Re: [PATCH v3 1/2] tee: optee: Fix supplicant based device enumeration

[Sumit Garg]

On Thu, 2 Nov 2023 at 13:28, Jan Kiszka <jan.kiszka@xxxxxxxxxxx> wrote:
>
> On 02.11.23 08:37, Sumit Garg wrote:
> > On Tue, 31 Oct 2023 at 17:14, Jan Kiszka <jan.kiszka@xxxxxxxxxxx> wrote:
> >>
> >> On 31.10.23 12:04, Jerome Forissier wrote:
> >>>
> >>>
> >>> On 10/30/23 16:59, Sumit Garg wrote:
> >>>> Currently supplicant dependent optee device enumeration only registers
> >>>> devices whenever tee-supplicant is invoked for the first time. But it
> >>>> forgets to remove devices when tee-supplicant daemon stops running and
> >>>> closes its context gracefully. This leads to following error for fTPM
> >>>> driver during reboot/shutdown:
> >>>>
> >>>> [   73.466791] tpm tpm0: ftpm_tee_tpm_op_send: SUBMIT_COMMAND invoke error: 0xffff3024
> >>>>
> >>>> Fix this by separating supplicant dependent devices so that the
> >>>> user-space service can detach supplicant devices before closing the
> >>>> supplicant.
> >>>>
> >>>> Reported-by: Jan Kiszka <jan.kiszka@xxxxxxxxxxx>
> >>>> Link: https://github.com/OP-TEE/optee_os/issues/6094
> >>>> Fixes: 5f178bb71e3a ("optee: enable support for multi-stage bus enumeration")
> >>>> Tested-by: Jan Kiszka <jan.kiszka@xxxxxxxxxxx>
> >>>> Tested-by: Masahisa Kojima <masahisa.kojima@xxxxxxxxxx>
> >>>> Signed-off-by: Sumit Garg <sumit.garg@xxxxxxxxxx>
> >>>> ---
> >>>>  drivers/tee/optee/device.c | 13 ++++++++++---
> >>>>  1 file changed, 10 insertions(+), 3 deletions(-)
> >>>>
> >>>> diff --git a/drivers/tee/optee/device.c b/drivers/tee/optee/device.c
> >>>> index 64f0e047c23d..78fc0a15c463 100644
> >>>> --- a/drivers/tee/optee/device.c
> >>>> +++ b/drivers/tee/optee/device.c
> >>>> @@ -60,9 +60,10 @@ static void optee_release_device(struct device *dev)
> >>>>      kfree(optee_device);
> >>>>  }
> >>>>
> >>>> -static int optee_register_device(const uuid_t *device_uuid)
> >>>> +static int optee_register_device(const uuid_t *device_uuid, u32 func)
> >>>>  {
> >>>>      struct tee_client_device *optee_device = NULL;
> >>>> +    const char *dev_name_fmt = NULL;
> >>>>      int rc;
> >>>>
> >>>>      optee_device = kzalloc(sizeof(*optee_device), GFP_KERNEL);
> >>>> @@ -71,7 +72,13 @@ static int optee_register_device(const uuid_t *device_uuid)
> >>>>
> >>>>      optee_device->dev.bus = &tee_bus_type;
> >>>>      optee_device->dev.release = optee_release_device;
> >>>> -    if (dev_set_name(&optee_device->dev, "optee-ta-%pUb", device_uuid)) {
> >>>> +
> >>>> +    if (func == PTA_CMD_GET_DEVICES_SUPP)
> >>>> +            dev_name_fmt = "optee-ta-supp-%pUb";
> >>>> +    else
> >>>> +            dev_name_fmt = "optee-ta-%pUb";
> >>>
> >>> That's an ABI change, isn't it?
> >>
> >
> > Indeed it is an ABI break although we would like this to be backported
> > but don't want to break existing users. So I brainstormed on it and
> > came up with an alternative fix via device attribute in v4. Please
> > have a look.
> >
> >> Oh, here did this come from! Yes, I recently had to adjust some systemd
> >> service due to carrying this patch but looking for the change only in
> >> upstream:
> >>
> >> https://github.com/ilbers/isar/commit/83644ddf694e51f11793e6107e4aaf68dc0043a5
> >>
> >
> > You don't need to unbind all of the optee devices. v4 would help you
> > to maintain backwards compatibility, can you retest it?
>
> How do I know from tee-supplicant perspective which devices I need to
> unbind? There could be one in the future that will also use storage and
> will therefore also fail once the supplicant is gone.
>

With v4, the devices where the below attribute is present need to
unbind before closing tee-supplicant.

/sys/bus/tee/devices/optee-ta-<uuid>/need_supplicant

-Sumit

> Jan
>
> --
> Siemens AG, Technology
> Linux Expert Center
>