RE: [PATCH V7 00/11] misc: xilinx sd-fec drive

[Dragan Cvetic]

> -----Original Message-----
> From: Greg KH [mailto:gregkh@xxxxxxxxxxxxxxxxxxx]
> Sent: Saturday 22 June 2019 07:02
> To: Dragan Cvetic <draganc@xxxxxxxxxx>
> Cc: arnd@xxxxxxxx; Michal Simek <michals@xxxxxxxxxx>; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; robh+dt@xxxxxxxxxx;
> mark.rutland@xxxxxxx; devicetree@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Derek Kiernan <dkiernan@xxxxxxxxxx>
> Subject: Re: [PATCH V7 00/11] misc: xilinx sd-fec drive
> 
> On Fri, Jun 21, 2019 at 05:49:45PM +0000, Dragan Cvetic wrote:
> >
> >
> > > -----Original Message-----
> > > From: Greg KH [mailto:gregkh@xxxxxxxxxxxxxxxxxxx]
> > > Sent: Friday 21 June 2019 15:16
> > > To: Dragan Cvetic <draganc@xxxxxxxxxx>
> > > Cc: arnd@xxxxxxxx; Michal Simek <michals@xxxxxxxxxx>; linux-arm-kernel@xxxxxxxxxxxxxxxxxxx; robh+dt@xxxxxxxxxx;
> > > mark.rutland@xxxxxxx; devicetree@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Derek Kiernan <dkiernan@xxxxxxxxxx>
> > > Subject: Re: [PATCH V7 00/11] misc: xilinx sd-fec drive
> > >
> > > On Tue, Jun 11, 2019 at 06:29:34PM +0100, Dragan Cvetic wrote:
> > > > This patchset is adding the full Soft Decision Forward Error
> > > > Correction (SD-FEC) driver implementation, driver DT binding and
> > > > driver documentation.
> > > >
> > > > Forward Error Correction (FEC) codes such as Low Density Parity
> > > > Check (LDPC) and turbo codes provide a means to control errors in
> > > > data transmissions over unreliable or noisy communication
> > > > channels. The SD-FEC Integrated Block is an optimized block for
> > > > soft-decision decoding of these codes. Fixed turbo codes are
> > > > supported directly, whereas custom and standardized LDPC codes
> > > > are supported through the ability to specify the parity check
> > > > matrix through an AXI4-Lite bus or using the optional programmable
> > > > (PL)-based support logic. For the further information see
> > > > https://www.xilinx.com/support/documentation/ip_documentation/
> > > > sd_fec/v1_1/pg256-sdfec-integrated-block.pdf
> > > >
> > > > This driver is a platform device driver which supports SDFEC16
> > > > (16nm) IP. SD-FEC driver supports LDPC decoding and encoding and
> > > > Turbo code decoding. LDPC codes can be specified on
> > > > a codeword-by-codeword basis, also a custom LDPC code can be used.
> > > >
> > > > The SD-FEC driver exposes a char device interface and supports
> > > > file operations: open(), close(), poll() and ioctl(). The driver
> > > > allows only one usage of the device, open() limits the number of
> > > > driver instances. The driver also utilize Common Clock Framework
> > > > (CCF).
> > > >
> > > > The control and monitoring is supported over ioctl system call.
> > > > The features supported by ioctl():
> > > > - enable or disable data pipes to/from device
> > > > - configure the FEC algorithm parameters
> > > > - set the order of data
> > > > - provide a control of a SDFEC bypass option
> > > > - activates/deactivates SD-FEC
> > > > - collect and provide statistical data
> > > > - enable/disable interrupt mode
> > >
> > > Is there any userspace tool that talks to this device using these custom
> > > ioctls yet?
> > >
> > Tools no, but could be the customer who is using the driver.
> 
> I don't understand this.  Who has written code to talk to these
> special ioctls from userspace?  Is there a pointer to that code
> anywhere?
> 
> > > Doing a one-off ioctl api is always a risky thing, you are pretty much
> > > just creating brand new system calls for one piece of hardware.
> > >
> >
> > Why is that wrong and what is the risk?
> 
> You now have custom syscalls for one specfic piece of hardware that you
> now have to maintain working properly for the next 40+ years.  You have
> to make sure those calls are correct and that this is the correct api to
> talk to this hardware.
> 


The only idea I have got from the comments are to do more abstraction
eg. have a few ioctls with the abstraction done through the passing arguments?




> > What would you propose?
> > Definitely, I have to read about this.
> 
> What is this hardware and what is it used for?  Who will be talking to
> it from userspace?  What userspace workload uses it?  What tools need to
> talk to it?  Where is the code that uses these new apis?
> 
> thanks,
> 
> greg k-h