LoadFragment 2 OK

include/load.h

@@ -0,0 +1,38 @@
+#ifndef LOAD_H
+#define LOAD_H
+
+#include "ultra64.h"
+
+extern s32 gOverlayLogSeverity;
+
+#define RELOC_SECTION(reloc) ((reloc) >> 30)
+#define RELOC_OFFSET(reloc) ((reloc) & 0xFFFFFF)
+#define RELOC_TYPE_MASK(reloc) ((reloc) & 0x3F000000)
+#define RELOC_TYPE_SHIFT 24
+
+/* MIPS Relocation Types, matches the MIPS ELF spec */
+#define R_MIPS_32 2
+#define R_MIPS_26 4
+#define R_MIPS_HI16 5
+#define R_MIPS_LO16 6
+
+typedef enum {
+    /* 0 */ RELOC_SECTION_NULL,
+    /* 1 */ RELOC_SECTION_TEXT,
+    /* 2 */ RELOC_SECTION_DATA,
+    /* 3 */ RELOC_SECTION_RODATA,
+    /* 4 */ RELOC_SECTION_MAX
+} RelocSectionId;
+
+typedef struct OverlayRelocationSection {
+    /* 0x00 */ u32 textSize;
+    /* 0x04 */ u32 dataSize;
+    /* 0x08 */ u32 rodataSize;
+    /* 0x0C */ u32 bssSize;
+    /* 0x10 */ u32 nRelocations;
+    /* 0x14 */ u32 relocations[1]; // size is nRelocations
+} OverlayRelocationSection; // size >= 0x18
+
+s32 Overlay_Load(void *vromStart, void *vromEnd, void *ovlStart, void *ovlEnd, void *vramStart, void *vramEnd, void *allocatedRamAddr, OverlayRelocationSection *ovlRelocs);
+
+#endif

include/z_std_dma.h

@@ -13,7 +13,7 @@
 // void func_80026828_jp();
 // void func_800269E4_jp();
 // void func_80026A64_jp();
-// void func_80026B44_jp();
+void DmaMgr_RequestSync(void* ram, void* vrom, size_t size);
 // void func_80026BC0_jp();
 // void func_80026C00_jp();
 // void func_80026C28_jp();

linker_scripts/jp/symbol_addrs_boot.txt

@@ -27,7 +27,7 @@ func_80026814_jp = 0x80026814; // type:func
 func_80026828_jp = 0x80026828; // type:func
 func_800269E4_jp = 0x800269E4; // type:func
 func_80026A64_jp = 0x80026A64; // type:func
-func_80026B44_jp = 0x80026B44; // type:func
+DmaMgr_RequestSync = 0x80026B44; // type:func
 func_80026BC0_jp = 0x80026BC0; // type:func
 func_80026C00_jp = 0x80026C00; // type:func
 func_80026C28_jp = 0x80026C28; // type:func
@@ -153,7 +153,7 @@ gfxprint_printf = 0x8002B770; // type:func
 MtxConv_F2L = 0x8002B7A0; // type:func
 MtxConv_L2F = 0x8002B8E8; // type:func
 
-func_8002B910_jp = 0x8002B910; // type:func
+Overlay_Load = 0x8002B910; // type:func
 DoRelocation = 0x8002B9C0; // type:func
 
 _dbg_hungup = 0x8002BC00; // type:func
@@ -279,6 +279,8 @@ D_8003BC50_jp = 0x8003BC50; //
 D_8003BC90_jp = 0x8003BC90; //
 D_8003BC94_jp = 0x8003BC94; //
 
+gOverlayLogSeverity = 0x8003BCA0 // type:s32 size:0x4
+
 gfxprint_moji_tlut = 0x8003BCB0; // type:u8 size:0x80
 gfxprint_rainbow_tlut = 0x8003BD30; // type:u8 size:0x20
 gfxprint_rainbow_txtr = 0x8003BD50; // type:u8 size:0x8

src/boot/loadfragment2.c

@@ -0,0 +1,179 @@
+/**
+ * @file loadfragment2.c
+ *
+ * Functions used to process and relocate dynamically loadable code segments (overlays).
+ *
+ */
+#include "global.h"
+#include "load.h"
+#include "z_std_dma.h"
+
+void DoRelocation(void *allocatedRamAddr, OverlayRelocationSection *ovlRelocs, void *vramStart);
+
+s32 Overlay_Load(void *vromStart, void *vromEnd, void *ovlStart, void *ovlEnd, void *vramStart, void *vramEnd,
+                 void *allocatedRamAddr, OverlayRelocationSection *ovlRelocs) {
+    OverlayRelocationSection *ovl = ovlRelocs;
+    s32 vromSize = (uintptr_t)vromEnd - (uintptr_t)vromStart;
+    s32 ovlSize = (uintptr_t)ovlEnd - (uintptr_t)ovlStart;
+    s32 vramSize = (uintptr_t)vramEnd - (uintptr_t)vramStart;
+    void *end = (void *)((uintptr_t)allocatedRamAddr + vromSize);
+
+    DmaMgr_RequestSync(allocatedRamAddr, vromStart, vromSize);
+    DmaMgr_RequestSync(ovl, ovlStart, ovlSize);
+    DoRelocation(allocatedRamAddr, ovl, vramStart);
+
+    if (ovl->bssSize != 0) {
+        bzero(end, ovl->bssSize);
+    }
+
+    osWritebackDCache(allocatedRamAddr, vramSize);
+    osInvalICache(allocatedRamAddr, vramSize);
+    return vramSize;
+}
+
+// Extract MIPS register rs from an instruction word
+#define MIPS_REG_RS(insn) (((insn) >> 0x15) & 0x1F)
+
+// Extract MIPS register rt from an instruction word
+#define MIPS_REG_RT(insn) (((insn) >> 0x10) & 0x1F)
+
+// Extract MIPS jump target from an instruction word
+#define MIPS_JUMP_TARGET(insn) (((insn)&0x03FFFFFF) << 2)
+
+/**
+ * Performs runtime relocation of overlay files, loadable code segments.
+ *
+ * Overlays are expected to be loadable anywhere in direct-mapped cached (KSEG0) memory, with some appropriate
+ * alignment requirements; memory addresses in such code must be updated once loaded in order to execute properly.
+ * When compiled, overlays are given 'fake' KSEG0 RAM addresses larger than the total possible available main memory
+ * (>= 0x80800000), such addresses are referred to as Virtual RAM (VRAM) to distinguish them. When loading the overlay
+ * the relocation table produced at compile time is consulted to determine where and how to update these VRAM addresses
+ * to correct RAM addresses based on the location the overlay was loaded at, enabling the code to execute at this
+ * address as if it were compiled to run at this address.
+ *
+ * Each relocation is represented by a packed 32-bit value, formatted in the following way:
+ *  - [31:30]  2-bit section id, taking values from the `RelocSectionId` enum.
+ *  - [29:24]  6-bit relocation type describing which relocation operation should be performed. Same as ELF32 MIPS.
+ *  - [23: 0]  24-bit section-relative offset indicating where in the section to apply this relocation.
+ *
+ * @param allocatedRamAddr Memory address the binary was loaded at.
+ * @param ovlRelocs Overlay relocation section containing overlay section layout and runtime relocations.
+ * @param vramStart Virtual RAM address that the overlay was compiled at.
+ */
+void DoRelocation(void *allocatedRamAddr, OverlayRelocationSection *ovlRelocs, void *vramStart) {
+    uintptr_t sections[RELOC_SECTION_MAX];
+    u32 *relocDataP;
+    u32 reloc;
+    u32 relocData;
+    u32 isLoNeg;
+    uintptr_t allocu32 = (uintptr_t)allocatedRamAddr;
+    u32 i;
+    u32 *regValP;
+    //! MIPS ELF relocation does not generally require tracking register values, so at first glance it appears this
+    //! register tracking was an unnecessary complication. However there is a bug in the IDO compiler that can cause
+    //! relocations to be emitted in the wrong order under rare circumstances when the compiler attempts to reuse a
+    //! previous HI16 relocation for a different LO16 relocation as an optimization. This register tracking is likely
+    //! a workaround to prevent improper matching of unrelated HI16 and LO16 relocations that would otherwise arise
+    //! due to the incorrect ordering.
+    u32 *luiRefs[32];
+    u32 luiVals[32];
+    u32 *luiInstRef;
+    UNUSED u32 dbg;
+    ptrdiff_t relocOffset = 0;
+    u32 relocatedValue = 0;
+    UNUSED uintptr_t unrelocatedAddress = 0;
+    uintptr_t relocatedAddress = 0;
+    UNUSED s32 pad;
+
+    if (gOverlayLogSeverity >= 3) {}
+
+    sections[RELOC_SECTION_NULL] = 0;
+    sections[RELOC_SECTION_TEXT] = allocu32;
+    sections[RELOC_SECTION_DATA] = ovlRelocs->textSize + allocu32;
+    sections[RELOC_SECTION_RODATA] = ovlRelocs->dataSize + sections[RELOC_SECTION_DATA];
+
+    for (i = 0; i < ovlRelocs->nRelocations; i++) {
+        reloc = ovlRelocs->relocations[i];
+        // This will always resolve to a 32-bit aligned address as each section
+        // containing code or pointers must be aligned to at least 4 bytes and the
+        // MIPS ABI defines the offset of both 16-bit and 32-bit relocations to be
+        // the start of the 32-bit word containing the target.
+        relocDataP = (u32 *)(sections[RELOC_SECTION(reloc)] + RELOC_OFFSET(reloc));
+        relocData = *relocDataP;
+
+        switch (RELOC_TYPE_MASK(reloc)) {
+            case R_MIPS_32 << RELOC_TYPE_SHIFT:
+                // Handles 32-bit address relocation, used for things such as jump tables and pointers in data.
+                // Just relocate the full address.
+
+                // Check address is valid for relocation
+                if ((*relocDataP & 0x0F000000) == 0) {
+                    relocOffset = *relocDataP - (uintptr_t)vramStart;
+                    relocatedValue = relocOffset + allocu32;
+                    unrelocatedAddress = relocData;
+                    relocatedAddress = relocatedValue;
+                    *relocDataP = relocatedAddress;
+                }
+                break;
+
+            case R_MIPS_26 << RELOC_TYPE_SHIFT:
+                // Handles 26-bit address relocation, used for jumps and jals.
+                // Extract the address from the target field of the J-type MIPS instruction.
+                // Relocate the address and update the instruction.
+                if (1) {
+                    relocOffset = PHYS_TO_K0(MIPS_JUMP_TARGET(*relocDataP)) - (uintptr_t)vramStart;
+                    unrelocatedAddress = PHYS_TO_K0(MIPS_JUMP_TARGET(*relocDataP));
+                    relocatedValue = (*relocDataP & 0xFC000000) | (((allocu32 + relocOffset) & 0x0FFFFFFF) >> 2);
+                    relocatedAddress = PHYS_TO_K0(MIPS_JUMP_TARGET(relocatedValue));
+                    *relocDataP = relocatedValue;
+                }
+                break;
+
+            case R_MIPS_HI16 << RELOC_TYPE_SHIFT:
+                // Handles relocation for a hi/lo pair, part 1.
+                // Store the reference to the LUI instruction (hi) using the `rt` register of the instruction.
+                // This will be updated later in the `R_MIPS_LO16` section.
+
+                luiRefs[MIPS_REG_RT(*relocDataP)] = relocDataP;
+                luiVals[MIPS_REG_RT(*relocDataP)] = *relocDataP;
+                break;
+
+            case R_MIPS_LO16 << RELOC_TYPE_SHIFT:
+                // Handles relocation for a hi/lo pair, part 2.
+                // Grab the stored LUI (hi) from the `R_MIPS_HI16` section using the `rs` register of the instruction.
+                // The full address is calculated, relocated, and then used to update both the LUI and lo instructions.
+                // If the lo part is negative, add 1 to the LUI value.
+                // Note: The lo instruction is assumed to have a signed immediate.
+
+                luiInstRef = luiRefs[MIPS_REG_RS(*relocDataP)];
+                regValP = &luiVals[MIPS_REG_RS(*relocDataP)];
+
+                // Check address is valid for relocation
+                if ((((*regValP << 0x10) + (s16)*relocDataP) & 0x0F000000) == 0) {
+                    relocOffset = ((*regValP << 0x10) + (s16)*relocDataP) - (uintptr_t)vramStart;
+                    isLoNeg = ((relocOffset + allocu32) & 0x8000) ? 1 : 0; // adjust for signed immediate
+                    unrelocatedAddress = (*luiInstRef << 0x10) + (s16)relocData;
+                    *luiInstRef =
+                        (*luiInstRef & 0xFFFF0000) | ((((relocOffset + allocu32) >> 0x10) & 0xFFFF) + isLoNeg);
+                    relocatedValue = (*relocDataP & 0xFFFF0000) | ((relocOffset + allocu32) & 0xFFFF);
+
+                    relocatedAddress = (*luiInstRef << 0x10) + (s16)relocatedValue;
+                    *relocDataP = relocatedValue;
+                }
+                break;
+        }
+
+        dbg = 16;
+        switch (RELOC_TYPE_MASK(reloc)) {
+            case R_MIPS_32 << RELOC_TYPE_SHIFT:
+                dbg += 6;
+                FALLTHROUGH;
+            case R_MIPS_26 << RELOC_TYPE_SHIFT:
+                dbg += 10;
+                FALLTHROUGH;
+            case R_MIPS_LO16 << RELOC_TYPE_SHIFT:
+                if (gOverlayLogSeverity >= 3) {}
+                // Adding a break prevents matching
+        }
+    }
+}

src/boot/logseverity.c

@@ -0,0 +1,3 @@
+#include "global.h"
+
+s32 gOverlayLogSeverity = 2;

yamls/jp/boot.yaml

@@ -20,7 +20,7 @@
       - [0x005BE0, asm, boot/stackcheck]
       - [0x005E30, asm, boot/gfxprint]
       - [0x006BA0, asm, boot/mtxuty-cvt]
-      - [0x006D10, asm, boot/006D10]
+      - [0x006D10, c, boot/loadfragment2]
       - [0x007000, asm, boot/debug]
       - [0x007060, asm, boot/007060]
       - [0x0073A0, asm, boot/0073A0]
@@ -202,6 +202,7 @@
       - [0x016EF0, data, libultra/rsp/rspboot]
 
       - [0x016FC0, data, boot/016FC0]
+      - [0x0170A0, .data, boot/logseverity]
       - [0x0170B0, data, boot/gfxprint]
       - [0x017960, data, boot/017960] # boot/007060 starts around here, but no idea where it ends