Allow blacking out parts of the address space

Author: fwsGonzo

lib/tinykvm/amd64/paging.cpp

@@ -66,11 +66,12 @@ static void add_remappings(vMemory& memory,
 		throw MachineException("Invalid remapping address", remapping.virt);
 	if (remapping.size % vMemory::PageSize() != 0)
 		throw MachineException("Invalid remapping size", remapping.size);
-	const auto virt_tera_page = (remapping.virt >> 39UL) & 511;
-	const auto virt_giga_page = (remapping.virt >> 30UL) & 511;
+	const auto virt_tera_page = (remapping.virt >> 39U) & 511;
+	auto virt_giga_page = (remapping.virt >> 30U) & 511;
 
 	uint64_t paddr_base = remapping.phys;
-	if (paddr_base == 0x0) {
+	// Blackout remappings are used to reserve/create unmapped virtual memory space
+	if (paddr_base == 0x0 && !remapping.blackout) {
 		constexpr auto PD_ALIGN_MASK = (1ULL << 21U) - 1;
 		// Over-allocate rounding up to nearest 2MB
 		paddr_base = memory.machine.mmap_allocate(remapping.size + PD_ALIGN_MASK);
@@ -89,25 +90,33 @@ static void add_remappings(vMemory& memory,
 	auto  pdpt_addr = pml4[virt_tera_page] & PDE64_ADDR_MASK;
 	auto* pdpt = memory.page_at(pdpt_addr);
 
-	// Allocate the gigapage with 512x 2MB entries
-	if (pdpt[virt_giga_page] == 0) {
-		const auto giga_page = free_page;
-		free_page += 0x1000;
-		pdpt[virt_giga_page] = PDE64_PRESENT | PDE64_USER | PDE64_RW | giga_page;
-	}
+	constexpr auto PDPT_ALIGN_MASK = (1ULL << 30U) - 1;
+	const auto n_pd_pages = ((remapping.size + PDPT_ALIGN_MASK) >> 30UL) & 511;
+	for (uint64_t n_pd = 0; n_pd < n_pd_pages; n_pd++)
+	{
+		// Allocate the gigapage with 512x 2MB entries
+		if (pdpt[virt_giga_page] == 0) {
+			const auto giga_page = free_page;
+			free_page += 0x1000;
+			pdpt[virt_giga_page] = PDE64_PRESENT | PDE64_USER | PDE64_RW | giga_page;
+		}
 
-	auto  pd_addr = pdpt[virt_giga_page] & PDE64_ADDR_MASK;
-	auto* pd = memory.page_at(pd_addr);
+		auto  pd_addr = pdpt[virt_giga_page] & PDE64_ADDR_MASK;
+		auto* pd = memory.page_at(pd_addr);
 
-	// Create 2MB entries for remapping size
-	const auto n_2mb_pages = (remapping.size >> 21UL) & 511;
-	for (uint64_t i = 0; i < 512; i++)
-	{
-		const auto paddr = paddr_base + (i << 21UL);
-		if (i < n_2mb_pages)
-			pd[i] = PDE64_PRESENT | flags | PDE64_PS | paddr;
-		else
-			pd[i] = 0;
+		// Create 2MB entries for remapping size
+		constexpr auto PD_ALIGN_MASK = (1ULL << 21U) - 1;
+		const auto n_2mb_pages = ((remapping.size + PD_ALIGN_MASK) >> 21UL) & 511;
+		for (uint64_t i = 0; i < 512; i++)
+		{
+			if (i < n_2mb_pages && !remapping.blackout) {
+				const auto paddr = paddr_base + (i << 21UL);
+				pd[i] = PDE64_PRESENT | flags | PDE64_PS | paddr;
+			}
+			else pd[i] = 0; // Unmapped 2MB page (remainder or blackout)
+		}
+
+		virt_giga_page++; // Next gigapage
 	}
 
 	// Track the first seen executable mapping, allowing mmap to use it for
@@ -306,6 +315,11 @@ uint64_t setup_amd64_paging(vMemory& memory,
 		uint64_t flags = PDE64_USER | PDE64_NX;
 		if (vmem.writable) flags |= PDE64_RW;
 		if (vmem.executable) flags &= ~PDE64_NX;
+		if (vmem.blackout) flags = 0;
+		if constexpr (true) {
+			printf("* Remapping 0x%lX -> 0x%lX  size 0x%lX write=%d exec=%d blackout=%d\n",
+				vmem.virt, vmem.phys, vmem.size, vmem.writable, vmem.executable, vmem.blackout);
+		}
 		add_remappings(memory, vmem, pml4, flags, free_page);
 	}
 

lib/tinykvm/common.hpp

@@ -27,6 +27,7 @@ namespace tinykvm
 		size_t   size;
 		bool     writable = false;
 		bool     executable = false;
+		bool     blackout = false; /* Unmapped virtual area */
 	};
 
 	struct MachineOptions {
@@ -51,8 +52,8 @@ namespace tinykvm
 		/* When enabled, reset_to() will accept a different
 		   master VM than the original, but at a steep cost. */
 		bool allow_reset_to_new_master = false;
-		/* Allow fixed addresses with mmap(). */
-		bool allow_fixed_mmap = false;
+		/* Force-relocate fixed addresses with mmap(). */
+		bool relocate_fixed_mmap = true;
 		/* Make heap executable, to support JIT. */
 		bool executable_heap = false;
 	};

lib/tinykvm/machine.cpp

@@ -29,7 +29,7 @@ __attribute__ ((cold))
 Machine::Machine(std::string_view binary, const MachineOptions& options)
 	: m_forked {false},
 	  m_just_reset {false},
-	  m_allow_fixed_mmap {options.allow_fixed_mmap},
+	  m_relocate_fixed_mmap {options.relocate_fixed_mmap},
 	  memory { vMemory::New(*this, options,
 	  	options.vmem_base_address, options.vmem_base_address + 0x100000, options.max_mem)
 	  },
@@ -59,7 +59,7 @@ Machine::Machine(const Machine& other, const MachineOptions& options)
 	: m_prepped {false},
 	  m_forked  {true},
 	  m_just_reset {true},
-	  m_allow_fixed_mmap {options.allow_fixed_mmap},
+	  m_relocate_fixed_mmap {options.relocate_fixed_mmap},
 	  m_binary {options.binary.empty() ? other.m_binary : options.binary},
 	  memory   {*this, options, other.memory},
 	  m_image_base    {other.m_image_base},

lib/tinykvm/machine.hpp

@@ -147,7 +147,7 @@ struct Machine
 	address_t mmap_start() const noexcept { return this->m_heap_address + BRK_MAX; }
 	address_t mmap_allocate(size_t bytes);
 	bool      mmap_unmap(uint64_t addr, size_t size);
-	bool allow_fixed_mmap() const noexcept { return m_allow_fixed_mmap; }
+	bool relocate_fixed_mmap() const noexcept { return m_relocate_fixed_mmap; }
 	bool mmap_relax(uint64_t addr, size_t size, size_t new_size);
 	auto& mmap_cache() noexcept { return m_mmap_cache; }
 
@@ -239,7 +239,7 @@ struct Machine
 	bool  m_prepped = false;
 	bool  m_forked = false;
 	bool  m_just_reset = false;
-	bool  m_allow_fixed_mmap = false;
+	bool  m_relocate_fixed_mmap = false;
 	void* m_userdata = nullptr;
 
 	std::string_view m_binary;

src/functions.cpp

@@ -134,7 +134,7 @@ void setup_kvm_system_calls()
 				} else {
 					regs.rax = ~0LL; /* MAP_FAILED */
 				}
-			} else if (regs.rdi != 0x0 && cpu.machine().allow_fixed_mmap()) {
+			} else if (regs.rdi != 0x0 && !cpu.machine().relocate_fixed_mmap()) {
 				regs.rax = regs.rdi;
 			} else if (regs.rdi != 0x0 && regs.rdi >= cpu.machine().heap_address() && regs.rdi < cpu.machine().mmap_start()) {
 				// Existing range already mmap'ed

src/simple.cpp

@@ -68,7 +68,7 @@ int main(int argc, char** argv)
 		.remappings {remappings},
 		.verbose_loader = (getenv("VERBOSE") != nullptr),
 		.hugepages = (getenv("HUGE") != nullptr),
-		.allow_fixed_mmap = (getenv("GO") != nullptr),
+		.relocate_fixed_mmap = (getenv("GO") == nullptr),
 	};
 	tinykvm::Machine master_vm {binary, options};
 	//master_vm.print_pagetables();