Kernel/LibC: Replace SYS_{GET,SET}_TLS with SYS_{SET,GET}_{FS,GS}BASE

This allows userspace to use both registers

kernel/arch/i686/Thread.S

@@ -45,11 +45,11 @@ start_userspace_thread:
 	movw $(0x20 | 3), %bx
 	movw %bx, %ds
 	movw %bx, %es
-	# gs = thread local
+	# fs
 	movw $(0x30 | 3), %bx
-	movw %bx, %gs
-	# fs = 0
-	xorw %bx, %bx
 	movw %bx, %fs
+	# gs
+	movw $(0x38 | 3), %bx
+	movw %bx, %gs
 
 	iret

kernel/arch/x86_64/Syscall.S

@@ -2,6 +2,7 @@
 // System V ABI: RDI, RSI, RDX, RCX, R8, R9
 .global asm_syscall_handler
 asm_syscall_handler:
+	swapgs
 	pushq %rbx
 	pushq %rcx
 	pushq %rdx
@@ -42,9 +43,9 @@ asm_syscall_handler:
 	popq %rdx
 	popq %rcx
 	popq %rbx
+	swapgs
 	iretq
 
-
 .global sys_fork_trampoline
 sys_fork_trampoline:
 	pushq %rbx

kernel/arch/x86_64/Thread.S

@@ -32,4 +32,6 @@ start_userspace_thread:
 	call get_thread_start_sp
 	movq %rax, %rsp
 
+	swapgs
+
 	iretq

kernel/arch/x86_64/interrupts.S

@@ -1,4 +1,12 @@
-.macro pushaq
+.macro swapgs_if_necessary, n
+	cmpb $0x08, \n(%rsp)
+	je 1f
+	swapgs
+1:
+.endm
+
+.macro pushaq, n
+	swapgs_if_necessary \n
 	pushq %rax
 	pushq %rcx
 	pushq %rdx
@@ -16,7 +24,7 @@
 	pushq %r15
 .endm
 
-.macro popaq
+.macro popaq, n
 	popq %r15
 	popq %r14
 	popq %r13
@@ -32,10 +40,11 @@
 	popq %rdx
 	popq %rcx
 	popq %rax
+	swapgs_if_necessary \n
 .endm
 
 isr_stub:
-	pushaq
+	pushaq 24
 	cld
 	movq %cr0, %rax; pushq %rax
 	movq %cr2, %rax; pushq %rax
@@ -49,43 +58,43 @@ isr_stub:
 	call cpp_isr_handler
 	addq $32, %rsp
 
-	popaq
+	popaq 24
 	addq $16, %rsp
 	iretq
 
 irq_stub:
-	pushaq
+	pushaq 24
 	cld
 	movq 120(%rsp), %rdi	# irq number
 	call cpp_irq_handler
-	popaq
+	popaq 24
 	addq $16, %rsp
 	iretq
 
 .global asm_yield_handler
 asm_yield_handler:
-	pushaq
+	pushaq 8
 	cld
 	leaq 120(%rsp), %rdi # interrupt stack ptr
 	movq     %rsp,  %rsi # interrupt register ptr
 	call cpp_yield_handler
-	popaq
+	popaq 8
 	iretq
 
 .global asm_ipi_handler
 asm_ipi_handler:
-	pushaq
+	pushaq 8
 	cld
 	call cpp_ipi_handler
-	popaq
+	popaq 8
 	iretq
 
 .global asm_timer_handler
 asm_timer_handler:
-	pushaq
+	pushaq 8
 	cld
 	call cpp_timer_handler
-	popaq
+	popaq 8
 	iretq
 
 .macro isr n

kernel/include/kernel/GDT.h

@@ -129,7 +129,8 @@ namespace Kernel
 		}
 
 #if ARCH(i686)
-		void set_tls(uintptr_t addr);
+		void set_fsbase(uintptr_t addr);
+		void set_gsbase(uintptr_t addr);
 #endif
 
 	private:
@@ -153,8 +154,8 @@ namespace Kernel
 		BAN::Array<SegmentDescriptor, 7> m_gdt; // null, kernel code, kernel data, user code, user data, tss low, tss high
 		static constexpr uint16_t m_tss_offset = 0x28;
 #elif ARCH(i686)
-		BAN::Array<SegmentDescriptor, 8> m_gdt; // null, kernel code, kernel data, user code, user data, processor data, tls, tss
+		BAN::Array<SegmentDescriptor, 9> m_gdt; // null, kernel code, kernel data, user code, user data, processor data, fsbase, gsbase, tss
-		static constexpr uint16_t m_tss_offset = 0x38;
+		static constexpr uint16_t m_tss_offset = 0x40;
 #endif
 		TaskStateSegment m_tss;
 		const GDTR m_gdtr {

kernel/include/kernel/Process.h

@@ -203,8 +203,12 @@ namespace Kernel
 
 		BAN::ErrorOr<long> sys_futex(int op, const uint32_t* addr, uint32_t val, const timespec* abstime);
 		BAN::ErrorOr<long> sys_yield();
-		BAN::ErrorOr<long> sys_set_tls(void*);
+
-		BAN::ErrorOr<long> sys_get_tls();
+		BAN::ErrorOr<long> sys_set_fsbase(void*);
+		BAN::ErrorOr<long> sys_get_fsbase();
+		BAN::ErrorOr<long> sys_set_gsbase(void*);
+		BAN::ErrorOr<long> sys_get_gsbase();
+
 		BAN::ErrorOr<long> sys_pthread_create(const pthread_attr_t* attr, void (*entry)(void*), void* arg);
 		BAN::ErrorOr<long> sys_pthread_exit(void* value);
 		BAN::ErrorOr<long> sys_pthread_join(pthread_t thread, void** value);

kernel/include/kernel/Processor.h

@@ -112,7 +112,9 @@ namespace Kernel
 		static void send_smp_message(ProcessorID, const SMPMessage&, bool send_ipi = true);
 		static void broadcast_smp_message(const SMPMessage&);
 
-		static void load_tls();
+		static void load_segments();
+		static void load_fsbase();
+		static void load_gsbase();
 
 	private:
 		Processor() = default;

kernel/include/kernel/Thread.h

@@ -122,8 +122,10 @@ namespace Kernel
 		void set_cpu_time_start();
 		void set_cpu_time_stop();
 
-		void set_tls(vaddr_t tls) { m_tls = tls; }
+		void set_fsbase(vaddr_t base) { m_fsbase = base; }
-		vaddr_t get_tls() const { return m_tls; }
+		vaddr_t get_fsbase() const { return m_fsbase; }
+		void set_gsbase(vaddr_t base) { m_gsbase = base; }
+		vaddr_t get_gsbase() const { return m_gsbase; }
 
 		size_t virtual_page_count() const { return (m_kernel_stack ? (m_kernel_stack->size() / PAGE_SIZE) : 0) + (m_userspace_stack ? (m_userspace_stack->size() / PAGE_SIZE) : 0); }
 		size_t physical_page_count() const { return virtual_page_count(); }
@@ -164,7 +166,8 @@ namespace Kernel
 		bool                       m_is_userspace         { false };
 		bool                       m_delete_process       { false };
 
-		vaddr_t                    m_tls                  { 0 };
+		vaddr_t                    m_fsbase               { 0 };
+		vaddr_t                    m_gsbase               { 0 };
 
 		SchedulerQueue::Node*      m_scheduler_node       { nullptr };
 

kernel/kernel/GDT.cpp

@@ -29,7 +29,8 @@ namespace Kernel
 		gdt->write_entry(0x20, 0x00000000, 0xFFFFF, 0xF2, data_flags);	// user data
 #if ARCH(i686)
 		gdt->write_entry(0x28, reinterpret_cast<uint32_t>(processor), sizeof(Processor), 0x92, 0x4); // processor data
-		gdt->write_entry(0x30, 0x00000000, 0x00000, 0x00, 0x0);			// tls
+		gdt->write_entry(0x30, 0x00000000, 0x00000, 0x00, 0x0);			// fsbase
+		gdt->write_entry(0x38, 0x00000000, 0x00000, 0x00, 0x0);			// gsbase
 #endif
 		gdt->write_tss();
 
@@ -37,10 +38,14 @@ namespace Kernel
 	}
 
 #if ARCH(i686)
-	void GDT::set_tls(uintptr_t addr)
+	void GDT::set_fsbase(uintptr_t addr)
 	{
 		write_entry(0x30, addr, 0xFFFF, 0xF2, 0xC);
 	}
+	void GDT::set_gsbase(uintptr_t addr)
+	{
+		write_entry(0x38, addr, 0xFFFF, 0xF2, 0xC);
+	}
 #endif
 
 	void GDT::write_entry(uint8_t offset, uint32_t base, uint32_t limit, uint8_t access, uint8_t flags)

kernel/kernel/Process.cpp

@@ -173,7 +173,15 @@ namespace Kernel
 			auxiliary_vector.span()
 		));
 		if (tls_addr.has_value())
-			thread->set_tls(*tls_addr);
+		{
+#if ARCH(x86_64)
+			thread->set_fsbase(*tls_addr);
+#elif ARCH(i686)
+			thread->set_gsbase(*tls_addr);
+#else
+#error
+#endif
+		}
 
 		process->add_thread(thread);
 		process->register_to_scheduler();
@@ -730,7 +738,13 @@ namespace Kernel
 			{
 				auto tls_result = TRY(initialize_thread_local_storage(*new_page_table, *executable.master_tls));
 				TRY(new_mapped_regions.emplace_back(BAN::move(tls_result.region)));
-				new_thread->set_tls(tls_result.addr);
+#if ARCH(x86_64)
+				new_thread->set_fsbase(tls_result.addr);
+#elif ARCH(i686)
+				new_thread->set_gsbase(tls_result.addr);
+#else
+#error
+#endif
 			}
 
 			// NOTE: this is done before disabling interrupts and moving the threads as
@@ -3016,16 +3030,28 @@ namespace Kernel
 		return 0;
 	}
 
-	BAN::ErrorOr<long> Process::sys_set_tls(void* addr)
+	BAN::ErrorOr<long> Process::sys_set_fsbase(void* addr)
 	{
-		Thread::current().set_tls(reinterpret_cast<vaddr_t>(addr));
+		Thread::current().set_fsbase(reinterpret_cast<vaddr_t>(addr));
-		Processor::load_tls();
+		Processor::load_fsbase();
 		return 0;
 	}
 
-	BAN::ErrorOr<long> Process::sys_get_tls()
+	BAN::ErrorOr<long> Process::sys_get_fsbase()
 	{
-		return Thread::current().get_tls();
+		return Thread::current().get_fsbase();
+	}
+
+	BAN::ErrorOr<long> Process::sys_set_gsbase(void* addr)
+	{
+		Thread::current().set_gsbase(reinterpret_cast<vaddr_t>(addr));
+		Processor::load_gsbase();
+		return 0;
+	}
+
+	BAN::ErrorOr<long> Process::sys_get_gsbase()
+	{
+		return Thread::current().get_gsbase();
 	}
 
 	BAN::ErrorOr<long> Process::sys_pthread_create(const pthread_attr_t* attr, void (*entry)(void*), void* arg)

kernel/kernel/Processor.cpp

@@ -12,6 +12,7 @@ namespace Kernel
 #if ARCH(x86_64)
 	static constexpr uint32_t MSR_IA32_FS_BASE = 0xC0000100;
 	static constexpr uint32_t MSR_IA32_GS_BASE = 0xC0000101;
+	static constexpr uint32_t MSR_IA32_KERNEL_GS_BASE = 0xC0000102;
 #endif
 
 	ProcessorID          Processor::s_bsp_id                { PROCESSOR_NONE };
@@ -265,15 +266,52 @@ namespace Kernel
 		set_interrupt_state(state);
 	}
 
-	void Processor::load_tls()
+	void Processor::load_segments()
 	{
-		const auto addr = scheduler().current_thread().get_tls();
+		{
+			const auto addr = scheduler().current_thread().get_fsbase();
 #if ARCH(x86_64)
 			uint32_t ptr_hi = addr >> 32;
 			uint32_t ptr_lo = addr & 0xFFFFFFFF;
 			asm volatile("wrmsr" :: "d"(ptr_hi), "a"(ptr_lo), "c"(MSR_IA32_FS_BASE));
 #elif ARCH(i686)
-		gdt().set_tls(addr);
+			gdt().set_fsbase(addr);
+#endif
+		}
+
+		{
+			const auto addr = scheduler().current_thread().get_gsbase();
+#if ARCH(x86_64)
+			uint32_t ptr_hi = addr >> 32;
+			uint32_t ptr_lo = addr & 0xFFFFFFFF;
+			asm volatile("wrmsr" :: "d"(ptr_hi), "a"(ptr_lo), "c"(MSR_IA32_KERNEL_GS_BASE));
+#elif ARCH(i686)
+			gdt().set_gsbase(addr);
+#endif
+		}
+	}
+
+	void Processor::load_fsbase()
+	{
+		const auto addr = scheduler().current_thread().get_fsbase();
+#if ARCH(x86_64)
+		uint32_t ptr_hi = addr >> 32;
+		uint32_t ptr_lo = addr & 0xFFFFFFFF;
+		asm volatile("wrmsr" :: "d"(ptr_hi), "a"(ptr_lo), "c"(MSR_IA32_FS_BASE));
+#elif ARCH(i686)
+		gdt().set_fsbase(addr);
+#endif
+	}
+
+	void Processor::load_gsbase()
+	{
+		const auto addr = scheduler().current_thread().get_gsbase();
+#if ARCH(x86_64)
+		uint32_t ptr_hi = addr >> 32;
+		uint32_t ptr_lo = addr & 0xFFFFFFFF;
+		asm volatile("wrmsr" :: "d"(ptr_hi), "a"(ptr_lo), "c"(MSR_IA32_KERNEL_GS_BASE));
+#elif ARCH(i686)
+		gdt().set_gsbase(addr);
 #endif
 	}
 

kernel/kernel/Scheduler.cpp

@@ -282,7 +282,8 @@ namespace Kernel
 		}
 
 		Processor::gdt().set_tss_stack(thread->kernel_stack_top());
-		Processor::load_tls();
+		if (thread->is_userspace())
+			Processor::load_segments();
 		*interrupt_stack     = thread->interrupt_stack();
 		*interrupt_registers = thread->interrupt_registers();
 

kernel/kernel/Thread.cpp

@@ -328,7 +328,8 @@ namespace Kernel
 		thread->m_kernel_stack = TRY(m_kernel_stack->clone(new_process->page_table()));
 		thread->m_userspace_stack = TRY(m_userspace_stack->clone(new_process->page_table()));
 
-		thread->m_tls = m_tls;
+		thread->m_fsbase = m_fsbase;
+		thread->m_gsbase = m_gsbase;
 
 		thread->m_state = State::NotStarted;
 

userspace/libraries/LibC/include/sys/syscall.h

@@ -101,8 +101,10 @@ __BEGIN_DECLS
 	O(SYS_HARDLINKAT,		hardlinkat)		\
 	O(SYS_UTIMENSAT,		utimensat)		\
 	O(SYS_YIELD,			yield)			\
-	O(SYS_SET_TLS,			set_tls)		\
+	O(SYS_SET_FSBASE,		set_fsbase)		\
-	O(SYS_GET_TLS,			get_tls)		\
+	O(SYS_GET_FSBASE,		get_fsbase)		\
+	O(SYS_SET_GSBASE,		set_gsbase)		\
+	O(SYS_GET_GSBASE,		get_gsbase)		\
 	O(SYS_PTHREAD_CREATE,	pthread_create)	\
 	O(SYS_PTHREAD_EXIT,		pthread_exit)	\
 	O(SYS_PTHREAD_JOIN,		pthread_join)	\

userspace/libraries/LibC/pthread.cpp

@@ -49,7 +49,13 @@ extern "C" void _pthread_trampoline_cpp(void* arg)
 {
 	auto info = *reinterpret_cast<pthread_trampoline_info_t*>(arg);
 	info.uthread->id = syscall(SYS_PTHREAD_SELF);
-	syscall(SYS_SET_TLS, info.uthread);
+#if defined(__x86_64__)
+	syscall(SYS_SET_FSBASE, info.uthread);
+#elif defined(__i686__)
+	syscall(SYS_SET_GSBASE, info.uthread);
+#else
+#error
+#endif
 	free(arg);
 	pthread_exit(info.start_routine(info.arg));
 	ASSERT_NOT_REACHED();

userspace/libraries/LibC/unistd.cpp

@@ -34,7 +34,13 @@ extern "C" void _init_libc(char** environ, init_funcs_t init_funcs, init_funcs_t
 	if (::environ == nullptr)
 		::environ = environ;
 
-	if (uthread* self = reinterpret_cast<uthread*>(syscall(SYS_GET_TLS)))
+#if defined(__x86_64__)
+	if (uthread* self = reinterpret_cast<uthread*>(syscall(SYS_GET_FSBASE)))
+#elif defined(__i686__)
+	if (uthread* self = reinterpret_cast<uthread*>(syscall(SYS_GET_GSBASE)))
+#else
+#error
+#endif
 	{
 		self->cleanup_stack = nullptr;
 		self->id = syscall(SYS_PTHREAD_SELF);
@@ -61,7 +67,13 @@ extern "C" void _init_libc(char** environ, init_funcs_t init_funcs, init_funcs_t
 		};
 		uthread.dtv[0] = 0;
 
-		syscall(SYS_SET_TLS, &uthread);
+#if defined(__x86_64__)
+		syscall(SYS_SET_FSBASE, &uthread);
+#elif defined(__i686__)
+		syscall(SYS_SET_GSBASE, &uthread);
+#else
+#error
+#endif
 	}
 
 	// call global constructors

userspace/programs/DynamicLoader/main.cpp

@@ -1366,7 +1366,13 @@ static void initialize_tls(MasterTLS master_tls)
 		uthread.dtv[elf.tls_module] = reinterpret_cast<uintptr_t>(tls_addr) + uthread.master_tls_size - elf.tls_offset;
 	}
 
-	syscall(SYS_SET_TLS, &uthread);
+#if defined(__x86_64__)
+	syscall(SYS_SET_FSBASE, &uthread);
+#elif defined(__i686__)
+	syscall(SYS_SET_GSBASE, &uthread);
+#else
+#error
+#endif
 }
 
 static void initialize_environ(char** envp)