diff mbox series

[v3] EDAC/bluefield: Use Arm SMC for EMI access on BlueField-2

Message ID 20241021233013.18405-1-davthompson@nvidia.com (mailing list archive)
State New
Headers show
Series [v3] EDAC/bluefield: Use Arm SMC for EMI access on BlueField-2 | expand

Commit Message

David Thompson Oct. 21, 2024, 11:30 p.m. UTC
The BlueField EDAC driver supports the first generation BlueField-1 SoC,
but not the second generation BlueField-2 SoC.  The BlueField-2 SoC is
different in that only secure accesses are allowed to the External Memory
Interface (EMI) register block. On BlueField-2, all read/write accesses
from Linux to EMI registers are routed via Arm Secure Monitor Call (SMC)
through Arm Trusted Firmware (ATF), which runs at EL3 privileged state.

On BlueField-1, EMI registers are mapped and accessed directly. In order
to support BlueField-2, the driver's read and write access methods must
be extended with additional logic to include secure access to the EMI
registers via SMCs.

Signed-off-by: David Thompson <davthompson@nvidia.com>
Reviewed-by: Shravan Kumar Ramani <shravankr@nvidia.com>
---
v2 -> v3
a) added comments for new members of bluefield_edac_priv structure
b) renamed static functions "bluefield_edac_secure_[readl|writel]"
   to "secure_[readl|writel]". Note: the static functions
   "bluefield_edac_[readl|writel]" are not renamed to "readl|writel"
   because that would conflict with IO primitives in kernel.
c) re-wrote switch statements in "secure_[readl|writel]" to if-else
d) re-wrote function signature of "bluefield_edac_[readl|writel]"
   so that "priv" is first argument
e) fixed line break in "bluefield_edac_mc_probe"
f) re-worded commit message to focus on the patch's reason

v1 -> v2
a) removed #defines for SMC function IDs that are not used
b) added "bluefield_edac_" prefix to "secure_readl/writel" functions
c) added "bluefield_" prefix to "edac_readl/writel" functions
d) changed logic to use "uN" typedefs instead of "uintN_t"
e) added initialization of "priv->dev" in probe()
f) added more details to commit message

v0 -> v1
Updated commit message
---
 drivers/edac/bluefield_edac.c | 169 ++++++++++++++++++++++++++++++----
 1 file changed, 150 insertions(+), 19 deletions(-)

Comments

Borislav Petkov Oct. 22, 2024, 12:32 p.m. UTC | #1
On Mon, Oct 21, 2024 at 07:30:13PM -0400, David Thompson wrote:
>  drivers/edac/bluefield_edac.c | 169 ++++++++++++++++++++++++++++++----
>  1 file changed, 150 insertions(+), 19 deletions(-)

Some trivial simplifications ontop along with moving the struct member
comments above them (not sideways) as requested.

I'll queue it with them ontop unless you see issues.

---
diff --git a/drivers/edac/bluefield_edac.c b/drivers/edac/bluefield_edac.c
index 126efb96deee..739132e5ed8a 100644
--- a/drivers/edac/bluefield_edac.c
+++ b/drivers/edac/bluefield_edac.c
@@ -89,7 +89,7 @@ struct bluefield_edac_priv {
 	/* access to secure regs supported */
 	bool svc_sreg_support;
 	/* SMC table# for secure regs access */
-	u32 sreg_tbl_edac;
+	u32 sreg_tbl;
 };
 
 static u64 smc_call1(u64 smc_op, u64 smc_arg)
@@ -138,15 +138,13 @@ static int secure_writel(void __iomem *addr, u32 data, u32 sreg_tbl)
 
 static int bluefield_edac_readl(struct bluefield_edac_priv *priv, u32 offset, u32 *result)
 {
-	bool sreg_support = priv->svc_sreg_support;
-	u32 sreg_tbl = priv->sreg_tbl_edac;
 	void __iomem *addr;
 	int err = 0;
 
 	addr = priv->emi_base + offset;
 
-	if (sreg_support)
-		err = secure_readl(addr, result, sreg_tbl);
+	if (priv->svc_sreg_support)
+		err = secure_readl(addr, result, priv->sreg_tbl);
 	else
 		*result = readl(addr);
 
@@ -155,15 +153,13 @@ static int bluefield_edac_readl(struct bluefield_edac_priv *priv, u32 offset, u3
 
 static int bluefield_edac_writel(struct bluefield_edac_priv *priv, u32 offset, u32 data)
 {
-	bool sreg_support = priv->svc_sreg_support;
-	u32 sreg_tbl = priv->sreg_tbl_edac;
 	void __iomem *addr;
 	int err = 0;
 
 	addr = priv->emi_base + offset;
 
-	if (sreg_support)
-		err = secure_writel(addr, data, sreg_tbl);
+	if (priv->svc_sreg_support)
+		err = secure_writel(addr, data, priv->sreg_tbl);
 	else
 		writel(data, addr);
 
@@ -393,7 +389,7 @@ static int bluefield_edac_mc_probe(struct platform_device *pdev)
 	 * b) property is present - indirectly access registers via SMC calls
 	 *    (assuming required Silicon Provider service version found)
 	 */
-	if (device_property_read_u32(dev, "sec_reg_block", &priv->sreg_tbl_edac)) {
+	if (device_property_read_u32(dev, "sec_reg_block", &priv->sreg_tbl)) {
 		priv->svc_sreg_support = false;
 	} else {
 		/*
David Thompson Oct. 22, 2024, 1:33 p.m. UTC | #2
> -----Original Message-----
> From: Borislav Petkov <bp@alien8.de>
> Sent: Tuesday, October 22, 2024 8:33 AM
> To: David Thompson <davthompson@nvidia.com>
> Cc: Shravan Ramani <shravankr@nvidia.com>; tony.luck@intel.com;
> james.morse@arm.com; mchehab@kernel.org; rric@kernel.org; linux-
> edac@vger.kernel.org; linux-kernel@vger.kernel.org
> Subject: Re: [PATCH v3] EDAC/bluefield: Use Arm SMC for EMI access on
> BlueField-2
> 
> On Mon, Oct 21, 2024 at 07:30:13PM -0400, David Thompson wrote:
> >  drivers/edac/bluefield_edac.c | 169
> > ++++++++++++++++++++++++++++++----
> >  1 file changed, 150 insertions(+), 19 deletions(-)
> 
> Some trivial simplifications ontop along with moving the struct member
> comments above them (not sideways) as requested.
> 
> I'll queue it with them ontop unless you see issues.
> 

I reviewed the changes you have made to my v3.  Looks good to me.

Thank you Borislav.

Regards, Dave
Borislav Petkov Oct. 22, 2024, 4:49 p.m. UTC | #3
On Tue, Oct 22, 2024 at 01:33:01PM +0000, David Thompson wrote:
> I reviewed the changes you have made to my v3.  Looks good to me.

Thanks.

Now queued.
diff mbox series

Patch

diff --git a/drivers/edac/bluefield_edac.c b/drivers/edac/bluefield_edac.c
index 0e539c107351..6d52f129f914 100644
--- a/drivers/edac/bluefield_edac.c
+++ b/drivers/edac/bluefield_edac.c
@@ -47,13 +47,22 @@ 
 #define MLXBF_EDAC_MAX_DIMM_PER_MC	2
 #define MLXBF_EDAC_ERROR_GRAIN		8
 
+#define MLXBF_WRITE_REG_32		(0x82000009)
+#define MLXBF_READ_REG_32		(0x8200000A)
+#define MLXBF_SIP_SVC_VERSION		(0x8200ff03)
+
+#define MLXBF_SMCCC_ACCESS_VIOLATION	(-4)
+
+#define MLXBF_SVC_REQ_MAJOR		0
+#define MLXBF_SVC_REQ_MINOR		3
+
 /*
- * Request MLNX_SIP_GET_DIMM_INFO
+ * Request MLXBF_SIP_GET_DIMM_INFO
  *
  * Retrieve information about DIMM on a certain slot.
  *
  * Call register usage:
- * a0: MLNX_SIP_GET_DIMM_INFO
+ * a0: MLXBF_SIP_GET_DIMM_INFO
  * a1: (Memory controller index) << 16 | (Dimm index in memory controller)
  * a2-7: not used.
  *
@@ -61,7 +70,7 @@ 
  * a0: MLXBF_DIMM_INFO defined below describing the DIMM.
  * a1-3: not used.
  */
-#define MLNX_SIP_GET_DIMM_INFO		0x82000008
+#define MLXBF_SIP_GET_DIMM_INFO		0x82000008
 
 /* Format for the SMC response about the memory information */
 #define MLXBF_DIMM_INFO__SIZE_GB GENMASK_ULL(15, 0)
@@ -72,9 +81,12 @@ 
 #define MLXBF_DIMM_INFO__PACKAGE_X GENMASK_ULL(31, 24)
 
 struct bluefield_edac_priv {
+	struct device *dev;  /* pointer to device structure */
 	int dimm_ranks[MLXBF_EDAC_MAX_DIMM_PER_MC];
 	void __iomem *emi_base;
 	int dimm_per_mc;
+	bool svc_sreg_support; /* access to secure regs supported */
+	u32 sreg_tbl_edac;     /* SMC table# for secure regs access */
 };
 
 static u64 smc_call1(u64 smc_op, u64 smc_arg)
@@ -86,6 +98,75 @@  static u64 smc_call1(u64 smc_op, u64 smc_arg)
 	return res.a0;
 }
 
+static int secure_readl(void __iomem *addr, u32 *result, u32 sreg_tbl)
+{
+	struct arm_smccc_res res;
+	int status;
+
+	arm_smccc_smc(MLXBF_READ_REG_32, sreg_tbl, (uintptr_t)addr,
+		      0, 0, 0, 0, 0, &res);
+
+	status = res.a0;
+
+	if (status == SMCCC_RET_NOT_SUPPORTED ||
+	    status == MLXBF_SMCCC_ACCESS_VIOLATION)
+		return -1;
+
+	*result = (u32)res.a1;
+	return 0;
+}
+
+static int secure_writel(void __iomem *addr, u32 data, u32 sreg_tbl)
+{
+	struct arm_smccc_res res;
+	int status;
+
+	arm_smccc_smc(MLXBF_WRITE_REG_32, sreg_tbl, data, (uintptr_t)addr,
+		      0, 0, 0, 0, &res);
+
+	status = res.a0;
+
+	if (status == SMCCC_RET_NOT_SUPPORTED ||
+	    status == MLXBF_SMCCC_ACCESS_VIOLATION)
+		return -1;
+	else
+		return 0;
+}
+
+static int bluefield_edac_readl(struct bluefield_edac_priv *priv, u32 offset, u32 *result)
+{
+	bool sreg_support = priv->svc_sreg_support;
+	u32 sreg_tbl = priv->sreg_tbl_edac;
+	void __iomem *addr;
+	int err = 0;
+
+	addr = priv->emi_base + offset;
+
+	if (sreg_support)
+		err = secure_readl(addr, result, sreg_tbl);
+	else
+		*result = readl(addr);
+
+	return err;
+}
+
+static int bluefield_edac_writel(struct bluefield_edac_priv *priv, u32 offset, u32 data)
+{
+	bool sreg_support = priv->svc_sreg_support;
+	u32 sreg_tbl = priv->sreg_tbl_edac;
+	void __iomem *addr;
+	int err = 0;
+
+	addr = priv->emi_base + offset;
+
+	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 +180,7 @@  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 +190,19 @@  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 = bluefield_edac_writel(priv, MLXBF_ECC_LATCH_SEL, ecc_latch_select);
+	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 = bluefield_edac_readl(priv, MLXBF_SYNDROM, &dram_syndrom);
+	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 +213,21 @@  static void bluefield_gather_report_ecc(struct mem_ctl_info *mci,
 		return;
 	}
 
-	dram_additional_info = readl(priv->emi_base + MLXBF_ADD_INFO);
+	err = bluefield_edac_readl(priv, MLXBF_ADD_INFO, &dram_additional_info);
+	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 = bluefield_edac_readl(priv, MLXBF_ERR_ADDR_0, &edea0);
+	if (err)
+		dev_err(priv->dev, "Error addr 0 read failed.\n");
+
+	err = bluefield_edac_readl(priv, MLXBF_ERR_ADDR_1, &edea1);
+	if (err)
+		dev_err(priv->dev, "Error addr 1 read failed.\n");
 
 	ecc_dimm_addr = ((u64)edea1 << 32) | edea0;
 
@@ -147,6 +241,7 @@  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 +250,10 @@  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 = bluefield_edac_readl(priv, MLXBF_ECC_CNT, &ecc_count);
+	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 +270,11 @@  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 = bluefield_edac_writel(priv, MLXBF_ECC_ERR, ecc_error);
+		if (err)
+			dev_err(priv->dev, "ECC Error write failed.\n");
+	}
 }
 
 /* Initialize the DIMMs information for the given memory controller. */
@@ -189,7 +290,7 @@  static void bluefield_edac_init_dimms(struct mem_ctl_info *mci)
 		dimm = mci->dimms[i];
 
 		smc_arg = mem_ctrl_idx << 16 | i;
-		smc_info = smc_call1(MLNX_SIP_GET_DIMM_INFO, smc_arg);
+		smc_info = smc_call1(MLXBF_SIP_GET_DIMM_INFO, smc_arg);
 
 		if (!FIELD_GET(MLXBF_DIMM_INFO__SIZE_GB, smc_info)) {
 			dimm->mtype = MEM_EMPTY;
@@ -244,6 +345,7 @@  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;
@@ -279,13 +381,43 @@  static int bluefield_edac_mc_probe(struct platform_device *pdev)
 		return -ENOMEM;
 
 	priv = mci->pvt_info;
+	priv->dev = dev;
+
+	/*
+	 * The "sec_reg_block" property in the ACPI table determines the method
+	 * the driver uses to access the EMI registers:
+	 * a) property is not present - directly access registers via readl/writel
+	 * b) property is present - indirectly access registers via SMC calls
+	 *    (assuming required Silicon Provider service version found)
+	 */
+	if (device_property_read_u32(dev, "sec_reg_block", &priv->sreg_tbl_edac)) {
+		priv->svc_sreg_support = false;
+	} else {
+		/*
+		 * Check for minimum required Arm Silicon Provider (SiP) service
+		 * version, ensuring support of required SMC function IDs.
+		 */
+		arm_smccc_smc(MLXBF_SIP_SVC_VERSION, 0, 0, 0, 0, 0, 0, 0, &res);
+		if (res.a0 == MLXBF_SVC_REQ_MAJOR &&
+		    res.a1 >= MLXBF_SVC_REQ_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;
@@ -320,7 +452,6 @@  static int bluefield_edac_mc_probe(struct platform_device *pdev)
 	edac_mc_free(mci);
 
 	return ret;
-
 }
 
 static void bluefield_edac_mc_remove(struct platform_device *pdev)