diff mbox series

[v7,1/3] riscv: Avoid unaligned access when relocating modules

Message ID 20231031-module_relocations-v7-1-6f4719b64bf7@rivosinc.com (mailing list archive)
State Superseded
Headers show
Series riscv: Add remaining module relocations and tests | expand

Checks

Context Check Description
conchuod/vmtest-for-next-PR fail PR summary
conchuod/patch-1-test-1 success .github/scripts/patches/build_rv32_defconfig.sh
conchuod/patch-1-test-2 success .github/scripts/patches/build_rv64_clang_allmodconfig.sh
conchuod/patch-1-test-3 success .github/scripts/patches/build_rv64_gcc_allmodconfig.sh
conchuod/patch-1-test-4 success .github/scripts/patches/build_rv64_nommu_k210_defconfig.sh
conchuod/patch-1-test-5 success .github/scripts/patches/build_rv64_nommu_virt_defconfig.sh
conchuod/patch-1-test-6 success .github/scripts/patches/checkpatch.sh
conchuod/patch-1-test-7 success .github/scripts/patches/dtb_warn_rv64.sh
conchuod/patch-1-test-8 success .github/scripts/patches/header_inline.sh
conchuod/patch-1-test-9 success .github/scripts/patches/kdoc.sh
conchuod/patch-1-test-10 success .github/scripts/patches/module_param.sh
conchuod/patch-1-test-11 success .github/scripts/patches/verify_fixes.sh
conchuod/patch-1-test-12 success .github/scripts/patches/verify_signedoff.sh

Commit Message

Charlie Jenkins Oct. 31, 2023, 7:24 a.m. UTC
From: Emil Renner Berthing <kernel@esmil.dk>

With the C-extension regular 32bit instructions are not
necessarily aligned on 4-byte boundaries. RISC-V instructions
are in fact an ordered list of 16bit little-endian
"parcels", so access the instruction as such.

This should also make the code work in case someone builds
a big-endian RISC-V machine.

Signed-off-by: Emil Renner Berthing <kernel@esmil.dk>
Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
---
 arch/riscv/kernel/module.c | 153 +++++++++++++++++++++++----------------------
 1 file changed, 77 insertions(+), 76 deletions(-)

Comments

Emil Renner Berthing Oct. 31, 2023, 1:11 p.m. UTC | #1
Charlie Jenkins wrote:
> From: Emil Renner Berthing <kernel@esmil.dk>
>
> With the C-extension regular 32bit instructions are not
> necessarily aligned on 4-byte boundaries. RISC-V instructions
> are in fact an ordered list of 16bit little-endian
> "parcels", so access the instruction as such.
>
> This should also make the code work in case someone builds
> a big-endian RISC-V machine.
>
> Signed-off-by: Emil Renner Berthing <kernel@esmil.dk>
> Signed-off-by: Charlie Jenkins <charlie@rivosinc.com>
> ---
>  arch/riscv/kernel/module.c | 153 +++++++++++++++++++++++----------------------
>  1 file changed, 77 insertions(+), 76 deletions(-)
>
> diff --git a/arch/riscv/kernel/module.c b/arch/riscv/kernel/module.c
> index 7c651d55fcbd..a9e94e939cb5 100644
> --- a/arch/riscv/kernel/module.c
> +++ b/arch/riscv/kernel/module.c
> @@ -27,68 +27,86 @@ static bool riscv_insn_valid_32bit_offset(ptrdiff_t val)
>  #endif
>  }
>
> -static int apply_r_riscv_32_rela(struct module *me, u32 *location, Elf_Addr v)
> +static int riscv_insn_rmw(void *location, u32 keep, u32 set)
> +{
> +	u16 *parcel = location;
> +	u32 insn = (u32)le16_to_cpu(parcel[0]) | (u32)le16_to_cpu(parcel[1]) << 16;
> +
> +	insn &= keep;
> +	insn |= set;
> +
> +	parcel[0] = cpu_to_le32(insn);

Why cpu_to_le32(insn)? Unless I've misunderstood something downcasting unsigned
to unsigned values in C (eg. from u32 to u16) is defined to always discard the
most signifcant bits, so cpu_to_le16(insn) should be fine.

> +	parcel[1] = cpu_to_le16(insn >> 16);
> +	return 0;
> +}
> +
> +static int riscv_insn_rvc_rmw(void *location, u16 keep, u16 set)
> +{
> +	u16 *parcel = location;
> +
> +	*parcel = cpu_to_le16((le16_to_cpu(*parcel) & keep) | set);

In this case, maybe consider writing it out like above just so it's easy to see
that the two functions does the same just for differently sized instructions.
The compiler should generate the same code.

> +	return 0;
> +}
> +
> ...
Andreas Schwab Oct. 31, 2023, 4:35 p.m. UTC | #2
On Okt 31 2023, Emil Renner Berthing wrote:

>> +static int riscv_insn_rmw(void *location, u32 keep, u32 set)
>> +{
>> +	u16 *parcel = location;
>> +	u32 insn = (u32)le16_to_cpu(parcel[0]) | (u32)le16_to_cpu(parcel[1]) << 16;
>> +
>> +	insn &= keep;
>> +	insn |= set;
>> +
>> +	parcel[0] = cpu_to_le32(insn);
>
> Why cpu_to_le32(insn)? Unless I've misunderstood something downcasting unsigned
> to unsigned values in C (eg. from u32 to u16) is defined to always discard the
> most signifcant bits, so cpu_to_le16(insn) should be fine.

cpu_to_le32(insn) can't be right here anyway, since it also swaps the
two u16 halves and would be the same as cpu_to_le16(insn >> 16) on big
endian.
Charlie Jenkins Oct. 31, 2023, 8:06 p.m. UTC | #3
On Tue, Oct 31, 2023 at 05:35:57PM +0100, Andreas Schwab wrote:
> On Okt 31 2023, Emil Renner Berthing wrote:
> 
> >> +static int riscv_insn_rmw(void *location, u32 keep, u32 set)
> >> +{
> >> +	u16 *parcel = location;
> >> +	u32 insn = (u32)le16_to_cpu(parcel[0]) | (u32)le16_to_cpu(parcel[1]) << 16;
> >> +
> >> +	insn &= keep;
> >> +	insn |= set;
> >> +
> >> +	parcel[0] = cpu_to_le32(insn);
> >
> > Why cpu_to_le32(insn)? Unless I've misunderstood something downcasting unsigned
> > to unsigned values in C (eg. from u32 to u16) is defined to always discard the
> > most signifcant bits, so cpu_to_le16(insn) should be fine.
> 
> cpu_to_le32(insn) can't be right here anyway, since it also swaps the
> two u16 halves and would be the same as cpu_to_le16(insn >> 16) on big
> endian.

Yes, not sure why I did that... I will fix that up.

- Charlie

> 
> -- 
> Andreas Schwab, schwab@linux-m68k.org
> GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
> "And now for something completely different."
diff mbox series

Patch

diff --git a/arch/riscv/kernel/module.c b/arch/riscv/kernel/module.c
index 7c651d55fcbd..a9e94e939cb5 100644
--- a/arch/riscv/kernel/module.c
+++ b/arch/riscv/kernel/module.c
@@ -27,68 +27,86 @@  static bool riscv_insn_valid_32bit_offset(ptrdiff_t val)
 #endif
 }
 
-static int apply_r_riscv_32_rela(struct module *me, u32 *location, Elf_Addr v)
+static int riscv_insn_rmw(void *location, u32 keep, u32 set)
+{
+	u16 *parcel = location;
+	u32 insn = (u32)le16_to_cpu(parcel[0]) | (u32)le16_to_cpu(parcel[1]) << 16;
+
+	insn &= keep;
+	insn |= set;
+
+	parcel[0] = cpu_to_le32(insn);
+	parcel[1] = cpu_to_le16(insn >> 16);
+	return 0;
+}
+
+static int riscv_insn_rvc_rmw(void *location, u16 keep, u16 set)
+{
+	u16 *parcel = location;
+
+	*parcel = cpu_to_le16((le16_to_cpu(*parcel) & keep) | set);
+	return 0;
+}
+
+static int apply_r_riscv_32_rela(struct module *me, void *location, Elf_Addr v)
 {
 	if (v != (u32)v) {
 		pr_err("%s: value %016llx out of range for 32-bit field\n",
 		       me->name, (long long)v);
 		return -EINVAL;
 	}
-	*location = v;
+	*(u32 *)location = v;
 	return 0;
 }
 
-static int apply_r_riscv_64_rela(struct module *me, u32 *location, Elf_Addr v)
+static int apply_r_riscv_64_rela(struct module *me, void *location, Elf_Addr v)
 {
 	*(u64 *)location = v;
 	return 0;
 }
 
-static int apply_r_riscv_branch_rela(struct module *me, u32 *location,
+static int apply_r_riscv_branch_rela(struct module *me, void *location,
 				     Elf_Addr v)
 {
-	ptrdiff_t offset = (void *)v - (void *)location;
+	ptrdiff_t offset = (void *)v - location;
 	u32 imm12 = (offset & 0x1000) << (31 - 12);
 	u32 imm11 = (offset & 0x800) >> (11 - 7);
 	u32 imm10_5 = (offset & 0x7e0) << (30 - 10);
 	u32 imm4_1 = (offset & 0x1e) << (11 - 4);
 
-	*location = (*location & 0x1fff07f) | imm12 | imm11 | imm10_5 | imm4_1;
-	return 0;
+	return riscv_insn_rmw(location, 0x1fff07f, imm12 | imm11 | imm10_5 | imm4_1);
 }
 
-static int apply_r_riscv_jal_rela(struct module *me, u32 *location,
+static int apply_r_riscv_jal_rela(struct module *me, void *location,
 				  Elf_Addr v)
 {
-	ptrdiff_t offset = (void *)v - (void *)location;
+	ptrdiff_t offset = (void *)v - location;
 	u32 imm20 = (offset & 0x100000) << (31 - 20);
 	u32 imm19_12 = (offset & 0xff000);
 	u32 imm11 = (offset & 0x800) << (20 - 11);
 	u32 imm10_1 = (offset & 0x7fe) << (30 - 10);
 
-	*location = (*location & 0xfff) | imm20 | imm19_12 | imm11 | imm10_1;
-	return 0;
+	return riscv_insn_rmw(location, 0xfff, imm20 | imm19_12 | imm11 | imm10_1);
 }
 
-static int apply_r_riscv_rvc_branch_rela(struct module *me, u32 *location,
+static int apply_r_riscv_rvc_branch_rela(struct module *me, void *location,
 					 Elf_Addr v)
 {
-	ptrdiff_t offset = (void *)v - (void *)location;
+	ptrdiff_t offset = (void *)v - location;
 	u16 imm8 = (offset & 0x100) << (12 - 8);
 	u16 imm7_6 = (offset & 0xc0) >> (6 - 5);
 	u16 imm5 = (offset & 0x20) >> (5 - 2);
 	u16 imm4_3 = (offset & 0x18) << (12 - 5);
 	u16 imm2_1 = (offset & 0x6) << (12 - 10);
 
-	*(u16 *)location = (*(u16 *)location & 0xe383) |
-		    imm8 | imm7_6 | imm5 | imm4_3 | imm2_1;
-	return 0;
+	return riscv_insn_rvc_rmw(location, 0xe383,
+			imm8 | imm7_6 | imm5 | imm4_3 | imm2_1);
 }
 
-static int apply_r_riscv_rvc_jump_rela(struct module *me, u32 *location,
+static int apply_r_riscv_rvc_jump_rela(struct module *me, void *location,
 				       Elf_Addr v)
 {
-	ptrdiff_t offset = (void *)v - (void *)location;
+	ptrdiff_t offset = (void *)v - location;
 	u16 imm11 = (offset & 0x800) << (12 - 11);
 	u16 imm10 = (offset & 0x400) >> (10 - 8);
 	u16 imm9_8 = (offset & 0x300) << (12 - 11);
@@ -98,16 +116,14 @@  static int apply_r_riscv_rvc_jump_rela(struct module *me, u32 *location,
 	u16 imm4 = (offset & 0x10) << (12 - 5);
 	u16 imm3_1 = (offset & 0xe) << (12 - 10);
 
-	*(u16 *)location = (*(u16 *)location & 0xe003) |
-		    imm11 | imm10 | imm9_8 | imm7 | imm6 | imm5 | imm4 | imm3_1;
-	return 0;
+	return riscv_insn_rvc_rmw(location, 0xe003,
+			imm11 | imm10 | imm9_8 | imm7 | imm6 | imm5 | imm4 | imm3_1);
 }
 
-static int apply_r_riscv_pcrel_hi20_rela(struct module *me, u32 *location,
+static int apply_r_riscv_pcrel_hi20_rela(struct module *me, void *location,
 					 Elf_Addr v)
 {
-	ptrdiff_t offset = (void *)v - (void *)location;
-	s32 hi20;
+	ptrdiff_t offset = (void *)v - location;
 
 	if (!riscv_insn_valid_32bit_offset(offset)) {
 		pr_err(
@@ -116,23 +132,20 @@  static int apply_r_riscv_pcrel_hi20_rela(struct module *me, u32 *location,
 		return -EINVAL;
 	}
 
-	hi20 = (offset + 0x800) & 0xfffff000;
-	*location = (*location & 0xfff) | hi20;
-	return 0;
+	return riscv_insn_rmw(location, 0xfff, (offset + 0x800) & 0xfffff000);
 }
 
-static int apply_r_riscv_pcrel_lo12_i_rela(struct module *me, u32 *location,
+static int apply_r_riscv_pcrel_lo12_i_rela(struct module *me, void *location,
 					   Elf_Addr v)
 {
 	/*
 	 * v is the lo12 value to fill. It is calculated before calling this
 	 * handler.
 	 */
-	*location = (*location & 0xfffff) | ((v & 0xfff) << 20);
-	return 0;
+	return riscv_insn_rmw(location, 0xfffff, (v & 0xfff) << 20);
 }
 
-static int apply_r_riscv_pcrel_lo12_s_rela(struct module *me, u32 *location,
+static int apply_r_riscv_pcrel_lo12_s_rela(struct module *me, void *location,
 					   Elf_Addr v)
 {
 	/*
@@ -142,15 +155,12 @@  static int apply_r_riscv_pcrel_lo12_s_rela(struct module *me, u32 *location,
 	u32 imm11_5 = (v & 0xfe0) << (31 - 11);
 	u32 imm4_0 = (v & 0x1f) << (11 - 4);
 
-	*location = (*location & 0x1fff07f) | imm11_5 | imm4_0;
-	return 0;
+	return riscv_insn_rmw(location, 0x1fff07f, imm11_5 | imm4_0);
 }
 
-static int apply_r_riscv_hi20_rela(struct module *me, u32 *location,
+static int apply_r_riscv_hi20_rela(struct module *me, void *location,
 				   Elf_Addr v)
 {
-	s32 hi20;
-
 	if (IS_ENABLED(CONFIG_CMODEL_MEDLOW)) {
 		pr_err(
 		  "%s: target %016llx can not be addressed by the 32-bit offset from PC = %p\n",
@@ -158,22 +168,20 @@  static int apply_r_riscv_hi20_rela(struct module *me, u32 *location,
 		return -EINVAL;
 	}
 
-	hi20 = ((s32)v + 0x800) & 0xfffff000;
-	*location = (*location & 0xfff) | hi20;
-	return 0;
+	return riscv_insn_rmw(location, 0xfff, ((s32)v + 0x800) & 0xfffff000);
 }
 
-static int apply_r_riscv_lo12_i_rela(struct module *me, u32 *location,
+static int apply_r_riscv_lo12_i_rela(struct module *me, void *location,
 				     Elf_Addr v)
 {
 	/* Skip medlow checking because of filtering by HI20 already */
 	s32 hi20 = ((s32)v + 0x800) & 0xfffff000;
 	s32 lo12 = ((s32)v - hi20);
-	*location = (*location & 0xfffff) | ((lo12 & 0xfff) << 20);
-	return 0;
+
+	return riscv_insn_rmw(location, 0xfffff, (lo12 & 0xfff) << 20);
 }
 
-static int apply_r_riscv_lo12_s_rela(struct module *me, u32 *location,
+static int apply_r_riscv_lo12_s_rela(struct module *me, void *location,
 				     Elf_Addr v)
 {
 	/* Skip medlow checking because of filtering by HI20 already */
@@ -181,20 +189,18 @@  static int apply_r_riscv_lo12_s_rela(struct module *me, u32 *location,
 	s32 lo12 = ((s32)v - hi20);
 	u32 imm11_5 = (lo12 & 0xfe0) << (31 - 11);
 	u32 imm4_0 = (lo12 & 0x1f) << (11 - 4);
-	*location = (*location & 0x1fff07f) | imm11_5 | imm4_0;
-	return 0;
+
+	return riscv_insn_rmw(location, 0x1fff07f, imm11_5 | imm4_0);
 }
 
-static int apply_r_riscv_got_hi20_rela(struct module *me, u32 *location,
+static int apply_r_riscv_got_hi20_rela(struct module *me, void *location,
 				       Elf_Addr v)
 {
-	ptrdiff_t offset = (void *)v - (void *)location;
-	s32 hi20;
+	ptrdiff_t offset = (void *)v - location;
 
 	/* Always emit the got entry */
 	if (IS_ENABLED(CONFIG_MODULE_SECTIONS)) {
-		offset = module_emit_got_entry(me, v);
-		offset = (void *)offset - (void *)location;
+		offset = (void *)module_emit_got_entry(me, v) - location;
 	} else {
 		pr_err(
 		  "%s: can not generate the GOT entry for symbol = %016llx from PC = %p\n",
@@ -202,22 +208,19 @@  static int apply_r_riscv_got_hi20_rela(struct module *me, u32 *location,
 		return -EINVAL;
 	}
 
-	hi20 = (offset + 0x800) & 0xfffff000;
-	*location = (*location & 0xfff) | hi20;
-	return 0;
+	return riscv_insn_rmw(location, 0xfff, (offset + 0x800) & 0xfffff000);
 }
 
-static int apply_r_riscv_call_plt_rela(struct module *me, u32 *location,
+static int apply_r_riscv_call_plt_rela(struct module *me, void *location,
 				       Elf_Addr v)
 {
-	ptrdiff_t offset = (void *)v - (void *)location;
+	ptrdiff_t offset = (void *)v - location;
 	u32 hi20, lo12;
 
 	if (!riscv_insn_valid_32bit_offset(offset)) {
 		/* Only emit the plt entry if offset over 32-bit range */
 		if (IS_ENABLED(CONFIG_MODULE_SECTIONS)) {
-			offset = module_emit_plt_entry(me, v);
-			offset = (void *)offset - (void *)location;
+			offset = (void *)module_emit_plt_entry(me, v) - location;
 		} else {
 			pr_err(
 			  "%s: target %016llx can not be addressed by the 32-bit offset from PC = %p\n",
@@ -228,15 +231,14 @@  static int apply_r_riscv_call_plt_rela(struct module *me, u32 *location,
 
 	hi20 = (offset + 0x800) & 0xfffff000;
 	lo12 = (offset - hi20) & 0xfff;
-	*location = (*location & 0xfff) | hi20;
-	*(location + 1) = (*(location + 1) & 0xfffff) | (lo12 << 20);
-	return 0;
+	riscv_insn_rmw(location, 0xfff, hi20);
+	return riscv_insn_rmw(location + 4, 0xfffff, lo12 << 20);
 }
 
-static int apply_r_riscv_call_rela(struct module *me, u32 *location,
+static int apply_r_riscv_call_rela(struct module *me, void *location,
 				   Elf_Addr v)
 {
-	ptrdiff_t offset = (void *)v - (void *)location;
+	ptrdiff_t offset = (void *)v - location;
 	u32 hi20, lo12;
 
 	if (!riscv_insn_valid_32bit_offset(offset)) {
@@ -248,18 +250,17 @@  static int apply_r_riscv_call_rela(struct module *me, u32 *location,
 
 	hi20 = (offset + 0x800) & 0xfffff000;
 	lo12 = (offset - hi20) & 0xfff;
-	*location = (*location & 0xfff) | hi20;
-	*(location + 1) = (*(location + 1) & 0xfffff) | (lo12 << 20);
-	return 0;
+	riscv_insn_rmw(location, 0xfff, hi20);
+	return riscv_insn_rmw(location + 4, 0xfffff, lo12 << 20);
 }
 
-static int apply_r_riscv_relax_rela(struct module *me, u32 *location,
+static int apply_r_riscv_relax_rela(struct module *me, void *location,
 				    Elf_Addr v)
 {
 	return 0;
 }
 
-static int apply_r_riscv_align_rela(struct module *me, u32 *location,
+static int apply_r_riscv_align_rela(struct module *me, void *location,
 				    Elf_Addr v)
 {
 	pr_err(
@@ -268,49 +269,49 @@  static int apply_r_riscv_align_rela(struct module *me, u32 *location,
 	return -EINVAL;
 }
 
-static int apply_r_riscv_add16_rela(struct module *me, u32 *location,
+static int apply_r_riscv_add16_rela(struct module *me, void *location,
 				    Elf_Addr v)
 {
 	*(u16 *)location += (u16)v;
 	return 0;
 }
 
-static int apply_r_riscv_add32_rela(struct module *me, u32 *location,
+static int apply_r_riscv_add32_rela(struct module *me, void *location,
 				    Elf_Addr v)
 {
 	*(u32 *)location += (u32)v;
 	return 0;
 }
 
-static int apply_r_riscv_add64_rela(struct module *me, u32 *location,
+static int apply_r_riscv_add64_rela(struct module *me, void *location,
 				    Elf_Addr v)
 {
 	*(u64 *)location += (u64)v;
 	return 0;
 }
 
-static int apply_r_riscv_sub16_rela(struct module *me, u32 *location,
+static int apply_r_riscv_sub16_rela(struct module *me, void *location,
 				    Elf_Addr v)
 {
 	*(u16 *)location -= (u16)v;
 	return 0;
 }
 
-static int apply_r_riscv_sub32_rela(struct module *me, u32 *location,
+static int apply_r_riscv_sub32_rela(struct module *me, void *location,
 				    Elf_Addr v)
 {
 	*(u32 *)location -= (u32)v;
 	return 0;
 }
 
-static int apply_r_riscv_sub64_rela(struct module *me, u32 *location,
+static int apply_r_riscv_sub64_rela(struct module *me, void *location,
 				    Elf_Addr v)
 {
 	*(u64 *)location -= (u64)v;
 	return 0;
 }
 
-static int (*reloc_handlers_rela[]) (struct module *me, u32 *location,
+static int (*reloc_handlers_rela[]) (struct module *me, void *location,
 				Elf_Addr v) = {
 	[R_RISCV_32]			= apply_r_riscv_32_rela,
 	[R_RISCV_64]			= apply_r_riscv_64_rela,
@@ -342,9 +343,9 @@  int apply_relocate_add(Elf_Shdr *sechdrs, const char *strtab,
 		       struct module *me)
 {
 	Elf_Rela *rel = (void *) sechdrs[relsec].sh_addr;
-	int (*handler)(struct module *me, u32 *location, Elf_Addr v);
+	int (*handler)(struct module *me, void *location, Elf_Addr v);
 	Elf_Sym *sym;
-	u32 *location;
+	void *location;
 	unsigned int i, type;
 	Elf_Addr v;
 	int res;