lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  linux-hardening  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Sun, 10 Oct 2021 17:10:47 +0200
From:   Hans de Goede <hdegoede@...hat.com>
To:     "Rafael J . Wysocki" <rjw@...ysocki.net>,
        Mark Gross <markgross@...nel.org>,
        Andy Shevchenko <andy@...radead.org>,
        Daniel Scally <djrscally@...il.com>,
        Laurent Pinchart <laurent.pinchart@...asonboard.com>,
        Mauro Carvalho Chehab <mchehab@...nel.org>,
        Liam Girdwood <lgirdwood@...il.com>,
        Mark Brown <broonie@...nel.org>,
        Michael Turquette <mturquette@...libre.com>,
        Stephen Boyd <sboyd@...nel.org>
Cc:     Len Brown <lenb@...nel.org>, linux-acpi@...r.kernel.org,
        platform-driver-x86@...r.kernel.org, linux-kernel@...r.kernel.org,
        Sakari Ailus <sakari.ailus@...ux.intel.com>,
        Kate Hsuan <hpa@...hat.com>, linux-media@...r.kernel.org,
        linux-clk@...r.kernel.org
Subject: Re: [PATCH v2 02/13] ACPI: honor dependencies for devices with a _DEP
 pointing to an INT3472 device

Hi,

On 10/9/21 6:05 PM, Hans de Goede wrote:
> The clk and regulator frameworks expect clk/regulator consumer-devices
> to have info about the consumed clks/regulators described in the device's
> fw_node.
> 
> To work around cases where this info is not present in the firmware tables,
> which is often the case on x86/ACPI devices, both frameworks allow the
> provider-driver to attach info about consumers to the clks/regulators
> when registering these.
> 
> This causes problems with the probe ordering wrt drivers for consumers
> of these clks/regulators. Since the lookups are only registered when the
> provider-driver binds, trying to get these clks/regulators before then
> results in a -ENOENT error for clks and a dummy regulator for regulators.
> 
> One case where we hit this issue is camera sensors such as e.g. the OV8865
> sensor found on the Microsoft Surface Go. The sensor uses clks, regulators
> and GPIOs provided by a TPS68470 PMIC which is described in an INT3472
> ACPI device. There is special platform code handling this and setting
> platform_data with the necessary consumer info on the MFD cells
> instantiated for the PMIC under: drivers/platform/x86/intel/int3472.
> 
> For this to work properly the ov8865 driver must not bind to the i2c-client
> for the OV8865 sensor until after the TPS68470 PMIC gpio, regulator and
> clk MFD cells have all been fully setup.
> 
> The OV8865 on the Microsoft Surface Go is just one example, all X86
> devices using the Intel IPU3 camera block found on recent Intel SoCs
> have similar issues where there is an INT3472 HID ACPI-device which
> describes the clks and regulators and the driver for this INT3472 device
> must be fully initialized before the sensor driver (any sensor driver)
> binds for things to work properly.
> 
> On these devices the ACPI nodes describing the sensors all have a _DEP
> dependency on the matching INT3472 ACPI device (there is one per sensor).
> 
> This allows solving the probe-ordering problem by making ACPI-devices
> honor the _DEP dependencies (like we already do for batteries) when they
> have a dependency on an INT3472 device.
> 
> Note the matching is done on there being a _DEP on an INT3472 device,
> rather then matching on the HID of the sensor device itself to avoid
> having to maintain an ever growing list of HIDs of sensors which need
> the honor_dep behavior.
> 
> Signed-off-by: Hans de Goede <hdegoede@...hat.com>

In the end I was not entirely happy with the approach in patch 1 + 2
(this patch) myself, reporting 0 for the status of some devices until
the _DEP-s are met has some troublesome implications for code which
behavior depends on acpi_dev_is_present() calls (or equivalent code).

Which led to the: "[PATCH v2 13/13] media: ipu3-cio2: Add module
soft-deps for the INT3472 drivers" adding unnecessary soft
module-dependencies to the ipu3-cio2 code to make sure that the
status was not reported as 0 because of unmet deps when that code
runs.

Besides these soft-deps being non-sense / undesirable. This also
is a bit racy since the status only gets updated to reflect the
unmet_deps==0 once the acpi_scan_clear_dep() workqueue work
has run and there is no guarantee the work has fully run
once the modprobe-s of the INT3472 are done (it should run
soon afterwards, but there is a race there).

So I'm working on a rewrite of patch 1 + 2 now which replaces
the approach where we "lie" about the status with deferring the
enumeration step (1) until all the _DEP-s are met (for select
devices).

Expect a v3 with this new approach soon...

Regards,

Hans



1) The step where a platform_device or e.g. i2c-client is
instantiated as physical_device_node for the ACPI-device




> ---
>  drivers/acpi/scan.c     | 16 ++++++++++++++--
>  include/acpi/acpi_bus.h |  1 +
>  2 files changed, 15 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/acpi/scan.c b/drivers/acpi/scan.c
> index 4e0a946b35ed..976724540197 100644
> --- a/drivers/acpi/scan.c
> +++ b/drivers/acpi/scan.c
> @@ -796,6 +796,12 @@ static const char * const acpi_ignore_dep_ids[] = {
>  	NULL
>  };
>  
> +/* List of HIDs for which we honor deps of matching ACPI devs, when checking _DEP lists. */
> +static const char * const acpi_honor_dep_ids[] = {
> +	"INT3472", /* Camera sensor PMIC / clk and regulator info */
> +	NULL
> +};
> +
>  static struct acpi_device *acpi_bus_get_parent(acpi_handle handle)
>  {
>  	struct acpi_device *device = NULL;
> @@ -1761,8 +1767,12 @@ static void acpi_scan_dep_init(struct acpi_device *adev)
>  		adev->honor_deps = true;
>  
>  	list_for_each_entry(dep, &acpi_dep_list, node) {
> -		if (dep->consumer == adev->handle)
> +		if (dep->consumer == adev->handle) {
> +			if (dep->honor_dep)
> +				adev->honor_deps = true;
> +
>  			adev->dep_unmet++;
> +		}
>  	}
>  }
>  
> @@ -1966,7 +1976,7 @@ static u32 acpi_scan_check_dep(acpi_handle handle, bool check_dep)
>  	for (count = 0, i = 0; i < dep_devices.count; i++) {
>  		struct acpi_device_info *info;
>  		struct acpi_dep_data *dep;
> -		bool skip;
> +		bool skip, honor_dep;
>  
>  		status = acpi_get_object_info(dep_devices.handles[i], &info);
>  		if (ACPI_FAILURE(status)) {
> @@ -1975,6 +1985,7 @@ static u32 acpi_scan_check_dep(acpi_handle handle, bool check_dep)
>  		}
>  
>  		skip = acpi_info_matches_ids(info, acpi_ignore_dep_ids);
> +		honor_dep = acpi_info_matches_ids(info, acpi_honor_dep_ids);
>  		kfree(info);
>  
>  		if (skip)
> @@ -1988,6 +1999,7 @@ static u32 acpi_scan_check_dep(acpi_handle handle, bool check_dep)
>  
>  		dep->supplier = dep_devices.handles[i];
>  		dep->consumer = handle;
> +		dep->honor_dep = honor_dep;
>  
>  		mutex_lock(&acpi_dep_list_lock);
>  		list_add_tail(&dep->node , &acpi_dep_list);
> diff --git a/include/acpi/acpi_bus.h b/include/acpi/acpi_bus.h
> index 0ba344a5f4f8..b6fb050e77bb 100644
> --- a/include/acpi/acpi_bus.h
> +++ b/include/acpi/acpi_bus.h
> @@ -284,6 +284,7 @@ struct acpi_dep_data {
>  	struct list_head node;
>  	acpi_handle supplier;
>  	acpi_handle consumer;
> +	bool honor_dep;
>  };
>  
>  /* Performance Management */
> 

Powered by blists - more mailing lists