sonic-buildimage/platform/mellanox/non-upstream-patches/patches/0243-UBUNTU-SAUCE-bluefield_edac-Add-SMC-support.patch
Kebo Liu e286869b24
[Mellanox] Update HW-MGMT package to new version V.7.0030.1011 (#16239)
- Why I did it
1. Update Mellanox HW-MGMT package to newer version V.7.0030.1011
2. Replace the SONiC PMON Thermal control algorithm with the one inside the HW-MGMT package on all Nvidia platforms
3. Support Spectrum-4 systems

- How I did it
1. Update the HW-MGMT package version number and submodule pointer
2. Remove the thermal control algorithm implementation from Mellanox platform API
3. Revise the patch to HW-MGMT package which will disable HW-MGMT from running on SIMX
4. Update the downstream kernel patch list

Signed-off-by: Kebo Liu <kebol@nvidia.com>
2023-09-06 11:32:08 +03:00

292 lines
8.6 KiB
Diff

From 9bebe7236f4e3d956feda9911ffee5f31dfbdb13 Mon Sep 17 00:00:00 2001
From: Shravan Kumar Ramani <shravankr@nvidia.com>
Date: Wed, 6 Jul 2022 03:37:38 -0400
Subject: [PATCH backport 5.10 44/63] UBUNTU: SAUCE: bluefield_edac: Add SMC
support
BugLink: https://launchpad.net/bugs/1980812
This patch adds secure read/write calls to bluefield_edac. The
ACPI table entry decides whether the secure calls need to be
used for accessing the EMI registers.
Signed-off-by: Shravan Kumar Ramani <shravankr@nvidia.com>
Signed-off-by: Ike Panhc <ike.pan@canonical.com>
---
drivers/edac/bluefield_edac.c | 168 +++++++++++++++++++++++++++++++---
1 file changed, 154 insertions(+), 14 deletions(-)
diff --git a/drivers/edac/bluefield_edac.c b/drivers/edac/bluefield_edac.c
index e4736eb37..8e1127a56 100644
--- a/drivers/edac/bluefield_edac.c
+++ b/drivers/edac/bluefield_edac.c
@@ -12,6 +12,7 @@
#include <linux/io.h>
#include <linux/module.h>
#include <linux/platform_device.h>
+#include <linux/version.h>
#include "edac_module.h"
@@ -47,6 +48,18 @@
#define MLXBF_EDAC_MAX_DIMM_PER_MC 2
#define MLXBF_EDAC_ERROR_GRAIN 8
+#define MLNX_WRITE_REG_32 (0x82000009)
+#define MLNX_READ_REG_32 (0x8200000A)
+#define MLNX_WRITE_REG_64 (0x8200000B)
+#define MLNX_READ_REG_64 (0x8200000C)
+#define MLNX_SIP_SVC_UID (0x8200ff01)
+#define MLNX_SIP_SVC_VERSION (0x8200ff03)
+
+#define SMCCC_ACCESS_VIOLATION (-4)
+
+#define MLNX_EDAC_SVC_REQ_MAJOR 0
+#define MLNX_EDAC_SVC_MIN_MINOR 3
+
/*
* Request MLNX_SIP_GET_DIMM_INFO
*
@@ -72,9 +85,12 @@
#define MLXBF_DIMM_INFO__PACKAGE_X GENMASK_ULL(31, 24)
struct bluefield_edac_priv {
+ struct device *dev;
int dimm_ranks[MLXBF_EDAC_MAX_DIMM_PER_MC];
void __iomem *emi_base;
int dimm_per_mc;
+ bool svc_sreg_support;
+ uint32_t sreg_tbl_edac;
};
static u64 smc_call1(u64 smc_op, u64 smc_arg)
@@ -86,6 +102,73 @@ static u64 smc_call1(u64 smc_op, u64 smc_arg)
return res.a0;
}
+static int secure_readl(void __iomem *addr, uint32_t *result, uint32_t sreg_tbl)
+{
+ struct arm_smccc_res res;
+ int status;
+
+ arm_smccc_smc(MLNX_READ_REG_32, sreg_tbl, (uintptr_t) addr,
+ 0, 0, 0, 0, 0, &res);
+
+ status = res.a0;
+
+ switch (status) {
+ case SMCCC_RET_NOT_SUPPORTED:
+ case SMCCC_ACCESS_VIOLATION:
+ return -1;
+ default:
+ *result = (uint32_t)res.a1;
+ return 0;
+ }
+
+}
+
+static int secure_writel(void __iomem *addr, uint32_t data, uint32_t sreg_tbl)
+{
+ struct arm_smccc_res res;
+ int status;
+
+ arm_smccc_smc(MLNX_WRITE_REG_32, sreg_tbl, data, (uintptr_t) addr,
+ 0, 0, 0, 0, &res);
+
+ status = res.a0;
+
+ switch (status) {
+ case SMCCC_RET_NOT_SUPPORTED:
+ case SMCCC_ACCESS_VIOLATION:
+ return -1;
+ default:
+ return 0;
+ }
+
+}
+
+static int edac_readl(void __iomem *addr, uint32_t *result,
+ bool sreg_support, uint32_t sreg_tbl)
+{
+ int err = 0;
+
+ if (sreg_support)
+ err = secure_readl(addr, result, sreg_tbl);
+ else
+ *result = readl(addr);
+
+ return err;
+}
+
+static int edac_writel(void __iomem *addr, uint32_t data,
+ bool sreg_support, uint32_t sreg_tbl)
+{
+ int err = 0;
+
+ if (sreg_support)
+ err = secure_writel(addr, data, sreg_tbl);
+ else
+ writel(data, addr);
+
+ return err;
+}
+
/*
* Gather the ECC information from the External Memory Interface registers
* and report it to the edac handler.
@@ -99,7 +182,7 @@ static void bluefield_gather_report_ecc(struct mem_ctl_info *mci,
u32 ecc_latch_select, dram_syndrom, serr, derr, syndrom;
enum hw_event_mc_err_type ecc_type;
u64 ecc_dimm_addr;
- int ecc_dimm;
+ int ecc_dimm, err;
ecc_type = is_single_ecc ? HW_EVENT_ERR_CORRECTED :
HW_EVENT_ERR_UNCORRECTED;
@@ -109,14 +192,22 @@ static void bluefield_gather_report_ecc(struct mem_ctl_info *mci,
* registers with information about the last ECC error occurrence.
*/
ecc_latch_select = MLXBF_ECC_LATCH_SEL__START;
- writel(ecc_latch_select, priv->emi_base + MLXBF_ECC_LATCH_SEL);
+ err = edac_writel(priv->emi_base + MLXBF_ECC_LATCH_SEL,
+ ecc_latch_select, priv->svc_sreg_support,
+ priv->sreg_tbl_edac);
+ if (err)
+ dev_err(priv->dev, "ECC latch select write failed.\n");
/*
* Verify that the ECC reported info in the registers is of the
* same type as the one asked to report. If not, just report the
* error without the detailed information.
*/
- dram_syndrom = readl(priv->emi_base + MLXBF_SYNDROM);
+ err = edac_readl(priv->emi_base + MLXBF_SYNDROM, &dram_syndrom,
+ priv->svc_sreg_support, priv->sreg_tbl_edac);
+ if (err)
+ dev_err(priv->dev, "DRAM syndrom read failed.\n");
+
serr = FIELD_GET(MLXBF_SYNDROM__SERR, dram_syndrom);
derr = FIELD_GET(MLXBF_SYNDROM__DERR, dram_syndrom);
syndrom = FIELD_GET(MLXBF_SYNDROM__SYN, dram_syndrom);
@@ -127,13 +218,24 @@ static void bluefield_gather_report_ecc(struct mem_ctl_info *mci,
return;
}
- dram_additional_info = readl(priv->emi_base + MLXBF_ADD_INFO);
+ err = edac_readl(priv->emi_base + MLXBF_ADD_INFO, &dram_additional_info,
+ priv->svc_sreg_support, priv->sreg_tbl_edac);
+ if (err)
+ dev_err(priv->dev, "DRAM additional info read failed.\n");
+
err_prank = FIELD_GET(MLXBF_ADD_INFO__ERR_PRANK, dram_additional_info);
ecc_dimm = (err_prank >= 2 && priv->dimm_ranks[0] <= 2) ? 1 : 0;
- edea0 = readl(priv->emi_base + MLXBF_ERR_ADDR_0);
- edea1 = readl(priv->emi_base + MLXBF_ERR_ADDR_1);
+ err = edac_readl(priv->emi_base + MLXBF_ERR_ADDR_0, &edea0,
+ priv->svc_sreg_support, priv->sreg_tbl_edac);
+ if (err)
+ dev_err(priv->dev, "Error addr 0 read failed.\n");
+
+ err = edac_readl(priv->emi_base + MLXBF_ERR_ADDR_1, &edea1,
+ priv->svc_sreg_support, priv->sreg_tbl_edac);
+ if (err)
+ dev_err(priv->dev, "Error addr 1 read failed.\n");
ecc_dimm_addr = ((u64)edea1 << 32) | edea0;
@@ -147,6 +249,7 @@ static void bluefield_edac_check(struct mem_ctl_info *mci)
{
struct bluefield_edac_priv *priv = mci->pvt_info;
u32 ecc_count, single_error_count, double_error_count, ecc_error = 0;
+ int err;
/*
* The memory controller might not be initialized by the firmware
@@ -155,7 +258,11 @@ static void bluefield_edac_check(struct mem_ctl_info *mci)
if (mci->edac_cap == EDAC_FLAG_NONE)
return;
- ecc_count = readl(priv->emi_base + MLXBF_ECC_CNT);
+ err = edac_readl(priv->emi_base + MLXBF_ECC_CNT, &ecc_count,
+ priv->svc_sreg_support, priv->sreg_tbl_edac);
+ if (err)
+ dev_err(priv->dev, "ECC count read failed.\n");
+
single_error_count = FIELD_GET(MLXBF_ECC_CNT__SERR_CNT, ecc_count);
double_error_count = FIELD_GET(MLXBF_ECC_CNT__DERR_CNT, ecc_count);
@@ -172,8 +279,12 @@ static void bluefield_edac_check(struct mem_ctl_info *mci)
}
/* Write to clear reported errors. */
- if (ecc_count)
- writel(ecc_error, priv->emi_base + MLXBF_ECC_ERR);
+ if (ecc_count) {
+ err = edac_writel(priv->emi_base + MLXBF_ECC_ERR, ecc_error,
+ priv->svc_sreg_support, priv->sreg_tbl_edac);
+ if (err)
+ dev_err(priv->dev, "ECC Error write failed.\n");
+ }
}
/* Initialize the DIMMs information for the given memory controller. */
@@ -244,6 +355,7 @@ static int bluefield_edac_mc_probe(struct platform_device *pdev)
struct bluefield_edac_priv *priv;
struct device *dev = &pdev->dev;
struct edac_mc_layer layers[1];
+ struct arm_smccc_res res;
struct mem_ctl_info *mci;
struct resource *emi_res;
unsigned int mc_idx, dimm_count;
@@ -280,12 +392,40 @@ static int bluefield_edac_mc_probe(struct platform_device *pdev)
priv = mci->pvt_info;
+ /*
+ * ACPI indicates whether we use SMCs to access registers or not.
+ * If sreg_tbl_perf is not present, just assume we're not using SMCs.
+ */
+ if (device_property_read_u32(dev,
+ "sec_reg_block", &priv->sreg_tbl_edac)) {
+ priv->svc_sreg_support = false;
+ } else {
+ /*
+ * Check service version to see if we actually do support the
+ * needed SMCs. If we have the calls we need, mark support for
+ * them in the pmc struct.
+ */
+ arm_smccc_smc(MLNX_SIP_SVC_VERSION, 0, 0, 0, 0, 0, 0, 0, &res);
+ if (res.a0 == MLNX_EDAC_SVC_REQ_MAJOR &&
+ res.a1 >= MLNX_EDAC_SVC_MIN_MINOR)
+ priv->svc_sreg_support = true;
+ else {
+ dev_err(dev, "Required SMCs are not supported.\n");
+ ret = -EINVAL;
+ goto err;
+ }
+ }
+
priv->dimm_per_mc = dimm_count;
- priv->emi_base = devm_ioremap_resource(dev, emi_res);
- if (IS_ERR(priv->emi_base)) {
- dev_err(dev, "failed to map EMI IO resource\n");
- ret = PTR_ERR(priv->emi_base);
- goto err;
+ if (!priv->svc_sreg_support) {
+ priv->emi_base = devm_ioremap_resource(dev, emi_res);
+ if (IS_ERR(priv->emi_base)) {
+ dev_err(dev, "failed to map EMI IO resource\n");
+ ret = PTR_ERR(priv->emi_base);
+ goto err;
+ }
+ } else {
+ priv->emi_base = (void __iomem *) emi_res->start;
}
mci->pdev = dev;
--
2.20.1