Re: [patch v3 3/4] EDAC: synopsys: Add edac driver support for i.MX8MP

[Robert Richter]

On 02.04.20 13:09:57, Sherry Sun wrote:
> > From: Robert Richter <rrichter@xxxxxxxxxxx>
> > On 02.04.20 09:06:27, Sherry Sun wrote:
> > > > From: Robert Richter <rrichter@xxxxxxxxxxx> On 02.04.20 09:20:32,

> > > > > +/* Interrupt Handler for ECC interrupts on imx8mp platform. */
> > > > > +static irqreturn_t intr_handler_imx8mp(int irq, void *dev_id) {
> > > > > +	const struct synps_platform_data *p_data;
> > > > > +	struct mem_ctl_info *mci = dev_id;
> > > > > +	struct synps_edac_priv *priv;
> > > > > +	int status, regval;
> > > > > +
> > > > > +	priv = mci->pvt_info;
> > > > > +	p_data = priv->p_data;
> > > > > +
> > > > > +	regval = readl(priv->baseaddr + ECC_STAT_OFST);
> > > > > +	if (!(regval & ECC_INTR_MASK))
> > > > > +		return IRQ_NONE;
> > > > > +
> > > > > +	status = p_data->get_error_info(priv);
> > > > > +	if (status)
> > > > > +		return IRQ_NONE;
> > > > > +
> > > > > +	priv->ce_cnt += priv->stat.ce_cnt;
> > > > > +	priv->ue_cnt += priv->stat.ue_cnt;
> > > > > +	handle_error(mci, &priv->stat);
> > > > > +
> > > > > +	edac_dbg(3, "Total error count CE %d UE %d\n",
> > > > > +		 priv->ce_cnt, priv->ue_cnt);
 > > > > > +	enable_intr_imx8mp(priv);
> > > >
> > > > Why do you enable interrupts here?
> > >
> > > Because zynqmp_get_error_info() wrote 0 to ECC_CLR_OFST, so here have
> > to re-enable the interrupts.
> > 
> > This does not seem to be the right place for it.
> > 
> > > As said in the commit, the main difference between ZynqMP and i.MX8MP
> > ddr controller is the interrupt registers.
> > > ZynqMP use DDR QOS Interrupt registers,  but i.MX8MP use ECC_CLR_OFST
> > Register(bit8 and bit9) to enable/disable the ce/ue interrupts.
> > >
> > > In zynqmp_get_error_info(), Zynqmp wrote 0 to ECC_CLR_OFST register to
> > clear CE/UE error flags and counts, it has no effect on Zynqmp interrupts.
> > > But for i.MX8MP, wirte 0 to ECC_CLR_OFST will disable i.MX8MP CE/UE
> > interrupt, so need re-enable the interrupts.
> > 
> > All this shows one more time there should be separate handlers. You should
> > get rid most callbacks in struct synps_platform_data and instead have
> > separate probe functions for both flavors that share common code.
> 
> Do you mean that I should create specific callbacks in struct synps_platform_data for i.MX8MP instead reuse the ZynqMP functions?
> But except this operation of write 0 to ECC_CLR_OFST in zynqmp_get_error_info(), 
> the other callbacks in struct synps_platform_data are all the same between ZynqMP and i.MX8MP.

Most callbacks in struct synps_platform_data will become obsolete if
there are either device tailored functions or if the width parameter
would be stored in private data and used by a common function.

-Robert