Message-ID: <20240608150340.06e8c1e2@jic23-huawei>
Date: Sat, 8 Jun 2024 15:03:40 +0100
From: Jonathan Cameron <jic23@...nel.org>
To: Sean Anderson <sean.anderson@...ux.dev>
Cc: Jonathan Cameron <Jonathan.Cameron@...wei.com>, "O'Griofa, Conall"
 <conall.ogriofa@....com>, "linux-iio@...r.kernel.org"
 <linux-iio@...r.kernel.org>, "linux-arm-kernel@...ts.infradead.org"
 <linux-arm-kernel@...ts.infradead.org>, "linux-kernel@...r.kernel.org"
 <linux-kernel@...r.kernel.org>, Lars-Peter Clausen <lars@...afoo.de>
Subject: Re: [PATCH] iio: xilinx-ams: Don't include ams_ctrl_channels in
 scan_mask

On Fri, 7 Jun 2024 13:42:19 -0400
Sean Anderson <sean.anderson@...ux.dev> wrote:

> On 3/19/24 09:42, Jonathan Cameron wrote:
> > On Mon, 18 Mar 2024 11:28:49 -0400
> > Sean Anderson <sean.anderson@...ux.dev> wrote:
> >   
> >> On 3/18/24 11:24, Jonathan Cameron wrote:  
> >> > On Mon, 18 Mar 2024 11:18:43 -0400
> >> > Sean Anderson <sean.anderson@...ux.dev> wrote:
> >> >     
> >> >> On 3/16/24 09:36, Jonathan Cameron wrote:    
> >> >> > On Fri, 15 Mar 2024 13:47:40 -0400
> >> >> > Sean Anderson <sean.anderson@...ux.dev> wrote:
> >> >> >       
> >> >> >> Hi Conall,
> >> >> >> 
> >> >> >> On 3/15/24 09:18, O'Griofa, Conall wrote:      
> >> >> >> > [AMD Official Use Only - General]
> >> >> >> > 
> >> >> >> > Hi,
> >> >> >> > 
> >> >> >> > I think there was a fix for this issue applied to the version that was running on 5.15 that didn't seem to make it into the upstream driver.
> >> >> >> > Please see link for reference https://github.com/Xilinx/linux-xlnx/commit/608426961f16ab149b1b699f1c35f7ad244c0720
> >> >> >> > 
> >> >> >> > I think a similar fix to the above patch is may be beneficial?        
> >> >> >> 
> >> >> >> These patches look functionally identical to me.      
> >> >> > 
> >> >> > Because there are no channels with scan index between
> >> >> > 22 * 2 + 16 (that patch) and 22 * 3 (your patch) that is
> >> >> > the effect is indeed the same. But given the issues is the
> >> >> > 64 limit on maximum scan index, 22 * 3 = 66 is an ugly value
> >> >> > to compare with.
> >> >> > 
> >> >> > I'm still very against the use of scan_index for anything other
> >> >> > than scan indices (which is why partly how this bug wasn't noticed
> >> >> > in the first palce). So the check should be scan_index != -1
> >> >> > and uses of those values elsewhere in the driver should be fixed
> >> >> > (which looks simple to do from a quick glance at the code).      
> >> >> 
> >> >> OK, so how do the sysfs files get named then?    
> >> > 
> >> > Using channel and channel2 as appropriate (+ index and modified
> >> > which change the meaning of channel2) - scan_index never had
> >> > anything to do with sysfs file names - just the value in
> >> > bufferX/in_xyz_scan_index    
> >> 
> >> I tried to prototype setting scan_index to -1, but when registering channels I saw
> >> 
> >> [    1.637049] iio iio:device0: tried to double register : in_voltage_raw
> >> [    1.637245] xilinx-ams ffa50000.ams: Failed to register sysfs interfaces
> >> [    1.637433] xilinx-ams: probe of ffa50000.ams failed with error -16
> >> 
> >> And AIUI .channel is filled in by ams_parse_firmware.  
> > 
> > Is indexed set for the channel?  Check it at the point of calling
> > devm_iio_device_register() as the code that builds the channels in this
> > driver is complex, so maybe it's getting overwritten?
> > 
> > There might be a core bug somewhere, but there are other drivers using
> > -1 scan index without hitting this problem so my first instinct is
> > something is getting wrongly set in the driver.  
> 
> Upon further review, I think scan_index should remain the same, and this
> patch should be applied as-is.
> 
> address is the only driver-private data in all of iio_chan_spec.
> Unfortunately, it is suggestively named "address" and not "priv" or
> "driver_id" or something similar. So the original author of this driver
> went "Ah, I should put the channel address offsets in this register."
> Except, because this driver has three address spaces, this is not enough
> to uniquely identify the channel. So he then stuck an actual unique
> identifier in scan_index. Now, you may object to this since the driver
> doesn't actually support scans, but that is the current situation.
> 
> So there is really nothing wrong with scan_index semantically in the
> context of the driver. We should not convert one address space's
> channels to use -1 scan_index, since it is used as a unique identifier
> elsewhere in the channel.
> 
> Future patches could convert scan_index to address, and store the
> address offsets in an array. So e.g. reading a channel would go from
> e.g.
> 
> 	if (chan->scan_index >= AMS_PS_SEQ_MAX)
> 		*val = readl(ams->pl_base + chan->address);
> 	else
> 		*val = readl(ams->ps_base + chan->address);
> 
> to
> 
> 	if (chan->address >= AMS_PS_SEQ_MAX)
> 		*val = readl(ams->pl_base + ams_chan_addr[chan->address]);
> 	else
> 		*val = readl(ams->ps_base + ams_chan_addr[chan->address]);
> 
> which while strictly less perfmant due to another level of indirection
> does conform to existing semantics for scan_index. But TBH I don't see
> much point in this.
> 
> But the above change would be pretty significant and has a chance of
> causing bugs of its own. So I would rather this bug fix be applied as-is
> and the scan_index semantics be modified at some other time.
Ok.  I'll pick this one up, but I'd either like the change you mention above
or maybe as a lesser nice but easier solution, a patch adding comments
on the use of scan_index in this driver.

Applied to the fixes-togreg branch of iio.git.  Thanks for figuring out
what happened here and the clear explanation.

Jonathan

> 
> --Sean
> 
> >> 
> >> --Sean
> >>   
> >> >> 
> >> >> --Sean
> >> >>     
> >> >> >> 
> >> >> >> --Sean
> >> >> >>       
> >> >> >> >> -----Original Message-----
> >> >> >> >> From: Sean Anderson <sean.anderson@...ux.dev>
> >> >> >> >> Sent: Thursday, March 14, 2024 5:30 PM
> >> >> >> >> To: Jonathan Cameron <jic23@...nel.org>
> >> >> >> >> Cc: linux-iio@...r.kernel.org; O'Griofa, Conall <conall.ogriofa@....com>;
> >> >> >> >> linux-arm-kernel@...ts.infradead.org; linux-kernel@...r.kernel.org; Lars-Peter
> >> >> >> >> Clausen <lars@...afoo.de>
> >> >> >> >> Subject: Re: [PATCH] iio: xilinx-ams: Don't include ams_ctrl_channels in
> >> >> >> >> scan_mask
> >> >> >> >>
> >> >> >> >> Caution: This message originated from an External Source. Use proper caution
> >> >> >> >> when opening attachments, clicking links, or responding.
> >> >> >> >>
> >> >> >> >>
> >> >> >> >> On 3/14/24 11:48, Jonathan Cameron wrote:        
> >> >> >> >> > On Mon, 11 Mar 2024 12:28:00 -0400
> >> >> >> >> > Sean Anderson <sean.anderson@...ux.dev> wrote:
> >> >> >> >> >        
> >> >> >> >> >> ams_enable_channel_sequence constructs a "scan_mask" for all the PS
> >> >> >> >> >> and PL channels. This works out fine, since scan_index for these
> >> >> >> >> >> channels is less than 64. However, it also includes the
> >> >> >> >> >> ams_ctrl_channels, where scan_index is greater than 64, triggering
> >> >> >> >> >> undefined behavior. Since we don't need these channels anyway, just        
> >> >> >> >> exclude them.        
> >> >> >> >> >>
> >> >> >> >> >> Fixes: d5c70627a794 ("iio: adc: Add Xilinx AMS driver")
> >> >> >> >> >> Signed-off-by: Sean Anderson <sean.anderson@...ux.dev>        
> >> >> >> >> >
> >> >> >> >> > Hi Sean,
> >> >> >> >> >
> >> >> >> >> > I'd ideally like to understand why we have channels with such large
> >> >> >> >> > scan indexes.  Those values should only be used for buffered capture.
> >> >> >> >> > It feels like they are being abused here.  Can we set them to -1
> >> >> >> >> > instead and check based on that?
> >> >> >> >> > For a channel, a scan index of -1 means it can't be captured via the
> >> >> >> >> > buffered interfaces but only accessed via sysfs reads.
> >> >> >> >> > I think that's what we have here?        
> >> >> >> >>
> >> >> >> >> From what I can tell, none of the channels support buffered reads. And we can't
> >> >> >> >> naïvely convert the scan_index to -1, since that causes sysfs naming conflicts
> >> >> >> >> (not to mention the compatibility break).
> >> >> >> >>        
> >> >> >> >> >
> >> >> >> >> > I just feel like if we leave these as things stand, we will get bitten
> >> >> >> >> > by similar bugs in the future.  At least with -1 it should be obvious why!        
> >> >> >> >>
> >> >> >> >> There are just as likely to be bugs confusing the PL/PS subdevices...
> >> >> >> >>
> >> >> >> >> FWIW I had no trouble identifying the channels involved with this bug.
> >> >> >> >>
> >> >> >> >> --Sean
> >> >> >> >>        
> >> >> >> >> > Jonathan
> >> >> >> >> >
> >> >> >> >> >        
> >> >> >> >> >> ---
> >> >> >> >> >>
> >> >> >> >> >>  drivers/iio/adc/xilinx-ams.c | 8 ++++++--
> >> >> >> >> >>  1 file changed, 6 insertions(+), 2 deletions(-)
> >> >> >> >> >>
> >> >> >> >> >> diff --git a/drivers/iio/adc/xilinx-ams.c
> >> >> >> >> >> b/drivers/iio/adc/xilinx-ams.c index a55396c1f8b2..4de7ce598e4d
> >> >> >> >> >> 100644
> >> >> >> >> >> --- a/drivers/iio/adc/xilinx-ams.c
> >> >> >> >> >> +++ b/drivers/iio/adc/xilinx-ams.c
> >> >> >> >> >> @@ -414,8 +414,12 @@ static void ams_enable_channel_sequence(struct
> >> >> >> >> >> iio_dev *indio_dev)
> >> >> >> >> >>
> >> >> >> >> >>      /* Run calibration of PS & PL as part of the sequence */
> >> >> >> >> >>      scan_mask = BIT(0) | BIT(AMS_PS_SEQ_MAX);
> >> >> >> >> >> -    for (i = 0; i < indio_dev->num_channels; i++)
> >> >> >> >> >> -            scan_mask |= BIT_ULL(indio_dev->channels[i].scan_index);
> >> >> >> >> >> +    for (i = 0; i < indio_dev->num_channels; i++) {
> >> >> >> >> >> +            const struct iio_chan_spec *chan =
> >> >> >> >> >> + &indio_dev->channels[i];
> >> >> >> >> >> +
> >> >> >> >> >> +            if (chan->scan_index < AMS_CTRL_SEQ_BASE)
> >> >> >> >> >> +                    scan_mask |= BIT_ULL(chan->scan_index);
> >> >> >> >> >> +    }
> >> >> >> >> >>
> >> >> >> >> >>      if (ams->ps_base) {
> >> >> >> >> >>              /* put sysmon in a soft reset to change the sequence */        
> >> >> >> >> >        
> >> >> >>       
> >> >> >       
> >> >> 
> >> >>     
> >> >     
> >>   
> >   

Powered by blists - more mailing lists