Lior, Simon,

On Wed, Jan 08, 2014 at 03:45:31PM +0200, Lior Amsalem wrote:

Hi Andrew,

quoted

From: Andrew Lunn [mailto:andrew at lunn.ch]
Sent: Tuesday, December 31, 2013 7:05 PM

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

quoted

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;

quoted

quoted

 		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)

quoted

quoted

 			 */
 			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?

This register (0x58) was introduced in a370/axp (with a new SATA PHY version).
In other SoCs the register does not exist.

Writing/reading the registers on A300/A310 will not cause any harm (I haven't
tried A510).
But, I think it's better to check if we're running on A370/AXP. (maybe with a new
compatibility string in DT?)

Lior, thanks for the clarification.  Simon, care to respin this with a
check for "marvell,armada-370-xp" root compatible string?  It should be
safe to say that if there is no DT, don't write the register.

Alternatively, we could do as Lior suggests, and create a new sata
compatible string.  But I think that is overkill.

Also, I'm growing more leery creating compatible strings for IP blocks
which are tied to the SoC revision.  If the IP block doesn't get issued
it's own version number/codename, we should just use the root compatible
strings to determine which SoC we are on.  I'll expand on this though as
I get caught up with Gregory's series's. 

thx,

Jason.