Re: [PATCH 1/2] pinctrl: Add driver for Awinic AW9523/B I2C GPIO Expander

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Sun, Jan 10, 2021 at 3:32 PM AngeloGioacchino Del Regno
<angelogioacchino.delregno@xxxxxxxxxxxxxx> wrote:

> So, I've retried some basic usage of the regcache, relevant snippets here:
> static bool aw9523_volatile_reg(struct device *dev, unsigned int reg)
> {
>
>         return reg == AW9523_REG_IN_STATE(0) ||
>                reg == AW9523_REG_IN_STATE(AW9523_PINS_PER_PORT) ||
>                reg == AW9523_REG_CHIPID;
> }
(...)
> Since REG_IN_STATE is used to read the GPIO input level, it's not
> cacheable,

Fair enough.

> then CHIPID was set as not cacheable for safety: that may be
> avoided, but that may make no sense.. since it's a one-time readout for
> init putposes, it'd be useless to keep it cached.

I guess.

> Then, the set_bit/clear_bit in aw9523_irq_mask(), aw9523_irq_unmask were
> replaced with calls to regmap_update_bits_async, example:
>
>         regmap_update_bits_async(awi->regmap,
>                                  AW9523_REG_INTR_DIS(d->hwirq),
>                                  BIT(n), BIT(n));
>
> Where of course the value is either BIT(n) or 0 for mask and unmask
> respectively.
> Also, the bus_sync_unlock callback was changed as follows:
>
> static void aw9523_irq_bus_sync_unlock(struct irq_data *d)
>
> {
>       struct aw9523 *awi = gpiochip_get_data(irq_data_get_irq_chip_data(d));
>       regcache_mark_dirty(awi->regmap);
>       regcache_sync_region(awi->regmap, AW9523_REG_INTR_DIS(0),
>                            AW9523_REG_INTR_DIS(AW9523_PINS_PER_PORT));
>        mutex_unlock(&awi->irq->lock);
(...)
> One of the biggest / oddest issues that I get when trying to use
> regcache is that I'm getting badbadbad scheduling while atomic warnings
> all over and I don't get why, since regcache_default_sync is just
> calling _regmap_write, which is exactly what (non _prefix) regmap_write
> also calls...

OK that is the real problem to solve then.

> As a reference, this is one out of "many" (as you can imagine) stacktraces:
>
> <3>[    1.061428] BUG: scheduling while atomic: kworker/3:1/119/0x00000000
(...)
> <4>[    1.063134]  wait_for_completion_timeout+0x8c/0x110
> <4>[    1.063257]  qup_i2c_wait_for_complete.isra.18+0x1c/0x80
> <4>[    1.063429]  qup_i2c_xfer_v2_msg+0x2d4/0x3f0
> <4>[    1.063543]  qup_i2c_xfer_v2+0x290/0xa28
> <4>[    1.063652]  __i2c_transfer+0x16c/0x380
> <4>[    1.063798]  i2c_transfer+0x5c/0x138
> <4>[    1.063903]  i2c_transfer_buffer_flags+0x58/0x80
> <4>[    1.064060]  regmap_i2c_write+0x1c/0x50
> <4>[    1.064168]  _regmap_raw_write_impl+0x35c/0x688
> <4>[    1.064285]  _regmap_bus_raw_write+0x64/0x80
> <4>[    1.064440]  _regmap_write+0x58/0xa8
> <4>[    1.064545]  regcache_default_sync+0xcc/0x1a0
> <4>[    1.064660]  regcache_sync_region+0xdc/0xe8
> <4>[    1.064811]  aw9523_irq_bus_sync_unlock+0x30/0x48
> <4>[    1.064931]  __setup_irq+0x798/0x890
> <4>[    1.065034]  request_threaded_irq+0xe0/0x198
> <4>[    1.065188]  devm_request_threaded_irq+0x78/0xf8
> <4>[    1.065311]  gpio_keyboard_probe+0x2a8/0x468

scheduling while atomic happens when this trace gets called with interrupts
disabled, usually because someone has taken a spinlock.

Looking in __setup_irq() it looks safe.

I would turn on lock debugging (lockdep) and see if I can find it that way.

Yours,
Linus Walleij



[Index of Archives]     [Device Tree Compilter]     [Device Tree Spec]     [Linux Driver Backports]     [Video for Linux]     [Linux USB Devel]     [Linux PCI Devel]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Yosemite Backpacking]


  Powered by Linux