Add support for loadable kernel modules.

Submitted by:	Yukishige Shibata <y-shibat@mtd.biglobe.ne.jp>

2 files changed, 368 insertions, 3 deletions

diff --git a/sys/riscv/conf/GENERIC b/sys/riscv/conf/GENERIC
index 12c1714..13f595a4 100644
--- a/sys/riscv/conf/GENERIC
+++ b/sys/riscv/conf/GENERIC
@@ -23,7 +23,9 @@ ident		GENERIC
 
 makeoptions	DEBUG=-g		# Build kernel with gdb(1) debug symbols
 # makeoptions	WITH_CTF=1		# Run ctfconvert(1) for DTrace support
-makeoptions	NO_MODULES=1		# We don't yet support modules on RISC-V
+
+# FIXME: linker error. "--relax and -r may not be used together"
+makeoptions	WITHOUT_MODULES="usb otusfw mwlfw ispfw mwlfw ralfw rtwnfw urtwnfw"
 
 options 	SCHED_ULE		# ULE scheduler
 options 	PREEMPTION		# Enable kernel thread preemption
diff --git a/sys/riscv/riscv/elf_machdep.c b/sys/riscv/riscv/elf_machdep.c
index 1e44801..1b8c7a9 100644
--- a/sys/riscv/riscv/elf_machdep.c
+++ b/sys/riscv/riscv/elf_machdep.c
@@ -1,6 +1,7 @@
 /*-
  * Copyright 1996-1998 John D. Polstra.
  * Copyright (c) 2015 Ruslan Bukin <br@bsdpad.com>
+ * Copyright (c) 2016 Yukishige SHibata <y-shibat@mtd.biglobe.ne.jp>
  * All rights reserved.
  *
  * Portions of this software were developed by SRI International and the
@@ -43,6 +44,7 @@ __FBSDID("$FreeBSD$");
 #include <sys/imgact.h>
 #include <sys/linker.h>
 #include <sys/proc.h>
+#include <sys/sysctl.h>
 #include <sys/sysent.h>
 #include <sys/imgact_elf.h>
 #include <sys/syscall.h>
@@ -123,19 +125,380 @@ SYSINIT(oelf64, SI_SUB_EXEC, SI_ORDER_ANY,
 	(sysinit_cfunc_t) elf64_insert_brand_entry,
 	&freebsd_brand_oinfo);
 
+static int debug_kld;
+SYSCTL_INT(_kern, OID_AUTO, debug_kld,
+	   CTLFLAG_RW, &debug_kld, 0,
+	   "Activate debug prints in elf_reloc_internal()");
+
+struct type2str_ent {
+	int type;
+	const char* str;
+};
+
 void
 elf64_dump_thread(struct thread *td, void *dst, size_t *off)
 {
 
 }
 
-/* Process one elf relocation with addend. */
+/*
+ * Following 4 functions are used to manupilate bits on 32bit interger value.
+ * FIXME: I implemetend for ease-to-understand rather than for well-optimized.
+ */
+static uint32_t
+gen_bitmask(int msb, int lsb)
+{
+	uint32_t mask;
+
+	if (msb == sizeof(mask) * 8 - 1)
+		mask = ~0;
+	else
+		mask = (1U << (msb + 1)) - 1;
+
+	if (lsb > 0)
+		mask &= ~((1U << lsb) - 1);
+
+	return (mask);
+}
+
+static uint32_t
+extract_bits(uint32_t x, int msb, int lsb)
+{
+	uint32_t mask;
+
+	mask = gen_bitmask(msb, lsb);
+
+	x &= mask;
+	x >>= lsb;
+
+	return (x);
+}
+
+static uint32_t
+insert_bits(uint32_t d, uint32_t s, int msb, int lsb)
+{
+	uint32_t mask;
+
+	mask = gen_bitmask(msb, lsb);
+
+	d &= ~mask;
+
+	s <<= lsb;
+	s &= mask;
+
+	return (d | s);
+}
+
+static uint32_t
+insert_imm(uint32_t insn, uint32_t imm, int imm_msb, int imm_lsb,
+    int insn_lsb)
+{
+	int insn_msb;
+	uint32_t v;
+
+	v = extract_bits(imm, imm_msb, imm_lsb);
+	insn_msb = (imm_msb - imm_lsb) + insn_lsb;
+
+	return (insert_bits(insn, v, insn_msb, insn_lsb));
+}
+
+/*
+ * The RISCV ISA is designed so that all of immediate value is
+ * sign-extened.
+ * An immediate value is sometimes generated at runtime by adding
+ * 12bit sign integer and 20bit signed integer. This requests 20bit
+ * immediate value to be ajusted if the MSB of the 12bit immediate
+ * value is asserted (sign extened value is treated as negative value).
+ *
+ * For example, 0x123800 can be calculated by adding upper 20 bit of
+ * 0x124000 and signed-extended 12bit immediate whose bit pattern is
+ * 0x800 as follows;
+ *   0x123800
+ *     = 0x123000 + 0x800
+ *     = (0x123000 + 0x1000) + (-0x1000 + 0x800)
+ *     = (0x123000 + 0x1000) + (0xff...ff800)
+ *     = 0x124000            + sign-exntend(0x800)
+ */
+static uint32_t
+calc_hi20_imm(uint32_t value)
+{
+	/*
+	 * There is the arithmetical hack that can remove conditional
+	 * statement. But I implement it in straghtforward way.
+	 */
+	if ((value & 0x800) != 0)
+		value += 0x1000;
+	return (value & ~0xfff);
+}
+
+static const struct type2str_ent t2s[] = {
+	{ R_RISCV_NONE,		"R_RISCV_NONE"		},
+	{ R_RISCV_64,		"R_RISCV_64"		},
+	{ R_RISCV_JUMP_SLOT,	"R_RISCV_JUMP_SLOT"	},
+	{ R_RISCV_RELATIVE,	"R_RISCV_RELATIVE"	},
+	{ R_RISCV_JAL,		"R_RISCV_JAL"		},
+	{ R_RISCV_CALL,		"R_RISCV_CALL"		},
+	{ R_RISCV_PCREL_HI20,	"R_RISCV_PCREL_HI20"	},
+	{ R_RISCV_PCREL_LO12_I,	"R_RISCV_PCREL_LO12_I"	},
+	{ R_RISCV_PCREL_LO12_S,	"R_RISCV_PCREL_LO12_S"	},
+	{ R_RISCV_HI20,		"R_RISCV_HI20"		},
+	{ R_RISCV_LO12_I,	"R_RISCV_LO12_I"	},
+	{ R_RISCV_LO12_S,	"R_RISCV_LO12_S"	},
+};
+
+static const char*
+reloctype_to_str(int type)
+{
+	int i;
+
+	for (i = 0; i < sizeof(t2s) / sizeof(t2s[0]); ++i) {
+		if (type == t2s[i].type)
+			return t2s[i].str;
+	}
+
+	return "*unknown*";
+}
+
+/*
+ * Currently kernel loadable module for RISCV is compiled with -fPIC option.
+ * (see also additional CFLAGS definition for RISCV in sys/conf/kmod.mk)
+ * Only R_RISCV_64, R_RISCV_JUMP_SLOT and RISCV_RELATIVE are emitted in
+ * the module. Other relocations will be processed when kernel loadable
+ * modules are built in non-PIC.
+ *
+ * FIXME: only RISCV64 is supported.
+ */
 static int
 elf_reloc_internal(linker_file_t lf, Elf_Addr relocbase, const void *data,
     int type, int local, elf_lookup_fn lookup)
 {
+	Elf_Size rtype, symidx;
+	const Elf_Rela *rela;
+	Elf_Addr val, addr;
+	Elf64_Addr *where;
+	Elf_Addr addend;
+	uint32_t before32_1;
+	uint32_t before32;
+	uint64_t before64;
+	uint32_t* insn32p;
+	uint32_t imm20;
+	int error;
+
+	switch (type) {
+	case ELF_RELOC_RELA:
+		rela = (const Elf_Rela *)data;
+		where = (Elf_Addr *)(relocbase + rela->r_offset);
+		insn32p = (uint32_t*)where;
+		addend = rela->r_addend;
+		rtype = ELF_R_TYPE(rela->r_info);
+		symidx = ELF_R_SYM(rela->r_info);
+		break;
+	default:
+		printf("%s:%d unknown reloc type %d\n",
+		       __FUNCTION__, __LINE__, type);
+		return -1;
+	}
+
+	switch (rtype) {
+	case R_RISCV_NONE:
+		break;
+
+	case R_RISCV_64:
+	case R_RISCV_JUMP_SLOT:
+		error = lookup(lf, symidx, 1, &addr);
+		if (error != 0)
+			return -1;
+
+		val = addr;
+		before64 = *where;
+		if (*where != val)
+			*where = val;
+
+		if (debug_kld)
+			printf("%p %c %-24s %016lx -> %016lx\n",
+			       where,
+			       (local? 'l': 'g'),
+			       reloctype_to_str(rtype),
+			       before64, *where);
+		break;
+
+	case R_RISCV_RELATIVE:
+		val = relocbase + addend;
+
+		before64 = *where;
+		if (*where != val)
+			*where = val;
+
+		if (debug_kld)
+			printf("%p %c %-24s %016lx -> %016lx\n",
+			       where,
+			       (local? 'l': 'g'),
+			       reloctype_to_str(rtype),
+			       before64, *where);
+		break;
+
+	case R_RISCV_JAL:
+		error = lookup(lf, symidx, 1, &addr);
+		if (error != 0)
+			return -1;
+
+		val = addr - (Elf_Addr)where;
+		if ((val <= -(1UL << 20) || (1UL << 20) <= val)) {
+			printf("kldload: huge offset against R_RISCV_JAL\n");
+			return -1;
+		}
+
+		before32 = *insn32p;
+		*insn32p = insert_imm(*insn32p, val, 20, 20, 31);
+		*insn32p = insert_imm(*insn32p, val, 10,  1, 21);
+		*insn32p = insert_imm(*insn32p, val, 11, 11, 20);
+		*insn32p = insert_imm(*insn32p, val, 19, 12, 12);
+
+		if (debug_kld)
+			printf("%p %c %-24s %08x -> %08x\n",
+			       where,
+			       (local? 'l': 'g'),
+			       reloctype_to_str(rtype),
+			       before32, *insn32p);
+		break;
+
+	case R_RISCV_CALL:
+		/*
+		 * R_RISCV_CALL relocates 8-byte region that consists
+		 * of the sequence of AUIPC and JALR.
+		 */
+		/* calculate and check the pc relative offset. */
+		error = lookup(lf, symidx, 1, &addr);
+		if (error != 0)
+			return -1;
+		val = addr - (Elf_Addr)where;
+		if ((val <= -(1UL << 32) || (1UL << 32) <= val)) {
+			printf("kldload:%s: huge offset against R_RISCV_CALL\n");
+			return -1;
+		}
+
+		/* Relocate AUIPC. */
+		before32 = insn32p[0];
+		imm20 = calc_hi20_imm(val);
+		insn32p[0] = insert_imm(insn32p[0], imm20, 31, 12, 12);
 
-	panic("elf_reloc_internal");
+		/* Relocate JALR. */
+		before32_1 = insn32p[1];
+		insn32p[1] = insert_imm(insn32p[1], val, 11,  0, 20);
+
+		if (debug_kld)
+			printf("%p %c %-24s %08x %08x -> %08x %08x\n",
+			       where,
+			       (local? 'l': 'g'),
+			       reloctype_to_str(rtype),
+			       before32,   insn32p[0],
+			       before32_1, insn32p[1]);
+		break;
+
+	case R_RISCV_PCREL_HI20:
+		val = addr - (Elf_Addr)where;
+		insn32p = (uint32_t*)where;
+		before32 = *insn32p;
+		imm20 = calc_hi20_imm(val);
+		*insn32p = insert_imm(*insn32p, imm20, 31, 12, 12);
+
+		if (debug_kld)
+			printf("%p %c %-24s %08x -> %08x\n",
+			       where,
+			       (local? 'l': 'g'),
+			       reloctype_to_str(rtype),
+			       before32, *insn32p);
+		break;
+
+	case R_RISCV_PCREL_LO12_I:
+		val = addr - (Elf_Addr)where;
+		insn32p = (uint32_t*)where;
+		before32 = *insn32p;
+		*insn32p = insert_imm(*insn32p, addr, 11,  0, 20);
+
+		if (debug_kld)
+			printf("%p %c %-24s %08x -> %08x\n",
+			       where,
+			       (local? 'l': 'g'),
+			       reloctype_to_str(rtype),
+			       before32, *insn32p);
+		break;
+
+	case R_RISCV_PCREL_LO12_S:
+		val = addr - (Elf_Addr)where;
+		insn32p = (uint32_t*)where;
+		before32 = *insn32p;
+		*insn32p = insert_imm(*insn32p, addr, 11,  5, 25);
+		*insn32p = insert_imm(*insn32p, addr,  4,  0,  7);
+		if (debug_kld)
+			printf("%p %c %-24s %08x -> %08x\n",
+			       where,
+			       (local? 'l': 'g'),
+			       reloctype_to_str(rtype),
+			       before32, *insn32p);
+		break;
+
+	case R_RISCV_HI20:
+		error = lookup(lf, symidx, 1, &addr);
+		if (error != 0)
+			return -1;
+
+		insn32p = (uint32_t*)where;
+		before32 = *insn32p;
+		imm20 = calc_hi20_imm(val);
+		*insn32p = insert_imm(*insn32p, imm20, 31, 12, 12);
+
+		if (debug_kld)
+			printf("%p %c %-24s %08x -> %08x\n",
+			       where,
+			       (local? 'l': 'g'),
+			       reloctype_to_str(rtype),
+			       before32, *insn32p);
+		break;
+
+	case R_RISCV_LO12_I:
+		error = lookup(lf, symidx, 1, &addr);
+		if (error != 0)
+			return -1;
+
+		val = addr;
+		insn32p = (uint32_t*)where;
+		before32 = *insn32p;
+		*insn32p = insert_imm(*insn32p, addr, 11,  0, 20);
+
+		if (debug_kld)
+			printf("%p %c %-24s %08x -> %08x\n",
+			       where,
+			       (local? 'l': 'g'),
+			       reloctype_to_str(rtype),
+			       before32, *insn32p);
+		break;
+
+	case R_RISCV_LO12_S:
+		error = lookup(lf, symidx, 1, &addr);
+		if (error != 0)
+			return -1;
+
+		val = addr;
+		insn32p = (uint32_t*)where;
+		before32 = *insn32p;
+		*insn32p = insert_imm(*insn32p, addr, 11,  5, 25);
+		*insn32p = insert_imm(*insn32p, addr,  4,  0,  7);
+
+		if (debug_kld)
+			printf("%p %c %-24s %08x -> %08x\n",
+			       where,
+			       (local? 'l': 'g'),
+			       reloctype_to_str(rtype),
+			       before32, *insn32p);
+		break;
+
+	default:
+		printf("kldload: unexpected relocation type %ld\n", rtype);
+		return (-1);
+	}
+
+	return (0);
 }
 
 int