Core: Properly configure address space when loading a binary

The code now properly configures the process image to match the loaded
binary segments (code, rodata, data) instead of just blindly allocating
a large chunk of dummy memory.

3 files changed, 123 insertions, 27 deletions

diff --git a/src/core/loader/3dsx.cpp b/src/core/loader/3dsx.cpp
index e57d7274..05566136 100644
--- a/src/core/loader/3dsx.cpp
+++ b/src/core/loader/3dsx.cpp
@@ -101,7 +101,10 @@ static u32 TranslateAddr(u32 addr, const THREEloadinfo *loadinfo, u32* offsets)
     return loadinfo->seg_addrs[2] + addr - offsets[1];
 }
 
-static THREEDSX_Error Load3DSXFile(FileUtil::IOFile& file, u32 base_addr)
+using Kernel::SharedPtr;
+using Kernel::CodeSet;
+
+static THREEDSX_Error Load3DSXFile(FileUtil::IOFile& file, u32 base_addr, SharedPtr<CodeSet>* out_codeset)
 {
     if (!file.IsOpen())
         return ERROR_FILE;
@@ -201,13 +204,29 @@ static THREEDSX_Error Load3DSXFile(FileUtil::IOFile& file, u32 base_addr)
         }
     }
 
-    // Write the data
-    memcpy(Memory::GetPointer(base_addr), program_image.data(), program_image.size());
+    // Create the CodeSet
+    SharedPtr<CodeSet> code_set = CodeSet::Create("", 0);
+
+    code_set->code.offset = loadinfo.seg_ptrs[0] - program_image.data();
+    code_set->code.addr   = loadinfo.seg_addrs[0];
+    code_set->code.size   = loadinfo.seg_sizes[0];
+
+    code_set->rodata.offset = loadinfo.seg_ptrs[1] - program_image.data();
+    code_set->rodata.addr   = loadinfo.seg_addrs[1];
+    code_set->rodata.size   = loadinfo.seg_sizes[1];
+
+    code_set->data.offset = loadinfo.seg_ptrs[2] - program_image.data();
+    code_set->data.addr   = loadinfo.seg_addrs[2];
+    code_set->data.size   = loadinfo.seg_sizes[2];
+
+    code_set->entrypoint = code_set->code.addr;
+    code_set->memory = std::make_shared<std::vector<u8>>(std::move(program_image));
 
     LOG_DEBUG(Loader, "code size:   0x%X", loadinfo.seg_sizes[0]);
     LOG_DEBUG(Loader, "rodata size: 0x%X", loadinfo.seg_sizes[1]);
     LOG_DEBUG(Loader, "data size:   0x%X (including 0x%X of bss)", loadinfo.seg_sizes[2], hdr.bss_size);
 
+    *out_codeset = code_set;
     return ERROR_NONE;
 }
 
@@ -230,17 +249,19 @@ ResultStatus AppLoader_THREEDSX::Load() {
     if (!file->IsOpen())
         return ResultStatus::Error;
 
-    Kernel::g_current_process = Kernel::Process::Create(filename, 0);
+    SharedPtr<CodeSet> codeset;
+    if (Load3DSXFile(*file, Memory::PROCESS_IMAGE_VADDR, &codeset) != ERROR_NONE)
+        return ResultStatus::Error;
+    codeset->name = filename;
+
+    Kernel::g_current_process = Kernel::Process::Create(std::move(codeset));
     Kernel::g_current_process->svc_access_mask.set();
     Kernel::g_current_process->address_mappings = default_address_mappings;
 
     // Attach the default resource limit (APPLICATION) to the process
     Kernel::g_current_process->resource_limit = Kernel::ResourceLimit::GetForCategory(Kernel::ResourceLimitCategory::APPLICATION);
 
-    if (Load3DSXFile(*file, Memory::PROCESS_IMAGE_VADDR) != ERROR_NONE)
-        return ResultStatus::Error;
-
-    Kernel::g_current_process->Run(Memory::PROCESS_IMAGE_VADDR, 48, Kernel::DEFAULT_STACK_SIZE);
+    Kernel::g_current_process->Run(48, Kernel::DEFAULT_STACK_SIZE);
 
     is_loaded = true;
     return ResultStatus::Success;
diff --git a/src/core/loader/elf.cpp b/src/core/loader/elf.cpp
index a7eea78a..ca3c18a9 100644
--- a/src/core/loader/elf.cpp
+++ b/src/core/loader/elf.cpp
@@ -16,6 +16,9 @@
 #include "core/loader/elf.h"
 #include "core/memory.h"
 
+using Kernel::SharedPtr;
+using Kernel::CodeSet;
+
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // ELF Header Constants
 
@@ -97,6 +100,12 @@ enum ElfSectionFlags
 #define PT_LOPROC  0x70000000
 #define PT_HIPROC  0x7FFFFFFF
 
+// Segment flags
+#define PF_X               0x1
+#define PF_W               0x2
+#define PF_R               0x4
+#define PF_MASKPROC 0xF0000000
+
 typedef unsigned int   Elf32_Addr;
 typedef unsigned short Elf32_Half;
 typedef unsigned int   Elf32_Off;
@@ -193,7 +202,7 @@ public:
     ElfMachine GetMachine() const { return (ElfMachine)(header->e_machine); }
     u32 GetEntryPoint() const { return entryPoint; }
     u32 GetFlags() const { return (u32)(header->e_flags); }
-    void LoadInto(u32 vaddr);
+    SharedPtr<CodeSet> LoadInto(u32 vaddr);
     bool LoadSymbols();
 
     int GetNumSegments() const { return (int)(header->e_phnum); }
@@ -249,7 +258,7 @@ const char *ElfReader::GetSectionName(int section) const {
     return nullptr;
 }
 
-void ElfReader::LoadInto(u32 vaddr) {
+SharedPtr<CodeSet> ElfReader::LoadInto(u32 vaddr) {
     LOG_DEBUG(Loader, "String section: %i", header->e_shstrndx);
 
     // Should we relocate?
@@ -267,19 +276,61 @@ void ElfReader::LoadInto(u32 vaddr) {
     u32 segment_addr[32];
     u32 base_addr = relocate ? vaddr : 0;
 
-    for (unsigned i = 0; i < header->e_phnum; i++) {
-        Elf32_Phdr* p = segments + i;
-        LOG_DEBUG(Loader, "Type: %i Vaddr: %08x Filesz: %i Memsz: %i ", p->p_type, p->p_vaddr,
+    u32 total_image_size = 0;
+    for (unsigned int i = 0; i < header->e_phnum; ++i) {
+        Elf32_Phdr* p = &segments[i];
+        if (p->p_type == PT_LOAD) {
+            total_image_size += (p->p_memsz + 0xFFF) & ~0xFFF;
+        }
+    }
+
+    std::vector<u8> program_image(total_image_size);
+    size_t current_image_position = 0;
+
+    SharedPtr<CodeSet> codeset = CodeSet::Create("", 0);
+
+    for (unsigned int i = 0; i < header->e_phnum; ++i) {
+        Elf32_Phdr* p = &segments[i];
+        LOG_DEBUG(Loader, "Type: %i Vaddr: %08X Filesz: %8X Memsz: %8X ", p->p_type, p->p_vaddr,
                   p->p_filesz, p->p_memsz);
 
         if (p->p_type == PT_LOAD) {
-            segment_addr[i] = base_addr + p->p_vaddr;
-            memcpy(Memory::GetPointer(segment_addr[i]), GetSegmentPtr(i), p->p_filesz);
-            LOG_DEBUG(Loader, "Loadable Segment Copied to %08x, size %08x", segment_addr[i],
-                      p->p_memsz);
+            CodeSet::Segment* codeset_segment;
+            u32 permission_flags = p->p_flags & (PF_R | PF_W | PF_X);
+            if (permission_flags == (PF_R | PF_X)) {
+                codeset_segment = &codeset->code;
+            } else if (permission_flags == (PF_R)) {
+                codeset_segment = &codeset->rodata;
+            } else if (permission_flags == (PF_R | PF_W)) {
+                codeset_segment = &codeset->data;
+            } else {
+                LOG_ERROR(Loader, "Unexpected ELF PT_LOAD segment id %u with flags %X", i, p->p_flags);
+                continue;
+            }
+
+            if (codeset_segment->size != 0) {
+                LOG_ERROR(Loader, "ELF has more than one segment of the same type. Skipping extra segment (id %i)", i);
+                continue;
+            }
+
+            u32 segment_addr = base_addr + p->p_vaddr;
+            u32 aligned_size = (p->p_memsz + 0xFFF) & ~0xFFF;
+
+            codeset_segment->offset = current_image_position;
+            codeset_segment->addr = segment_addr;
+            codeset_segment->size = aligned_size;
+
+            memcpy(&program_image[current_image_position], GetSegmentPtr(i), p->p_filesz);
+            current_image_position += aligned_size;
         }
     }
+
+    codeset->entrypoint = base_addr + header->e_entry;
+    codeset->memory = std::make_shared<std::vector<u8>>(std::move(program_image));
+
     LOG_DEBUG(Loader, "Done loading.");
+
+    return codeset;
 }
 
 SectionID ElfReader::GetSectionByName(const char *name, int firstSection) const {
@@ -352,18 +403,18 @@ ResultStatus AppLoader_ELF::Load() {
     if (file->ReadBytes(&buffer[0], size) != size)
         return ResultStatus::Error;
 
-    Kernel::g_current_process = Kernel::Process::Create(filename, 0);
+    ElfReader elf_reader(&buffer[0]);
+    SharedPtr<CodeSet> codeset = elf_reader.LoadInto(Memory::PROCESS_IMAGE_VADDR);
+    codeset->name = filename;
+
+    Kernel::g_current_process = Kernel::Process::Create(std::move(codeset));
     Kernel::g_current_process->svc_access_mask.set();
     Kernel::g_current_process->address_mappings = default_address_mappings;
 
     // Attach the default resource limit (APPLICATION) to the process
     Kernel::g_current_process->resource_limit = Kernel::ResourceLimit::GetForCategory(Kernel::ResourceLimitCategory::APPLICATION);
 
-    ElfReader elf_reader(&buffer[0]);
-    elf_reader.LoadInto(Memory::PROCESS_IMAGE_VADDR);
-    // TODO: Fill application title
-
-    Kernel::g_current_process->Run(elf_reader.GetEntryPoint(), 48, Kernel::DEFAULT_STACK_SIZE);
+    Kernel::g_current_process->Run(48, Kernel::DEFAULT_STACK_SIZE);
 
     is_loaded = true;
     return ResultStatus::Success;
diff --git a/src/core/loader/ncch.cpp b/src/core/loader/ncch.cpp
index 2b26b31c..87603d19 100644
--- a/src/core/loader/ncch.cpp
+++ b/src/core/loader/ncch.cpp
@@ -118,6 +118,9 @@ FileType AppLoader_NCCH::IdentifyType(FileUtil::IOFile& file) {
 }
 
 ResultStatus AppLoader_NCCH::LoadExec() const {
+    using Kernel::SharedPtr;
+    using Kernel::CodeSet;
+
     if (!is_loaded)
         return ResultStatus::ErrorNotLoaded;
 
@@ -126,7 +129,30 @@ ResultStatus AppLoader_NCCH::LoadExec() const {
         std::string process_name = Common::StringFromFixedZeroTerminatedBuffer(
                 (const char*)exheader_header.codeset_info.name, 8);
         u64 program_id = *reinterpret_cast<u64_le const*>(&ncch_header.program_id[0]);
-        Kernel::g_current_process = Kernel::Process::Create(process_name, program_id);
+
+        SharedPtr<CodeSet> codeset = CodeSet::Create(process_name, program_id);
+
+        codeset->code.offset = 0;
+        codeset->code.addr = exheader_header.codeset_info.text.address;
+        codeset->code.size = exheader_header.codeset_info.text.num_max_pages * Memory::PAGE_SIZE;
+
+        codeset->rodata.offset = codeset->code.offset + codeset->code.size;
+        codeset->rodata.addr = exheader_header.codeset_info.ro.address;
+        codeset->rodata.size = exheader_header.codeset_info.ro.num_max_pages * Memory::PAGE_SIZE;
+
+        // TODO(yuriks): Not sure if the bss size is added to the page-aligned .data size or just
+        //               to the regular size. Playing it safe for now.
+        u32 bss_page_size = (exheader_header.codeset_info.bss_size + 0xFFF) & ~0xFFF;
+        code.resize(code.size() + bss_page_size, 0);
+
+        codeset->data.offset = codeset->rodata.offset + codeset->rodata.size;
+        codeset->data.addr = exheader_header.codeset_info.data.address;
+        codeset->data.size = exheader_header.codeset_info.data.num_max_pages * Memory::PAGE_SIZE + bss_page_size;
+
+        codeset->entrypoint = codeset->code.addr;
+        codeset->memory = std::make_shared<std::vector<u8>>(std::move(code));
+
+        Kernel::g_current_process = Kernel::Process::Create(std::move(codeset));
 
         // Attach a resource limit to the process based on the resource limit category
         Kernel::g_current_process->resource_limit = Kernel::ResourceLimit::GetForCategory(
@@ -137,11 +163,9 @@ ResultStatus AppLoader_NCCH::LoadExec() const {
         std::copy_n(exheader_header.arm11_kernel_caps.descriptors, kernel_caps.size(), begin(kernel_caps));
         Kernel::g_current_process->ParseKernelCaps(kernel_caps.data(), kernel_caps.size());
 
-        Memory::WriteBlock(entry_point, &code[0], code.size());
-
         s32 priority = exheader_header.arm11_system_local_caps.priority;
         u32 stack_size = exheader_header.codeset_info.stack_size;
-        Kernel::g_current_process->Run(entry_point, priority, stack_size);
+        Kernel::g_current_process->Run(priority, stack_size);
         return ResultStatus::Success;
     }
     return ResultStatus::Error;