Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:7 as permitted sender) client-ip=2620:137:e000::3:7;
Received-SPF: Pass (protection.outlook.com: domain of amd.com designates
 165.204.84.17 as permitted sender) receiver=protection.outlook.com;
 client-ip=165.204.84.17; helo=SATLEXMB04.amd.com; pr=C
From:   Muralidhara M K <muralimk@amd.com>
To:     <linux-edac@vger.kernel.org>, <x86@kernel.org>
CC:     <linux-kernel@vger.kernel.org>, <bp@alien8.de>,
        <mchehab@kernel.org>, Muralidhara M K <muralidhara.mk@amd.com>
Subject: [PATCH 3/7] RAS: Add MCA Error address conversion for UMC
Date:   Wed, 25 Oct 2023 07:33:35 +0000
Message-ID: <20231025073339.630093-4-muralimk@amd.com>
In-Reply-To: <20231025073339.630093-1-muralimk@amd.com>
References: <20231025073339.630093-1-muralimk@amd.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Content-Type: text/plain
X-MS-Exchange-CrossTenant-OriginalArrivalTime: 25 Oct 2023 07:35:18.6956
 (UTC)
X-MS-Exchange-CrossTenant-Network-Message-Id: 7074fb77-3bf6-46e6-5b7c-08dbd52cf035
X-MS-Exchange-CrossTenant-Id: 3dd8961f-e488-4e60-8e11-a82d994e183d
X-MS-Exchange-CrossTenant-OriginalAttributedTenantConnectingIp: TenantId=3dd8961f-e488-4e60-8e11-a82d994e183d;Ip=[165.204.84.17];Helo=[SATLEXMB04.amd.com]
X-MS-Exchange-CrossTenant-AuthSource: CO1PEPF000044F0.namprd05.prod.outlook.com
X-MS-Exchange-CrossTenant-AuthAs: Anonymous
X-MS-Exchange-CrossTenant-FromEntityHeader: HybridOnPrem
X-MS-Exchange-Transport-CrossTenantHeadersStamped: CH0PR12MB5236
X-Spam-Status: No, score=-1.1 required=5.0 tests=BAYES_00,DKIMWL_WL_HIGH,
        DKIM_SIGNED,DKIM_VALID,DKIM_VALID_AU,DKIM_VALID_EF,FORGED_SPF_HELO,
        RCVD_IN_DNSWL_BLOCKED,RCVD_IN_MSPIKE_H2,SPF_HELO_PASS,SPF_NONE
        autolearn=no autolearn_force=no version=3.4.6
X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on
        lindbergh.monkeyblade.net
Precedence: bulk
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org
X-Greylist: Sender passed SPF test, not delayed by milter-greylist-4.6.4 (snail.vger.email [0.0.0.0]); Wed, 25 Oct 2023 00:38:09 -0700 (PDT)

From: Muralidhara M K <muralidhara.mk@amd.com>

On AMD systems with HBM3 memory, The reported MCA address is
DRAM address which needs to be converted to normalized address
before the data fabric address translation.

MI300A models have on-chip HBM3 memory capable of On-Die ECC support.
On-Die ECC error address to MCA is a encoded address reported with
a DRAM address (PC/SID/Bank/ROW/COL) instead of normalized address
unlike MI200s UMC ECC, as the implementation difference between
HBM3 On-Die ECC and HBM2 host ECC.
Because On-Die ECC address reporting is done in the back-end of UMC
and it no longer has normalized address at that point.
So software needs to convert the reported MCA Error Address back to
normalized address.

Signed-off-by: Muralidhara M K <muralidhara.mk@amd.com>
---
Link: https://lore.kernel.org/linux-edac/20230720125425.3735538-1-muralimk@amd.com/T/#m225efdf5812820efd084158bd8cdf40cad1a5af6

 drivers/ras/amd/atl/umc.c | 145 ++++++++++++++++++++++++++++++++++++++
 1 file changed, 145 insertions(+)

diff --git a/drivers/ras/amd/atl/umc.c b/drivers/ras/amd/atl/umc.c
index f334be0dc034..fa8c3072a54f 100644
--- a/drivers/ras/amd/atl/umc.c
+++ b/drivers/ras/amd/atl/umc.c
@@ -12,6 +12,147 @@
 
 #include "internal.h"
 
+static bool internal_bit_wise_xor(u32 inp)
+{
+	bool tmp = 0;
+	int i;
+
+	for (i = 0; i < 32; i++)
+		tmp = tmp ^ ((inp >> i) & 0x1);
+
+	return tmp;
+}
+
+/*
+ * Mapping of MCA decoded error address bit location to
+ * normalized address on MI300A systems.
+ */
+static const u8 umc_mca2na_mapping[] = {
+	0,  5,  6,  8,  9,  14, 12, 13,
+	10, 11, 15, 16, 17, 18, 19, 20,
+	21, 22, 23, 24, 25, 26, 27, 28,
+	7,  29, 30,
+};
+
+/* AddrHashBank and AddrHashPC/PC2 umc register bit fields */
+static struct {
+	u32 xor_enable	:1;
+	u32 col_xor	:13;
+	u32 row_xor	:18;
+} addr_hash_pc, addr_hash_bank[4];
+
+static struct {
+	u32 bank_xor	:6;
+} addr_hash_pc2;
+
+#define COLUMN_LOCATION		GENMASK(5, 1)
+#define ROW_LOCATION		GENMASK(23, 10)
+/*
+ * The location of bank, column and row are fixed.
+ * location of column bit must be NA[5].
+ * Row bits are always placed in a contiguous stretch of NA above the
+ * column and bank bits.
+ * Bits below the row bits can be either column or bank in any order,
+ * with the exception that NA[5] must be a column bit.
+ * Stack ID(SID) bits are placed in the MSB position of the NA.
+ */
+static int umc_ondie_addr_to_normaddr(u64 mca_addr, u16 nid)
+{
+	u32 bank[4], bank_hash[4], pc_hash;
+	u32 col, row, rawbank = 0, pc;
+	int i, temp = 0, err;
+	u64 mca2na;
+
+	/* Default umc base address on MI300A systems */
+	u32 gpu_umc_base = 0x90000;
+
+	/*
+	 * Error address logged on MI300A systems is ondie MCA address
+	 * in the format MCA_Addr[27:0] =
+	 *	{SID[1:0],PC[0],row[14:0],bank[3:0],col[4:0],1'b0}
+	 * The bit locations are calculated as per umc_mca2na_mapping[]
+	 * to find normalized address.
+	 * Refer F19 M90h BKDG Section 20.3.1.3 for clarifications
+	 *
+	 * XORs need to be applied based on the hash settings below.
+	 */
+
+	/* Calculate column and row */
+	col = FIELD_GET(COLUMN_LOCATION, mca_addr);
+	row = FIELD_GET(ROW_LOCATION, mca_addr);
+
+	/* Apply hashing on below banks for bank calculation */
+	for (i = 0; i < 4; i++)
+		bank_hash[i] = (mca_addr >> (6 + i)) & 0x1;
+
+	/* bank hash algorithm */
+	for (i = 0; i < 4; i++) {
+		/* Read AMD PPR UMC::AddrHashBank register */
+		err = amd_smn_read(nid, gpu_umc_base + 0xC8 + (i * 4), &temp);
+		if (err)
+			return err;
+
+		addr_hash_bank[i].xor_enable = temp & 1;
+		addr_hash_bank[i].col_xor = FIELD_GET(GENMASK(13, 1), temp);
+		addr_hash_bank[i].row_xor = FIELD_GET(GENMASK(31, 14), temp);
+		/* bank hash selection */
+		bank[i] = bank_hash[i] ^ (addr_hash_bank[i].xor_enable &
+			  (internal_bit_wise_xor(col & addr_hash_bank[i].col_xor) ^
+			  internal_bit_wise_xor(row & addr_hash_bank[i].row_xor)));
+	}
+
+	/* To apply hash on pc bit */
+	pc_hash = (mca_addr >> 25) & 0x1;
+
+	/* Read AMD PPR UMC::CH::AddrHashPC register */
+	err = amd_smn_read(nid, gpu_umc_base + 0xE0, &temp);
+	if (err)
+		return err;
+
+	addr_hash_pc.xor_enable = temp & 1;
+	addr_hash_pc.col_xor = FIELD_GET(GENMASK(13, 1), temp);
+	addr_hash_pc.row_xor = FIELD_GET(GENMASK(31, 14), temp);
+
+	/* Read AMD PPR UMC::CH::AddrHashPC2 register*/
+	err = amd_smn_read(nid, gpu_umc_base + 0xE4, &temp);
+	if (err)
+		return err;
+
+	addr_hash_pc2.bank_xor = FIELD_GET(GENMASK(5, 0), temp);
+
+	/* Calculate bank value from bank[0..3], bank[4] and bank[5] */
+	for (i = 0; i < 4; i++)
+		rawbank |= (bank[i] & 1) << i;
+
+	rawbank |= (mca_addr >> 22) & 0x30;
+
+	/* pseudochannel(pc) hash selection */
+	pc = pc_hash ^ (addr_hash_pc.xor_enable &
+		(internal_bit_wise_xor(col & addr_hash_pc.col_xor) ^
+		internal_bit_wise_xor(row & addr_hash_pc.row_xor) ^
+		internal_bit_wise_xor(rawbank & addr_hash_pc2.bank_xor)));
+
+	/* Mask b'25(pc_bit) and b'[9:6](bank) */
+	mca_addr &= ~0x20003c0ULL;
+
+	for (i = 0; i < 4; i++)
+		mca_addr |= (bank[i] << (6 + i));
+
+	 mca_addr |= (pc << 25);
+
+	/* NA[4..0] is fixed */
+	mca2na = 0x0;
+	/* convert mca error address to normalized address */
+	for (i = 1; i < ARRAY_SIZE(umc_mca2na_mapping); i++)
+		mca2na |= ((mca_addr >> i) & 0x1) << umc_mca2na_mapping[i];
+
+	mca_addr = mca2na;
+	pr_info("Error Addr: 0x%016llx\n", mca_addr);
+	pr_info("Error hit on: Bank %d Row %d Column %d\n", rawbank, row, col);
+
+	return mca_addr;
+}
+
 static u8 get_socket_id(struct mce *m)
 {
 	return m->socketid;
@@ -36,6 +177,10 @@ static u8 get_die_id(struct mce *m)
 
 static u64 get_norm_addr(struct mce *m)
 {
+	/* MI300: DRAM->Normalized translation */
+	if (df_cfg.rev == DF4p5 && df_cfg.flags.heterogeneous)
+		return umc_ondie_addr_to_normaddr(m->addr, get_socket_id(m));
+
 	return m->addr;
 }
 
-- 
2.25.1