Hey Russ,

On Mon, Feb 27, 2023 at 10:56:07PM +0000, Conor Dooley wrote:
> On Mon, Feb 27, 2023 at 02:42:30PM -0800, Russ Weight wrote:
> > On 2/27/23 14:16, Conor Dooley wrote:
> > > On Mon, Feb 27, 2023 at 02:04:36PM -0800, Russ Weight wrote:
> > >> On 2/24/23 00:28, Conor Dooley wrote:
> > >>> On Fri, Feb 24, 2023 at 03:57:09PM +0800, Xu Yilun wrote:
> > >>>> On 2023-02-17 at 16:40:17 +0000, Conor Dooley wrote:
> > >>>>> This patchset adds support for the "Auto Update" feature on PolarFire
> > >>>>> SoC that allows for writing an FPGA bistream to the SPI flash connected
> > >>>>> to the system controller.
> > >>>> I haven't fully checked the patches yet, just some quick comments:
> > >>>>
> > >>>> Since this feature is just to R/W the flash, and would not affect the
> > >>>> runtime FPGA region, I don't think an FPGA manager is actually needed.
> > >>>> Why not just use the MTD uAPI? There is a set of exsiting MTD uAPI &
> > >>>> MTD tool if I remember correctly.
> > >>> A lack of interest in opening up the system controller to userspace!
> > >>> You're right in that the writing of the image can be done that way, and
> > >>> while I was testing I used the userspace bits of mtd along the way - but
> > >>> for validating that the image we are writing we rely on the system
> > >>> controller.
> > >>> I'm really not interested in exposing the system controller's
> > >>> functionality, especially the bitstream manipulation parts, to userspace
> > >>> due to the risk of input validation bugs, so at least that side of
> > >>> things should remain in the kernel.
> > >>> I suppose I could implement something custom in drivers/soc that does
> > >>> the validation only, and push the rest out to userspace. Just seemed
> > >>> fitting to do the whole lot in drivers/fpga.
> > >> In case you haven't already looked at the new firmware-upload
> > >> support in the firmware-loader, I'll give you some references
> > >> here to see if it fit you needs. This would only support the
> > >> write (not the read) but it would allow the controller to do
> > >> validation on the write.
> > >>
> > >> The is the cover letter which shows a usage example:
> > >> https://lore.kernel.org/lkml/20220421212204.36052-1-russell.h.weight@intel.com/
> > >>
> > >> And this is a pointer to the latest documentation for it:
> > >> https://docs.kernel.org/driver-api/firmware/fw_upload.html
> > >>
> > >> The only current user is: drivers/fpga/intel-m10-bmc-sec-update.c
> > > Sounds interesting, I shall go and take a look. Just quickly, when you
> > > say it wouldn't support the read, what exactly do you mean?
> > > The only read that I am really interested in doing is reading the first
> > > 1K of flash as I need to RMW it, but I don't think that that's what you
> > > mean.
> > > Do you mean that I would not be able to dump the firmware using your
> > > fw_upload interface? If so, that's an acceptable constraint IMO.
> > 
> > Yes - I mean that you couldn't dump the firmware image from userspace
> > using the fw_upload interface. The sysfs interface allows you to read
> > and write a temporary buffer, but once you "echo 0 > loading", there
> > is no sysfs interface associated with the firmware-loader that would
> > allow you to read the image from flash. Your controller actually does
> > the writes, so RMW in that context is fine.
> 
> Ahh cool. I don't really care about dumping the firmware via such a
> mechanism, so that sounds good to me.
> I'll check out your approach, the immediate thought is that it sounds
> suitable to my use case, so thanks!

Taken me a while to get around to it, but thanks for your suggestion.
Looks like it is suitable for what I am trying to do, so in the middle
of working on another version of this using fw_upload.
The enum return codes from write are a little clumsy, but oh well, could
be worse I suppose.

Just one thing I noted (although I rarely pay much attention to/rely on
the driver-api docs when recent drivers exist as usable examples) is
that the docs for this stuff is a wee bit out of date due to some API
changes.
In the code example in this document:
https://docs.kernel.org/driver-api/firmware/fw_upload.html
firmware_upload_register() has fewer arguments than it does when you
look at the signature of the function in the documentation right below
it.

Cheers,
Conor.