Division by zero caused by CCF

Tue Jul 16 13:45:38 EDT 2013

On Tue, Jul 16, 2013 at 6:10 AM, Eduardo Valentin
<eduardo.valentin at ti.com> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA256
>
>
> Hi,
>
> Adding Mike's correct address.
>
> On 16-07-2013 08:37, Felipe Balbi wrote:
>> Hi,
>>
>> trying to get USB host verified on OMAP5 uEVM with v3.11-rc1. The
>> clk_set_rate() call ends up in a division by zero, which is quite
>> interesting provided the driver will only call clk_set_rate() if the
>> clock is valid and clk_rate is != 0.
>>
>>
>> [   22.009238] Division by zero in kernel.
>> [   22.009250] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G        W    3.11.0-rc1-00081-g3310d44-dirty #118
>> [   22.009275] [<c001c83c>] (unwind_backtrace+0x0/0xf0) from [<c0018a1c>] (show_stack+0x10/0x14)
>> [   22.009289] [<c0018a1c>] (show_stack+0x10/0x14) from [<c057403c>] (dump_stack+0x70/0x8c)
>> [   22.009304] [<c057403c>] (dump_stack+0x70/0x8c) from [<c02e4154>] (Ldiv0+0x8/0x10)
>> [   22.009319] [<c02e4154>] (Ldiv0+0x8/0x10) from [<c048d460>] (clk_divider_set_rate+0x10/0xdc)
>> [   22.009331] [<c048d460>] (clk_divider_set_rate+0x10/0xdc) from [<c048c124>] (clk_change_rate+0x38/0xb0)
>> [   22.009341] [<c048c124>] (clk_change_rate+0x38/0xb0) from [<c048c20c>] (clk_set_rate+0x70/0xa8)
>> [   22.009354] [<c048c20c>] (clk_set_rate+0x70/0xa8) from [<c042b244>] (nop_usb_xceiv_probe+0x1fc/0x2f8)
>> [   22.009369] [<c042b244>] (nop_usb_xceiv_probe+0x1fc/0x2f8) from [<c036b47c>] (platform_drv_probe+0x18/0x1c)
>> [   22.009380] [<c036b47c>] (platform_drv_probe+0x18/0x1c) from [<c0369f44>] (really_probe+0x70/0x1f4)
>> [   22.009390] [<c0369f44>] (really_probe+0x70/0x1f4) from [<c036a1dc>] (driver_probe_device+0x30/0x48)
>> [   22.009401] [<c036a1dc>] (driver_probe_device+0x30/0x48) from [<c036a288>] (__driver_attach+0x94/0x98)
>> [   22.009411] [<c036a288>] (__driver_attach+0x94/0x98) from [<c0368748>] (bus_for_each_dev+0x54/0x88)
>> [   22.009420] [<c0368748>] (bus_for_each_dev+0x54/0x88) from [<c036972c>] (bus_add_driver+0xdc/0x29c)
>> [   22.009430] [<c036972c>] (bus_add_driver+0xdc/0x29c) from [<c036a760>] (driver_register+0x78/0x190)
>> [   22.009440] [<c036a760>] (driver_register+0x78/0x190) from [<c00087b0>] (do_one_initcall+0x34/0x164)
>> [   22.009453] [<c00087b0>] (do_one_initcall+0x34/0x164) from [<c07b18f4>] (do_basic_setup+0x90/0xc4)
>> [   22.009466] [<c07b18f4>] (do_basic_setup+0x90/0xc4) from [<c07b199c>] (kernel_init_freeable+0x74/0x110)
>> [   22.009478] [<c07b199c>] (kernel_init_freeable+0x74/0x110) from [<c05676c4>] (kernel_init+0x8/0xe4)
>> [   22.009491] [<c05676c4>] (kernel_init+0x8/0xe4) from [<c0014648>] (ret_from_fork+0x14/0x2c)
>>
>> I believe the problem is the actual division reaching
>> clk_divider_set_rate().
>>
>> drivers/clk/clk-divider.c::clk_divider_set_rate()
>>
>> | static int clk_divider_set_rate(struct clk_hw *hw, unsigned long rate,
>> |                                 unsigned long parent_rate)
>> | {
>> |         struct clk_divider *divider = to_clk_divider(hw);
>> |         unsigned int div, value;
>> |         unsigned long flags = 0;
>> |         u32 val;
>> |
>> |         div = parent_rate / rate;
>>
>> right here, but how come rate would zero provided driver checks for it
>> as below.
>>
>> drivers/usb/phy/phy-nop.c::nop_usb_xceiv_probe()
>>
>> |         if (!IS_ERR(nop->clk) && clk_rate) {
>> |                 err = clk_set_rate(nop->clk, clk_rate);
>> |                 if (err) {
>> |                         dev_err(&pdev->dev, "Error setting clock rate\n");
>> |                         return err;
>> |                 }
>> |         }
>>
>> I've added a few prints around CCF to try and track what's going on:
>>
>> [   21.592690] ====> nop_usb_xceiv_probe rate 19200000
>> [   21.592700] ====> clk_set_rate rate 19200000
>> [   21.592707] ====> clk_calc_new_rates rate 19200000
>> [   21.592713] ====> clk_divider_round_rate rate 19200000
>> [   21.592719] ====> clk_divider_bestdiv rate 19200000
>> [   21.592726] ====> clk_change_rate best_parent_rate 0
>
> or because we reach:
>         if (clk->ops->set_rate)
>                 clk->ops->set_rate(clk->hw, clk->new_rate, best_parent_rate);
>
> with clk->new_rate == 0.

Hmm, I'll look into this. We used to have a check which would at least
WARN on division by zero, but looks like that was replaced by some
other code at some point.

Also does your clock have the CLK_SET_RATE_PARENT flag set? If so then
you could be propagating a rate request of zero up to the next parent,
which would be a neat trick... however based on the dump that doesn't
seem to be what is happening.

Regards,
Mike

>
>
>> [   21.592732] ====> clk_divider_set_rate rate 0
>> [   21.592737] Division by zero in kernel.
>> [   21.592747] CPU: 0 PID: 1 Comm: swapper/0 Tainted: G        W    3.11.0-rc1-00081-g3310d44-dirty #121
>> [   21.592773] [<c001c83c>] (unwind_backtrace+0x0/0xf0) from [<c0018a1c>] (show_stack+0x10/0x14)
>> [   21.592787] [<c0018a1c>] (show_stack+0x10/0x14) from [<c057400c>] (dump_stack+0x70/0x8c)
>> [   21.592803] [<c057400c>] (dump_stack+0x70/0x8c) from [<c02e4154>] (Ldiv0+0x8/0x10)
>> [   21.592819] [<c02e4154>] (Ldiv0+0x8/0x10) from [<c048d3e0>] (clk_divider_set_rate+0x2c/0x100)
>> [   21.592831] [<c048d3e0>] (clk_divider_set_rate+0x2c/0x100) from [<c048c050>] (clk_change_rate+0x48/0xe0)
>> [   21.592841] [<c048c050>] (clk_change_rate+0x48/0xe0) from [<c048c174>] (clk_set_rate+0x8c/0xc0)
>> [   21.592855] [<c048c174>] (clk_set_rate+0x8c/0xc0) from [<c042b254>] (nop_usb_xceiv_probe+0x20c/0x304)
>> [   21.592869] [<c042b254>] (nop_usb_xceiv_probe+0x20c/0x304) from [<c036b47c>] (platform_drv_probe+0x18/0x1c)
>> [   21.592880] [<c036b47c>] (platform_drv_probe+0x18/0x1c) from [<c0369f44>] (really_probe+0x70/0x1f4)
>> [   21.592891] [<c0369f44>] (really_probe+0x70/0x1f4) from [<c036a1dc>] (driver_probe_device+0x30/0x48)
>> [   21.592901] [<c036a1dc>] (driver_probe_device+0x30/0x48) from [<c036a288>] (__driver_attach+0x94/0x98)
>> [   21.592911] [<c036a288>] (__driver_attach+0x94/0x98) from [<c0368748>] (bus_for_each_dev+0x54/0x88)
>> [   21.592921] [<c0368748>] (bus_for_each_dev+0x54/0x88) from [<c036972c>] (bus_add_driver+0xdc/0x29c)
>> [   21.592930] [<c036972c>] (bus_add_driver+0xdc/0x29c) from [<c036a760>] (driver_register+0x78/0x190)
>> [   21.592941] [<c036a760>] (driver_register+0x78/0x190) from [<c00087b0>] (do_one_initcall+0x34/0x164)
>> [   21.592954] [<c00087b0>] (do_one_initcall+0x34/0x164) from [<c07b18f4>] (do_basic_setup+0x90/0xc4)
>> [   21.592966] [<c07b18f4>] (do_basic_setup+0x90/0xc4) from [<c07b199c>] (kernel_init_freeable+0x74/0x110)
>> [   21.592980] [<c07b199c>] (kernel_init_freeable+0x74/0x110) from [<c0567694>] (kernel_init+0x8/0xe4)
>> [   21.592992] [<c0567694>] (kernel_init+0x8/0xe4) from [<c0014648>] (ret_from_fork+0x14/0x2c)
>>
>> even though driver passed 19.2MHz, best_parent_rate ends up being zero
>> which triggers the division by zero above.
>>
>> cheers
>>
>
>
> - --
> You have got to be excited about what you are doing. (L. Lamport)
>
> Eduardo Valentin
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.12 (GNU/Linux)
> Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
>
> iF4EAREIAAYFAlHlRl8ACgkQCXcVR3XQvP00XQEAtQgDEJLt8OFCJiIhUj46Zq1h
> PvNq67RSFTRXcq/zHa8A/0IZSPitTXt1TqDfalTKof/DR6n9/W6md8/C2Ovqb59o
> =AKnu
> -----END PGP SIGNATURE-----