From mboxrd@z Thu Jan  1 00:00:00 1970
From: jiri.prchal@aksignal.cz (=?UTF-8?B?SmnFmcOtIFByY2hhbA==?=)
Date: Mon, 10 Mar 2014 09:11:55 +0100
Subject: [BUG] atmel: spi: scheduling while atomic
In-Reply-To: <CAH+eYFCT-VSP0Ej8jOYVGNgf=ONUjcFs1M8O0HOGNRNX23SW6A@mail.gmail.com>
References: <5315E33C.5000205@aksignal.cz> <20140304214039.GA4441@piout.net>
 <5316CA84.4080907@aksignal.cz> <20140307225810.GB2918@piout.net>
 <CAH+eYFCT-VSP0Ej8jOYVGNgf=ONUjcFs1M8O0HOGNRNX23SW6A@mail.gmail.com>
Message-ID: <531D73CB.7040402@aksignal.cz>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

Hi all,
at first: warning is on all spi devices.
at second: its with CONFIG_PREEMPT=y, without (# CONFIG_PREEMPT_RCU is not set) is OK, but I wonder if this config will 
be fast enough for our usage (nearly RT)?


Dne 8.3.2014 01:41, Rabin Vincent napsal(a):
> 2014-03-07 23:58 GMT+01:00 Alexandre Belloni
> <alexandre.belloni@free-electrons.com>:
>> On 05/03/2014 at 07:56:04 +0100, Ji?? Prchal wrote :
>>> [    0.902343] BUG: scheduling while atomic: spi0/383/0x00000002
>>> [    0.906250] Modules linked in:
>>> [    0.906250] CPU: 0 PID: 383 Comm: spi0 Not tainted 3.14.0-rc4_cpm9g25+ #1
>>> [    0.906250] [<c000d9f4>] (unwind_backtrace) from [<c000bdc0>] (show_stack+0x10/0x14)
>>> [    0.906250] [<c000bdc0>] (show_stack) from [<c003a224>] (__schedule_bug+0x48/0x60)
>>> [    0.906250] [<c003a224>] (__schedule_bug) from [<c0390294>] (__schedule+0x60/0x484)
>>> [    0.906250] [<c0390294>] (__schedule) from [<c038feac>] (schedule_timeout+0x17c/0x1ac)
>>> [    0.906250] [<c038feac>] (schedule_timeout) from [<c039111c>] (wait_for_common+0x10c/0x1f0)
>>> [    0.906250] [<c039111c>] (wait_for_common) from [<c0249228>] (atmel_spi_transfer_one_message+0x71c/0xa48)
>>> [    0.906250] [<c0249228>] (atmel_spi_transfer_one_message) from [<c0246624>] (spi_pump_messages+0x210/0x238)
>>> [    0.906250] [<c0246624>] (spi_pump_messages) from [<c00343d0>] (kthread_worker_fn+0x15c/0x1b4)
>>> [    0.906250] [<c00343d0>] (kthread_worker_fn) from [<c0034534>] (kthread+0xb8/0xcc)
>>> [    0.906250] [<c0034534>] (kthread) from [<c0009510>] (ret_from_fork+0x14/0x24)
>>> [    0.906250] at25 spi0.0: 128 KByte at25 eeprom, pagesize 512
>>
>> Actually, I'm not quite sure how spi_pump_messages can be called from an
>> atomic context.
>
> spi_pump_messages() is not called from atomic context.  Rather,
> atmel_spi_transfer_one_message() does a spin_lock_irqsave() (via
> atmel_spi_lock()) and then calls atmel_spi_one_transfer(), which calls
> wait_for_completion_timeout().  That's why schedule is being called
> from atomic context.  This was introduced by 8090d6d1a4 ("spi: atmel:
> Refactor spi-atmel to use SPI framework queue"). I think you should
> see the warning with CONFIG_PREEMPT=y.
>