Re: [PATCH v2 4/4] EDAC: synopsys: Add useful debug and output information for 64bit systems

[Robert Richter]

On 01.04.20 14:32:58, Sherry Sun wrote:
> > From: Robert Richter <rrichter@xxxxxxxxxxx>
> > On 01.04.20 15:39:09, Sherry Sun wrote:

> > > --- a/drivers/edac/synopsys_edac.c
> > > +++ b/drivers/edac/synopsys_edac.c
> > > @@ -278,18 +278,22 @@
> > >   * @col:	Column number.
> > >   * @bank:	Bank number.
> > >   * @bitpos:	Bit position.
> > > - * @data:	Data causing the error.
> > > + * @data_low:	Low bit data causing the error.
> > > + * @data_high:	High bit data causing the error(used for 64 bit
> > systems).
> > >   * @bankgrpnr:	Bank group number.
> > >   * @blknr:	Block number.
> > > + * @syndrome:	Syndrome of the error.
> > >   */
> > >  struct ecc_error_info {
> > >  	u32 row;
> > >  	u32 col;
> > >  	u32 bank;
> > >  	u32 bitpos;
> > > -	u32 data;
> > > +	u32 data_low;
> > > +	u32 data_high;
> > 
> > Where are 16, 32 and 64 bit widths. You could handle them all the same in a
> > u64.
> > 
> > If I am not wrong, the width is fix for the whole mci. So you could create
> > various .get_error_info() functions depending on the data width without run
> > time width checks.
> > 
> 
> Yes, here we can use u64. Will correct it in v3.
> For the zynq, it's 16-bit bus width, for imx8mp, it's 64-bit bus width.
> But for zynqmp, it's optional between 32-bit and 64-bit bus width,
> So we had better run time check the bus width, what do you think?

I am wondering a bit here. *get_dtype() is used only in init_csrows()
which sets up the mci. So it will be a fix value all the time after
init. With that you easily can setup functions depending on the width,
alternatively the width could be stored in struct synps_platform_data
or struct synps_edac_priv.

> 
> > >  	u32 bankgrpnr;
> > >  	u32 blknr;
> > > +	u32 syndrome;
> > >  };

> > @@ -480,20
> > > +559,27 @@ static void handle_error(struct mem_ctl_info *mci, struct
> > > synps_ecc_status *p)  {
> > >  	struct synps_edac_priv *priv = mci->pvt_info;
> > >  	struct ecc_error_info *pinf;
> > > +	int n;
> > >
> > >  	if (p->ce_cnt) {
> > >  		pinf = &p->ceinfo;
> > >  		if (priv->p_data->quirks & DDR_ECC_INTR_SUPPORT) {
> > > -			snprintf(priv->message, SYNPS_EDAC_MSG_SIZE,
> > > -				 "DDR ECC error type:%s Row %d Bank %d
> > BankGroup Number %d Block Number %d Bit Position: %d Data: 0x%08x",
> > > -				 "CE", pinf->row, pinf->bank,
> > > -				 pinf->bankgrpnr, pinf->blknr,
> > > -				 pinf->bitpos, pinf->data);
> > > +			n = snprintf(priv->message, SYNPS_EDAC_MSG_SIZE,
> > > +				     "DDR ECC error type:%s Row %d Bank %d
> > BankGroup Number %d Block Number %d Bit Position: %d Data: 0x%08x",
> > > +				     "CE", pinf->row, pinf->bank,
> > > +				     pinf->bankgrpnr, pinf->blknr,
> > > +				     pinf->bitpos, pinf->data_low);
> > > +
> > > +			if (zynqmp_get_dtype(priv->baseaddr) == DEV_X8)
> > 
> > This is zynqmp specific, right? but you call it in the generic function
> > handle_error().
> 
> Yes, although handle_error() is a generic function, 
> but this operation is checked by priv->p_data->quirks,
> so here is zynqmp and imx8mp specific.

Naah, this is a bit hacky, just make 2 variants of handlers and set
them up during init. Storing the width in some private data would be
an alternative.

Thanks,

-Robert

> 
> Best regards
> Sherry Sun
> 
> > 
> > -Robert
> > 
> > > +				snprintf(priv->message + n,
> > > +					 SYNPS_EDAC_MSG_SIZE - n,
> > > +					 " Data_high: 0x%08x Syndrome:
> > 0x%08x",
> > > +					 pinf->data_high, pinf->syndrome);
> > >  		} else {
> > >  			snprintf(priv->message, SYNPS_EDAC_MSG_SIZE,
> > >  				 "DDR ECC error type:%s Row %d Bank %d
> > Col %d Bit Position: %d Data: 0x%08x",
> > >  				 "CE", pinf->row, pinf->bank, pinf->col,
> > > -				 pinf->bitpos, pinf->data);
> > > +				 pinf->bitpos, pinf->data_low);
> > >  		}
> > >
> > >  		edac_mc_handle_error(HW_EVENT_ERR_CORRECTED, mci,