Tegra186: mce: Uncore Perfmon ARI Programming

Uncore perfmon appears to the CPU as a set of uncore perfmon registers
which can be read and written using the ARI interface. The MCE code
sequence handles reads and writes to these registers by manipulating
the underlying T186 uncore hardware.

To access an uncore perfmon register, CPU software writes the ARI
request registers to specify

* whether the operation is a read or a write,
* which uncore perfmon register to access,
* the uncore perfmon unit, group, and counter number (if necessary),
* the data to write (if the operation is a write).

It then initiates an ARI request to run the uncore perfmon sequence in
the MCE and reads the resulting value of the uncore perfmon register
and any status information from the ARI response registers.

The NS world's MCE driver issues MCE_CMD_UNCORE_PERFMON_REQ command
for the EL3 layer to start the entire sequence. Once the request
completes, the NS world would receive the command status in the X0
register and the command data in the X1 register.

Change-Id: I20bf2eca2385f7c8baa81e9445617ae711ecceea
Signed-off-by: Varun Wadekar <vwadekar@nvidia.com>

diff --git a/plat/nvidia/tegra/soc/t186/drivers/include/mce.h b/plat/nvidia/tegra/soc/t186/drivers/include/mce.h
index 38ca32c37d352001807bd2bf0faf9dd84ddfd3a2..66e212bff64b4793ea91a4e45fcd945bba769474 100644 (file)
--- a/plat/nvidia/tegra/soc/t186/drivers/include/mce.h
+++ b/plat/nvidia/tegra/soc/t186/drivers/include/mce.h
@@ -95,6 +95,7 @@ typedef enum mce_cmd {
        MCE_CMD_ROC_FLUSH_CACHE,
        MCE_CMD_ROC_CLEAN_CACHE,
        MCE_CMD_ENABLE_LATIC,
+       MCE_CMD_UNCORE_PERFMON_REQ,
        MCE_CMD_IS_CCX_ALLOWED = 0xFE,
        MCE_CMD_MAX = 0xFF,
 } mce_cmd_t;
@@ -201,6 +202,54 @@ typedef union mca_arg {
        uint64_t data;
 } mca_arg_t;
 
+/*******************************************************************************
+ * Uncore PERFMON ARI struct
+ ******************************************************************************/
+typedef union uncore_perfmon_req {
+       struct perfmon_command {
+               /*
+                * Commands: 0 = READ, 1 = WRITE
+                */
+               uint64_t cmd:8;
+               /*
+                * The unit group: L2=0, L3=1, ROC=2, MC=3, IOB=4
+                */
+               uint64_t grp:4;
+               /*
+                * Unit selector: Selects the unit instance, with 0 = Unit
+                * = (number of units in group) - 1.
+                */
+               uint64_t unit:4;
+               /*
+                * Selects the uncore perfmon register to access
+                */
+               uint64_t reg:8;
+               /*
+                * Counter number. Selects which counter to use for
+                * registers NV_PMEVCNTR and NV_PMEVTYPER.
+                */
+               uint64_t counter:8;
+       } perfmon_command;
+       struct perfmon_status {
+               /*
+                * Resulting command status
+                */
+               uint64_t val:8;
+               uint64_t unused:24;
+       } perfmon_status;
+       uint64_t data;
+} uncore_perfmon_req_t;
+
+#define UNCORE_PERFMON_CMD_READ                        0
+#define UNCORE_PERFMON_CMD_WRITE               1
+
+#define UNCORE_PERFMON_CMD_MASK                        0xFF
+#define UNCORE_PERFMON_UNIT_GRP_MASK           0xF
+#define UNCORE_PERFMON_SELECTOR_MASK           0xF
+#define UNCORE_PERFMON_REG_MASK                        0xFF
+#define UNCORE_PERFMON_CTR_MASK                        0xFF
+#define UNCORE_PERFMON_RESP_STATUS_MASK                0xFF
+
 /*******************************************************************************
  * Structure populated by arch specific code to export routines which perform
  * common low level MCE functions
@@ -331,6 +380,12 @@ typedef struct arch_mce_ops {
         * reset the entire system
         */
        void (*enter_ccplex_state)(uint32_t ari_base, uint32_t state_idx);
+       /*
+        * This ARI request reads/writes data from/to Uncore PERFMON
+        * registers
+        */
+       int (*read_write_uncore_perfmon)(uint32_t ari_base,
+                       uncore_perfmon_req_t req, uint64_t *data);
 } arch_mce_ops_t;
 
 int mce_command_handler(mce_cmd_t cmd, uint64_t arg0, uint64_t arg1,
@@ -363,6 +418,8 @@ int ari_roc_clean_cache(uint32_t ari_base);
 uint64_t ari_read_write_mca(uint32_t ari_base, mca_cmd_t cmd, uint64_t *data);
 int ari_update_ccplex_gsc(uint32_t ari_base, uint32_t gsc_idx);
 void ari_enter_ccplex_state(uint32_t ari_base, uint32_t state_idx);
+int ari_read_write_uncore_perfmon(uint32_t ari_base,
+               uncore_perfmon_req_t req, uint64_t *data);
 
 int nvg_enter_cstate(uint32_t ari_base, uint32_t state, uint32_t wake_time);
 int nvg_update_cstate_info(uint32_t ari_base, uint32_t cluster, uint32_t ccplex,

diff --git a/plat/nvidia/tegra/soc/t186/drivers/mce/ari.c b/plat/nvidia/tegra/soc/t186/drivers/mce/ari.c
index 147a358aec4fe59fe5f90c0db60fcb8f6f822fc5..e11d1600005aa509b0a890cede85c7c81e2809bd 100644 (file)
--- a/plat/nvidia/tegra/soc/t186/drivers/mce/ari.c
+++ b/plat/nvidia/tegra/soc/t186/drivers/mce/ari.c
@@ -389,3 +389,41 @@ void ari_enter_ccplex_state(uint32_t ari_base, uint32_t state_idx)
         */
        (void)ari_request_wait(ari_base, 0, TEGRA_ARI_MISC_CCPLEX, state_idx, 0);
 }
+
+int ari_read_write_uncore_perfmon(uint32_t ari_base,
+               uncore_perfmon_req_t req, uint64_t *data)
+{
+       int ret;
+       uint32_t val;
+
+       /* sanity check input parameters */
+       if (req.perfmon_command.cmd == UNCORE_PERFMON_CMD_READ && !data) {
+               ERROR("invalid parameters\n");
+               return EINVAL;
+       }
+
+       /*
+        * For "write" commands get the value that has to be written
+        * to the uncore perfmon registers
+        */
+       val = (req.perfmon_command.cmd == UNCORE_PERFMON_CMD_WRITE) ?
+               *data : 0;
+
+       ret = ari_request_wait(ari_base, 0, TEGRA_ARI_PERFMON, val, req.data);
+       if (ret)
+               return ret;
+
+       /* read the command status value */
+       req.perfmon_status.val = ari_get_response_high(ari_base) &
+                                UNCORE_PERFMON_RESP_STATUS_MASK;
+
+       /*
+        * For "read" commands get the data from the uncore
+        * perfmon registers
+        */
+       if ((req.perfmon_status.val == 0) && (req.perfmon_command.cmd ==
+            UNCORE_PERFMON_CMD_READ))
+               *data = ari_get_response_low(ari_base);
+
+       return (int)req.perfmon_status.val;
+}

diff --git a/plat/nvidia/tegra/soc/t186/drivers/mce/mce.c b/plat/nvidia/tegra/soc/t186/drivers/mce/mce.c
index 1a712dcca69aa6283d1c7c16c4d0b921e0d565a8..981545d3fdb4d2284e3fcb3cd6cd25fa380cb301 100644 (file)
--- a/plat/nvidia/tegra/soc/t186/drivers/mce/mce.c
+++ b/plat/nvidia/tegra/soc/t186/drivers/mce/mce.c
@@ -61,7 +61,8 @@ static arch_mce_ops_t nvg_mce_ops = {
        .roc_clean_cache = ari_roc_clean_cache,
        .read_write_mca = ari_read_write_mca,
        .update_ccplex_gsc = ari_update_ccplex_gsc,
-       .enter_ccplex_state = ari_enter_ccplex_state
+       .enter_ccplex_state = ari_enter_ccplex_state,
+       .read_write_uncore_perfmon = ari_read_write_uncore_perfmon
 };
 
 /* ARI functions handlers */
@@ -82,7 +83,8 @@ static arch_mce_ops_t ari_mce_ops = {
        .roc_clean_cache = ari_roc_clean_cache,
        .read_write_mca = ari_read_write_mca,
        .update_ccplex_gsc = ari_update_ccplex_gsc,
-       .enter_ccplex_state = ari_enter_ccplex_state
+       .enter_ccplex_state = ari_enter_ccplex_state,
+       .read_write_uncore_perfmon = ari_read_write_uncore_perfmon
 };
 
 typedef struct mce_config {
@@ -173,6 +175,7 @@ int mce_command_handler(mce_cmd_t cmd, uint64_t arg0, uint64_t arg1,
        uint64_t ret64 = 0, arg3, arg4, arg5;
        int ret = 0;
        mca_cmd_t mca_cmd;
+       uncore_perfmon_req_t req;
        cpu_context_t *ctx = cm_get_context(NON_SECURE);
        gp_regs_t *gp_regs = get_gpregs_ctx(ctx);
 
@@ -374,6 +377,15 @@ int mce_command_handler(mce_cmd_t cmd, uint64_t arg0, uint64_t arg1,
 
                break;
 #endif
+
+       case MCE_CMD_UNCORE_PERFMON_REQ:
+               memcpy(&req, &arg0, sizeof(arg0));
+               ret = ops->read_write_uncore_perfmon(cpu_ari_base, req, &arg1);
+
+               /* update context to return data */
+               write_ctx_reg(gp_regs, CTX_GPREG_X1, arg1);
+               break;
+
        default:
                ERROR("unknown MCE command (%d)\n", cmd);
                return EINVAL;

diff --git a/plat/nvidia/tegra/soc/t186/plat_sip_calls.c b/plat/nvidia/tegra/soc/t186/plat_sip_calls.c
index c7a2c416761816fa2d951efa0712e219f0e994ec..9f48ddd905d08484aa194263bfd1cf2a1fe30033 100644 (file)
--- a/plat/nvidia/tegra/soc/t186/plat_sip_calls.c
+++ b/plat/nvidia/tegra/soc/t186/plat_sip_calls.c
@@ -65,6 +65,7 @@ extern uint32_t tegra186_system_powerdn_state;
 #define TEGRA_SIP_MCE_CMD_ROC_FLUSH_CACHE              0x82FFFF0E
 #define TEGRA_SIP_MCE_CMD_ROC_CLEAN_CACHE              0x82FFFF0F
 #define TEGRA_SIP_MCE_CMD_ENABLE_LATIC                 0x82FFFF10
+#define TEGRA_SIP_MCE_CMD_UNCORE_PERFMON_REQ           0x82FFFF11
 
 /*******************************************************************************
  * This function is responsible for handling all T186 SiP calls
@@ -102,6 +103,7 @@ int plat_sip_handler(uint32_t smc_fid,
        case TEGRA_SIP_MCE_CMD_ROC_FLUSH_CACHE:
        case TEGRA_SIP_MCE_CMD_ROC_CLEAN_CACHE:
        case TEGRA_SIP_MCE_CMD_ENABLE_LATIC:
+       case TEGRA_SIP_MCE_CMD_UNCORE_PERFMON_REQ:
 
                /* clean up the high bits */
                smc_fid &= MCE_CMD_MASK;