Patchwork MCP67 SPI detection/debugging

login
register
about
Submitter Carl-Daniel Hailfinger
Date 2010-02-01 02:31:05
Message ID <4B663CE9.5090004@gmx.net>
Download mbox | patch
Permalink /patch/860/
State Superseded
Headers show

Comments

Carl-Daniel Hailfinger - 2010-02-01 02:31:05
Here's a new iteration of the patch.
Alex, it would be great if you could supply the output (as described
below) from your Asus M2N68-VM board. Feel free to wait until this patch
is merged, though.

I'd like to merge this ASAP and then ping everyone who had post-MCP55
Nvidia chipset boards to get flashrom -V output.

-----------------
Add SPI mode diagnostics for all post-MCP55 (nForce 5) chipsets from Nvidia.

Huge thanks to Michael Karcher for reverse engineering the MCP67 chipset
and writing a spec. Due to this, we were able to use the chinese wall
technique for 100% clean room reverse engineering.

This patch doesn't touch any of the new registers, it only reads them.
Assuming that read has no side effects, this patch is a no-op and safe.

We need "flashrom -V" output from all post-MCP55 (nForce 5) chipset
boards. Please indicate if your board uses SPI flash or LPC flash (if
you know it). Note: That output is only helpful if it is created with
patched flashrom and if is from the first run of flashrom after a cold
boot (reset or Ctrl-Alt-Del is not sufficient). There is a pattern based
on which we can probably detect which flash type is present on the board.

Thanks to Alessandro Polverini for testing an earlier iteration of this
patch.

Note: The MCP67 should work. I guessed that the other recent Nvidia
chipsets would work in a similar way, and created a simplified
do-nothing catchall chipset enable function which dumps some info and
instructs the user to send more info.

Signed-off-by: Carl-Daniel Hailfinger <c-d.hailfinger.devel.2006@gmx.net>
Andrew Morgan - 2010-02-01 23:27:47
Carl-Daniel Hailfinger wrote:
> We need "flashrom -V" output from all post-MCP55 (nForce 5) chipset
> boards. Please indicate if your board uses SPI flash or LPC flash (if
> you know it). Note: That output is only helpful if it is created with
> patched flashrom and if is from the first run of flashrom after a cold
> boot (reset or Ctrl-Alt-Del is not sufficient). There is a pattern based
> on which we can probably detect which flash type is present on the board.
>   

I've patched, compiled and run flashrom -V on my Acer Aspire Revo R3600, 
nVidia MCP79 chipset.
Powered off and even unplugged from the mains before testing.
Out of curiosity I ran it twice, the only difference was the 
'Calibrating delay loop...' line.

Attached is the output of 'flashrom -V' the first run after booting. 
Board has a W25X80AVAIZ SPI flash chip.
Michael Karcher - 2010-02-03 17:32:11
Am Montag, den 01.02.2010, 03:31 +0100 schrieb Carl-Daniel Hailfinger:
> +/**
> + * The MCP67 code is guesswork based on cleanroom reverse engineering.
> + * Due to that, it only reads info and doesn't change any settings.
> + * It is assumed that LPC chips need the MCP55 code and SPI chips need the
> + * code provided in this function. Until we know for sure, call
> + * enable_flash_mcp55 from this function.
> + */
We don't know whether we can switch the chip between SPI and LPC mode
(does it support LPC at all?), and running correct initialization for
SPI might make LPC inaccessible.
[...]
> +	msg_pdbg("ISA bridge reg 0x8a contents: 0x%02x, bit 6 is %i, bit 5 is "
> +		 "%i\n", byte, (byte >> 6) & 0x1, (byte >> 5) & 0x1);
> +	msg_pdbg("Guessed flash bus type is %s\n", ((byte >> 5) & 0x3) == 0x2 ?
> +		 "SPI" : "unknown, probably LPC");
> +	/* Disable the write code for now until we have more info. */
Looks fine.

> +	/* Locate the BAR where the GPIOs live. */
Maybe GPIO should be changed to something else, as the standard GPIOs
are I/O mapped. Make that "where the SPI interface lives"
> +	gpiobaraddr = pci_read_long(smbusdev, 0x74);
> +	msg_pdbg("GPIO BAR is at 0x%08x, ", gpiobaraddr);
> +	/* We hope this has native alignment. We know the GPIOs are at offset
> +	 * 0x530, so we expect a size of at least 0x800. Clear the lower bits.
> +	 * It is entirely possible that the BAR is 64k big and the low bits are
> +	 * reserved for an entirely different purpose.
> +	 */
> +	gpiobaraddr &= ~0x7ff;
> +	msg_pdbg("after clearing low bits BAR is at 0x%08x\n", gpiobaraddr);
> +
> +	/* Accessing a NULL pointer BAR is evil. Don't do it. */
> +	if (gpiobaraddr) {
> +		/* Map the BAR. We access bytewise and wordwise at 0x530. */
> +		gpiobar = physmap("MCP67 GPIO", gpiobaraddr, 0x534);
We might need gpiobaraddr + 0x10 later. Some BIOSses access it, too. Go
directly for 0x544.

> +/* Guessed. If this is correct, migrate to a separate MCP67 SPI driver. */
> +#define MCP67_SPI_CS		(1 << 1)
> +#define MCP67_SPI_SCK		(1 << 2)
> +#define MCP67_SPI_MOSI		(1 << 3)
> +#define MCP67_SPI_MISO		(1 << 4)
> +#define MCP67_SPI_ENABLE	(1 << 0)
> +#define MCP67_SPI_IDLE		(1 << 8)
> +		status = mmio_readw(gpiobar + 0x530);
> +		msg_pdbg("SPI control is 0x%04x, enable=%i, idle=%i\n",
> +			 status, status & 0x1, (status >> 8) & 0x1);
> +	}
Add error handling here [else msg_perr("bad base address of memory
mapped interface")]

> +/* This is a shot in the dark. Even if the code is totally bogus for some
> + * chipsets, users will at least start to send in reports.
> + */
Mostly a ripoff of the mcp67 function, bit without any writing. OK.

Acked-By: Michael Karcher <flashrom@mkarcher.dialup.fu-berlin.de>

Patch

Index: flashrom-mcp67_spi_detect/chipset_enable.c
===================================================================
--- flashrom-mcp67_spi_detect/chipset_enable.c	(Revision 886)
+++ flashrom-mcp67_spi_detect/chipset_enable.c	(Arbeitskopie)
@@ -1054,6 +1054,118 @@ 
 	return 0;
 }
 
+/**
+ * The MCP67 code is guesswork based on cleanroom reverse engineering.
+ * Due to that, it only reads info and doesn't change any settings.
+ * It is assumed that LPC chips need the MCP55 code and SPI chips need the
+ * code provided in this function. Until we know for sure, call
+ * enable_flash_mcp55 from this function.
+ */
+static int enable_flash_mcp67(struct pci_dev *dev, const char *name)
+{
+	int result = 0;
+	uint8_t byte;
+	uint16_t status;
+	uint32_t gpiobaraddr;
+	void *gpiobar;
+	struct pci_dev *smbusdev;
+
+	/* dev is the ISA bridge. No idea what the stuff below does. */
+	byte = pci_read_byte(dev, 0x8a);
+	msg_pdbg("ISA bridge reg 0x8a contents: 0x%02x, bit 6 is %i, bit 5 is "
+		 "%i\n", byte, (byte >> 6) & 0x1, (byte >> 5) & 0x1);
+	msg_pdbg("Guessed flash bus type is %s\n", ((byte >> 5) & 0x3) == 0x2 ?
+		 "SPI" : "unknown, probably LPC");
+	/* Disable the write code for now until we have more info. */
+#if 0
+	byte |= (1 << 6);
+	byte &= ~(1 << 5);
+	pci_write_byte(dev, 0x8a, byte);
+#endif
+
+	/* Look for the SMBus device (SMBus PCI class) */
+	smbusdev = pci_dev_find_vendorclass(0x10de, 0x0c05);
+	if (!smbusdev) {
+		msg_perr("ERROR: SMBus device not found. Aborting.\n");
+		exit(1);
+	}
+	msg_pdbg("Found SMBus device %04x:%04x at %02x:%02x:%01x\n",
+		smbusdev->vendor_id, smbusdev->device_id,
+		smbusdev->bus, smbusdev->dev, smbusdev->func);
+
+	/* Locate the BAR where the GPIOs live. */
+	gpiobaraddr = pci_read_long(smbusdev, 0x74);
+	msg_pdbg("GPIO BAR is at 0x%08x, ", gpiobaraddr);
+	/* We hope this has native alignment. We know the GPIOs are at offset
+	 * 0x530, so we expect a size of at least 0x800. Clear the lower bits.
+	 * It is entirely possible that the BAR is 64k big and the low bits are
+	 * reserved for an entirely different purpose.
+	 */
+	gpiobaraddr &= ~0x7ff;
+	msg_pdbg("after clearing low bits BAR is at 0x%08x\n", gpiobaraddr);
+
+	/* Accessing a NULL pointer BAR is evil. Don't do it. */
+	if (gpiobaraddr) {
+		/* Map the BAR. We access bytewise and wordwise at 0x530. */
+		gpiobar = physmap("MCP67 GPIO", gpiobaraddr, 0x534);
+/* Guessed. If this is correct, migrate to a separate MCP67 SPI driver. */
+#define MCP67_SPI_CS		(1 << 1)
+#define MCP67_SPI_SCK		(1 << 2)
+#define MCP67_SPI_MOSI		(1 << 3)
+#define MCP67_SPI_MISO		(1 << 4)
+#define MCP67_SPI_ENABLE	(1 << 0)
+#define MCP67_SPI_IDLE		(1 << 8)
+		status = mmio_readw(gpiobar + 0x530);
+		msg_pdbg("SPI control is 0x%04x, enable=%i, idle=%i\n",
+			 status, status & 0x1, (status >> 8) & 0x1);
+	}
+	msg_pinfo("Please send the output of \"flashrom -V\" to "
+		  "flashrom@flashrom.org to help us finish support for your "
+		  "chipset. Thanks.\n");
+
+	/* Not sure if this is still correct. No docs as usual. */
+	result = enable_flash_mcp55(dev, name);
+
+	return result;
+}
+
+/* This is a shot in the dark. Even if the code is totally bogus for some
+ * chipsets, users will at least start to send in reports.
+ */
+static int enable_flash_mcp7x(struct pci_dev *dev, const char *name)
+{
+	uint8_t byte;
+	uint32_t gpiobaraddr;
+	struct pci_dev *smbusdev;
+
+	msg_pinfo("This chipset is not really supported yet. Guesswork...\n");
+
+	/* dev is the ISA bridge. No idea what the stuff below does. */
+	byte = pci_read_byte(dev, 0x8a);
+	msg_pdbg("ISA/LPC bridge reg 0x8a contents: 0x%02x, bit 6 is %i, bit 5 "
+		 "is %i\n", byte, (byte >> 6) & 0x1, (byte >> 5) & 0x1);
+
+	/* Look for the SMBus device (SMBus PCI class) */
+	smbusdev = pci_dev_find_vendorclass(0x10de, 0x0c05);
+	if (!smbusdev) {
+		msg_perr("ERROR: SMBus device not found. Aborting.\n");
+		exit(1);
+	}
+	msg_pdbg("Found SMBus device %04x:%04x at %02x:%02x:%01x\n",
+		smbusdev->vendor_id, smbusdev->device_id,
+		smbusdev->bus, smbusdev->dev, smbusdev->func);
+
+	/* Locate the BAR where the GPIOs live. */
+	gpiobaraddr = pci_read_long(smbusdev, 0x74);
+	msg_pdbg("GPIO BAR is at 0x%08x, ", gpiobaraddr);
+
+	msg_pinfo("Please send the output of \"flashrom -V\" to "
+		  "flashrom@flashrom.org to help us finish support for your "
+		  "chipset. Thanks.\n");
+
+	return 0;
+}
+
 static int enable_flash_ht1000(struct pci_dev *dev, const char *name)
 {
 	uint8_t byte;
@@ -1193,7 +1305,22 @@ 
 	{0x10de, 0x0365, OK, "NVIDIA", "MCP55",		enable_flash_mcp55}, /* LPC */
 	{0x10de, 0x0366, OK, "NVIDIA", "MCP55",		enable_flash_mcp55}, /* LPC */
 	{0x10de, 0x0367, OK, "NVIDIA", "MCP55",		enable_flash_mcp55}, /* Pro */
-	{0x10de, 0x0548, OK, "NVIDIA", "MCP67",		enable_flash_mcp55},
+	{0x10de, 0x03e0, NT, "NVIDIA", "MCP61",		enable_flash_mcp7x},
+	{0x10de, 0x03e1, NT, "NVIDIA", "MCP61",		enable_flash_mcp7x},
+	{0x10de, 0x03e2, NT, "NVIDIA", "MCP61",		enable_flash_mcp7x},
+	{0x10de, 0x03e3, NT, "NVIDIA", "MCP61",		enable_flash_mcp7x},
+	{0x10de, 0x0440, NT, "NVIDIA", "MCP65",		enable_flash_mcp7x},
+	{0x10de, 0x0441, NT, "NVIDIA", "MCP65",		enable_flash_mcp7x},
+	{0x10de, 0x0442, NT, "NVIDIA", "MCP65",		enable_flash_mcp7x},
+	{0x10de, 0x0443, NT, "NVIDIA", "MCP65",		enable_flash_mcp7x},
+	{0x10de, 0x0548, OK, "NVIDIA", "MCP67",		enable_flash_mcp67},
+	{0x10de, 0x075c, NT, "NVIDIA", "MCP78S",	enable_flash_mcp7x},
+	{0x10de, 0x075d, NT, "NVIDIA", "MCP78S",	enable_flash_mcp7x},
+	{0x10de, 0x07d7, NT, "NVIDIA", "MCP73",		enable_flash_mcp7x},
+	{0x10de, 0x0aac, NT, "NVIDIA", "MCP79",		enable_flash_mcp7x},
+	{0x10de, 0x0aad, NT, "NVIDIA", "MCP79",		enable_flash_mcp7x},
+	{0x10de, 0x0aae, NT, "NVIDIA", "MCP79",		enable_flash_mcp7x},
+	{0x10de, 0x0aaf, NT, "NVIDIA", "MCP79",		enable_flash_mcp7x},
 	{0x1039, 0x0496, NT, "SiS", "85C496+497",	enable_flash_sis85c496},
 	{0x1039, 0x0406, NT, "SiS", "501/5101/5501",	enable_flash_sis501},
 	{0x1039, 0x5511, NT, "SiS", "5511",		enable_flash_sis5511},