Re: [PATCH] HPT3xx: fix PCI clock detection

[Sergei Shtylyov <sshtylyov@ru.mvista.com>]

   Use the f_CNT value saved by the HighPoint BIOS if available as reading it 
directly would give us a wrong PCI frequency after DPLL has already been 
calibrated by BIOS.

   Should apply on top of my recent patches.

MBR, Sergei

Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>

Index: linus/drivers/ide/pci/hpt366.c
===================================================================
--- linus.orig/drivers/ide/pci/hpt366.c
+++ linus/drivers/ide/pci/hpt366.c
@@ -71,6 +71,8 @@
  *   needed and had many modes over- and  underclocked,  HPT372 33 MHz table was
  *   for 66 MHz and 50 MHz table missed UltraDMA mode 6, HPT374 33 MHz table was
  *   really for 50 MHz; switch to using HPT372 tables for HPT374...
+ * - use f_CNT value saved by  the HighPoint BIOS as reading it directly gives
+ *   the wrong PCI frequency since DPLL has already been calibrated by BIOS
  * - fix the hotswap code:  it caused RESET- to glitch when tristating the bus,
  *   and for HPT36x the obsolete HDIO_TRISTATE_HWIF handler was called instead
  * - pass to init_chipset() handlers a copy of the IDE PCI device structure as
@@ -1050,8 +1052,8 @@ static void __devinit hpt37x_clocking(id
 	struct hpt_info *info = ide_get_hwifdata(hwif);
 	struct pci_dev *dev = hwif->pci_dev;
 	int adjust, i;
-	u16 freq;
-	u32 pll;
+	u16 freq = 0;
+	u32 pll, temp = 0;
 	u8 reg5bh = 0, mcr1 = 0;
 	
 	/*
@@ -1065,15 +1067,34 @@ static void __devinit hpt37x_clocking(id
 	pci_write_config_byte(dev, 0x5b, 0x23);
 
 	/*
-	 * set up the PLL. we need to adjust it so that it's stable. 
-	 * freq = Tpll * 192 / Tpci
+	 * We'll have to read f_CNT value in order to determine
+	 * the PCI clock frequency according to the following ratio:
 	 *
-	 * Todo. For non x86 should probably check the dword is
-	 * set to 0xABCDExxx indicating the BIOS saved f_CNT
+	 * f_CNT = Fpci * 192 / Fdpll
+	 *
+	 * First try reading the register in which the HighPoint BIOS
+	 * saves f_CNT value before  reprogramming the DPLL from its
+	 * default setting (which differs for the various chips).
+	 * In case the signature check fails, we'll have to resort to
+	 * reading the f_CNT register itself in hopes that nobody has
+	 * touched the DPLL yet...
 	 */
-	pci_read_config_word(dev, 0x78, &freq);
-	freq &= 0x1FF;
-	
+	pci_read_config_dword(dev, 0x70, &temp);
+	if ((temp & 0xFFFFF000) != 0xABCDE000) {
+		int i;
+
+		printk(KERN_WARNING "HPT37X: no clock data saved by BIOS\n");
+
+		/* Calculate the average value of f_CNT */
+		for (temp = i = 0; i < 128; i++) {
+			pci_read_config_word(dev, 0x78, &freq);
+			temp += freq & 0x1ff;
+			mdelay(1);
+		}
+		freq = temp / 128;
+	} else
+		freq = temp & 0x1ff;
+
 	/*
 	 * HPT3xxN chips use different PCI clock information.
 	 * Currently we always set up the PLL for them.
@@ -1146,11 +1167,8 @@ static void __devinit hpt37x_clocking(id
 	info->flags |= PLL_MODE;
 	
 	/*
-	 * FIXME: make this work correctly, esp with 372N as per
-	 * reference driver code.
-	 *
-	 * adjust PLL based upon PCI clock, enable it, and wait for
-	 * stabilization.
+	 * Adjust the PLL based upon the PCI clock, enable it, and
+	 * wait for stabilization...
 	 */
 	adjust = 0;
 	freq = (pll < F_LOW_PCI_50) ? 2 : 4;