x38/0x40)
[  899.292479] [<c04ad8fc>] (arch_timer_handler_virt) from [<c00c3b48>] 
(handle_percpu_devid_irq+0x80/0x194)
[  899.702210] [<c00c3b48>] (handle_percpu_devid_irq) from [<c00bf738>] 
(generic_handle_irq+0x3c/0x4c)
[  900.082647] [<c00bf738>] (generic_handle_irq) from [<c0023ba4>] 
(handle_IRQ+0x50/0xa0)
[  900.427431] [<c0023ba4>] (handle_IRQ) from [<c0008948>] 
(gic_handle_irq+0x3c/0x70)
[  900.767678] [<c0008948>] (gic_handle_irq) from [<c05f4180>] 
(__irq_svc+0x40/0x54)
[  901.037948] Exception stack(0xee05bd80 to 0xee05bdc8)
[  901.317607] bd80: c0a3dd98 60000153 60000153 c09b44cc c0a3dd98 c0a3dd98 
00000028 00000000
[  901.677252] bda0: 00000006 00000000 00000001 ee05be2c ee05bd70 ee05bdc8 
c00bd7a8 c00bdd18
[  901.970690] bdc0: 60000153 ffffffff
[  902.218760] [<c05f4180>] (__irq_svc) from [<c00bdd18>] 
(vprintk_emit+0x2c8/0x650)
[  902.571360] [<c00bdd18>] (vprintk_emit) from [<c05eaeb8>] (printk+0x44/0x4c)
[  902.916079] [<c05eaeb8>] (printk) from [<c00d14c0>] 
(__clocksource_select+0x12c/0x170)
[  903.290503] [<c00d14c0>] (__clocksource_select) from [<c08da7d4>] 
(clocksource_done_booting+0x38/0x4c)
[  903.671545] [<c08da7d4>] (clocksource_done_booting) from [<c0008d14>] 
(do_one_initcall+0xf0/0x234)
[  904.050606] [<c0008d14>] (do_one_initcall) from [<c08b2ec0>] 
(kernel_init_freeable+0x1c8/0x268)
[  904.433926] [<c08b2ec0>] (kernel_init_freeable) from [<c05e82ac>] 
(kernel_init+0x1c/0x100)
[  904.797782] [<c05e82ac>] (kernel_init) from [<c0023280>] 
(ret_from_fork+0x14/0x20)



** Changed in: ubuntu
       Status: New => Incomplete

-- 
You received this bug notification because you are a member of qemu-
devel-ml, which is subscribed to QEMU.
https://bugs.launchpad.net/bugs/1783422

Title:
  qemu_clock_get_ns does not take into account icount_time_shift

Status in QEMU:
  Incomplete

Bug description:
  Hello,

  If you check the qemu/util/qemu-timer.c you will find the following
  function:

  597: int64_t qemu_clock_get_ns(QEMUClockType type)
  598: {
  ....
  602:    switch (type) {
  ....
  606:    case QEMU_CLOCK_VIRTUAL:
  607:        if (use_icount) {
  608:            return cpu_get_icount(); 

  
  Now on line 606, in case we requested QEMU_CLOCK_VIRTUAL, and we are using 
icount, the value of cpu_get_icount(); will be returned.

  However if I understand correctly, in order to convert icount to ns,
  you must use take into account the icount shift -- as defined in the
  documentation: "The virtual cpu will execute one instruction every 2^N
  ns of virtual time.".

  Therefor, the correct value to return would be
  cpu_icount_to_ns(cpu_get_icount()), where cpu_icount_to_ns is defined
  in cpus.c:

  296: int64_t cpu_icount_to_ns(int64_t icount)
  297: {
  298:    return icount << icount_time_shift;
  299: }

  Best Regards,
  Humberto "SilverOne" Carvalho

To manage notifications about this bug go to:
https://bugs.launchpad.net/qemu/+bug/1783422/+subscriptions