[PATCH edk2-platforms v2 6/8] Silicon/AMD/StyxDtbLoaderLib: add description of the cache topology

["Ard Biesheuvel" <ard.biesheuvel@linaro.org>]

Emit the cache topology into the device tree too when generating the
CPU nodes and the cpu-map. Note that the cache geometries are all
fixed and thus hardcoded - the only runtime variable aspect is how
many L2 nodes to generate (one per detected cluster)

Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
---
 Silicon/AMD/Styx/Library/StyxDtbLoaderLib/StyxDtbLoaderLib.c | 49 ++++++++++++++++++++
 1 file changed, 49 insertions(+)

diff --git a/Silicon/AMD/Styx/Library/StyxDtbLoaderLib/StyxDtbLoaderLib.c b/Silicon/AMD/Styx/Library/StyxDtbLoaderLib/StyxDtbLoaderLib.c
index 2f7b5e2a7b25..e723e77c7965 100644
--- a/Silicon/AMD/Styx/Library/StyxDtbLoaderLib/StyxDtbLoaderLib.c
+++ b/Silicon/AMD/Styx/Library/StyxDtbLoaderLib/StyxDtbLoaderLib.c
@@ -305,6 +305,10 @@ PrepareFdt (
   UINT32                      ClusterCount;
   UINT32                      CoresInCluster;
   UINT32                      ClusterId;
+  INT32                       L2Node;
+  INT32                       L3Node;
+  INT32                       L2Phandle[NUM_CORES / 2];
+  INT32                       L3Phandle;
   UINTN                       MpId;
   CHAR8                       Name[10];
   AMD_MP_CORE_INFO_PROTOCOL   *AmdMpCoreInfoProtocol;
@@ -328,6 +332,42 @@ PrepareFdt (
   ASSERT (ArmCoreInfoTable != NULL);
   ASSERT (ArmCoreCount <= NUM_CORES);
 
+  // Create the L3 cache node
+  L3Node = fdt_add_subnode (Fdt, 0, "l3cache");
+  if (L3Node < 0) {
+    DEBUG ((DEBUG_ERROR, "FDT: Error creating 'l3cache' node\n"));
+    return EFI_INVALID_PARAMETER;
+  }
+
+  L3Phandle = fdt_alloc_phandle (Fdt);
+  fdt_setprop_cell (Fdt, L3Node, "cache-level", 3);
+  fdt_setprop_cell (Fdt, L3Node, "cache-size", SIZE_8MB);
+  fdt_setprop_cell (Fdt, L3Node, "cache-line-size", 64);
+  fdt_setprop_cell (Fdt, L3Node, "cache-sets", 8192);
+  fdt_setprop_empty (Fdt, L3Node, "cache-unified");
+  fdt_setprop_cell (Fdt, L3Node, "phandle", L3Phandle);
+
+  ClusterCount = NumberOfClustersInTable (ArmCoreInfoTable, ArmCoreCount);
+  ASSERT (ClusterCount <= ARRAY_SIZE (L2Phandle));
+
+  for (Index = 0; Index < ClusterCount; Index++) {
+    AsciiSPrint (Name, sizeof (Name), "l2cache%d", Index);
+
+    L2Node = fdt_add_subnode (Fdt, 0, Name);
+    if (L2Node < 0) {
+      DEBUG ((DEBUG_ERROR, "FDT: Error creating '%a' node\n", Name));
+      return EFI_INVALID_PARAMETER;
+    }
+
+    L2Phandle[Index] = fdt_alloc_phandle (Fdt);
+    fdt_setprop_cell (Fdt, L2Node, "cache-size", SIZE_1MB);
+    fdt_setprop_cell (Fdt, L2Node, "cache-line-size", 64);
+    fdt_setprop_cell (Fdt, L2Node, "cache-sets", 1024);
+    fdt_setprop_empty (Fdt, L2Node, "cache-unified");
+    fdt_setprop_cell (Fdt, L2Node, "next-level-cache", L3Phandle);
+    fdt_setprop_cell (Fdt, L2Node, "phandle", L2Phandle[Index]);
+  }
+
   // Get Id from primary CPU
   MpId = (UINTN)ArmReadMpidr ();
 
@@ -367,6 +407,15 @@ PrepareFdt (
     fdt_setprop (Fdt, CpuNode, "reg", &MpId, sizeof (MpId));
     fdt_setprop_string (Fdt, CpuNode, "compatible", "arm,armv8");
     fdt_setprop_string (Fdt, CpuNode, "device_type", "cpu");
+
+    fdt_setprop_cell (Fdt, CpuNode, "i-cache-size", 3 * SIZE_16KB);
+    fdt_setprop_cell (Fdt, CpuNode, "i-cache-line-size", 64);
+    fdt_setprop_cell (Fdt, CpuNode, "i-cache-sets", 256);
+    fdt_setprop_cell (Fdt, CpuNode, "d-cache-size", 2 * SIZE_16KB);
+    fdt_setprop_cell (Fdt, CpuNode, "d-cache-line-size", 64);
+    fdt_setprop_cell (Fdt, CpuNode, "d-cache-sets", 256);
+    fdt_setprop_cell (Fdt, CpuNode, "l2-cache",
+      L2Phandle[ArmCoreInfoTable[Index].ClusterId]);
   }
 
   // Create /cpu-map node
-- 
2.17.1