[PATCH] ARM: imx: add cpuidle support for i.mx6sl

[John Tobias]

Does the time_after function inside the loop is not enough to handle
the situation?

On Thu, Jan 23, 2014 at 6:05 PM, Anson.Huang at freescale.com
<Anson.Huang at freescale.com> wrote:
> Hi, Tobias
>         Yes, the root cause is the reschedule introduced by the usleep, for our internal release, we use busy loop for PLL wait lock, just removing the usleep is not good enough, as the PLL hardware may need about 60~90us to lock, during this period, there will be no clock output in PLL. So, we need to figure out a way to fix this issue, either using busy loop instead of usleep or access the ccm register to adjust ARM divider instead of calling clk APIs.
>
> Best Regards.
> Anson huang 黄勇才
>
> Freescale Semiconductor Shanghai
> 上海浦东新区亮景路192号A座2楼
> 201203
> Tel:021-28937058
>
>
>>-----Original Message-----
>>From: John Tobias [mailto:john.tobias.ph at gmail.com]
>>Sent: Friday, January 24, 2014 4:17 AM
>>To: Huang Yongcai-B20788
>>Cc: <linux-arm-kernel at lists.infradead.org>; Shawn Guo
>>Subject: Re: [PATCH] ARM: imx: add cpuidle support for i.mx6sl
>>
>>Hi Anson / Shawn,
>>
>>I compared the clk_pllv3_wait_lock to the clk_pllv3_wait_for_lock of Freescale
>>alpha release. In alpha, the usleep_range did not exist. I tried removing it,
>>build the kernel again and it works.
>>I didn't get crashes anymore...
>>
>>Regards,
>>
>>john
>>
>>
>>
>>On Mon, Jan 20, 2014 at 1:03 AM, Anson.Huang at freescale.com
>><Anson.Huang at freescale.com> wrote:
>>> Hi, Tobias and Shawn
>>>         I debug into this issue today, the root cause is that we have a
>>usleep_range(50, 500); in arch/arm/mach-imx/clk-pllv3.c's clk_pllv3_wait_lock
>>function, which will cause kernel schedule during cpufreq change, and the idle
>>thread has another clk operation which will cause mutex nest, and it will try
>>to wake up previous mutex hold by cpufreq change's pll1 clk_set_rate.
>>>         So, to fix this issue, we should not use any usleep in the wait
>>function of pllv3's code, or we should not call any clk API in the
>>imx6sl_set_wait_clk function. What do you think?
>>>
>>>         BTW, I found another issue, if EVK board boot up with 396MHz, then
>>the initialized clock parent status is wrong, as ARM is from PFD 396MHz, I
>>will generate another patch to fix that once I have bandwidth.
>>>
>>> Best Regards.
>>> Anson huang 黄勇才
>>>
>>> Freescale Semiconductor Shanghai
>>> 上海浦东新区亮景路192号A座2楼
>>> 201203
>>> Tel:021-28937058
>>>
>>>
>>>>-----Original Message-----
>>>>From: Huang Yongcai-B20788
>>>>Sent: Saturday, January 18, 2014 7:53 AM
>>>>To: John Tobias
>>>>Cc: <linux-arm-kernel at lists.infradead.org>; Shawn Guo
>>>>Subject: Re: [PATCH] ARM: imx: add cpuidle support for i.mx6sl
>>>>
>>>>Hi, Tobias
>>>>        I will debug it next week and feedback to you. Thanks.
>>>>
>>>>Sent from Anson's iPhone
>>>>
>>>>> 在 2014年1月18日，7:42，"John Tobias" <john.tobias.ph at gmail.com> 写道：
>>>>>
>>>>> Hi Anson,
>>>>>
>>>>> My kernel for iMX6SL has a imx6q-cpufreq and I used your patch.
>>>>> Unfortunately, the kernel crashes if the imx6sl_set_wait_clk being
>>>>> called in imx6sl_enter_wait.
>>>>>
>>>>> [  288.166905] [<80044b80>] (dequeue_task+0x0/0xc8) from
>>>>> [<80045650>]
>>>>> (deactivate_task+0x30/0x34)
>>>>> [  288.176403] [<80045620>] (deactivate_task+0x0/0x34) from
>>>>> [<8050ac88>] (__schedule+0x318/0x58c) [  288.185848] [<8050a970>]
>>>>> (__schedule+0x0/0x58c) from [<8050af34>]
>>>>> (schedule+0x38/0x88)
>>>>> [  288.194758] [<8050aefc>] (schedule+0x0/0x88) from [<8050b1c8>]
>>>>> (schedule_preempt_disabled+0x10/0x14)
>>>>> [  288.204858] [<8050b1b8>] (schedule_preempt_disabled+0x0/0x14)
>>>>> from [<8050bcd4>] (mutex_lock_nested+0x16c/0x334) [  288.215850]
>>>>> [<8050bb68>] (mutex_lock_nested+0x0/0x334) from [<8034763c>]
>>>>> (clk_prepare_lock+0x90/0x104) [  288.226041] [<803475ac>]
>>>>> (clk_prepare_lock+0x0/0x104) from [<80349254>]
>>>>> (clk_set_rate+0x1c/0xbc) [  288.235535]  r6:806fd360 r5:00000000
>>>>> r4:bf817f80 r3:80713d80 [  288.242679] [<80349238>]
>>>>> (clk_set_rate+0x0/0xbc) from [<8001e574>]
>>>>> (imx6sl_set_wait_clk+0x28/0x70)
>>>>> [  288.252354]  r5:00000043 r4:00000001 [  288.256585] [<8001e54c>]
>>>>> (imx6sl_set_wait_clk+0x0/0x70) from [<8001def4>]
>>>>> (imx6sl_enter_wait+0x24/0x2c) [  288.266684]  r5:00000043
>>>>> r4:00000001 [  288.270949] [<8001ded0>] (imx6sl_enter_wait+0x0/0x2c)
>>>>> from [<80317608>] (cpuidle_enter_state+0x44/0xfc) [  288.281051]
>>>>> r4:13bf2645 r3:8001ded0 [  288.285682] [<803175c4>]
>>>>> (cpuidle_enter_state+0x0/0xfc) from [<803177bc>]
>>>>> (cpuidle_idle_call+0xfc/0x150) [  288.295871]  r8:806e00d8
>>>>> r7:00000001
>>>>> r6:00000000 r5:80c58034 r4:806fd360 [  288.304428] [<803176c0>]
>>>>> (cpuidle_idle_call+0x0/0x150) from [<8000fa68>]
>>>>> (arch_cpu_idle+0x10/0x44) [  288.314108]  r9:8070970a r8:8070970a
>>>>> r7:806d2000 r6:806da0dc r5:806d2000
>>>>> r4:806d2000
>>>>> [  288.323611] [<8000fa58>] (arch_cpu_idle+0x0/0x44) from
>>>>> [<800590f4>]
>>>>> (cpu_startup_entry+0xe0/0x120)
>>>>> [  288.333460] [<80059014>] (cpu_startup_entry+0x0/0x120) from
>>>>> [<80503c8c>] (rest_init+0xcc/0xdc) [  288.342810]  r7:ffffffff
>>>>> r3:00000000 [  288.347115] [<80503bc0>] (rest_init+0x0/0xdc) from
>>>>> [<806a2b00>]
>>>>> (start_kernel+0x348/0x354)
>>>>> [  288.356099]  r6:806da040 r5:806da040 r4:806da150 [  288.361972]
>>>>> [<806a27b8>] (start_kernel+0x0/0x354) from [<80008070>]
>>>>> (0x80008070)
>>>>>
>>>>>
>>>>> Regards,
>>>>>
>>>>> john
>>>>>
>>>>>
>>
>