Re: [PATCH v8 08/14] mfd: rk8xx: add rk806 support

[Sebastian Reichel]

Hi,

On Thu, Nov 16, 2023 at 09:41:03AM +0100, Neil Armstrong wrote:
> On 15/11/2023 19:00, Sebastian Reichel wrote:
> > On Wed, Nov 15, 2023 at 06:17:50PM +0100, Neil Armstrong wrote:
> > > On 04/05/2023 19:36, Sebastian Reichel wrote:
> > > > Add support for SPI connected rk806, which is used by the RK3588
> > > > evaluation boards. The PMIC is advertised to support I2C and SPI,
> > > > but the evaluation boards all use SPI. Thus only SPI support is
> > > > added here.
> > > > 
> > > > Acked-for-MFD-by: Lee Jones <lee@xxxxxxxxxx>
> > > > Tested-by: Diederik de Haas <didi.debian@xxxxxxxxx> # Rock64, Quartz64 Model A + B
> > > > Tested-by: Vincent Legoll <vincent.legoll@xxxxxxxxx> # Pine64 QuartzPro64
> > > > Signed-off-by: Sebastian Reichel <sebastian.reichel@xxxxxxxxxxxxx>
> > > > ---
> > > >    drivers/mfd/Kconfig       |  14 ++
> > > >    drivers/mfd/Makefile      |   1 +
> > > >    drivers/mfd/rk8xx-core.c  |  69 ++++++-
> > > >    drivers/mfd/rk8xx-spi.c   | 124 ++++++++++++
> > > >    include/linux/mfd/rk808.h | 409 ++++++++++++++++++++++++++++++++++++++
> > > >    5 files changed, 614 insertions(+), 3 deletions(-)
> > > >    create mode 100644 drivers/mfd/rk8xx-spi.c
> > > > 
> > > 
> > > <snip>
> > > 
> > > > -	ret = devm_mfd_add_devices(dev, PLATFORM_DEVID_NONE,
> > > > -			      cells, nr_cells, NULL, 0,
> > > > +	ret = devm_mfd_add_devices(dev, 0, cells, nr_cells, NULL, 0,
> > > >    			      regmap_irq_get_domain(rk808->irq_data));
> > > 
> > > It seems you replaced PLATFORM_DEVID_NONE by 0, triggering again the bug preventing
> > > having multiples RK pmics on the same system I fixed earlier at [1].
> > 
> > All cells have PLATFORM_DEVID_NONE specified and thus are registered
> > without an ID. I changed this bit to avoid overriding the
> > information, since I did not want to have PLATFORM_DEVID_NONE for
> > rk806.
> > 
> > > This gives (again):
> > > <4>[ 0.664107] sysfs: cannot create duplicate filename '/bus/platform/devices/rk808-clkout'
> > 
> > Which means, you do not want PLATFORM_DEVID_NONE (-1), but
> > PLATFORM_DEVID_AUTO (-2). The above path is the expected path
> > for PLATFORM_DEVID_NONE.
> > 
> > > <4>[ 0.664120] CPU: 3 PID: 97 Comm: kworker/u12:2 Not tainted 6.6.1 #1
> > > <4>[ 0.664131] Hardware name: Hardkernel ODROID-GO-Ultra (DT)
> > > <4>[ 0.664139] Workqueue: events_unbound deferred_probe_work_func
> > > <4>[ 0.664160] Call trace:
> > > <4>[ 0.664165] dump_backtrace+0x9c/0x11c
> > > <4>[ 0.664181] show_stack+0x18/0x24
> > > <4>[ 0.664193] dump_stack_lvl+0x78/0xc4
> > > <4>[ 0.664205] dump_stack+0x18/0x24
> > > <4>[ 0.664215] sysfs_warn_dup+0x64/0x80
> > > <4>[ 0.664227] sysfs_do_create_link_sd+0xf0/0xf8
> > > <4>[ 0.664239] sysfs_create_link+0x20/0x40
> > > <4>[ 0.664250] bus_add_device+0x114/0x160
> > > <4>[ 0.664259] device_add+0x3f0/0x7cc
> > > <4>[ 0.664267] platform_device_add+0x180/0x270
> > > <4>[ 0.664278] mfd_add_device+0x390/0x4a8
> > > <4>[ 0.664290] devm_mfd_add_devices+0xb0/0x150
> > > <4>[ 0.664301] rk8xx_probe+0x26c/0x410
> > > <4>[ 0.664312] rk8xx_i2c_probe+0x64/0x98
> > > <4>[ 0.664323] i2c_device_probe+0x104/0x2e8
> > > <4>[ 0.664333] really_probe+0x184/0x3c8
> > > <4>[ 0.664342] __driver_probe_device+0x7c/0x16c
> > > <4>[ 0.664351] driver_probe_device+0x3c/0x10c
> > > <4>[ 0.664360] __device_attach_driver+0xbc/0x158
> > > <4>[ 0.664369] bus_for_each_drv+0x80/0xdc
> > > <4>[ 0.664377] __device_attach+0x9c/0x1ac
> > > <4>[ 0.664386] device_initial_probe+0x14/0x20
> > > <4>[ 0.664395] bus_probe_device+0xac/0xb0
> > > <4>[ 0.664403] deferred_probe_work_func+0xa0/0xf4
> > > <4>[ 0.664412] process_one_work+0x1bc/0x378
> > > <4>[ 0.664421] worker_thread+0x1dc/0x3d4
> > > <4>[ 0.664429] kthread+0x104/0x118
> > > <4>[ 0.664440] ret_from_fork+0x10/0x20
> > > <3>[ 0.664494] rk8xx-i2c 0-001c: error -EEXIST: failed to add MFD devices
> > > <4>[ 0.666769] rk8xx-i2c: probe of 0-001c failed with error -17
> > 
> > I didn't notice when working on rk806, but after analyzing it now:
> > 
> > Your patch effectively set the cells to PLATFORM_DEVID_AUTO, because
> > you set all cells to PLATFORM_DEVID_NONE (-1) and additionally used
> > PLATFORM_DEVID_NONE (-1) for the devm_mfd_add_devices() call. But
> > that uses the sum of both IDs. Adding -1 to -1 is -2 and thus
> > PLATFORM_DEVID_AUTO. This is of course very confusing and just
> > worked by chance. There are two options:
> > 
> > 1. Modify all cells to use PLATFORM_DEVID_AUTO instead of
> > PLATFORM_DEVID_NONE
> > 2. Drop the .id from all cells and use PLATFORM_DEVID_AUTO in the
> > call to devm_mfd_add_devices()
> > 
> > Note, that switching from PLATFORM_DEVID_NONE to PLATFORM_DEVID_AUTO
> > modifies sysfs paths and thus might break people's scripts; that's why
> > I tried not to modify any existing platform. I will let you deal
> > with that, since I cannot even test any !rk806 platform supported by
> > this driver :)
> 
> Yes it will modify sysfs path, but it's a regression since this before this patch
> everything was registered with PLATFORM_DEVID_AUTO anyway,
> so I'll provide a fix adding back PLATFORM_DEVID_NONE to devm_mfd_add_devices
> in a first time...

If you just add back PLATFORM_DEVID_NONE to devm_mfd_add_devices(),
it will break rk806, since that correctly states PLATFORM_DEVID_AUTO
in its .id and that would become -3.

> > Also mfd_add_device should probably get special handling for
> > PLATFORM_DEVID_NONE, just like it already has special handling
> > for PLATFORM_DEVID_AUTO.
> 
> ... and yes thanks for the great analysis I'll provide a change
> cleaning the mess.

Thanks,

-- Sebastian

Attachment: signature.asc
Description: PGP signature