Re: Bug with dwc3 id detect and regulators

[Chanwoo Choi]

Hi Roger,

On 2018ë 06ì 11ì 17:53, Roger Quadros wrote:
> Chanwoo,
> 
> On 11/06/18 11:33, H. Nikolaus Schaller wrote:
>> Hi Tony,
>> another bug...
>>
>> [  174.540313] BUG: scheduling while atomic: kworker/0:4/1327/0x00000002
>> [  174.547353] Modules linked in: omapdrm drm_kms_helper syscopyarea sysfillrect sysimgblt fb_sys_fops drm drm_panel_orientation_quirks bnep bluetooth ecdh_generic usb_f_ecm g_ether usb_f_rndis u_ether libcomposite configfs ipv6 arc4 wl18xx wlcore mac80211 panel_boe_w677l snd_soc_omap_hdmi_audio snd_soc_dmic cfg80211 dwc3 snd_soc_omap_abe_twl6040 snd_soc_twl6040 leds_gpio wwan_on_off connector_hdmi omapdss encoder_tpd12s015 cec pwm_omap_dmtimer omapdss_base pwm_bl generic_adc_battery ehci_omap wlcore_sdio dwc3_omap bmp280_spi snd_soc_ts3a227e crtouch_mt leds_is31fl319x tsc2007 bq2429x_charger bq27xxx_battery_i2c bq27xxx_battery ina2xx as5013 tca8418_keypad twl6040_vibra palmas_pwrbutton gpio_twl6040 palmas_gpadc usb3503 bmp280_i2c bmc150_accel_i2c w2cbw003_bluetooth bmc150_magn_i2c bmp280 bmc150_accel_core
>> [  174.624601]  bmc150_magn bno055 industrialio_triggered_buffer kfifo_buf industrialio snd_soc_omap_mcbsp snd_soc_omap_mcpdm snd_soc_omap snd_pcm_dmaengine [last unloaded: syscopyarea]
>> [  174.642327] CPU: 0 PID: 1327 Comm: kworker/0:4 Tainted: G        W         4.17.0-letux+ #2408
>> [  174.651541] Hardware name: Generic OMAP5 (Flattened Device Tree)
>> [  174.658004] Workqueue: events_power_efficient palmas_gpio_id_detect
>> [  174.664780] [<c01106ec>] (unwind_backtrace) from [<c010c058>] (show_stack+0x10/0x14)
>> [  174.673109] [<c010c058>] (show_stack) from [<c074ba68>] (dump_stack+0x7c/0x9c)
>> [  174.680881] [<c074ba68>] (dump_stack) from [<c0154050>] (__schedule_bug+0x60/0x84)
>> [  174.689006] [<c0154050>] (__schedule_bug) from [<c075e3c0>] (__schedule+0x50/0x694)
>> [  174.697217] [<c075e3c0>] (__schedule) from [<c075eab4>] (schedule+0xb0/0xcc)
>> [  174.704790] [<c075eab4>] (schedule) from [<c0761e58>] (schedule_timeout+0x354/0x3d4)
>> [  174.713100] [<c0761e58>] (schedule_timeout) from [<c075f544>] (wait_for_common+0x118/0x158)
>> [  174.722064] [<c075f544>] (wait_for_common) from [<c05de908>] (omap_i2c_xfer+0x354/0x48c)
>> [  174.730749] [<c05de908>] (omap_i2c_xfer) from [<c05d8410>] (__i2c_transfer+0x238/0x550)
>> [  174.739350] [<c05d8410>] (__i2c_transfer) from [<c05d87ac>] (i2c_transfer+0x84/0xb4)
>> [  174.747685] [<c05d87ac>] (i2c_transfer) from [<bf15b1f0>] (bq24296_i2c_reg8_read.constprop.8+0x54/0x64 [bq2429x_charger])
>> [  174.759465] [<bf15b1f0>] (bq24296_i2c_reg8_read.constprop.8 [bq2429x_charger]) from [<bf15b8f4>] (bq24296_update_reg+0x28/0xf8 [bq2429x_charger])
>> [  174.773437] [<bf15b8f4>] (bq24296_update_reg [bq2429x_charger]) from [<c0493b2c>] (_regulator_do_disable+0x100/0x238)
>> [  174.784804] [<c0493b2c>] (_regulator_do_disable) from [<c0493cec>] (_regulator_disable+0x88/0x120)
>> [  174.794404] [<c0493cec>] (_regulator_disable) from [<c0493db4>] (regulator_disable+0x30/0x60)
>> [  174.803556] [<c0493db4>] (regulator_disable) from [<bf1950c8>] (dwc3_omap_set_mailbox+0x84/0xf8 [dwc3_omap])
>> [  174.814124] [<bf1950c8>] (dwc3_omap_set_mailbox [dwc3_omap]) from [<bf195198>] (dwc3_omap_id_notifier+0x14/0x1c [dwc3_omap])
>> [  174.826149] [<bf195198>] (dwc3_omap_id_notifier [dwc3_omap]) from [<c014e458>] (notifier_call_chain+0x40/0x68)
>> [  174.836867] [<c014e458>] (notifier_call_chain) from [<c014e4a0>] (raw_notifier_call_chain+0x14/0x1c)
>> [  174.846661] [<c014e4a0>] (raw_notifier_call_chain) from [<c063bda4>] (extcon_sync+0x54/0x19c)
>> [  174.855804] [<c063bda4>] (extcon_sync) from [<c014874c>] (process_one_work+0x244/0x464)
>> [  174.864386] [<c014874c>] (process_one_work) from [<c0148f20>] (worker_thread+0x2c0/0x3ec)
>> [  174.873156] [<c0148f20>] (worker_thread) from [<c014cef4>] (kthread+0x134/0x150)
>> [  174.881087] [<c014cef4>] (kthread) from [<c01010e8>] (ret_from_fork+0x14/0x2c)
>>
>> It turns out that extcon_sync() holds a spinlock while sending its notifiers and this
>> is ending up in our regulator.en/disable() which wants to use blocking i2c.
>>
>> Do you see similar things on the OMAP5EVM when using OTG mode?
>> The Palmas SMPS10 is also handled through i2c. Or is this magically hidden by regmap?
>>
>> Well, as a workaround, I can make the regulator.en/disable() in the bq2429x driver
>> just trigger a worker, but IMHO it is not expected for regulator ops to be spinlock safe.
>>
>> So I think extcon should not spinlock (which might be against the extcon design) or
> 
> I think this something that should be addressed in the Extcon layer.
> Do you really need to call the raw_notifier_call_chain() function with spinlock held in extcon_sync()?
> if yes why?

To reduce the latency of notification is important of extcon.

So, extcon used the spinlock before calling the notifier_call_chain
to prevent the scheduled out for a moment. Also, the notifier_call
calls the registered notifier function sequentially. It means that
if some registered notifier function spends a lot of time,
the next registered notifier might receive the notification
with a delay. So, extcon used the spinlock. Unitl now, 

But, as you commented and previously, sometimes it caused the similar issues.
(To fix similar issues, the consumer device used the workqueue)

I will call notifier function without spinlock as following
and I'll consider the other way to notify the changed state to consumer device
with broadcasting way.

diff --git a/drivers/extcon/extcon.c b/drivers/extcon/extcon.c
index 8bff5fd18185..1e35b8ca3951 100644
--- a/drivers/extcon/extcon.c
+++ b/drivers/extcon/extcon.c
@@ -433,8 +433,8 @@ int extcon_sync(struct extcon_dev *edev, unsigned int id)
                return index;
 
        spin_lock_irqsave(&edev->lock, flags);
-
        state = !!(edev->state & BIT(index));
+       spin_unlock_irqrestore(&edev->lock, flags);
 
        /*
         * Call functions in a raw notifier chain for the specific one
@@ -448,6 +448,8 @@ int extcon_sync(struct extcon_dev *edev, unsigned int id)
         */
        raw_notifier_call_chain(&edev->nh_all, state, edev);
 
+       spin_lock_irqsave(&edev->lock, flags);
+

> 
> I think we don't want to call all notifiers in atomic context as this would keep interrupts disabled
> for quite some time and is suboptimal.
> 
>> dwc3_omap_set_mailbox should move dis/enabling regulator to some worker thread so
>> that they can block.
>>
>> The best would be to make dwc3_omap_set_mailbox call regulator_enable_deferred(omap->vbus_reg, 0)
>> but that function does not exist.
>>
>> Any ideas?
>>
>> BR,
>> Nikolaus
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-omap" in
>> the body of a message to majordomo@xxxxxxxxxxxxxxx
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
> 


-- 
Best Regards,
Chanwoo Choi
Samsung Electronics