On Tue, Dec 31, 2013 at 07:12:14AM -0500, Tejun Heo wrote:

On Mon, Dec 23, 2013 at 01:07:35PM +0100, Simon Guinot wrote:

quoted

@@ -1358,6 +1359,7 @@ static int mv_scr_write(struct ata_link *link, unsigned int sc_reg_in, u32 val)
 
 	if (ofs != 0xffffffffU) {
 		void __iomem *addr = mv_ap_base(link->ap) + ofs;
+		void __iomem *lp_phy_addr = mv_ap_base(link->ap) + LP_PHY_CTL;
 		if (sc_reg_in == SCR_CONTROL) {
 			/*
 			 * Workaround for 88SX60x1 FEr SATA#26:

@@ -1374,6 +1376,14 @@ static int mv_scr_write(struct ata_link *link, unsigned int sc_reg_in, u32 val)
 			 */
 			if ((val & 0xf) == 1 || (readl(addr) & 0xf) == 1)
 				val |= 0xf000;
+
+			/*
+			 * Setting PHY speed according to SControl speed
+			 */
+			if ((val & 0xf0) == 0x10)
+				writelfl(0x7, lp_phy_addr);
+			else
+				writelfl(0x227, lp_phy_addr);

Do we know that this is safe for all sata_mvs?  If other sata_mvs
haven't had the described issue, maybe this should be applied
selectively to the said soc?  I'd actually prefer to avoid such
conditionals but we need to confirm this is okay for other
implementations.

Hi Tejun

I've tested on Kirkwood, and not had problems. But i agree with
you. We need somebody in Marvell to say this is safe with all sata_mv
variants.

Lior, can you comment?

Thanks
	Andrew