Re: [PATCH v7 10/24] wfx: add fwio.c/fwio.h

[Jérôme Pouiller]

On Friday 1 October 2021 18:08:32 CEST Pali Rohár wrote:
> On Friday 01 October 2021 17:09:41 Jérôme Pouiller wrote:
> > On Friday 1 October 2021 13:58:38 CEST Kalle Valo wrote:
> > > Jerome Pouiller <Jerome.Pouiller@xxxxxxxxxx> writes:
> > >
> > > > From: Jérôme Pouiller <jerome.pouiller@xxxxxxxxxx>
> > > >
> > > > Signed-off-by: Jérôme Pouiller <jerome.pouiller@xxxxxxxxxx>
> > >
> > > [...]
> > >
> > > > +static int get_firmware(struct wfx_dev *wdev, u32 keyset_chip,
> > > > +                     const struct firmware **fw, int *file_offset)
> > > > +{
> > > > +     int keyset_file;
> > > > +     char filename[256];
> > > > +     const char *data;
> > > > +     int ret;
> > > > +
> > > > +     snprintf(filename, sizeof(filename), "%s_%02X.sec",
> > > > +              wdev->pdata.file_fw, keyset_chip);
> > > > +     ret = firmware_request_nowarn(fw, filename, wdev->dev);
> > > > +     if (ret) {
> > > > +             dev_info(wdev->dev, "can't load %s, falling back to %s.sec\n",
> > > > +                      filename, wdev->pdata.file_fw);
> > > > +             snprintf(filename, sizeof(filename), "%s.sec",
> > > > +                      wdev->pdata.file_fw);
> > > > +             ret = request_firmware(fw, filename, wdev->dev);
> > > > +             if (ret) {
> > > > +                     dev_err(wdev->dev, "can't load %s\n", filename);
> > > > +                     *fw = NULL;
> > > > +                     return ret;
> > > > +             }
> > > > +     }
> > >
> > > How is this firmware file loading supposed to work? If I'm reading the
> > > code right, the driver tries to load file "wfm_wf200_??.sec" but in
> > > linux-firmware the file is silabs/wfm_wf200_C0.sec:
> > >
> > > https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git/tree/silabs
> > >
> > > That can't work automatically, unless I'm missing something of course.
> >
> > The firmware are signed. "C0" is the key used to sign this firmware. This
> > key must match with the key burned into the chip. Fortunately, the driver
> > is able to read the key accepted by the chip and automatically choose the
> > right firmware.
> >
> > We could imagine to add a attribute in the DT to choose the firmware to
> > load. However, it would be a pity to have to specify it manually whereas
> > the driver is able to detect it automatically.
> >
> > Currently, the only possible key is C0. However, it exists some internal
> > parts with other keys. In addition, it is theoretically possible to ask
> > to Silabs to burn parts with a specific key in order to improve security
> > of a product.
> >
> > Obviously, for now, this feature mainly exists for the Silabs firmware
> > developers who have to work with other keys.
> >
> > > Also I would prefer to use directory name as the driver name wfx, but I
> > > guess silabs is also doable.
> >
> > I have no opinion.
> >
> >
> > > Also I'm not seeing the PDS files in linux-firmware. The idea is that
> > > when user installs an upstream kernel and the linux-firmware everything
> > > will work automatically, without any manual file installations.
> >
> > WF200 is just a chip. Someone has to design an antenna before to be able
> > to use.
> >
> > However, we have evaluation boards that have antennas and corresponding
> > PDS files[1]. Maybe linux-firmware should include the PDS for these boards
> 
> So chip vendor provides firmware and card vendor provides PDS files.

Exactly.

> In
> my opinion all files should go into linux-firmware repository. If Silabs
> has PDS files for its devel boards (which are basically cards) then I
> think these files should go also into linux-firmware repository.
> 
> And based on some parameter, driver should load correct PDS file. Seems
> like DT can be a place where to put something which indicates which PDS
> file should be used.
> 
> But should be in DT directly name of PDS file? Or should be in DT just
> additional compatible string with card vendor name and then in driver
> itself should be mapping table from compatible string to filename? I do
> not know what is better.

The DT already accepts the attribute silabs,antenna-config-file (see
patch #2).

I think that linux-firmware repository will reject the pds files if
no driver in the kernel directly point to it. Else how to detect orphans?
So, I think it is slightly better to use a mapping table.


> > and the DT should contains the name of the design. eg:
> >
> >     compatible = "silabs,brd4001a", "silabs,wf200";
> >
> > So the driver will know which PDS it should use.
> >
> > In fact, I am sure I had this idea in mind when I have started to write
> > the wfx driver. But with the time I have forgotten it.
> >
> > If you agree with that idea, I can work on it next week.
> >
> >
> > [1]: https://github.com/SiliconLabs/wfx-pds


-- 
Jérôme Pouiller