Re: [PATCH v2 Resent 6/6] i3c: master: svc: fix random hot join failure since timeout error

[Frank Li]

On Fri, Oct 20, 2023 at 04:06:45PM +0200, Miquel Raynal wrote:
> Hi Frank,
> 
> Frank.li@xxxxxxx wrote on Thu, 19 Oct 2023 11:39:42 -0400:
> 
> > On Thu, Oct 19, 2023 at 08:44:52AM +0200, Miquel Raynal wrote:
> > > Hi Frank,
> > > 
> > > Frank.Li@xxxxxxx wrote on Wed, 18 Oct 2023 11:59:26 -0400:
> > >   
> > > > master side report:
> > > >   silvaco-i3c-master 44330000.i3c-master: Error condition: MSTATUS 0x020090c7, MERRWARN 0x00100000
> > > > 
> > > > BIT 20: TIMEOUT error
> > > >   The module has stalled too long in a frame. This happens when:
> > > >   - The TX FIFO or RX FIFO is not handled and the bus is stuck in the
> > > > middle of a message,
> > > >   - No STOP was issued and between messages,
> > > >   - IBI manual is used and no decision was made.  
> > > 
> > > I am still not convinced this should be ignored in all cases.
> > > 
> > > Case 1 is a problem because the hardware failed somehow.  
> > 
> > But so far, no action to handle this case in current code.
> 
> Yes, but if you detect an issue and ignore it, it's not better than
> reporting it without handling it. Instead of totally ignoring this I
> would at least write a debug message (identical to what's below) before
> returning false, even though I am not convinced unconditionally
> returning false here is wise. If you fail a hardware sequence because
> you added a printk, it's a problem. Maybe you consider this line as
> noise, but I believe it's still an error condition. Maybe, however,
> this bit gets set after the whole sequence, and this is just a "bus
> is idle" condition. If that's the case, then you need some
> additional heuristics to properly ignore the bit?
> 

                dev_err(master->dev,                                       
                        "Error condition: MSTATUS 0x%08x, MERRWARN 0x%08x\n",
                        mstatus, merrwarn);
+
+		/* ignore timeout error */
+		if (merrwarn & SVC_I3C_MERRWARN_TIMEOUT)
+			return false;
+

Is it okay move SVC_I3C_MERRWARN_TIMEOUT after dev_err?

Frank


> > In svc_i3c_master_xfer() have not check this flags. also have not enable
> > ERRWARN irq.
> > 
> > If we met this case, we can add new functions/argument to handle this.
> > Then we can real debug the code and recover bus.
> > 
> > Without this patch, simplest add some debug message before issue
> > SVC_I3C_MCTRL_REQUEST_AUTO_IBI, TIMEOUT will be set.
> 
> Yes, and sometimes it won't be an issue, but sometimes it may. Maybe we
> can find more advanced heuristics there.
> 
> > And svc_i3c_master_error() was only called by svc_i3c_master_ibi_work().
> >
> > So I can think only case 3 happen in svc_i3c_master_ibi_work().
> 
> Case 3 cannot be handled by Linux (because of the natural latency of
> the OS).
> 
> > 
> > Frank
> > 
> > > Case 2 is fine I guess.
> > > Case 3 is not possible in Linux, this will not be supported.
> > >   
> > > >   The maximum stall period is 10 KHz or 100 μs.  
> > > 
> > > s/10 KHz//
> > >   
> > > > 
> > > > This is a just warning. System irq thread schedule latency is possible
> > > > bigger than 100us. Just omit this waring.  
> > > 
> > > This can be considered as being just a warning as the system IRQ
> > > latency can easily be greater than 100us.
> 
> This was skipped in your v3.
> 
> > > > Fixes: dd3c52846d59 ("i3c: master: svc: Add Silvaco I3C master driver")
> > > > Cc: stable@xxxxxxxxxxxxxxx
> > > > Signed-off-by: Frank Li <Frank.Li@xxxxxxx>
> > > > ---
> 
> Thanks,
> Miquèl