Re: [PATCH v6 08/10] phy: qualcomm: phy-qcom-qmp-ufs: Add High Speed Gear 5 support for SM8550

[Manivannan Sadhasivam <mani@kernel.org>]

On Thu, Nov 30, 2023 at 04:49:12PM +0800, Can Guo wrote:
> 
> 
> On 11/30/2023 4:38 PM, Manivannan Sadhasivam wrote:
> > On Thu, Nov 30, 2023 at 04:14:25PM +0800, Can Guo wrote:

[...]

> > > > > +static int qmp_ufs_get_gear_overlay(struct qmp_ufs *qmp, const struct qmp_phy_cfg *cfg)
> > > > > +{
> > > > > +	u32 max_gear, floor_max_gear = cfg->max_supported_gear;
> > > > > +	int i = NUM_OVERLAY - 1;
> > > > 
> > > > Just use i directly in the for loop. Also, please rename "i" with "idx" to make
> > > > it clear.
> > > 
> > > OK
> > > 
> > > > 
> > > > > +	int ret = -EINVAL;
> > > > > +
> > > > > +	for (; i >= 0; i --) {
> > > > 
> > > > i--
> > > > 
> > > > > +		max_gear = cfg->tbls_hs_overlay[i].max_gear;
> > > > > +
> > > > > +		if (max_gear == 0)
> > > > > +			continue;
> > > > 
> > > > You are setting max_gear even for targets with a single overlay. How can this
> > > > become 0?
> > > 
> > > Say 8550 has two overlays, 8450 has one overlay. We are sweeping all
> > > overlays as NUM_OVERLAY == 2, so for 8450, there is one overlay initialized,
> > > another one not initialized (max_gear == 0), we are skipping the one which
> > > is not initialized.
> > > 
> > 
> > This is confusing at its best :) Please check for the existence of the actual
> > table instead. Like,
> > 
> > 	for (idx = NUM_OVERLAY - 1; i >= 0, i--) {
> > 
> > 		/* Skip if the table is not available */
> > 		if (!cfg->tbls_hs_overlay[i].serdes)
> > 			continue;
> > 
> > 		...
> > 	}
> 
> We cannot expect overlay must has its own serdes, or tx/rx/pcs, hence I am
> checking max_gear intead of any specific member.
> 

Hmm, then please add the comment as I suggested above.

> > 
> > > > 
> > > > > +
> > > > > +		/* Direct matching, bail */
> > > > > +		if (qmp->submode == max_gear)
> > > > > +			return i;
> > > > > +
> > > > > +		/* If no direct matching, the lowest gear is the best matching */
> > > > > +		if (max_gear < floor_max_gear) {
> > > > > +			ret = i;
> > > > > +			floor_max_gear = max_gear;
> > > > > +		}
> > > > > +	}
> > > > > +
> > > > > +	return ret;
> > > > > +}
> > > > > +
> > > > >    static void qmp_ufs_init_registers(struct qmp_ufs *qmp, const struct qmp_phy_cfg *cfg)
> > > > >    {
> > > > > +	int i;
> > > > > +	bool apply_overlay = false;
> > > > > +
> > > > > +	i = qmp_ufs_get_gear_overlay(qmp, cfg);
> > > > > +	if (i >= 0)
> > > > > +		apply_overlay = true;
> > > > 
> > > > How about?
> > > > 
> > > > ```
> > > > 	int idx;
> > > > 
> > > > 	idx = qmp_ufs_get_gear_overlay(qmp, cfg);
> > > > 
> > > > 	qmp_ufs_serdes_init(qmp, &cfg->tbls);
> > > > 	qmp_ufs_lanes_init(qmp, &cfg->tbls);
> > > > 	...
> > > > 
> > > > 	if (idx >= 0) {
> > > > 		qmp_ufs_serdes_init(qmp, &cfg->tbls_hs_overlay[idx]);
> > > > 		qmp_ufs_lanes_init(qmp, &cfg->tbls_hs_overlay[idx]);
> > > > 		...
> > > > 	}
> > > > ```
> > > > 
> > > > Since the ordering doesn't matter for init sequence, you can program the overlay
> > > > tables under a single condition.
> > > 
> > > We can do that, but we need to be careful. When I say (in my previous reply)
> > > the ordering does not matter, that saying is from the UFS PHY HPG doc.
> > > However, in SW implementation, the 'tbls_hs_b' is actually overwriting one
> > > COM_VCO_TUNE_MAP register, the same register is also programmed by common
> > > table or overlay table. So qmp_ufs_serdes_init(qmp, &cfg->tbls_hs_b) should
> > > come after overlays.
> > > 
> > 
> > Then you can program tbls_hs_b after overlay tables. Wouldn't that work?
> 
> I am programming tbls_hs_b after overlay tables, just a heads up in case you
> are surprised :).
> 

Cool!

- Mani

-- 
மணிவண்ணன் சதாசிவம்