Switched to using mret for trap

risc_v/ch4/src/asm/boot.S

@@ -47,9 +47,14 @@ _start:
 	# to the very end of the stack range.
 	la		sp, _stack_end
 	# Setting `mstatus` register:
-	# 0b11 << 11: Machine's previous protection mode is 3 (MPP=3).
+	# 0b01 << 11: Machine's previous protection mode is 2 (MPP=2).
 	li		t0, 0b11 << 11
 	csrw	mstatus, t0
+	# li		t2, (1 << 1) | (1 << 5) | (1 << 9)
+	# li		t2, 0xffff
+	# csrw	mideleg, t2
+	li		t3, 0xaaa
+	csrw	mie, t3
 	# Machine's exception program counter (MEPC) is set to `kinit`.
 	la		t1, kinit
 	csrw	mepc, t1
@@ -71,10 +76,10 @@ _start:
 	# 1 << 1    : Supervisor's interrupt-enable bit will be set to 1 after sret.
 	# We set the "previous" bits because the sret will write the current bits
 	# with the previous bits.
-	li		t0, (1 << 8) | (1 << 5)
+	li		t0, (1 << 11)
-	csrw	sstatus, t0
+	csrw	mstatus, t0
 	la		t1, kmain
-	csrw	sepc, t1
+	csrw	mepc, t1
 	# Setting `mideleg` (machine interrupt delegate) register:
 	# 1 << 1   : Software interrupt delegated to supervisor mode
 	# 1 << 5   : Timer interrupt delegated to supervisor mode
@@ -83,20 +88,19 @@ _start:
 	# cause an elevation to the machine privilege mode (mode 3).
 	# When we delegate, we're telling the CPU to only elevate to
 	# the supervisor privilege mode (mode 1)
-	li		t2, (1 << 1) | (1 << 5) | (1 << 9)
-	csrw	mideleg, t2
 	# Setting `sie` (supervisor interrupt enable) register:
 	# This register takes the same bits as mideleg
 	# 1 << 1    : Supervisor software interrupt enable (SSIE=1 [Enabled])
 	# 1 << 5    : Supervisor timer interrupt enable (STIE=1 [Enabled])
 	# 1 << 9    : Supervisor external interrupt enable (SEIE=1 [Enabled])
-	csrw	sie, t2
+	# li		t2, (1 << 1) | (1 << 5) | (1 << 9)
+	# csrw	sie, t2
 	# Setting `stvec` (supervisor trap vector) register:
 	# Essentially this is a function pointer, but the last two bits can be 00 or 01
 	# 00        : All exceptions set pc to BASE
 	# 01        : Asynchronous interrupts set pc to BASE + 4 x scause
-	la		t3, s_trap_vector
+	# la		t3, s_trap
-	csrw	stvec, t3
+	# csrw	stvec, t3
 	# kinit() is required to return back the SATP value (including MODE) via a0
 	csrw	satp, a0
 	# Force the CPU to take our SATP register.
@@ -105,8 +109,10 @@ _start:
 	# it in memory, it will be the old table. So, sfence.vma will ensure that the MMU always
 	# grabs a fresh copy of the SATP register and associated tables.
 	sfence.vma
+	# ret
+	# call	kmain
 	# sret will put us in supervisor mode and re-enable interrupts
-	sret
+	mret
 3:
 
 	# Parked harts go here. We need to set these
@@ -117,6 +123,27 @@ _start:
 	# where base address is 0x0200_0000 (MMIO CLINT base address)
 	# We only use additional harts to run user-space programs, although this may
 	# change.
+
+	la		sp, _stack_end
+	li		t0, 0x10000
+	csrr	t1, mhartid
+	mul		t0, t0, t1
+	sub		sp, sp, t0
+
+	li		t0, 0b01 << 11 | (1 << 5)
+	csrw	mstatus, t0
+	li		t3, 0xaaa
+	csrw	mie, t3
+	# Machine's exception program counter (MEPC) is set to `kinit`.
+	la		t1, 4f
+	csrw	mepc, t1
+	# Machine's trap vector base address is set to `m_trap_vector`, for
+	# "machine" trap vector.
+	la		t2, m_trap_vector
+	csrw	mtvec, t2
+	# We use mret here so that the mstatus register is properly updated.
+	mret
+
 4:
 	wfi
 	j		4b

risc_v/ch4/src/asm/trap.S

@@ -5,13 +5,14 @@
 # 24 February 2019
 .option norvc
 .altmacro
-.set NUM_GP_REGS, 34  # Number of registers per context
+.set NUM_GP_REGS, 32  # Number of registers per context
 .set NUM_FP_REGS, 32
 .set REG_SIZE, 8   # Register size (in bytes)
+.set MAX_CPUS, 8   # Maximum number of CPUs
 
 .section .bss
 .global _GLOBAL_CTX
-.lcomm _GLOBAL_CTX, (NUM_FP_REGS + NUM_GP_REGS) * REG_SIZE
+.lcomm _GLOBAL_CTX, (NUM_FP_REGS + NUM_GP_REGS) * REG_SIZE * MAX_CPUS
 
 # Use macros for saving and restoring multiple registers
 .macro save_gp i, basereg=t6
@@ -39,21 +40,17 @@ m_trap_vector:
 	# base register for saving
 	csrw	mscratch, t6
 
-	# Disable the MMU since we're switching.
 	csrr	t6, satp
-	slli	t6, t6, 4
+	slli	t6, t6, 1
-	srli	t6, t6, 4
+	srli	t6, t6, 1
 	csrw	satp, t6
-
-	# Disable interrupts
-	li		t6, 1 << 3
-	csrrc	zero, mstatus, t6
-
     # Start saving the registers. We can't do much until the registers
     # have been saved.
 	la		t6, _GLOBAL_CTX
-	.set i, 0
+
-	.rept 31
+	# Set i = 1. i = 0 is the zero register and is a waste of time.
+	.set i, 1
+	.rept 30
 		save_gp	%i
 		.set	i, i+1
 	.endr
@@ -63,16 +60,15 @@ m_trap_vector:
 	csrr	t6, mscratch
 	save_gp	31, t5
 
-	# csrr	t0, satp
+	# We need to check the status of the floating
-	# sd		t0, 32*8(t5)
+	# point system. (FS bits are described in 3.1.6.5
-
+	# and table 3.3 in the RISC-V Privilege spec).
 	csrr	t1, mstatus
 	srli	t0, t1, 13
 	andi	t0, t0, 3
-	li	t1, 3
 	
 	# Skip saving the FP registers if the FP is turned off.
-	bne	t0, t1, 3f
+	beqz 	t0, 3f
 	la		t6, _GLOBAL_CTX
 	.set i, 0
 	.rept 32
@@ -86,22 +82,37 @@ m_trap_vector:
 	# trap_handler returns the new mepc
 	# via a0
 	la		sp, _stack_end
+	li		t0, 0x10000
+	csrr	t1, mhartid
+	mul		t0, t0, t1
+	sub		sp, sp, t0
 	csrr	a0, mepc
 	csrr	a1, mtval
 	csrr	a2, mcause
 	csrr	a3, mhartid
-	call	m_trap
+	csrr	a4, mstatus
+	la		t1, m_trap
+	csrw	mepc, t1
+	la		ra, 3f
+	li		t2, 0b11 << 11
+	csrw	mstatus, t2
+	mret
+3:
 	csrw	mepc, a0
 
 	# Move to the context for this CPU
 	la		t6, _GLOBAL_CTX
+	# Go to the context for this hart
+	csrr	t5, mhartid
+	slli	t5, t5, 6
+	add		t6, t6, t5
+
 	csrr	t1, mstatus
 	srli	t0, t1, 13
 	andi	t0, t0, 3
-	li		t1, 3
 
 	# Skip loading the FP registers if the FP is turned off.
-	bne	t0, t1, 3f
+	beqz	t0, 3f
 	.set i, 0
 	.rept 32
 		load_fp	%i
@@ -109,128 +120,22 @@ m_trap_vector:
 	.endr
 
 3:
-
-	# csrr	t5, mhartid
-	# slli	t5, t5, CTX_SHL
-	# add	t6, t6, t5
 	# Restore the registers
-	.set i, 0
+	.set i, 1
-	.rept 32
-		load_gp %i
-		.set i, i+1
-	.endr
-	# Get the MMU going
-	csrw	mscratch, t6
-	la		t6, _GLOBAL_CTX
-	ld		t6, 32*8(t6)
-	csrw	satp, t6
-	sfence.vma
-	csrr	t6, mscratch
-	# x31 gets restored at the last iteration
-	mret
-
-
-.global s_trap_vector
-# This must be aligned by 4 since the last two bits
-# of the mtvec register do not contribute to the address
-# of this vector.
-.align 4
-s_trap_vector:
-	# We have to save t6 (x31), since it will be our
-	# base register for saving
-	csrw	sscratch, t6
-
-	# Disable the MMU since we're switching.
-	csrr	t6, satp
-	slli	t6, t6, 4
-	srli	t6, t6, 4
-	csrw	satp, t6
-
-	# Disable interrupts
-	li		t6, 1 << 3
-	csrrc	zero, sstatus, t6
-
-    # Start saving the registers. We can't do much until the registers
-    # have been saved.
-	la		t6, _GLOBAL_CTX
-	.set i, 0
 	.rept 31
-		save_gp	%i
-		.set	i, i+1
-	.endr
-	# We saved 31 registers. All but t6
-	# t5 is saved, so we can overwrite it.
-	mv		t5, t6
-	csrr	t6, sscratch
-	save_gp	31, t5
-
-	# csrr	t0, satp
-	# sd		t0, 32*8(t5)
-
-	csrr	t1, sstatus
-	srli	t0, t1, 13
-	andi	t0, t0, 3
-	li		t1, 3
-	
-	# Skip saving the FP registers if the FP is turned off.
-	bne	t0, t1, 3f
-	la	t6, _GLOBAL_CTX
-	.set i, 0
-	.rept 32
-		save_fp	%i
-		.set	i, i+1
-	.endr
-
-3:
-	# Go to Rust
-	# usize trap_handler(mepc, mcause)
-	# trap_handler returns the new mepc
-	# via a0
-    la      sp, _stack_end
-	csrr	a0, sepc
-	csrr	a1, stval
-	csrr	a2, scause
-	call	s_trap
-	csrw	sepc, a0
-
-	# Move to the context for this CPU
-	la		t6, _GLOBAL_CTX
-	csrr	t1, sstatus
-	srli	t0, t1, 13
-	andi	t0, t0, 3
-	li		t1, 3
-
-	# Skip loading the FP registers if the FP is turned off.
-	bne	t0, t1, 3f
-	.set i, 0
-	.rept 32
-		load_fp	%i
-		.set	i, i+1
-	.endr
-
-3:
-
-	# csrr	t5, mhartid
-	# slli	t5, t5, CTX_SHL
-	# add	t6, t6, t5
-	# Restore the registers
-	.set i, 0
-	.rept 32
 		load_gp %i
 		.set i, i+1
 	.endr
-	# Get the MMU going
+
-	csrw	sscratch, t6
+	# The MMU should be going at this point via
-	la		t6, _GLOBAL_CTX
+	# Rust.
-	ld		t6, 32*8(t6)
+
-	csrw	satp, t6
+	li		t2, (1 << 8) | (1 << 5)
-	sfence.vma
+	csrw	mstatus, t2
-	csrr	t6, mscratch
+
 	# x31 gets restored at the last iteration
 	sret
 
-
-
 .global make_syscall
 make_syscall:
 	ecall

risc_v/ch4/src/kmem.rs

@@ -79,10 +79,10 @@ pub fn get_num_allocations() -> usize {
 /// alloc/dealloc from the page crate.
 pub fn init() {
 	unsafe {
-		// Allocate 64 kernel pages (64 * 4096 = 262 KiB)
+		// Allocate kernel pages (KMEM_ALLOC)
-		let k_alloc = zalloc(64);
+		KMEM_ALLOC = 512;
+		let k_alloc = zalloc(KMEM_ALLOC);
 		assert!(!k_alloc.is_null());
-		KMEM_ALLOC = 64;
 		KMEM_HEAD = k_alloc as *mut AllocList;
 		(*KMEM_HEAD).set_free();
 		(*KMEM_HEAD).set_size(KMEM_ALLOC * PAGE_SIZE);

risc_v/ch4/src/lib.rs

@@ -146,12 +146,12 @@ extern "C" fn kinit() -> usize {
 		println!("DATA:   0x{:x} -> 0x{:x}", DATA_START, DATA_END);
 		println!("BSS:    0x{:x} -> 0x{:x}", BSS_START, BSS_END);
 		println!("STACK:  0x{:x} -> 0x{:x}", KERNEL_STACK_START, KERNEL_STACK_END);
-		println!("HEAP:   0x{:x} -> 0x{:x}", kheap_head, kheap_head + total_pages * 4096);
+		println!("HEAP:   0x{:x} -> 0x{:x}", kheap_head, kheap_head + total_pages * page::PAGE_SIZE);
 	}
 	id_map_range(
 	             &mut root,
 	             kheap_head,
-	             kheap_head + total_pages * 4096,
+	             kheap_head + total_pages * page::PAGE_SIZE,
 	             page::EntryBits::ReadWrite.val(),
 	);
 	unsafe {
@@ -203,38 +203,20 @@ extern "C" fn kinit() -> usize {
 	}
 
 	// UART
-	page::map(
+	id_map_range(
 				&mut root,
 				0x1000_0000,
-	          0x1000_0000,
+				0x1000_0100,
 				page::EntryBits::ReadWrite.val(),
-			  0
 	);
 
 	// CLINT
 	//  -> MSIP
-	page::map(
+	id_map_range(
 				&mut root,
 				0x0200_0000,
-	          0x0200_0000,
+				0x0200_ffff,
 				page::EntryBits::ReadWrite.val(),
-			  0
-	);
-	//  -> MTIMECMP
-	page::map(
-	          &mut root,
-	          0x0200_b000,
-	          0x0200_b000,
-	          page::EntryBits::ReadWrite.val(),
-			  0
-	);
-	//  -> MTIME
-	page::map(
-	          &mut root,
-	          0x0200_c000,
-	          0x0200_c000,
-	          page::EntryBits::ReadWrite.val(),
-			  0
 	);
 	// PLIC
 	id_map_range(
@@ -255,7 +237,7 @@ extern "C" fn kinit() -> usize {
 	// space application requires services. Since the user space application
 	// only knows virtual addresses, we have to translate silently behind
 	// the scenes.
-	let p = 0x0200_0000 as usize;
+	let p = 0x0200_4000 as usize;
 	let m = page::virt_to_phys(&root, p).unwrap_or(0);
 	println!("Walk 0x{:x} = 0x{:x}", p, m);
 	// When we return from here, we'll go back to boot.S and switch into
@@ -271,6 +253,7 @@ extern "C" fn kinit() -> usize {
 		// We have to store the kernel's table. The tables will be moved back
 		// and forth between the kernel's table and user applicatons' tables.
 		KERNEL_TABLE = root_u;
+		println!("Setting 0x{:x}", KERNEL_TABLE);
 	}
 	// table / 4096    Sv39 mode
 	(root_u >> 12)  | (8 << 60)
@@ -290,18 +273,23 @@ extern "C" fn kmain() {
 		// We have the global allocator, so let's see if that works!
 		let k = Box::<u32>::new(100);
 		println!("Boxed value = {}", *k);
-		kmem::print_table();
 		// The following comes from the Rust documentation:
 		// some bytes, in a vector
 		let sparkle_heart = vec![240, 159, 146, 150];
 		// We know these bytes are valid, so we'll use `unwrap()`.
+		// This will MOVE the vector.
 		let sparkle_heart = String::from_utf8(sparkle_heart).unwrap();
 		println!("String = {}", sparkle_heart);
+		println!("\n\nAllocations of a box, vector, and string");
+		kmem::print_table();
 	}
+	println!("\n\nEverything should now be free:");
+	kmem::print_table();
 	unsafe {
-		let val = 0x0200_0000 as *mut u32;
+		// asm!("csrw sip, $0" :: "r"(1));
-		val.write_volatile(1);
+		let mtimecmp = 0x0200_4000 as *mut u64;
-		asm!("ecall");
+		let mtime = 0x0200_bff8 as *const u64;
+		mtimecmp.write_volatile(mtime.read_volatile() + 10_000_000);
 	}
 	// If we get here, the Box, vec, and String should all be freed since
 	// they go out of scope. This calls their "Drop" trait.

risc_v/ch4/src/page.rs

@@ -285,12 +285,12 @@ pub fn print_page_allocations() {
 		}
 		println!("~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~");
 		println!(
-		         "Allocated: {:>5} pages ({:>9} bytes).",
+		         "Allocated: {:>6} pages ({:>10} bytes).",
 		         num,
 		         num * PAGE_SIZE
 		);
 		println!(
-		         "Free     : {:>5} pages ({:>9} bytes).",
+		         "Free     : {:>6} pages ({:>10} bytes).",
 		         num_pages - num,
 		         (num_pages - num) * PAGE_SIZE
 		);
@@ -459,7 +459,10 @@ pub fn map(root: &mut Table, vaddr: usize, paddr: usize, bits: i64, level: usize
 	            (ppn[1] << 19) as i64 |   // PPN[1] = [27:19]
 				(ppn[0] << 10) as i64 |   // PPN[0] = [18:10]
 				bits |                    // Specified bits, such as User, Read, Write, etc
-				EntryBits::Valid.val();   // Valid bit
+				EntryBits::Valid.val() |  // Valid bit
+				EntryBits::Dirty.val() |  // Some machines require this to =1
+				EntryBits::Access.val()   // Just like dirty, some machines require this
+				;   
 			 // Set the entry. V should be set to the correct pointer by the loop
 			 // above.
 	v.set_entry(entry);

risc_v/ch4/src/trap.rs

@@ -9,7 +9,7 @@ extern "C" {
 
 #[no_mangle]
 extern "C"
-fn s_trap(epc: usize, tval: usize, cause: usize) -> usize {
+fn s_trap(epc: usize, tval: usize, cause: isize) -> usize {
 	println!("STRAP (cause: 0x{:x} @ 0x{:x})", cause, epc);
 	unsafe {
 		// Switch to kernel's page table.
@@ -17,12 +17,47 @@ fn s_trap(epc: usize, tval: usize, cause: usize) -> usize {
 		let satp = KERNEL_TABLE >> 12 | 8 << 60;
 		asm!("csrw satp, $0" :: "r"(satp));
 	}
+	if cause < 0 {
+		epc
+	}
+	else {
 		epc  + 4
+	}
 }
 
 #[no_mangle]
 extern "C"
-fn m_trap(epc: usize, tval: usize, cause: usize, hart: usize) -> usize {
+fn m_trap(epc: usize, tval: usize, cause: isize, hart: usize, stat: usize) -> usize {
-	println!("MTRAP (cause: 0x{:x} @ 0x{:x})", cause, epc);
+	println!("MTRAP ({}) (cause: 0x{:x} @ 0x{:x}) [0x{:x}]", hart, cause, epc, stat);
-	epc + 4
+	unsafe {
+		if cause < 0 {
+			// Asynchronous
+			match cause & 0xff {
+				4 | 5 | 7 => {
+					let satp: usize = KERNEL_TABLE >> 12 | 8 << 60;
+					println!("Kernel table = 0x{:x}", KERNEL_TABLE);
+					// asm!("csrw satp, $0" :: "r"(satp) :: "volatile");
+					// asm!("sfence.vma" :::: "volatile");
+					// asm!("csrw mie, zero" :::: "volatile");
+				},
+				_ => { println!("Async cause\n"); }
+			}
+
+		}
+		else {
+			match cause {
+				2 => {
+					panic!("Illegal instruction");
+				},
+				12 => {
+					panic!("Instruction page fault.");
+				},
+				13 => {
+					panic!("Load page fault.");
+				},
+				_ => { println!("Sync cause\n"); }
+			}
+		}
+	}
+	epc
 }