From: Mark Lord <liml@rtr.ca>
Date: Mon, 6 Apr 2009 19:26:24 +0000 (-0400)
Subject: sata_mv: workaround errata SATA#13
X-Git-Url: https://git.karo-electronics.de/?a=commitdiff_plain;h=12f3b6d7551306c00cf834540a33184de67c9187;p=linux-beck.git

sata_mv: workaround errata SATA#13

Add remainder of workaround for errata SATA#13.

This prevents writes of certain adjacent 32-bit registers
from being combined into single 64-bit writes, which might
fail for the affected registers.

Most of sata_mv is already safe from this issue,
but adding this code to mv_write_cached_reg() will
catch the remaining cases and hopefully prevent future ones.

Signed-off-by: Mark Lord <mlord@pobox.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
---

diff --git a/drivers/ata/sata_mv.c b/drivers/ata/sata_mv.c
index 55d3ce087304..82d928a426ac 100644
--- a/drivers/ata/sata_mv.c
+++ b/drivers/ata/sata_mv.c
@@ -919,8 +919,26 @@ static void mv_save_cached_regs(struct ata_port *ap)
 static inline void mv_write_cached_reg(void __iomem *addr, u32 *old, u32 new)
 {
 	if (new != *old) {
+		unsigned long laddr;
 		*old = new;
-		writel(new, addr);
+		/*
+		 * Workaround for 88SX60x1-B2 FEr SATA#13:
+		 * Read-after-write is needed to prevent generating 64-bit
+		 * write cycles on the PCI bus for SATA interface registers
+		 * at offsets ending in 0x4 or 0xc.
+		 *
+		 * Looks like a lot of fuss, but it avoids an unnecessary
+		 * +1 usec read-after-write delay for unaffected registers.
+		 */
+		laddr = (long)addr & 0xffff;
+		if (laddr >= 0x300 && laddr <= 0x33c) {
+			laddr &= 0x000f;
+			if (laddr == 0x4 || laddr == 0xc) {
+				writelfl(new, addr); /* read after write */
+				return;
+			}
+		}
+		writel(new, addr); /* unaffected by the errata */
 	}
 }