Message-ID: <1240931173-17477-16-git-send-email-borislav.petkov@amd.com>
Date:	Tue, 28 Apr 2009 17:06:07 +0200
From:	Borislav Petkov <borislav.petkov@....com>
To:	akpm@...ux-foundation.org, greg@...ah.com
CC:	<linux-kernel@...r.kernel.org>,
	Doug Thompson <dougthompson@...ssion.com>,
	Borislav Petkov <borislav.petkov@....com>
Subject: [PATCH 15/21] amd64_edac: add x4 chipkill syndrome mapping table

From: Doug Thompson <dougthompson@...ssion.com>

Signed-off-by: Doug Thompson <dougthompson@...ssion.com>
Signed-off-by: Borislav Petkov <borislav.petkov@....com>
---
 drivers/edac/amd64_edac.c |  141 ++++++++++++++++++++++++++++++++++++++++++++-
 1 files changed, 140 insertions(+), 1 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 49c0ce0..2de797d 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -2171,7 +2171,6 @@ static void k8_map_sysaddr_to_csrow(struct mem_ctl_info *mci,
 
 	/* CHIPKILL enabled */
 	if (info->nbcfg & K8_NBCFG_CHIPKILL) {
-		/* x4 chipkill ecc mode - determine channel */
 		channel = get_channel_from_ecc_syndrome(syndrome);
 		if (channel < 0) {
 			/* Syndrome didn't map, so we don't know which of
@@ -3163,4 +3162,144 @@ static void do_rdmsr(int cpu, u32 reg, u32 *eax, u32 *edx)
 	*edx = cmd.data[1];
 }
 
+/*
+ * syndrome mapping table for ECC ChipKill devices
+ *
+ * The comment in each row is the token (nibble) number that is in error.
+ * The least significant nibble of the syndrome is the mask for the bits
+ * that are in error (need to be toggled) for the particular nibble.
+ *
+ * Each row contains 16 entries.
+ * The first entry (0th) is the channel number for that row of syndromes.
+ * The remaining 15 entries are the syndromes for the respective Error
+ * bit mask index.
+ *
+ * 1st index entry is 0x0001 mask, indicating that the rightmost bit is the
+ * bit in error.
+ * The 2nd index entry is 0x0010 that the second bit is damaged.
+ * The 3rd index entry is 0x0011 indicating that the rightmost 2 bits
+ * are damaged.
+ * Thus so on until index 15, 0x1111, whose entry has the syndrome
+ * indicating that all 4 bits are damaged.
+ *
+ * A search is performed on this table looking for a given syndrome.
+ *
+ * See the AMD documentation for ECC syndromes. This ECC table is valid
+ * across all the versions of the AMD64 processors.
+ *
+ * A fast lookup is to use the LAST four bits of the 16-bit syndrome as a
+ * COLUMN index, then search all ROWS of that column, looking for a match
+ * with the input syndrome. The ROW value will be the token number.
+ *
+ * The 0'th entry on that row, can be returned as the CHANNEL (0 or 1) of this
+ * error.
+ */
+#define NUMBER_ECC_ROWS  36
+static const unsigned short ecc_chipkill_syndromes[NUMBER_ECC_ROWS][16] = {
+	/* Channel 0 syndromes */
+	{/*0*/  0, 0xe821, 0x7c32, 0x9413, 0xbb44, 0x5365, 0xc776, 0x2f57,
+	   0xdd88, 0x35a9, 0xa1ba, 0x499b, 0x66cc, 0x8eed, 0x1afe, 0xf2df },
+	{/*1*/  0, 0x5d31, 0xa612, 0xfb23, 0x9584, 0xc8b5, 0x3396, 0x6ea7,
+	   0xeac8, 0xb7f9, 0x4cda, 0x11eb, 0x7f4c, 0x227d, 0xd95e, 0x846f },
+	{/*2*/  0, 0x0001, 0x0002, 0x0003, 0x0004, 0x0005, 0x0006, 0x0007,
+	   0x0008, 0x0009, 0x000a, 0x000b, 0x000c, 0x000d, 0x000e, 0x000f },
+	{/*3*/  0, 0x2021, 0x3032, 0x1013, 0x4044, 0x6065, 0x7076, 0x5057,
+	   0x8088, 0xa0a9, 0xb0ba, 0x909b, 0xc0cc, 0xe0ed, 0xf0fe, 0xd0df },
+	{/*4*/  0, 0x5041, 0xa082, 0xf0c3, 0x9054, 0xc015, 0x30d6, 0x6097,
+	   0xe0a8, 0xb0e9, 0x402a, 0x106b, 0x70fc, 0x20bd, 0xd07e, 0x803f },
+	{/*5*/  0, 0xbe21, 0xd732, 0x6913, 0x2144, 0x9f65, 0xf676, 0x4857,
+	   0x3288, 0x8ca9, 0xe5ba, 0x5b9b, 0x13cc, 0xaded, 0xc4fe, 0x7adf },
+	{/*6*/  0, 0x4951, 0x8ea2, 0xc7f3, 0x5394, 0x1ac5, 0xdd36, 0x9467,
+	   0xa1e8, 0xe8b9, 0x2f4a, 0x661b, 0xf27c, 0xbb2d, 0x7cde, 0x358f },
+	{/*7*/  0, 0x74e1, 0x9872, 0xec93, 0xd6b4, 0xa255, 0x4ec6, 0x3a27,
+	   0x6bd8, 0x1f39, 0xf3aa, 0x874b, 0xbd6c, 0xc98d, 0x251e, 0x51ff },
+	{/*8*/  0, 0x15c1, 0x2a42, 0x3f83, 0xcef4, 0xdb35, 0xe4b6, 0xf177,
+	   0x4758, 0x5299, 0x6d1a, 0x78db, 0x89ac, 0x9c6d, 0xa3ee, 0xb62f },
+	{/*9*/  0, 0x3d01, 0x1602, 0x2b03, 0x8504, 0xb805, 0x9306, 0xae07,
+	   0xca08, 0xf709, 0xdc0a, 0xe10b, 0x4f0c, 0x720d, 0x590e, 0x640f },
+	{/*a*/  0, 0x9801, 0xec02, 0x7403, 0x6b04, 0xf305, 0x8706, 0x1f07,
+	   0xbd08, 0x2509, 0x510a, 0xc90b, 0xd60c, 0x4e0d, 0x3a0e, 0xa20f },
+	{/*b*/  0, 0xd131, 0x6212, 0xb323, 0x3884, 0xe9b5, 0x5a96, 0x8ba7,
+	   0x1cc8, 0xcdf9, 0x7eda, 0xafeb, 0x244c, 0xf57d, 0x465e, 0x976f },
+	{/*c*/  0, 0xe1d1, 0x7262, 0x93b3, 0xb834, 0x59e5, 0xca56, 0x2b87,
+	   0xdc18, 0x3dc9, 0xae7a, 0x4fab, 0x542c, 0x85fd, 0x164e, 0xf79f },
+	{/*d*/  0, 0x6051, 0xb0a2, 0xd0f3, 0x1094, 0x70c5, 0xa036, 0xc067,
+	   0x20e8, 0x40b9, 0x904a, 0x601b, 0x307c, 0x502d, 0x80de, 0xe08f },
+	{/*e*/  0, 0xa4c1, 0xf842, 0x5c83, 0xe6f4, 0x4235, 0x1eb6, 0xba77,
+	   0x7b58, 0xdf99, 0x831a, 0x27db, 0x9dac, 0x396d, 0x65ee, 0xc12f },
+	{/*f*/  0, 0x11c1, 0x2242, 0x3383, 0xc8f4, 0xd935, 0xeab6, 0xfb77,
+	   0x4c58, 0x5d99, 0x6e1a, 0x7fdb, 0x84ac, 0x956d, 0xa6ee, 0xb72f },
+
+	/* Channel 1 syndromes */
+	{/*10*/ 1, 0x45d1, 0x8a62, 0xcfb3, 0x5e34, 0x1be5, 0xd456, 0x9187,
+	   0xa718, 0xe2c9, 0x2d7a, 0x68ab, 0xf92c, 0xbcfd, 0x734e, 0x369f },
+	{/*11*/ 1, 0x63e1, 0xb172, 0xd293, 0x14b4, 0x7755, 0xa5c6, 0xc627,
+	   0x28d8, 0x4b39, 0x99aa, 0xfa4b, 0x3c6c, 0x5f8d, 0x8d1e, 0xeeff },
+	{/*12*/ 1, 0xb741, 0xd982, 0x6ec3, 0x2254, 0x9515, 0xfbd6, 0x4c97,
+	   0x33a8, 0x84e9, 0xea2a, 0x5d6b, 0x11fc, 0xa6bd, 0xc87e, 0x7f3f },
+	{/*13*/ 1, 0xdd41, 0x6682, 0xbbc3, 0x3554, 0xe815, 0x53d6, 0xce97,
+	   0x1aa8, 0xc7e9, 0x7c2a, 0xa1fb, 0x2ffc, 0xf2bd, 0x497e, 0x943f },
+	{/*14*/ 1, 0x2bd1, 0x3d62, 0x16b3, 0x4f34, 0x64e5, 0x7256, 0x5987,
+	   0x8518, 0xaec9, 0xb87a, 0x93ab, 0xca2c, 0xe1fd, 0xf74e, 0xdc9f },
+	{/*15*/ 1, 0x83c1, 0xc142, 0x4283, 0xa4f4, 0x2735, 0x65b6, 0xe677,
+	   0xf858, 0x7b99, 0x391a, 0xbadb, 0x5cac, 0xdf6d, 0x9dee, 0x1e2f },
+	{/*16*/ 1, 0x8fd1, 0xc562, 0x4ab3, 0xa934, 0x26e5, 0x6c56, 0xe387,
+	   0xfe18, 0x71c9, 0x3b7a, 0xb4ab, 0x572c, 0xd8fd, 0x924e, 0x1d9f },
+	{/*17*/ 1, 0x4791, 0x89e2, 0xce73, 0x5264, 0x15f5, 0xdb86, 0x9c17,
+	   0xa3b8, 0xe429, 0x2a5a, 0x6dcb, 0xf1dc, 0xb64d, 0x783e, 0x3faf },
+	{/*18*/ 1, 0x5781, 0xa9c2, 0xfe43, 0x92a4, 0xc525, 0x3b66, 0x6ce7,
+	   0xe3f8, 0xb479, 0x4a3a, 0x1dbb, 0x715c, 0x26dd, 0xd89e, 0x8f1f },
+	{/*19*/ 1, 0xbf41, 0xd582, 0x6ac3, 0x2954, 0x9615, 0xfcd6, 0x4397,
+	   0x3ea8, 0x81e9, 0xeb2a, 0x546b, 0x17fc, 0xa8bd, 0xc27e, 0x7d3f },
+	{/*1a*/ 1, 0x9891, 0xe1e2, 0x7273, 0x6464, 0xf7f5, 0x8586, 0x1617,
+	   0xb8b8, 0x2b29, 0x595a, 0xcacb, 0xdcdc, 0x4f4d, 0x3d3e, 0xaeaf },
+	{/*1b*/ 1, 0xcce1, 0x4472, 0x8893, 0xfdb4, 0x3f55, 0xb9c6, 0x7527,
+	   0x56d8, 0x9a39, 0x12aa, 0xde4b, 0xab6c, 0x678d, 0xef1e, 0x23ff },
+	{/*1c*/ 1, 0xa761, 0xf9b2, 0x5ed3, 0xe214, 0x4575, 0x1ba6, 0xbcc7,
+	   0x7328, 0xd449, 0x8a9a, 0x2dfb, 0x913c, 0x365d, 0x688e, 0xcfef },
+	{/*1d*/ 1, 0xff61, 0x55b2, 0xaad3, 0x7914, 0x8675, 0x2ca6, 0xd3c7,
+	   0x9e28, 0x6149, 0xcb9a, 0x34fb, 0xe73c, 0x185d, 0xb28e, 0x4def },
+	{/*1e*/ 1, 0x5451, 0xa8a2, 0xfcf3, 0x9694, 0xc2c5, 0x3e36, 0x6a67,
+	   0xebe8, 0xbfb9, 0x434a, 0x171b, 0x7d7c, 0x292d, 0xd5de, 0x818f },
+	{/*1f*/ 1, 0x6fc1, 0xb542, 0xda83, 0x19f4, 0x7635, 0xacb6, 0xc377,
+	   0x2e58, 0x4199, 0x9b1a, 0xf4db, 0x37ac, 0x586d, 0x82ee, 0xed2f },
+
+	/* ECC bits are also in the set of tokens and they too can go bad
+	 * first 2 cover channel 0, while the second 2 cover channel 1
+	 */
+	{/*20*/ 0, 0xbe01, 0xd702, 0x6903, 0x2104, 0x9f05, 0xf606, 0x4807,
+	   0x3208, 0x8c09, 0xe50a, 0x5b0b, 0x130c, 0xad0d, 0xc40e, 0x7a0f },
+	{/*21*/ 0, 0x4101, 0x8202, 0xc303, 0x5804, 0x1905, 0xda06, 0x9b07,
+	   0xac08, 0xed09, 0x2e0a, 0x6f0b, 0x640c, 0xb50d, 0x760e, 0x370f },
+	{/*22*/ 1, 0xc441, 0x4882, 0x8cc3, 0xf654, 0x3215, 0xbed6, 0x7a97,
+	   0x5ba8, 0x9fe9, 0x132a, 0xd76b, 0xadfc, 0x69bd, 0xe57e, 0x213f },
+	{/*23*/ 1, 0x7621, 0x9b32, 0xed13, 0xda44, 0xac65, 0x4176, 0x3757,
+	   0x6f88, 0x19a9, 0xf4ba, 0x829b, 0xb5cc, 0xc3ed, 0x2efe, 0x58df }
+};
 
+/*
+ * get_channel_from_ecc_syndrome
+ *
+ *	given the syndrome argument, scan each of the channel tables
+ *	looking for a syndrome match. Depending on which table it is
+ *	found, return the channel number
+ */
+static int get_channel_from_ecc_syndrome(unsigned short syndrome)
+{
+	int row;
+	int column;
+
+	debugf0("%s()\n", __func__);
+
+	/* Determine column to scan */
+	column = syndrome & 0xF;
+
+	/* Scan all rows, looking for syndrome, or end of table */
+	for (row = 0; row < NUMBER_ECC_ROWS; row++) {
+		if (ecc_chipkill_syndromes[row][column] == syndrome)
+			return ecc_chipkill_syndromes[row][0];
+	}
+
+	debugf0("%s(): syndrome(%x) not found\n", __func__, syndrome);
+	return -1;
+}
-- 
1.6.2.4


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists