Skip to contentSkip to conversation

arm64: Optimise and update memcpy, user copy and string routines

PATCH Superseded Oliver Swede 3 participants 19 messages Jun 30, 2020 – Sep 11, 2020 arm64

v4 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v4 cover arm64: Optimise and update memcpy, user copy and string routines

Hi,

This contains an update to the cortex-strings patchset: the correctness of the fixup routines are improved, with the aim being to return the exact number of remaining bytes for all copy sizes. To ensure they are exact - which the current fixups are not for some copy sizes and are off by a few byes - is an extension to the original intention of fixing an issue reported by an LTP run last year, where the fixup routine in v2 of this patchset (which was importing the cortex-strings memcpy implementation) would over-report the number of bytes that successfully copied.
Version 3 addressed this but I later found some issues with the fixup correctness after further testing, and have partially re-written them here, and addressed some other behaviours of the copy algorithm.

Comments welcome,

Thanks
Oliver

v1: https://lore.kernel.org/linux-arm-kernel/[email protected]/
v2: https://lore.kernel.org/linux-arm-kernel/[email protected]/
v3: https://lore.kernel.org/linux-arm-kernel/[email protected]/

Changes since v3:
* Improves the accuracy of the fixups in response to issues that arose during futher testing
* Accounts for faults on store instructions on systems with UAO enabled
* Expands on comments detailing the implementation

Changes since v2:
* Adds Robin's separate patch that fixes a compilation issue with KProbes fixup [1]
* Imports the most recent memcpy implementation by updating Sam's patch (and moves this patch to occur after the cortex-strings importing so that it's closer to the patches containing its corresponding fixups)
* Uses the stack to preserve the initial parameters
* Replaces the usercopy fixup routine in v2 with multiple longer fixups that each make use of the fault address to return the exact number of bytes that haven't yet copied.

[1] https://lore.kernel.org/linux-arm-kernel/e70f7b9de7e601b9e4a6fedad8eaf64d304b1637.1571326276.git.robin.murphy@arm.com/

Oliver Swede (5):
arm64: Store the arguments to copy_*_user on the stack
arm64: Use additional memcpy macros and fixups
arm64: Add fixup routines for usercopy load exceptions
arm64: Add fixup routines for usercopy store exceptions
arm64: Improve accuracy of fixup for UAO cases

Robin Murphy (2):
arm64: kprobes: Drop open-coded exception fixup
arm64: Tidy up _asm_extable_faultaddr usage

Sam Tebbs (7):
arm64: Allow passing fault address to fixup handlers
arm64: Import latest version of Cortex Strings' memcmp
arm64: Import latest version of Cortex Strings' memmove
arm64: Import latest version of Cortex Strings' strcmp
arm64: Import latest version of Cortex Strings' strlen
arm64: Import latest version of Cortex Strings' strncmp
arm64: Import latest optimization of memcpy

 arch/arm64/include/asm/alternative.h |  36 ---
 arch/arm64/include/asm/assembler.h   |  13 +
 arch/arm64/include/asm/extable.h     |  10 +-
 arch/arm64/kernel/probes/kprobes.c   |   7 -
 arch/arm64/lib/copy_from_user.S      | 272 +++++++++++++++--
 arch/arm64/lib/copy_in_user.S        | 287 ++++++++++++++++--
 arch/arm64/lib/copy_template.S       | 377 +++++++++++++----------
 arch/arm64/lib/copy_template_user.S  |  50 ++++
 arch/arm64/lib/copy_to_user.S        | 273 +++++++++++++++--
 arch/arm64/lib/copy_user_fixup.S     | 433 +++++++++++++++++++++++++++
 arch/arm64/lib/memcmp.S              | 333 ++++++++------------
 arch/arm64/lib/memcpy.S              | 127 ++++++--
 arch/arm64/lib/memmove.S             | 232 +++++---------
 arch/arm64/lib/strcmp.S              | 272 +++++++----------
 arch/arm64/lib/strlen.S              | 247 ++++++++++-----
 arch/arm64/lib/strncmp.S             | 363 ++++++++++------------
 arch/arm64/mm/extable.c              |  13 +-
 arch/arm64/mm/fault.c                |   2 +-
 18 files changed, 2228 insertions(+), 1119 deletions(-)
 create mode 100644 arch/arm64/lib/copy_template_user.S
 create mode 100644 arch/arm64/lib/copy_user_fixup.S
Signature · 1 line
2.17.1
#1 lore raw

PATCH v4 1/14 arm64: Allow passing fault address to fixup handlers

From: Sam Tebbs <[email protected]>

Extend fixup_exception() to optionally place the faulting address in a register when returning to a fixup handler. Since A64 instructions must be 4-byte-aligned, we can mimic the IA-64 implementation and encode a flag in the lower bits of the offset field to indicate handlers which expect an address. This will allow us to use more efficient offset addressing modes in usercopy routines, rather than updating the base register on every access just for the sake of inferring where a fault occurred in order to compute the return value upon failure.

The choice of x15 is somewhat arbitrary, but with the consideration that as the highest-numbered temporary register with no possible 'special' role in the ABI, it is most likely not used by hand-written assembly code, and thus a minimally-invasive option for imported routines.

[ rm: split into separate patch, use UL(), expand commit message ]
---
 arch/arm64/include/asm/assembler.h |  9 +++++++++
 arch/arm64/include/asm/extable.h   | 10 +++++++++-
 arch/arm64/mm/extable.c            | 13 +++++++++----
 arch/arm64/mm/fault.c              |  2 +-
 4 files changed, 28 insertions(+), 6 deletions(-)
arch/arm64/include/asm/assembler.h +9 -0
index 54d181177656..438382a277c8 100644
@@ -18,6 +18,7 @@
 #include <asm/cpufeature.h>
 #include <asm/cputype.h>
 #include <asm/debug-monitors.h>
+#include <asm/extable.h>
 #include <asm/page.h>
 #include <asm/pgtable-hwdef.h>
 #include <asm/ptrace.h>
@@ -129,6 +130,14 @@ alternative_endif
 	.popsection
 	.endm
 
+/*
+ * Emit an entry into the exception table.
+ * The fixup handler will receive the faulting address in x15
+ */
+	.macro		_asm_extable_faultaddr, from, to
+	_asm_extable	\from, \to + FIXUP_WITH_ADDR
+	.endm
+
 #define USER(l, x...)				\
 9999:	x;					\
 	_asm_extable	9999b, l
arch/arm64/include/asm/extable.h +9 -1
index 56a4f68b262e..4c4955f2bb44 100644
@@ -2,6 +2,12 @@
 #ifndef __ASM_EXTABLE_H
 #define __ASM_EXTABLE_H
 
+#include <linux/const.h>
+
+#define FIXUP_WITH_ADDR UL(1)
+
+#ifndef __ASSEMBLY__
+
 /*
  * The exception table consists of pairs of relative offsets: the first
  * is the relative offset to an instruction that is allowed to fault,
@@ -22,5 +28,7 @@ struct exception_table_entry
 
 #define ARCH_HAS_RELATIVE_EXTABLE
 
-extern int fixup_exception(struct pt_regs *regs);
+extern int fixup_exception(struct pt_regs *regs, unsigned long addr);
+
+#endif
 #endif
arch/arm64/mm/extable.c +9 -4
index 81e694af5f8c..e6578c2814b5 100644
@@ -6,13 +6,18 @@
 #include <linux/extable.h>
 #include <linux/uaccess.h>
 
-int fixup_exception(struct pt_regs *regs)
+int fixup_exception(struct pt_regs *regs, unsigned long addr)
 {
 	const struct exception_table_entry *fixup;
 
 	fixup = search_exception_tables(instruction_pointer(regs));
-	if (fixup)
-		regs->pc = (unsigned long)&fixup->fixup + fixup->fixup;
-
+	if (fixup) {
+		unsigned long offset = fixup->fixup;
+		if (offset & FIXUP_WITH_ADDR) {
+			regs->regs[15] = addr;
+			offset &= ~FIXUP_WITH_ADDR;
+		}
+		regs->pc = (unsigned long)&fixup->fixup + offset;
+	}
 	return fixup != NULL;
 }
arch/arm64/mm/fault.c +1 -1
index 8afb238ff335..f80e299dc91b 100644
@@ -303,7 +303,7 @@ static void __do_kernel_fault(unsigned long addr, unsigned int esr,
 	 * Are we prepared to handle this kernel fault?
 	 * We are almost certainly not prepared to handle instruction faults.
 	 */
-	if (!is_el1_instruction_abort(esr) && fixup_exception(regs))
+	if (!is_el1_instruction_abort(esr) && fixup_exception(regs, addr))
 		return;
 
 	if (WARN_RATELIMIT(is_spurious_el1_translation_fault(addr, esr, regs),
Signature · 1 line
2.17.1
#2 lore raw

PATCH v4 2/14 arm64: kprobes: Drop open-coded exception fixup

In reply to Oliver Swede

From: Robin Murphy <[email protected]>

The short-circuit call to fixup_exception() from kprobe_fault_handler() poses a problem now that the former wants to consume the fault address too, since the common kprobes API offers us no way to pass it through. Fortunately, however, it works out to be unnecessary:

- uaccess instructions themselves are not probeable, so at most we should only ever expect to take a fixable fault from the pre or post handlers.
- the pre and post handler run with preemption disabled, thus for any fault they may cause, an unhandled return from kprobe_page_fault() will proceed directly to __do_kernel_fault() thanks to the faulthandler_disabled() check.
- __do_kernel_fault() will immediately call fixup_exception() unless we're in an EL1 instruction abort, and if we've somehow taken one of those on what we think is the middle of a uaccess routine, then the world is already very on fire.

Thus we can reasonably drop the call from kprobe_fault_handler() and leave uaccess fixups to the regular flow.

---
 arch/arm64/kernel/probes/kprobes.c | 7 -------
 1 file changed, 7 deletions(-)
arch/arm64/kernel/probes/kprobes.c +0 -7
index d1c95dcf1d78..771635360110 100644
@@ -334,13 +334,6 @@ int __kprobes kprobe_fault_handler(struct pt_regs *regs, unsigned int fsr)
 		 */
 		if (cur->fault_handler && cur->fault_handler(cur, regs, fsr))
 			return 1;
-
-		/*
-		 * In case the user-specified fault handler returned
-		 * zero, try to fix up.
-		 */
-		if (fixup_exception(regs))
-			return 1;
 	}
 	return 0;
 }
Signature · 1 line
2.17.1
#3 lore raw

PATCH v4 3/14 arm64: Import latest version of Cortex Strings' memcmp

In reply to Oliver Swede

From: Sam Tebbs <[email protected]>

Import the latest version of Cortex Strings' memcmp function.

The upstream source is src/aarch64/memcmp.S as of commit f77e4c932b4f in https://git.linaro.org/toolchain/cortex-strings.git.

[ rm: update attribution, expand commit message ]
---
 arch/arm64/lib/memcmp.S | 333 ++++++++++++++--------------------------
 1 file changed, 117 insertions(+), 216 deletions(-)
arch/arm64/lib/memcmp.S +117 -216
index c0671e793ea9..580dd0b12ccb 100644
@@ -1,13 +1,12 @@
 /* SPDX-License-Identifier: GPL-2.0-only */
 /*
- * Copyright (C) 2013 ARM Ltd.
- * Copyright (C) 2013 Linaro.
+ * Copyright (c) 2013, 2018 Linaro Limited. All rights reserved.
+ * Copyright (c) 2017 ARM Ltd. All rights reserved.
  *
- * This code is based on glibc cortex strings work originally authored by Linaro
- * be found @
+ * This code is based on glibc Cortex Strings work originally authored by
+ * Linaro, found at:
  *
- * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/
- * files/head:/src/aarch64/
+ * https://git.linaro.org/toolchain/cortex-strings.git
  */
 
 #include <linux/linkage.h>
@@ -25,223 +24,125 @@
 *  x0 - a compare result, maybe less than, equal to, or greater than ZERO
 */
 
+#define L(l) .L ## l
+
 /* Parameters and result.  */
-src1		.req	x0
-src2		.req	x1
-limit		.req	x2
-result		.req	x0
+#define src1		x0
+#define src2		x1
+#define limit		x2
+#define result		w0
 
 /* Internal variables.  */
-data1		.req	x3
-data1w		.req	w3
-data2		.req	x4
-data2w		.req	w4
-has_nul		.req	x5
-diff		.req	x6
-endloop		.req	x7
-tmp1		.req	x8
-tmp2		.req	x9
-tmp3		.req	x10
-pos		.req	x11
-limit_wd	.req	x12
-mask		.req	x13
+#define data1		x3
+#define data1w		w3
+#define data1h		x4
+#define data2		x5
+#define data2w		w5
+#define data2h		x6
+#define tmp1		x7
+#define tmp2		x8
 
 SYM_FUNC_START_WEAK_PI(memcmp)
-	cbz	limit, .Lret0
-	eor	tmp1, src1, src2
-	tst	tmp1, #7
-	b.ne	.Lmisaligned8
-	ands	tmp1, src1, #7
-	b.ne	.Lmutual_align
-	sub	limit_wd, limit, #1 /* limit != 0, so no underflow.  */
-	lsr	limit_wd, limit_wd, #3 /* Convert to Dwords.  */
-	/*
-	* The input source addresses are at alignment boundary.
-	* Directly compare eight bytes each time.
-	*/
-.Lloop_aligned:
-	ldr	data1, [src1], #8
-	ldr	data2, [src2], #8
-.Lstart_realigned:
-	subs	limit_wd, limit_wd, #1
-	eor	diff, data1, data2	/* Non-zero if differences found.  */
-	csinv	endloop, diff, xzr, cs	/* Last Dword or differences.  */
-	cbz	endloop, .Lloop_aligned
-
-	/* Not reached the limit, must have found a diff.  */
-	tbz	limit_wd, #63, .Lnot_limit
-
-	/* Limit % 8 == 0 => the diff is in the last 8 bytes. */
-	ands	limit, limit, #7
-	b.eq	.Lnot_limit
-	/*
-	* The remained bytes less than 8. It is needed to extract valid data
-	* from last eight bytes of the intended memory range.
-	*/
-	lsl	limit, limit, #3	/* bytes-> bits.  */
-	mov	mask, #~0
-CPU_BE( lsr	mask, mask, limit )
-CPU_LE( lsl	mask, mask, limit )
-	bic	data1, data1, mask
-	bic	data2, data2, mask
-
-	orr	diff, diff, mask
-	b	.Lnot_limit
-
-.Lmutual_align:
-	/*
-	* Sources are mutually aligned, but are not currently at an
-	* alignment boundary. Round down the addresses and then mask off
-	* the bytes that precede the start point.
-	*/
-	bic	src1, src1, #7
-	bic	src2, src2, #7
-	ldr	data1, [src1], #8
-	ldr	data2, [src2], #8
-	/*
-	* We can not add limit with alignment offset(tmp1) here. Since the
-	* addition probably make the limit overflown.
-	*/
-	sub	limit_wd, limit, #1/*limit != 0, so no underflow.*/
-	and	tmp3, limit_wd, #7
-	lsr	limit_wd, limit_wd, #3
-	add	tmp3, tmp3, tmp1
-	add	limit_wd, limit_wd, tmp3, lsr #3
-	add	limit, limit, tmp1/* Adjust the limit for the extra.  */
-
-	lsl	tmp1, tmp1, #3/* Bytes beyond alignment -> bits.*/
-	neg	tmp1, tmp1/* Bits to alignment -64.  */
-	mov	tmp2, #~0
-	/*mask off the non-intended bytes before the start address.*/
-CPU_BE( lsl	tmp2, tmp2, tmp1 )/*Big-endian.Early bytes are at MSB*/
-	/* Little-endian.  Early bytes are at LSB.  */
-CPU_LE( lsr	tmp2, tmp2, tmp1 )
-
-	orr	data1, data1, tmp2
-	orr	data2, data2, tmp2
-	b	.Lstart_realigned
-
-	/*src1 and src2 have different alignment offset.*/
-.Lmisaligned8:
-	cmp	limit, #8
-	b.lo	.Ltiny8proc /*limit < 8: compare byte by byte*/
-
-	and	tmp1, src1, #7
-	neg	tmp1, tmp1
-	add	tmp1, tmp1, #8/*valid length in the first 8 bytes of src1*/
-	and	tmp2, src2, #7
-	neg	tmp2, tmp2
-	add	tmp2, tmp2, #8/*valid length in the first 8 bytes of src2*/
-	subs	tmp3, tmp1, tmp2
-	csel	pos, tmp1, tmp2, hi /*Choose the maximum.*/
-
-	sub	limit, limit, pos
-	/*compare the proceeding bytes in the first 8 byte segment.*/
-.Ltinycmp:
-	ldrb	data1w, [src1], #1
-	ldrb	data2w, [src2], #1
-	subs	pos, pos, #1
-	ccmp	data1w, data2w, #0, ne  /* NZCV = 0b0000.  */
-	b.eq	.Ltinycmp
-	cbnz	pos, 1f /*diff occurred before the last byte.*/
-	cmp	data1w, data2w
-	b.eq	.Lstart_align
-1:
-	sub	result, data1, data2
-	ret
-
-.Lstart_align:
-	lsr	limit_wd, limit, #3
-	cbz	limit_wd, .Lremain8
-
-	ands	xzr, src1, #7
-	b.eq	.Lrecal_offset
-	/*process more leading bytes to make src1 aligned...*/
-	add	src1, src1, tmp3 /*backwards src1 to alignment boundary*/
-	add	src2, src2, tmp3
-	sub	limit, limit, tmp3
-	lsr	limit_wd, limit, #3
-	cbz	limit_wd, .Lremain8
-	/*load 8 bytes from aligned SRC1..*/
-	ldr	data1, [src1], #8
-	ldr	data2, [src2], #8
-
-	subs	limit_wd, limit_wd, #1
-	eor	diff, data1, data2  /*Non-zero if differences found.*/
-	csinv	endloop, diff, xzr, ne
-	cbnz	endloop, .Lunequal_proc
-	/*How far is the current SRC2 from the alignment boundary...*/
-	and	tmp3, tmp3, #7
-
-.Lrecal_offset:/*src1 is aligned now..*/
-	neg	pos, tmp3
-.Lloopcmp_proc:
-	/*
-	* Divide the eight bytes into two parts. First,backwards the src2
-	* to an alignment boundary,load eight bytes and compare from
-	* the SRC2 alignment boundary. If all 8 bytes are equal,then start
-	* the second part's comparison. Otherwise finish the comparison.
-	* This special handle can garantee all the accesses are in the
-	* thread/task space in avoid to overrange access.
-	*/
-	ldr	data1, [src1,pos]
-	ldr	data2, [src2,pos]
-	eor	diff, data1, data2  /* Non-zero if differences found.  */
-	cbnz	diff, .Lnot_limit
-
-	/*The second part process*/
-	ldr	data1, [src1], #8
-	ldr	data2, [src2], #8
-	eor	diff, data1, data2  /* Non-zero if differences found.  */
-	subs	limit_wd, limit_wd, #1
-	csinv	endloop, diff, xzr, ne/*if limit_wd is 0,will finish the cmp*/
-	cbz	endloop, .Lloopcmp_proc
-.Lunequal_proc:
-	cbz	diff, .Lremain8
-
-/* There is difference occurred in the latest comparison. */
-.Lnot_limit:
-/*
-* For little endian,reverse the low significant equal bits into MSB,then
-* following CLZ can find how many equal bits exist.
-*/
-CPU_LE( rev	diff, diff )
-CPU_LE( rev	data1, data1 )
-CPU_LE( rev	data2, data2 )
-
-	/*
-	* The MS-non-zero bit of DIFF marks either the first bit
-	* that is different, or the end of the significant data.
-	* Shifting left now will bring the critical information into the
-	* top bits.
-	*/
-	clz	pos, diff
-	lsl	data1, data1, pos
-	lsl	data2, data2, pos
-	/*
-	* We need to zero-extend (char is unsigned) the value and then
-	* perform a signed subtraction.
-	*/
-	lsr	data1, data1, #56
-	sub	result, data1, data2, lsr #56
+	subs	limit, limit, 8
+	b.lo	L(less8)
+
+	ldr	data1, [src1], 8
+	ldr	data2, [src2], 8
+	cmp	data1, data2
+	b.ne	L(return)
+
+	subs	limit, limit, 8
+	b.gt	L(more16)
+
+	ldr	data1, [src1, limit]
+	ldr	data2, [src2, limit]
+	b	L(return)
+
+L(more16):
+	ldr	data1, [src1], 8
+	ldr	data2, [src2], 8
+	cmp	data1, data2
+	bne	L(return)
+
+	/* Jump directly to comparing the last 16 bytes for 32 byte (or less)
+	   strings.  */
+	subs	limit, limit, 16
+	b.ls	L(last_bytes)
+
+	/* We overlap loads between 0-32 bytes at either side of SRC1 when we
+	   try to align, so limit it only to strings larger than 128 bytes.  */
+	cmp	limit, 96
+	b.ls	L(loop16)
+
+	/* Align src1 and adjust src2 with bytes not yet done.  */
+	and	tmp1, src1, 15
+	add	limit, limit, tmp1
+	sub	src1, src1, tmp1
+	sub	src2, src2, tmp1
+
+	/* Loop performing 16 bytes per iteration using aligned src1.
+	   Limit is pre-decremented by 16 and must be larger than zero.
+	   Exit if <= 16 bytes left to do or if the data is not equal.  */
+	.p2align 4
+L(loop16):
+	ldp	data1, data1h, [src1], 16
+	ldp	data2, data2h, [src2], 16
+	subs	limit, limit, 16
+	ccmp	data1, data2, 0, hi
+	ccmp	data1h, data2h, 0, eq
+	b.eq	L(loop16)
+
+	cmp	data1, data2
+	bne	L(return)
+	mov	data1, data1h
+	mov	data2, data2h
+	cmp	data1, data2
+	bne	L(return)
+
+	/* Compare last 1-16 bytes using unaligned access.  */
+L(last_bytes):
+	add	src1, src1, limit
+	add	src2, src2, limit
+	ldp	data1, data1h, [src1]
+	ldp	data2, data2h, [src2]
+	cmp	data1, data2
+	bne	L(return)
+	mov	data1, data1h
+	mov	data2, data2h
+	cmp	data1, data2
+
+	/* Compare data bytes and set return value to 0, -1 or 1.  */
+L(return):
+#ifndef __AARCH64EB__
+	rev	data1, data1
+	rev	data2, data2
+#endif
+	cmp	data1, data2
+L(ret_eq):
+	cset	result, ne
+	cneg	result, result, lo
 	ret
 
-.Lremain8:
-	/* Limit % 8 == 0 =>. all data are equal.*/
-	ands	limit, limit, #7
-	b.eq	.Lret0
-
-.Ltiny8proc:
-	ldrb	data1w, [src1], #1
-	ldrb	data2w, [src2], #1
-	subs	limit, limit, #1
-
-	ccmp	data1w, data2w, #0, ne  /* NZCV = 0b0000. */
-	b.eq	.Ltiny8proc
-	sub	result, data1, data2
-	ret
-.Lret0:
-	mov	result, #0
+	.p2align 4
+	/* Compare up to 8 bytes.  Limit is [-8..-1].  */
+L(less8):
+	adds	limit, limit, 4
+	b.lo	L(less4)
+	ldr	data1w, [src1], 4
+	ldr	data2w, [src2], 4
+	cmp	data1w, data2w
+	b.ne	L(return)
+	sub	limit, limit, 4
+L(less4):
+	adds	limit, limit, 4
+	beq	L(ret_eq)
+L(byte_loop):
+	ldrb	data1w, [src1], 1
+	ldrb	data2w, [src2], 1
+	subs	limit, limit, 1
+	ccmp	data1w, data2w, 0, ne	/* NZCV = 0b0000.  */
+	b.eq	L(byte_loop)
+	sub	result, data1w, data2w
 	ret
 SYM_FUNC_END_PI(memcmp)
 EXPORT_SYMBOL_NOKASAN(memcmp)
Signature · 1 line
2.17.1
#4 lore raw

PATCH v4 4/14 arm64: Import latest version of Cortex Strings' memmove

In reply to Oliver Swede

From: Sam Tebbs <[email protected]>

Import the latest version of Cortex Strings' memmove function.

The upstream source is src/aarch64/memmove.S as of commit 99b01ddb8e41 in https://git.linaro.org/toolchain/cortex-strings.git.

[ rm: update attribution, expand commit message ]
---
 arch/arm64/lib/memmove.S | 232 +++++++++++++--------------------------
 1 file changed, 78 insertions(+), 154 deletions(-)
arch/arm64/lib/memmove.S +78 -154
index 02cda2e33bde..d0977d0ad745 100644
@@ -1,13 +1,12 @@
 /* SPDX-License-Identifier: GPL-2.0-only */
 /*
- * Copyright (C) 2013 ARM Ltd.
- * Copyright (C) 2013 Linaro.
+ * Copyright (c) 2013 Linaro Limited. All rights reserved.
+ * Copyright (c) 2015 ARM Ltd. All rights reserved.
  *
- * This code is based on glibc cortex strings work originally authored by Linaro
- * be found @
+ * This code is based on glibc Cortex Strings work originally authored by
+ * Linaro, found at:
  *
- * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/
- * files/head:/src/aarch64/
+ * https://git.linaro.org/toolchain/cortex-strings.git
  */
 
 #include <linux/linkage.h>
@@ -25,165 +24,90 @@
  * Returns:
  *	x0 - dest
  */
-dstin	.req	x0
-src	.req	x1
-count	.req	x2
-tmp1	.req	x3
-tmp1w	.req	w3
-tmp2	.req	x4
-tmp2w	.req	w4
-tmp3	.req	x5
-tmp3w	.req	w5
-dst	.req	x6
+/* Parameters and result.  */
+#define dstin	x0
+#define src	x1
+#define count	x2
+#define srcend	x3
+#define dstend	x4
+#define tmp1	x5
+#define A_l	x6
+#define A_h	x7
+#define B_l	x8
+#define B_h	x9
+#define C_l	x10
+#define C_h	x11
+#define D_l	x12
+#define D_h	x13
+#define E_l	count
+#define E_h	tmp1
 
-A_l	.req	x7
-A_h	.req	x8
-B_l	.req	x9
-B_h	.req	x10
-C_l	.req	x11
-C_h	.req	x12
-D_l	.req	x13
-D_h	.req	x14
+/* All memmoves up to 96 bytes are done by memcpy as it supports overlaps.
+   Larger backwards copies are also handled by memcpy. The only remaining
+   case is forward large copies.  The destination is aligned, and an
+   unrolled loop processes 64 bytes per iteration.
+*/
 
-	.weak memmove
+    .weak memmove
 SYM_FUNC_START_ALIAS(__memmove)
 SYM_FUNC_START_PI(memmove)
-	cmp	dstin, src
-	b.lo	__memcpy
-	add	tmp1, src, count
-	cmp	dstin, tmp1
-	b.hs	__memcpy		/* No overlap.  */
+	sub	tmp1, dstin, src
+	cmp	count, 96
+	ccmp	tmp1, count, 2, hi
+	b.hs	__memcpy
 
-	add	dst, dstin, count
-	add	src, src, count
-	cmp	count, #16
-	b.lo	.Ltail15  /*probably non-alignment accesses.*/
+	cbz	tmp1, 3f
+	add	dstend, dstin, count
+	add	srcend, src, count
 
-	ands	tmp2, src, #15     /* Bytes to reach alignment.  */
-	b.eq	.LSrcAligned
-	sub	count, count, tmp2
-	/*
-	* process the aligned offset length to make the src aligned firstly.
-	* those extra instructions' cost is acceptable. It also make the
-	* coming accesses are based on aligned address.
-	*/
-	tbz	tmp2, #0, 1f
-	ldrb	tmp1w, [src, #-1]!
-	strb	tmp1w, [dst, #-1]!
-1:
-	tbz	tmp2, #1, 2f
-	ldrh	tmp1w, [src, #-2]!
-	strh	tmp1w, [dst, #-2]!
-2:
-	tbz	tmp2, #2, 3f
-	ldr	tmp1w, [src, #-4]!
-	str	tmp1w, [dst, #-4]!
-3:
-	tbz	tmp2, #3, .LSrcAligned
-	ldr	tmp1, [src, #-8]!
-	str	tmp1, [dst, #-8]!
-
-.LSrcAligned:
-	cmp	count, #64
-	b.ge	.Lcpy_over64
+	/* Align dstend to 16 byte alignment so that we don't cross cache line
+	   boundaries on both loads and stores.	 There are at least 96 bytes
+	   to copy, so copy 16 bytes unaligned and then align.	The loop
+	   copies 64 bytes per iteration and prefetches one iteration ahead.  */
 
-	/*
-	* Deal with small copies quickly by dropping straight into the
-	* exit block.
-	*/
-.Ltail63:
-	/*
-	* Copy up to 48 bytes of data. At this point we only need the
-	* bottom 6 bits of count to be accurate.
-	*/
-	ands	tmp1, count, #0x30
-	b.eq	.Ltail15
-	cmp	tmp1w, #0x20
-	b.eq	1f
-	b.lt	2f
-	ldp	A_l, A_h, [src, #-16]!
-	stp	A_l, A_h, [dst, #-16]!
+	and	tmp1, dstend, 15
+	ldp	D_l, D_h, [srcend, -16]
+	sub	srcend, srcend, tmp1
+	sub	count, count, tmp1
+	ldp	A_l, A_h, [srcend, -16]
+	stp	D_l, D_h, [dstend, -16]
+	ldp	B_l, B_h, [srcend, -32]
+	ldp	C_l, C_h, [srcend, -48]
+	ldp	D_l, D_h, [srcend, -64]!
+	sub	dstend, dstend, tmp1
+	subs	count, count, 128
+	b.ls	2f
+	nop
 1:
-	ldp	A_l, A_h, [src, #-16]!
-	stp	A_l, A_h, [dst, #-16]!
-2:
-	ldp	A_l, A_h, [src, #-16]!
-	stp	A_l, A_h, [dst, #-16]!
+	stp	A_l, A_h, [dstend, -16]
+	ldp	A_l, A_h, [srcend, -16]
+	stp	B_l, B_h, [dstend, -32]
+	ldp	B_l, B_h, [srcend, -32]
+	stp	C_l, C_h, [dstend, -48]
+	ldp	C_l, C_h, [srcend, -48]
+	stp	D_l, D_h, [dstend, -64]!
+	ldp	D_l, D_h, [srcend, -64]!
+	subs	count, count, 64
+	b.hi	1b
 
-.Ltail15:
-	tbz	count, #3, 1f
-	ldr	tmp1, [src, #-8]!
-	str	tmp1, [dst, #-8]!
-1:
-	tbz	count, #2, 2f
-	ldr	tmp1w, [src, #-4]!
-	str	tmp1w, [dst, #-4]!
+	/* Write the last full set of 64 bytes.	 The remainder is at most 64
+	   bytes, so it is safe to always copy 64 bytes from the start even if
+	   there is just 1 byte left.  */
 2:
-	tbz	count, #1, 3f
-	ldrh	tmp1w, [src, #-2]!
-	strh	tmp1w, [dst, #-2]!
-3:
-	tbz	count, #0, .Lexitfunc
-	ldrb	tmp1w, [src, #-1]
-	strb	tmp1w, [dst, #-1]
-
-.Lexitfunc:
-	ret
-
-.Lcpy_over64:
-	subs	count, count, #128
-	b.ge	.Lcpy_body_large
-	/*
-	* Less than 128 bytes to copy, so handle 64 bytes here and then jump
-	* to the tail.
-	*/
-	ldp	A_l, A_h, [src, #-16]
-	stp	A_l, A_h, [dst, #-16]
-	ldp	B_l, B_h, [src, #-32]
-	ldp	C_l, C_h, [src, #-48]
-	stp	B_l, B_h, [dst, #-32]
-	stp	C_l, C_h, [dst, #-48]
-	ldp	D_l, D_h, [src, #-64]!
-	stp	D_l, D_h, [dst, #-64]!
-
-	tst	count, #0x3f
-	b.ne	.Ltail63
-	ret
-
-	/*
-	* Critical loop. Start at a new cache line boundary. Assuming
-	* 64 bytes per line this ensures the entire loop is in one line.
-	*/
-	.p2align	L1_CACHE_SHIFT
-.Lcpy_body_large:
-	/* pre-load 64 bytes data. */
-	ldp	A_l, A_h, [src, #-16]
-	ldp	B_l, B_h, [src, #-32]
-	ldp	C_l, C_h, [src, #-48]
-	ldp	D_l, D_h, [src, #-64]!
-1:
-	/*
-	* interlace the load of next 64 bytes data block with store of the last
-	* loaded 64 bytes data.
-	*/
-	stp	A_l, A_h, [dst, #-16]
-	ldp	A_l, A_h, [src, #-16]
-	stp	B_l, B_h, [dst, #-32]
-	ldp	B_l, B_h, [src, #-32]
-	stp	C_l, C_h, [dst, #-48]
-	ldp	C_l, C_h, [src, #-48]
-	stp	D_l, D_h, [dst, #-64]!
-	ldp	D_l, D_h, [src, #-64]!
-	subs	count, count, #64
-	b.ge	1b
-	stp	A_l, A_h, [dst, #-16]
-	stp	B_l, B_h, [dst, #-32]
-	stp	C_l, C_h, [dst, #-48]
-	stp	D_l, D_h, [dst, #-64]!
+	ldp	E_l, E_h, [src, 48]
+	stp	A_l, A_h, [dstend, -16]
+	ldp	A_l, A_h, [src, 32]
+	stp	B_l, B_h, [dstend, -32]
+	ldp	B_l, B_h, [src, 16]
+	stp	C_l, C_h, [dstend, -48]
+	ldp	C_l, C_h, [src]
+	stp	D_l, D_h, [dstend, -64]
+	stp	E_l, E_h, [dstin, 48]
+	stp	A_l, A_h, [dstin, 32]
+	stp	B_l, B_h, [dstin, 16]
+	stp	C_l, C_h, [dstin]
+3:	ret
 
-	tst	count, #0x3f
-	b.ne	.Ltail63
-	ret
 SYM_FUNC_END_PI(memmove)
 EXPORT_SYMBOL(memmove)
 SYM_FUNC_END_ALIAS(__memmove)
Signature · 1 line
2.17.1
#5 lore raw

PATCH v4 5/14 arm64: Import latest version of Cortex Strings' strcmp

In reply to Oliver Swede

From: Sam Tebbs <[email protected]>

Import the latest version of Cortex Strings' strcmp function.

The upstream source is src/aarch64/strcmp.S as of commit 90b61261ceb4 in https://git.linaro.org/toolchain/cortex-strings.git.

[ rm: update attribution, expand commit message ]
---
 arch/arm64/lib/strcmp.S | 272 +++++++++++++++++-----------------------
 1 file changed, 113 insertions(+), 159 deletions(-)
arch/arm64/lib/strcmp.S +113 -159
index 4e79566726c8..e00ff46c4ffc 100644
@@ -1,13 +1,11 @@
 /* SPDX-License-Identifier: GPL-2.0-only */
 /*
- * Copyright (C) 2013 ARM Ltd.
- * Copyright (C) 2013 Linaro.
+ * Copyright (c) 2012,2018 Linaro Limited. All rights reserved.
  *
- * This code is based on glibc cortex strings work originally authored by Linaro
- * be found @
+ * This code is based on glibc Cortex Strings work originally authored by
+ * Linaro, found at:
  *
- * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/
- * files/head:/src/aarch64/
+ * https://git.linaro.org/toolchain/cortex-strings.git
  */
 
 #include <linux/linkage.h>
@@ -25,60 +23,106 @@
  * or be greater than s2.
  */
 
+#define L(label) .L ## label
+
 #define REP8_01 0x0101010101010101
 #define REP8_7f 0x7f7f7f7f7f7f7f7f
 #define REP8_80 0x8080808080808080
 
 /* Parameters and result.  */
-src1		.req	x0
-src2		.req	x1
-result		.req	x0
+#define src1		x0
+#define src2		x1
+#define result		x0
 
 /* Internal variables.  */
-data1		.req	x2
-data1w		.req	w2
-data2		.req	x3
-data2w		.req	w3
-has_nul		.req	x4
-diff		.req	x5
-syndrome	.req	x6
-tmp1		.req	x7
-tmp2		.req	x8
-tmp3		.req	x9
-zeroones	.req	x10
-pos		.req	x11
-
+#define data1		x2
+#define data1w		w2
+#define data2		x3
+#define data2w		w3
+#define has_nul		x4
+#define diff		x5
+#define syndrome	x6
+#define tmp1		x7
+#define tmp2		x8
+#define tmp3		x9
+#define zeroones	x10
+#define pos		x11
+
+	/* Start of performance-critical section  -- one 64B cache line.  */
 SYM_FUNC_START_WEAK_PI(strcmp)
 	eor	tmp1, src1, src2
 	mov	zeroones, #REP8_01
 	tst	tmp1, #7
-	b.ne	.Lmisaligned8
+	b.ne	L(misaligned8)
 	ands	tmp1, src1, #7
-	b.ne	.Lmutual_align
-
-	/*
-	* NUL detection works on the principle that (X - 1) & (~X) & 0x80
-	* (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
-	* can be done in parallel across the entire word.
-	*/
-.Lloop_aligned:
+	b.ne	L(mutual_align)
+	/* NUL detection works on the principle that (X - 1) & (~X) & 0x80
+	   (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
+	   can be done in parallel across the entire word.  */
+L(loop_aligned):
 	ldr	data1, [src1], #8
 	ldr	data2, [src2], #8
-.Lstart_realigned:
+L(start_realigned):
 	sub	tmp1, data1, zeroones
 	orr	tmp2, data1, #REP8_7f
 	eor	diff, data1, data2	/* Non-zero if differences found.  */
 	bic	has_nul, tmp1, tmp2	/* Non-zero if NUL terminator.  */
 	orr	syndrome, diff, has_nul
-	cbz	syndrome, .Lloop_aligned
-	b	.Lcal_cmpresult
-
-.Lmutual_align:
-	/*
-	* Sources are mutually aligned, but are not currently at an
-	* alignment boundary.  Round down the addresses and then mask off
-	* the bytes that preceed the start point.
-	*/
+	cbz	syndrome, L(loop_aligned)
+	/* End of performance-critical section  -- one 64B cache line.  */
+
+L(end):
+CPU_LE(rev	syndrome, syndrome)
+CPU_LE(rev	data1, data1)
+	/* The MS-non-zero bit of the syndrome marks either the first bit
+	   that is different, or the top bit of the first zero byte.
+	   Shifting left now will bring the critical information into the
+	   top bits.  */
+CPU_LE(clz	pos, syndrome)
+CPU_LE(rev	data2, data2)
+CPU_LE(lsl	data1, data1, pos)
+CPU_LE(lsl	data2, data2, pos)
+	/* But we need to zero-extend (char is unsigned) the value and then
+	   perform a signed 32-bit subtraction.  */
+CPU_LE(lsr	data1, data1, #56)
+CPU_LE(sub	result, data1, data2, lsr #56)
+CPU_LE(ret)
+	/* For big-endian we cannot use the trick with the syndrome value
+	   as carry-propagation can corrupt the upper bits if the trailing
+	   bytes in the string contain 0x01.  */
+	/* However, if there is no NUL byte in the dword, we can generate
+	   the result directly.  We can't just subtract the bytes as the
+	   MSB might be significant.  */
+CPU_BE(cbnz	has_nul, 1f)
+CPU_BE(cmp	data1, data2)
+CPU_BE(cset	result, ne)
+CPU_BE(cneg	result, result, lo)
+CPU_BE(ret)
+1:
+	/* Re-compute the NUL-byte detection, using a byte-reversed value.  */
+CPU_BE(rev	tmp3, data1)
+CPU_BE(sub	tmp1, tmp3, zeroones)
+CPU_BE(orr	tmp2, tmp3, #REP8_7f)
+CPU_BE(bic	has_nul, tmp1, tmp2)
+CPU_BE(rev	has_nul, has_nul)
+CPU_BE(orr	syndrome, diff, has_nul)
+CPU_BE(clz	pos, syndrome)
+	/* The MS-non-zero bit of the syndrome marks either the first bit
+	   that is different, or the top bit of the first zero byte.
+	   Shifting left now will bring the critical information into the
+	   top bits.  */
+CPU_BE(lsl	data1, data1, pos)
+CPU_BE(lsl	data2, data2, pos)
+	/* But we need to zero-extend (char is unsigned) the value and then
+	   perform a signed 32-bit subtraction.  */
+CPU_BE(lsr	data1, data1, #56)
+CPU_BE(sub	result, data1, data2, lsr #56)
+CPU_BE(ret)
+
+L(mutual_align):
+	/* Sources are mutually aligned, but are not currently at an
+	   alignment boundary.  Round down the addresses and then mask off
+	   the bytes that preceed the start point.  */
 	bic	src1, src1, #7
 	bic	src2, src2, #7
 	lsl	tmp1, tmp1, #3		/* Bytes beyond alignment -> bits.  */
@@ -87,137 +131,47 @@ SYM_FUNC_START_WEAK_PI(strcmp)
 	ldr	data2, [src2], #8
 	mov	tmp2, #~0
 	/* Big-endian.  Early bytes are at MSB.  */
-CPU_BE( lsl	tmp2, tmp2, tmp1 )	/* Shift (tmp1 & 63).  */
+CPU_BE(lsl	tmp2, tmp2, tmp1)    /* Shift (tmp1 & 63).  */
 	/* Little-endian.  Early bytes are at LSB.  */
-CPU_LE( lsr	tmp2, tmp2, tmp1 )	/* Shift (tmp1 & 63).  */
-
+CPU_LE(lsr	tmp2, tmp2, tmp1)	/* Shift (tmp1 & 63).  */
 	orr	data1, data1, tmp2
 	orr	data2, data2, tmp2
-	b	.Lstart_realigned
-
-.Lmisaligned8:
-	/*
-	* Get the align offset length to compare per byte first.
-	* After this process, one string's address will be aligned.
-	*/
-	and	tmp1, src1, #7
-	neg	tmp1, tmp1
-	add	tmp1, tmp1, #8
-	and	tmp2, src2, #7
-	neg	tmp2, tmp2
-	add	tmp2, tmp2, #8
-	subs	tmp3, tmp1, tmp2
-	csel	pos, tmp1, tmp2, hi /*Choose the maximum. */
-.Ltinycmp:
+	b	L(start_realigned)
+
+L(misaligned8):
+	/* Align SRC1 to 8 bytes and then compare 8 bytes at a time, always
+	   checking to make sure that we don't access beyond page boundary in
+	   SRC2.  */
+	tst	src1, #7
+	b.eq	L(loop_misaligned)
+L(do_misaligned):
 	ldrb	data1w, [src1], #1
 	ldrb	data2w, [src2], #1
-	subs	pos, pos, #1
-	ccmp	data1w, #1, #0, ne  /* NZCV = 0b0000.  */
-	ccmp	data1w, data2w, #0, cs  /* NZCV = 0b0000.  */
-	b.eq	.Ltinycmp
-	cbnz	pos, 1f /*find the null or unequal...*/
 	cmp	data1w, #1
-	ccmp	data1w, data2w, #0, cs
-	b.eq	.Lstart_align /*the last bytes are equal....*/
-1:
-	sub	result, data1, data2
-	ret
-
-.Lstart_align:
-	ands	xzr, src1, #7
-	b.eq	.Lrecal_offset
-	/*process more leading bytes to make str1 aligned...*/
-	add	src1, src1, tmp3
-	add	src2, src2, tmp3
-	/*load 8 bytes from aligned str1 and non-aligned str2..*/
+	ccmp	data1w, data2w, #0, cs	/* NZCV = 0b0000.  */
+	b.ne	L(done)
+	tst	src1, #7
+	b.ne	L(do_misaligned)
+
+L(loop_misaligned):
+	/* Test if we are within the last dword of the end of a 4K page.  If
+	   yes then jump back to the misaligned loop to copy a byte at a time.  */
+	and	tmp1, src2, #0xff8
+	eor	tmp1, tmp1, #0xff8
+	cbz	tmp1, L(do_misaligned)
 	ldr	data1, [src1], #8
 	ldr	data2, [src2], #8
 
 	sub	tmp1, data1, zeroones
 	orr	tmp2, data1, #REP8_7f
-	bic	has_nul, tmp1, tmp2
-	eor	diff, data1, data2 /* Non-zero if differences found.  */
-	orr	syndrome, diff, has_nul
-	cbnz	syndrome, .Lcal_cmpresult
-	/*How far is the current str2 from the alignment boundary...*/
-	and	tmp3, tmp3, #7
-.Lrecal_offset:
-	neg	pos, tmp3
-.Lloopcmp_proc:
-	/*
-	* Divide the eight bytes into two parts. First,backwards the src2
-	* to an alignment boundary,load eight bytes from the SRC2 alignment
-	* boundary,then compare with the relative bytes from SRC1.
-	* If all 8 bytes are equal,then start the second part's comparison.
-	* Otherwise finish the comparison.
-	* This special handle can garantee all the accesses are in the
-	* thread/task space in avoid to overrange access.
-	*/
-	ldr	data1, [src1,pos]
-	ldr	data2, [src2,pos]
-	sub	tmp1, data1, zeroones
-	orr	tmp2, data1, #REP8_7f
-	bic	has_nul, tmp1, tmp2
-	eor	diff, data1, data2  /* Non-zero if differences found.  */
-	orr	syndrome, diff, has_nul
-	cbnz	syndrome, .Lcal_cmpresult
-
-	/*The second part process*/
-	ldr	data1, [src1], #8
-	ldr	data2, [src2], #8
-	sub	tmp1, data1, zeroones
-	orr	tmp2, data1, #REP8_7f
-	bic	has_nul, tmp1, tmp2
-	eor	diff, data1, data2  /* Non-zero if differences found.  */
+	eor	diff, data1, data2	/* Non-zero if differences found.  */
+	bic	has_nul, tmp1, tmp2	/* Non-zero if NUL terminator.  */
 	orr	syndrome, diff, has_nul
-	cbz	syndrome, .Lloopcmp_proc
+	cbz	syndrome, L(loop_misaligned)
+	b	L(end)
 
-.Lcal_cmpresult:
-	/*
-	* reversed the byte-order as big-endian,then CLZ can find the most
-	* significant zero bits.
-	*/
-CPU_LE( rev	syndrome, syndrome )
-CPU_LE( rev	data1, data1 )
-CPU_LE( rev	data2, data2 )
-
-	/*
-	* For big-endian we cannot use the trick with the syndrome value
-	* as carry-propagation can corrupt the upper bits if the trailing
-	* bytes in the string contain 0x01.
-	* However, if there is no NUL byte in the dword, we can generate
-	* the result directly.  We cannot just subtract the bytes as the
-	* MSB might be significant.
-	*/
-CPU_BE( cbnz	has_nul, 1f )
-CPU_BE( cmp	data1, data2 )
-CPU_BE( cset	result, ne )
-CPU_BE( cneg	result, result, lo )
-CPU_BE( ret )
-CPU_BE( 1: )
-	/*Re-compute the NUL-byte detection, using a byte-reversed value. */
-CPU_BE(	rev	tmp3, data1 )
-CPU_BE(	sub	tmp1, tmp3, zeroones )
-CPU_BE(	orr	tmp2, tmp3, #REP8_7f )
-CPU_BE(	bic	has_nul, tmp1, tmp2 )
-CPU_BE(	rev	has_nul, has_nul )
-CPU_BE(	orr	syndrome, diff, has_nul )
-
-	clz	pos, syndrome
-	/*
-	* The MS-non-zero bit of the syndrome marks either the first bit
-	* that is different, or the top bit of the first zero byte.
-	* Shifting left now will bring the critical information into the
-	* top bits.
-	*/
-	lsl	data1, data1, pos
-	lsl	data2, data2, pos
-	/*
-	* But we need to zero-extend (char is unsigned) the value and then
-	* perform a signed 32-bit subtraction.
-	*/
-	lsr	data1, data1, #56
-	sub	result, data1, data2, lsr #56
+L(done):
+	sub	result, data1, data2
 	ret
 SYM_FUNC_END_PI(strcmp)
 EXPORT_SYMBOL_NOKASAN(strcmp)
Signature · 1 line
2.17.1
#6 lore raw

PATCH v4 6/14 arm64: Import latest version of Cortex Strings' strlen

In reply to Oliver Swede

From: Sam Tebbs <[email protected]>

Import latest version of Cortex Strings' strlen function.

The upstream source is src/aarch64/strlen.S as of commit eb80ac77a6cd in https://git.linaro.org/toolchain/cortex-strings.git.

[ rm: update attribution, expand commit message ]
---
 arch/arm64/lib/strlen.S | 247 +++++++++++++++++++++++++++-------------
 1 file changed, 168 insertions(+), 79 deletions(-)
arch/arm64/lib/strlen.S +168 -79
index ee3ed882dd79..974b67dcc186 100644
@@ -1,13 +1,11 @@
 /* SPDX-License-Identifier: GPL-2.0-only */
 /*
- * Copyright (C) 2013 ARM Ltd.
- * Copyright (C) 2013 Linaro.
+ * Copyright (c) 2013-2015 Linaro Limited. All rights reserved.
  *
- * This code is based on glibc cortex strings work originally authored by Linaro
- * be found @
+ * This code is based on glibc Cortex Strings work originally authored by
+ * Linaro, found at:
  *
- * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/
- * files/head:/src/aarch64/
+ * https://git.linaro.org/toolchain/cortex-strings.git
  */
 
 #include <linux/linkage.h>
@@ -23,93 +21,184 @@
  */
 
 /* Arguments and results.  */
-srcin		.req	x0
-len		.req	x0
+#define srcin		x0
+#define len		x0
 
 /* Locals and temporaries.  */
-src		.req	x1
-data1		.req	x2
-data2		.req	x3
-data2a		.req	x4
-has_nul1	.req	x5
-has_nul2	.req	x6
-tmp1		.req	x7
-tmp2		.req	x8
-tmp3		.req	x9
-tmp4		.req	x10
-zeroones	.req	x11
-pos		.req	x12
+#define src		x1
+#define data1		x2
+#define data2		x3
+#define has_nul1	x4
+#define has_nul2	x5
+#define tmp1		x4
+#define tmp2		x5
+#define tmp3		x6
+#define tmp4		x7
+#define zeroones	x8
+
+#define L(l) .L ## l
+
+	/* NUL detection works on the principle that (X - 1) & (~X) & 0x80
+	   (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
+	   can be done in parallel across the entire word. A faster check
+	   (X - 1) & 0x80 is zero for non-NUL ASCII characters, but gives
+	   false hits for characters 129..255.	*/
 
 #define REP8_01 0x0101010101010101
 #define REP8_7f 0x7f7f7f7f7f7f7f7f
 #define REP8_80 0x8080808080808080
 
+#ifdef TEST_PAGE_CROSS
+# define MIN_PAGE_SIZE 15
+#else
+# define MIN_PAGE_SIZE 4096
+#endif
+
+	/* Since strings are short on average, we check the first 16 bytes
+	   of the string for a NUL character.  In order to do an unaligned ldp
+	   safely we have to do a page cross check first.  If there is a NUL
+	   byte we calculate the length from the 2 8-byte words using
+	   conditional select to reduce branch mispredictions (it is unlikely
+	   strlen will be repeatedly called on strings with the same length).
+
+	   If the string is longer than 16 bytes, we align src so don't need
+	   further page cross checks, and process 32 bytes per iteration
+	   using the fast NUL check.  If we encounter non-ASCII characters,
+	   fallback to a second loop using the full NUL check.
+
+	   If the page cross check fails, we read 16 bytes from an aligned
+	   address, remove any characters before the string, and continue
+	   in the main loop using aligned loads.  Since strings crossing a
+	   page in the first 16 bytes are rare (probability of
+	   16/MIN_PAGE_SIZE ~= 0.4%), this case does not need to be optimized.
+
+	   AArch64 systems have a minimum page size of 4k.  We don't bother
+	   checking for larger page sizes - the cost of setting up the correct
+	   page size is just not worth the extra gain from a small reduction in
+	   the cases taking the slow path.  Note that we only care about
+	   whether the first fetch, which may be misaligned, crosses a page
+	   boundary.  */
+
 SYM_FUNC_START_WEAK_PI(strlen)
-	mov	zeroones, #REP8_01
-	bic	src, srcin, #15
-	ands	tmp1, srcin, #15
-	b.ne	.Lmisaligned
-	/*
-	* NUL detection works on the principle that (X - 1) & (~X) & 0x80
-	* (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
-	* can be done in parallel across the entire word.
-	*/
-	/*
-	* The inner loop deals with two Dwords at a time. This has a
-	* slightly higher start-up cost, but we should win quite quickly,
-	* especially on cores with a high number of issue slots per
-	* cycle, as we get much better parallelism out of the operations.
-	*/
-.Lloop:
-	ldp	data1, data2, [src], #16
-.Lrealigned:
+	and	tmp1, srcin, MIN_PAGE_SIZE - 1
+	mov	zeroones, REP8_01
+	cmp	tmp1, MIN_PAGE_SIZE - 16
+	b.gt	L(page_cross)
+	ldp	data1, data2, [srcin]
+	/* For big-endian, carry propagation (if the final byte in the
+	   string is 0x01) means we cannot use has_nul1/2 directly.
+	   Since we expect strings to be small and early-exit,
+	   byte-swap the data now so has_null1/2 will be correct.  */
+CPU_BE(rev	data1, data1)
+CPU_BE(rev	data2, data2)
+	sub	tmp1, data1, zeroones
+	orr	tmp2, data1, REP8_7f
+	sub	tmp3, data2, zeroones
+	orr	tmp4, data2, REP8_7f
+	bics	has_nul1, tmp1, tmp2
+	bic	has_nul2, tmp3, tmp4
+	ccmp	has_nul2, 0, 0, eq
+	beq	L(main_loop_entry)
+
+	/* Enter with C = has_nul1 == 0.  */
+	csel	has_nul1, has_nul1, has_nul2, cc
+	mov	len, 8
+	rev	has_nul1, has_nul1
+	clz	tmp1, has_nul1
+	csel	len, xzr, len, cc
+	add	len, len, tmp1, lsr 3
+	ret
+
+	/* The inner loop processes 32 bytes per iteration and uses the fast
+	   NUL check.  If we encounter non-ASCII characters, use a second
+	   loop with the accurate NUL check.  */
+	.p2align 4
+L(main_loop_entry):
+	bic	src, srcin, 15
+	sub	src, src, 16
+L(main_loop):
+	ldp	data1, data2, [src, 32]!
+.Lpage_cross_entry:
 	sub	tmp1, data1, zeroones
-	orr	tmp2, data1, #REP8_7f
 	sub	tmp3, data2, zeroones
-	orr	tmp4, data2, #REP8_7f
-	bic	has_nul1, tmp1, tmp2
-	bics	has_nul2, tmp3, tmp4
-	ccmp	has_nul1, #0, #0, eq	/* NZCV = 0000  */
-	b.eq	.Lloop
+	orr	tmp2, tmp1, tmp3
+	tst	tmp2, zeroones, lsl 7
+	bne	1f
+	ldp	data1, data2, [src, 16]
+	sub	tmp1, data1, zeroones
+	sub	tmp3, data2, zeroones
+	orr	tmp2, tmp1, tmp3
+	tst	tmp2, zeroones, lsl 7
+	beq	L(main_loop)
+	add	src, src, 16
+1:
+	/* The fast check failed, so do the slower, accurate NUL check.	 */
+	orr	tmp2, data1, REP8_7f
+	orr	tmp4, data2, REP8_7f
+	bics	has_nul1, tmp1, tmp2
+	bic	has_nul2, tmp3, tmp4
+	ccmp	has_nul2, 0, 0, eq
+	beq	L(nonascii_loop)
 
+	/* Enter with C = has_nul1 == 0.  */
+L(tail):
+	/* For big-endian, carry propagation (if the final byte in the
+	   string is 0x01) means we cannot use has_nul1/2 directly.  The
+	   easiest way to get the correct byte is to byte-swap the data
+	   and calculate the syndrome a second time.  */
+CPU_BE(csel	data1, data1, data2, cc)
+CPU_BE(rev	data1, data1)
+CPU_BE(sub	tmp1, data1, zeroones)
+CPU_BE(orr	tmp2, data1, REP8_7f)
+CPU_BE(bic	has_nul1, tmp1, tmp2)
+CPU_LE(csel	has_nul1, has_nul1, has_nul2, cc)
 	sub	len, src, srcin
-	cbz	has_nul1, .Lnul_in_data2
-CPU_BE(	mov	data2, data1 )	/*prepare data to re-calculate the syndrome*/
-	sub	len, len, #8
-	mov	has_nul2, has_nul1
-.Lnul_in_data2:
-	/*
-	* For big-endian, carry propagation (if the final byte in the
-	* string is 0x01) means we cannot use has_nul directly.  The
-	* easiest way to get the correct byte is to byte-swap the data
-	* and calculate the syndrome a second time.
-	*/
-CPU_BE( rev	data2, data2 )
-CPU_BE( sub	tmp1, data2, zeroones )
-CPU_BE( orr	tmp2, data2, #REP8_7f )
-CPU_BE( bic	has_nul2, tmp1, tmp2 )
-
-	sub	len, len, #8
-	rev	has_nul2, has_nul2
-	clz	pos, has_nul2
-	add	len, len, pos, lsr #3		/* Bits to bytes.  */
+	rev	has_nul1, has_nul1
+	add	tmp2, len, 8
+	clz	tmp1, has_nul1
+	csel	len, len, tmp2, cc
+	add	len, len, tmp1, lsr 3
 	ret
 
-.Lmisaligned:
-	cmp	tmp1, #8
-	neg	tmp1, tmp1
-	ldp	data1, data2, [src], #16
-	lsl	tmp1, tmp1, #3		/* Bytes beyond alignment -> bits.  */
-	mov	tmp2, #~0
-	/* Big-endian.  Early bytes are at MSB.  */
-CPU_BE( lsl	tmp2, tmp2, tmp1 )	/* Shift (tmp1 & 63).  */
-	/* Little-endian.  Early bytes are at LSB.  */
-CPU_LE( lsr	tmp2, tmp2, tmp1 )	/* Shift (tmp1 & 63).  */
+L(nonascii_loop):
+	ldp	data1, data2, [src, 16]!
+	sub	tmp1, data1, zeroones
+	orr	tmp2, data1, REP8_7f
+	sub	tmp3, data2, zeroones
+	orr	tmp4, data2, REP8_7f
+	bics	has_nul1, tmp1, tmp2
+	bic	has_nul2, tmp3, tmp4
+	ccmp	has_nul2, 0, 0, eq
+	bne	L(tail)
+	ldp	data1, data2, [src, 16]!
+	sub	tmp1, data1, zeroones
+	orr	tmp2, data1, REP8_7f
+	sub	tmp3, data2, zeroones
+	orr	tmp4, data2, REP8_7f
+	bics	has_nul1, tmp1, tmp2
+	bic	has_nul2, tmp3, tmp4
+	ccmp	has_nul2, 0, 0, eq
+	beq	L(nonascii_loop)
+	b	L(tail)
 
-	orr	data1, data1, tmp2
-	orr	data2a, data2, tmp2
-	csinv	data1, data1, xzr, le
-	csel	data2, data2, data2a, le
-	b	.Lrealigned
+	/* Load 16 bytes from [srcin & ~15] and force the bytes that precede
+	   srcin to 0x7f, so we ignore any NUL bytes before the string.
+	   Then continue in the aligned loop.  */
+L(page_cross):
+	bic	src, srcin, 15
+	ldp	data1, data2, [src]
+	lsl	tmp1, srcin, 3
+	mov	tmp4, -1
+	/* Big-endian.	Early bytes are at MSB.	 */
+CPU_BE(lsr	tmp1, tmp4, tmp1)	/* Shift (tmp1 & 63).  */
+	/* Little-endian.  Early bytes are at LSB.  */
+CPU_LE(lsl	tmp1, tmp4, tmp1)	/* Shift (tmp1 & 63).  */
+	orr	tmp1, tmp1, REP8_80
+	orn	data1, data1, tmp1
+	orn	tmp2, data2, tmp1
+	tst	srcin, 8
+	csel	data1, data1, tmp4, eq
+	csel	data2, data2, tmp2, eq
+	b	L(page_cross_entry)
 SYM_FUNC_END_PI(strlen)
 EXPORT_SYMBOL_NOKASAN(strlen)
Signature · 1 line
2.17.1
#7 lore raw

PATCH v4 7/14 arm64: Import latest version of Cortex Strings' strncmp

In reply to Oliver Swede

From: Sam Tebbs <[email protected]>

Import latest version of Cortex Strings' strncmp function.

The upstream source is src/aarch64/strncmp.S as of commit 071fe283b28d in https://git.linaro.org/toolchain/cortex-strings.git.

[ rm: update attribution, expand commit message ]
---
 arch/arm64/lib/strncmp.S | 363 ++++++++++++++++++---------------------
 1 file changed, 163 insertions(+), 200 deletions(-)
arch/arm64/lib/strncmp.S +163 -200
index 2a7ee949ed47..b954e0fd93be 100644
@@ -1,13 +1,11 @@
 /* SPDX-License-Identifier: GPL-2.0-only */
 /*
- * Copyright (C) 2013 ARM Ltd.
- * Copyright (C) 2013 Linaro.
+ * Copyright (c) 2013,2018 Linaro Limited. All rights reserved.
  *
- * This code is based on glibc cortex strings work originally authored by Linaro
- * be found @
+ * This code is based on glibc Cortex Strings work originally authored by
+ * Linaro, found at:
  *
- * http://bazaar.launchpad.net/~linaro-toolchain-dev/cortex-strings/trunk/
- * files/head:/src/aarch64/
+ * https://git.linaro.org/toolchain/cortex-strings.git
  */
 
 #include <linux/linkage.h>
@@ -30,49 +28,49 @@
 #define REP8_80 0x8080808080808080
 
 /* Parameters and result.  */
-src1		.req	x0
-src2		.req	x1
-limit		.req	x2
-result		.req	x0
+#define src1		x0
+#define src2		x1
+#define limit		x2
+#define result		x0
 
 /* Internal variables.  */
-data1		.req	x3
-data1w		.req	w3
-data2		.req	x4
-data2w		.req	w4
-has_nul		.req	x5
-diff		.req	x6
-syndrome	.req	x7
-tmp1		.req	x8
-tmp2		.req	x9
-tmp3		.req	x10
-zeroones	.req	x11
-pos		.req	x12
-limit_wd	.req	x13
-mask		.req	x14
-endloop		.req	x15
+#define data1		x3
+#define data1w		w3
+#define data2		x4
+#define data2w		w4
+#define has_nul		x5
+#define diff		x6
+#define syndrome	x7
+#define tmp1		x8
+#define tmp2		x9
+#define tmp3		x10
+#define zeroones	x11
+#define pos		x12
+#define limit_wd	x13
+#define mask		x14
+#define endloop		x15
+#define count		mask
 
+	.p2align 6
+	.rep 7
+	nop	/* Pad so that the loop below fits a cache line.  */
+	.endr
 SYM_FUNC_START_WEAK_PI(strncmp)
 	cbz	limit, .Lret0
 	eor	tmp1, src1, src2
 	mov	zeroones, #REP8_01
 	tst	tmp1, #7
+	and	count, src1, #7
 	b.ne	.Lmisaligned8
-	ands	tmp1, src1, #7
-	b.ne	.Lmutual_align
+	cbnz	count, .Lmutual_align
 	/* Calculate the number of full and partial words -1.  */
-	/*
-	* when limit is mulitply of 8, if not sub 1,
-	* the judgement of last dword will wrong.
-	*/
-	sub	limit_wd, limit, #1 /* limit != 0, so no underflow.  */
-	lsr	limit_wd, limit_wd, #3  /* Convert to Dwords.  */
+	sub	limit_wd, limit, #1	/* limit != 0, so no underflow.  */
+	lsr	limit_wd, limit_wd, #3	/* Convert to Dwords.  */
 
-	/*
-	* NUL detection works on the principle that (X - 1) & (~X) & 0x80
-	* (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
-	* can be done in parallel across the entire word.
-	*/
+	/* NUL detection works on the principle that (X - 1) & (~X) & 0x80
+	   (=> (X - 1) & ~(X | 0x7f)) is non-zero iff a byte is zero, and
+	   can be done in parallel across the entire word.  */
+	/* Start of performance-critical section  -- one 64B cache line.  */
 .Lloop_aligned:
 	ldr	data1, [src1], #8
 	ldr	data2, [src2], #8
@@ -80,23 +78,24 @@ SYM_FUNC_START_WEAK_PI(strncmp)
 	subs	limit_wd, limit_wd, #1
 	sub	tmp1, data1, zeroones
 	orr	tmp2, data1, #REP8_7f
-	eor	diff, data1, data2  /* Non-zero if differences found.  */
-	csinv	endloop, diff, xzr, pl  /* Last Dword or differences.*/
-	bics	has_nul, tmp1, tmp2 /* Non-zero if NUL terminator.  */
+	eor	diff, data1, data2	/* Non-zero if differences found.  */
+	csinv	endloop, diff, xzr, pl	/* Last Dword or differences.  */
+	bics	has_nul, tmp1, tmp2	/* Non-zero if NUL terminator.  */
 	ccmp	endloop, #0, #0, eq
 	b.eq	.Lloop_aligned
+	/* End of performance-critical section  -- one 64B cache line.  */
 
-	/*Not reached the limit, must have found the end or a diff.  */
+	/* Not reached the limit, must have found the end or a diff.  */
 	tbz	limit_wd, #63, .Lnot_limit
 
 	/* Limit % 8 == 0 => all bytes significant.  */
 	ands	limit, limit, #7
 	b.eq	.Lnot_limit
 
-	lsl	limit, limit, #3    /* Bits -> bytes.  */
+	lsl	limit, limit, #3	/* Bits -> bytes.  */
 	mov	mask, #~0
-CPU_BE( lsr	mask, mask, limit )
-CPU_LE( lsl	mask, mask, limit )
+CPU_BE(lsr	mask, mask, limit)
+CPU_LE(lsl	mask, mask, limit)
 	bic	data1, data1, mask
 	bic	data2, data2, mask
 
@@ -105,192 +104,156 @@ CPU_LE( lsl	mask, mask, limit )
 
 .Lnot_limit:
 	orr	syndrome, diff, has_nul
-	b	.Lcal_cmpresult
+
+	CPU_LE(rev	syndrome, syndrome)
+	CPU_LE(rev	data1, data1)
+	/* The MS-non-zero bit of the syndrome marks either the first bit
+	   that is different, or the top bit of the first zero byte.
+	   Shifting left now will bring the critical information into the
+	   top bits.  */
+	CPU_LE(clz	pos, syndrome)
+	CPU_LE(rev	data2, data2)
+	CPU_LE(lsl	data1, data1, pos)
+	CPU_LE(lsl	data2, data2, pos)
+	/* But we need to zero-extend (char is unsigned) the value and then
+	   perform a signed 32-bit subtraction.  */
+	CPU_LE(lsr	data1, data1, #56)
+	CPU_LE(sub	result, data1, data2, lsr #56)
+	CPU_LE(ret)
+	/* For big-endian we cannot use the trick with the syndrome value
+	   as carry-propagation can corrupt the upper bits if the trailing
+	   bytes in the string contain 0x01.  */
+	/* However, if there is no NUL byte in the dword, we can generate
+	   the result directly.  We can't just subtract the bytes as the
+	   MSB might be significant.  */
+	CPU_BE(cbnz	has_nul, 1f)
+	CPU_BE(cmp	data1, data2)
+	CPU_BE(cset	result, ne)
+	CPU_BE(cneg	result, result, lo)
+	CPU_BE(ret)
+1:
+	/* Re-compute the NUL-byte detection, using a byte-reversed value.  */
+	CPU_BE(rev	tmp3, data1)
+	CPU_BE(sub	tmp1, tmp3, zeroones)
+	CPU_BE(orr	tmp2, tmp3, #REP8_7f)
+	CPU_BE(bic	has_nul, tmp1, tmp2)
+	CPU_BE(rev	has_nul, has_nul)
+	CPU_BE(orr	syndrome, diff, has_nul)
+	CPU_BE(clz	pos, syndrome)
+	/* The MS-non-zero bit of the syndrome marks either the first bit
+	   that is different, or the top bit of the first zero byte.
+	   Shifting left now will bring the critical information into the
+	   top bits.  */
+	CPU_BE(lsl	data1, data1, pos)
+	CPU_BE(lsl	data2, data2, pos)
+	/* But we need to zero-extend (char is unsigned) the value and then
+	   perform a signed 32-bit subtraction.  */
+	CPU_BE(lsr	data1, data1, #56)
+	CPU_BE(sub	result, data1, data2, lsr #56)
+	CPU_BE(ret)
 
 .Lmutual_align:
-	/*
-	* Sources are mutually aligned, but are not currently at an
-	* alignment boundary.  Round down the addresses and then mask off
-	* the bytes that precede the start point.
-	* We also need to adjust the limit calculations, but without
-	* overflowing if the limit is near ULONG_MAX.
-	*/
+	/* Sources are mutually aligned, but are not currently at an
+	   alignment boundary.  Round down the addresses and then mask off
+	   the bytes that precede the start point.
+	   We also need to adjust the limit calculations, but without
+	   overflowing if the limit is near ULONG_MAX.  */
 	bic	src1, src1, #7
 	bic	src2, src2, #7
 	ldr	data1, [src1], #8
-	neg	tmp3, tmp1, lsl #3  /* 64 - bits(bytes beyond align). */
+	neg	tmp3, count, lsl #3	/* 64 - bits(bytes beyond align). */
 	ldr	data2, [src2], #8
 	mov	tmp2, #~0
-	sub	limit_wd, limit, #1 /* limit != 0, so no underflow.  */
+	sub	limit_wd, limit, #1	/* limit != 0, so no underflow.  */
 	/* Big-endian.  Early bytes are at MSB.  */
-CPU_BE( lsl	tmp2, tmp2, tmp3 )	/* Shift (tmp1 & 63).  */
+	CPU_BE(lsl	tmp2, tmp2, tmp3)	/* Shift (count & 63).  */
 	/* Little-endian.  Early bytes are at LSB.  */
-CPU_LE( lsr	tmp2, tmp2, tmp3 )	/* Shift (tmp1 & 63).  */
-
+	CPU_LE(lsr	tmp2, tmp2, tmp3)	/* Shift (count & 63).  */
 	and	tmp3, limit_wd, #7
 	lsr	limit_wd, limit_wd, #3
-	/* Adjust the limit. Only low 3 bits used, so overflow irrelevant.*/
-	add	limit, limit, tmp1
-	add	tmp3, tmp3, tmp1
+	/* Adjust the limit. Only low 3 bits used, so overflow irrelevant.  */
+	add	limit, limit, count
+	add	tmp3, tmp3, count
 	orr	data1, data1, tmp2
 	orr	data2, data2, tmp2
 	add	limit_wd, limit_wd, tmp3, lsr #3
 	b	.Lstart_realigned
 
-/*when src1 offset is not equal to src2 offset...*/
+	.p2align 6
+	/* Don't bother with dwords for up to 16 bytes.  */
 .Lmisaligned8:
-	cmp	limit, #8
-	b.lo	.Ltiny8proc /*limit < 8... */
-	/*
-	* Get the align offset length to compare per byte first.
-	* After this process, one string's address will be aligned.*/
-	and	tmp1, src1, #7
-	neg	tmp1, tmp1
-	add	tmp1, tmp1, #8
-	and	tmp2, src2, #7
-	neg	tmp2, tmp2
-	add	tmp2, tmp2, #8
-	subs	tmp3, tmp1, tmp2
-	csel	pos, tmp1, tmp2, hi /*Choose the maximum. */
-	/*
-	* Here, limit is not less than 8, so directly run .Ltinycmp
-	* without checking the limit.*/
-	sub	limit, limit, pos
-.Ltinycmp:
+	cmp	limit, #16
+	b.hs	.Ltry_misaligned_words
+
+.Lbyte_loop:
+	/* Perhaps we can do better than this.  */
 	ldrb	data1w, [src1], #1
 	ldrb	data2w, [src2], #1
-	subs	pos, pos, #1
-	ccmp	data1w, #1, #0, ne  /* NZCV = 0b0000.  */
-	ccmp	data1w, data2w, #0, cs  /* NZCV = 0b0000.  */
-	b.eq	.Ltinycmp
-	cbnz	pos, 1f /*find the null or unequal...*/
-	cmp	data1w, #1
-	ccmp	data1w, data2w, #0, cs
-	b.eq	.Lstart_align /*the last bytes are equal....*/
-1:
+	subs	limit, limit, #1
+	ccmp	data1w, #1, #0, hi	/* NZCV = 0b0000.  */
+	ccmp	data1w, data2w, #0, cs	/* NZCV = 0b0000.  */
+	b.eq	.Lbyte_loop
+.Ldone:
 	sub	result, data1, data2
 	ret
-
-.Lstart_align:
+	/* Align the SRC1 to a dword by doing a bytewise compare and then do
+	   the dword loop.  */
+.Ltry_misaligned_words:
 	lsr	limit_wd, limit, #3
-	cbz	limit_wd, .Lremain8
-	/*process more leading bytes to make str1 aligned...*/
-	ands	xzr, src1, #7
-	b.eq	.Lrecal_offset
-	add	src1, src1, tmp3	/*tmp3 is positive in this branch.*/
-	add	src2, src2, tmp3
-	ldr	data1, [src1], #8
-	ldr	data2, [src2], #8
+	cbz	count, .Ldo_misaligned
 
-	sub	limit, limit, tmp3
+	neg	count, count
+	and	count, count, #7
+	sub	limit, limit, count
 	lsr	limit_wd, limit, #3
-	subs	limit_wd, limit_wd, #1
 
-	sub	tmp1, data1, zeroones
-	orr	tmp2, data1, #REP8_7f
-	eor	diff, data1, data2  /* Non-zero if differences found.  */
-	csinv	endloop, diff, xzr, ne/*if limit_wd is 0,will finish the cmp*/
-	bics	has_nul, tmp1, tmp2
-	ccmp	endloop, #0, #0, eq /*has_null is ZERO: no null byte*/
-	b.ne	.Lunequal_proc
-	/*How far is the current str2 from the alignment boundary...*/
-	and	tmp3, tmp3, #7
-.Lrecal_offset:
-	neg	pos, tmp3
-.Lloopcmp_proc:
-	/*
-	* Divide the eight bytes into two parts. First,backwards the src2
-	* to an alignment boundary,load eight bytes from the SRC2 alignment
-	* boundary,then compare with the relative bytes from SRC1.
-	* If all 8 bytes are equal,then start the second part's comparison.
-	* Otherwise finish the comparison.
-	* This special handle can garantee all the accesses are in the
-	* thread/task space in avoid to overrange access.
-	*/
-	ldr	data1, [src1,pos]
-	ldr	data2, [src2,pos]
-	sub	tmp1, data1, zeroones
-	orr	tmp2, data1, #REP8_7f
-	bics	has_nul, tmp1, tmp2 /* Non-zero if NUL terminator.  */
-	eor	diff, data1, data2  /* Non-zero if differences found.  */
-	csinv	endloop, diff, xzr, eq
-	cbnz	endloop, .Lunequal_proc
+.Lpage_end_loop:
+	ldrb	data1w, [src1], #1
+	ldrb	data2w, [src2], #1
+	cmp	data1w, #1
+	ccmp	data1w, data2w, #0, cs	/* NZCV = 0b0000.  */
+	b.ne	.Ldone
+	subs	count, count, #1
+	b.hi	.Lpage_end_loop
+
+.Ldo_misaligned:
+	/* Prepare ourselves for the next page crossing.  Unlike the aligned
+	   loop, we fetch 1 less dword because we risk crossing bounds on
+	   SRC2.  */
+	mov	count, #8
+	subs	limit_wd, limit_wd, #1
+	b.lo	.Ldone_loop
+.Lloop_misaligned:
+	and	tmp2, src2, #0xff8
+	eor	tmp2, tmp2, #0xff8
+	cbz	tmp2, .Lpage_end_loop
 
-	/*The second part process*/
 	ldr	data1, [src1], #8
 	ldr	data2, [src2], #8
-	subs	limit_wd, limit_wd, #1
 	sub	tmp1, data1, zeroones
 	orr	tmp2, data1, #REP8_7f
-	eor	diff, data1, data2  /* Non-zero if differences found.  */
-	csinv	endloop, diff, xzr, ne/*if limit_wd is 0,will finish the cmp*/
-	bics	has_nul, tmp1, tmp2
-	ccmp	endloop, #0, #0, eq /*has_null is ZERO: no null byte*/
-	b.eq	.Lloopcmp_proc
-
-.Lunequal_proc:
-	orr	syndrome, diff, has_nul
-	cbz	syndrome, .Lremain8
-.Lcal_cmpresult:
-	/*
-	* reversed the byte-order as big-endian,then CLZ can find the most
-	* significant zero bits.
-	*/
-CPU_LE( rev	syndrome, syndrome )
-CPU_LE( rev	data1, data1 )
-CPU_LE( rev	data2, data2 )
-	/*
-	* For big-endian we cannot use the trick with the syndrome value
-	* as carry-propagation can corrupt the upper bits if the trailing
-	* bytes in the string contain 0x01.
-	* However, if there is no NUL byte in the dword, we can generate
-	* the result directly.  We can't just subtract the bytes as the
-	* MSB might be significant.
-	*/
-CPU_BE( cbnz	has_nul, 1f )
-CPU_BE( cmp	data1, data2 )
-CPU_BE( cset	result, ne )
-CPU_BE( cneg	result, result, lo )
-CPU_BE( ret )
-CPU_BE( 1: )
-	/* Re-compute the NUL-byte detection, using a byte-reversed value.*/
-CPU_BE( rev	tmp3, data1 )
-CPU_BE( sub	tmp1, tmp3, zeroones )
-CPU_BE( orr	tmp2, tmp3, #REP8_7f )
-CPU_BE( bic	has_nul, tmp1, tmp2 )
-CPU_BE( rev	has_nul, has_nul )
-CPU_BE( orr	syndrome, diff, has_nul )
-	/*
-	* The MS-non-zero bit of the syndrome marks either the first bit
-	* that is different, or the top bit of the first zero byte.
-	* Shifting left now will bring the critical information into the
-	* top bits.
-	*/
-	clz	pos, syndrome
-	lsl	data1, data1, pos
-	lsl	data2, data2, pos
-	/*
-	* But we need to zero-extend (char is unsigned) the value and then
-	* perform a signed 32-bit subtraction.
-	*/
-	lsr	data1, data1, #56
-	sub	result, data1, data2, lsr #56
-	ret
-
-.Lremain8:
-	/* Limit % 8 == 0 => all bytes significant.  */
-	ands	limit, limit, #7
-	b.eq	.Lret0
-.Ltiny8proc:
-	ldrb	data1w, [src1], #1
-	ldrb	data2w, [src2], #1
-	subs	limit, limit, #1
+	eor	diff, data1, data2	/* Non-zero if differences found.  */
+	bics	has_nul, tmp1, tmp2	/* Non-zero if NUL terminator.  */
+	ccmp	diff, #0, #0, eq
+	b.ne	.Lnot_limit
+	subs	limit_wd, limit_wd, #1
+	b.pl	.Lloop_misaligned
 
-	ccmp	data1w, #1, #0, ne  /* NZCV = 0b0000.  */
-	ccmp	data1w, data2w, #0, cs  /* NZCV = 0b0000.  */
-	b.eq	.Ltiny8proc
-	sub	result, data1, data2
-	ret
+.Ldone_loop:
+	/* We found a difference or a NULL before the limit was reached.  */
+	and	limit, limit, #7
+	cbz	limit, .Lnot_limit
+	/* Read the last word.  */
+	sub	src1, src1, 8
+	sub	src2, src2, 8
+	ldr	data1, [src1, limit]
+	ldr	data2, [src2, limit]
+	sub	tmp1, data1, zeroones
+	orr	tmp2, data1, #REP8_7f
+	eor	diff, data1, data2	/* Non-zero if differences found.  */
+	bics	has_nul, tmp1, tmp2	/* Non-zero if NUL terminator.  */
+	ccmp	diff, #0, #0, eq
+	b.ne	.Lnot_limit
 
 .Lret0:
 	mov	result, #0
Signature · 1 line
2.17.1
#8 lore raw

PATCH v4 8/14 arm64: Import latest optimization of memcpy

In reply to Oliver Swede

From: Sam Tebbs <[email protected]>

Import the latest memcpy implementation into memcpy, copy_{from, to and in}_user.
The implementation of the user routines is separated into two forms:
one for when UAO is enabled and one for when UAO is disabled, with the two being chosen between with a runtime patch.
This avoids executing the many NOPs emitted when UAO is disabled.

The project containing optimized implementations for various library functions has now been renamed from 'cortex-strings' to 'optimized-routines', and the new upstream source is string/aarch64/memcpy.S as of commit 4c175c8be12 in
https://github.com/ARM-software/optimized-routines.

[ rm: add UAO fixups, streamline copy_exit paths, expand commit message ]

[ os: import newer memcpy algorithm, replace innaccurate fixup routine
with placeholder, update commit message ]

---
 arch/arm64/include/asm/alternative.h |  36 ---
 arch/arm64/lib/copy_from_user.S      | 113 ++++++--
 arch/arm64/lib/copy_in_user.S        | 129 +++++++--
 arch/arm64/lib/copy_template.S       | 375 +++++++++++++++------------
 arch/arm64/lib/copy_template_user.S  |  24 ++
 arch/arm64/lib/copy_to_user.S        | 112 ++++++--
 arch/arm64/lib/copy_user_fixup.S     |   9 +
 arch/arm64/lib/memcpy.S              |  47 ++--
 8 files changed, 552 insertions(+), 293 deletions(-)
 create mode 100644 arch/arm64/lib/copy_template_user.S
 create mode 100644 arch/arm64/lib/copy_user_fixup.S
arch/arm64/include/asm/alternative.h +0 -36
index 5e5dc05d63a0..7ab752104170 100644
@@ -230,36 +230,6 @@ alternative_endif
  * unprivileged instructions, and USER() only works for single instructions.
  */
 #ifdef CONFIG_ARM64_UAO
-	.macro uao_ldp l, reg1, reg2, addr, post_inc
-		alternative_if_not ARM64_HAS_UAO
-8888:			ldp	\reg1, \reg2, [\addr], \post_inc;
-8889:			nop;
-			nop;
-		alternative_else
-			ldtr	\reg1, [\addr];
-			ldtr	\reg2, [\addr, #8];
-			add	\addr, \addr, \post_inc;
-		alternative_endif
-
-		_asm_extable	8888b,\l;
-		_asm_extable	8889b,\l;
-	.endm
-
-	.macro uao_stp l, reg1, reg2, addr, post_inc
-		alternative_if_not ARM64_HAS_UAO
-8888:			stp	\reg1, \reg2, [\addr], \post_inc;
-8889:			nop;
-			nop;
-		alternative_else
-			sttr	\reg1, [\addr];
-			sttr	\reg2, [\addr, #8];
-			add	\addr, \addr, \post_inc;
-		alternative_endif
-
-		_asm_extable	8888b,\l;
-		_asm_extable	8889b,\l;
-	.endm
-
 	.macro uao_user_alternative l, inst, alt_inst, reg, addr, post_inc
 		alternative_if_not ARM64_HAS_UAO
 8888:			\inst	\reg, [\addr], \post_inc;
@@ -272,12 +242,6 @@ alternative_endif
 		_asm_extable	8888b,\l;
 	.endm
 #else
-	.macro uao_ldp l, reg1, reg2, addr, post_inc
-		USER(\l, ldp \reg1, \reg2, [\addr], \post_inc)
-	.endm
-	.macro uao_stp l, reg1, reg2, addr, post_inc
-		USER(\l, stp \reg1, \reg2, [\addr], \post_inc)
-	.endm
 	.macro uao_user_alternative l, inst, alt_inst, reg, addr, post_inc
 		USER(\l, \inst \reg, [\addr], \post_inc)
 	.endm
arch/arm64/lib/copy_from_user.S +87 -26

137 lines not on this page · raw message

arch/arm64/lib/copy_in_user.S +104 -25

154 lines not on this page · raw message

arch/arm64/lib/copy_template.S +211 -164

385 lines not on this page · raw message

arch/arm64/lib/copy_template_user.S new file +24 -0
new file mode 100644
index 000000000000..3db24dcdab05
@@ -0,0 +1,24 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+
+#define L(l) .L ## l
+
+	alternative_if_not ARM64_HAS_UAO
+	b	L(copy_non_uao)
+	alternative_else_nop_endif
+#include "copy_template.S"
+
+#define ldp1 ldp1_nuao
+#define ldp1_pre ldp1_pre_nuao
+#define stp1 stp1_nuao
+#define stp1_pre stp1_pre_nuao
+#define ldr1 ldr1_nuao
+#define str1 str1_nuao
+#define ldrb1 ldrb1_nuao
+#define strb1 strb1_nuao
+#define ldrb1_reg ldrb1_nuao_reg
+#define strb1_reg strb1_nuao_reg
+
+L(copy_non_uao):
+#undef L
+#define L(l) .Lnuao ## l
+#include "copy_template.S"
arch/arm64/lib/copy_to_user.S +87 -25

136 lines not on this page · raw message

arch/arm64/lib/copy_user_fixup.S new file +9 -0
new file mode 100644
index 000000000000..117c37598691
@@ -0,0 +1,9 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+
+addr	.req	x15
+.section .fixup,"ax"
+.align	2
+9998:
+	// TODO: add accurate fixup
+	ret
+
arch/arm64/lib/memcpy.S +30 -17

73 lines not on this page · raw message

Signature · 1 line
2.17.1
#9 lore raw

PATCH v4 9/14 arm64: Tidy up _asm_extable_faultaddr usage

In reply to Oliver Swede

From: Robin Murphy <[email protected]>

To match the way the USER() shorthand wraps _asm_extable entries, introduce USER_F() to wrap _asm_extable_faultaddr and clean up a bit.

---
 arch/arm64/include/asm/assembler.h |  4 ++
 arch/arm64/lib/copy_from_user.S    | 37 ++++++---------
 arch/arm64/lib/copy_in_user.S      | 72 ++++++++++--------------------
 arch/arm64/lib/copy_to_user.S      | 33 +++++---------
 4 files changed, 52 insertions(+), 94 deletions(-)
arch/arm64/include/asm/assembler.h +4 -0
index 438382a277c8..84ddf25546eb 100644
@@ -142,6 +142,10 @@ alternative_endif
 9999:	x;					\
 	_asm_extable	9999b, l
 
+#define USER_F(l, x...)				\
+9999:	x;					\
+	_asm_extable_faultaddr	9999b, l
+
 /*
  * Register aliases.
  */
arch/arm64/lib/copy_from_user.S +13 -24
index 86945e84c009..fa319f27a42b 100644
@@ -19,8 +19,8 @@
  * Returns:
  *	x0 - bytes not copied
  */
-	8888: ldtrb \reg, [\ptr, \offset]
-	_asm_extable_faultaddr	8888b,9998f;
+	.macro ldrb1 reg, ptr, offset=0
+	USER_F(9998f, ldtrb \reg, [\ptr, \offset])
 	.endm
 
 	.macro strb1 reg, ptr, offset=0
@@ -29,9 +29,8 @@
 
 	.macro ldrb1_reg reg, ptr, offset
 	add \ptr, \ptr, \offset
-	8888: ldtrb \reg, [\ptr]
+	USER_F(9998f, ldtrb \reg, [\ptr])
 	sub \ptr, \ptr, \offset
-	_asm_extable_faultaddr	8888b,9998f;
 	.endm
 
 	.macro strb1_reg reg, ptr, offset
@@ -39,8 +38,7 @@
 	.endm
 
 	.macro ldr1 reg, ptr, offset=0
-	8888: ldtr \reg, [\ptr, \offset]
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, ldtr \reg, [\ptr, \offset])
 	.endm
 
 	.macro str1 reg, ptr, offset=0
@@ -48,10 +46,8 @@
 	.endm
 
 	.macro ldp1 regA, regB, ptr, offset=0
-	8888: ldtr \regA, [\ptr, \offset]
-	8889: ldtr \regB, [\ptr, \offset + 8]
-	_asm_extable_faultaddr	8888b,9998f;
-	_asm_extable_faultaddr	8889b,9998f;
+	USER_F(9998f, ldtr \regA, [\ptr, \offset])
+	USER_F(9998f, ldtr \regB, [\ptr, \offset + 8])
 	.endm
 
 	.macro stp1 regA, regB, ptr, offset=0
@@ -59,11 +55,9 @@
 	.endm
 
 	.macro ldp1_pre regA, regB, ptr, offset
-	8888: ldtr \regA, [\ptr, \offset]
-	8889: ldtr \regB, [\ptr, \offset + 8]
+	USER_F(9998f, ldtr \regA, [\ptr, \offset])
+	USER_F(9998f, ldtr \regB, [\ptr, \offset + 8])
 	add \ptr, \ptr, \offset
-	_asm_extable_faultaddr	8888b,9998f;
-	_asm_extable_faultaddr	8889b,9998f;
 	.endm
 
 	.macro stp1_pre regA, regB, ptr, offset
@@ -71,8 +65,7 @@
 	.endm
 
 	.macro ldrb1_nuao reg, ptr, offset=0
-	8888: ldrb \reg, [\ptr, \offset]
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, ldrb \reg, [\ptr, \offset])
 	.endm
 
 	.macro strb1_nuao reg, ptr, offset=0
@@ -80,8 +73,7 @@
 	.endm
 
 	.macro ldrb1_nuao_reg reg, ptr, offset=0
-	8888: ldrb \reg, [\ptr, \offset]
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, ldrb \reg, [\ptr, \offset])
 	.endm
 
 	.macro strb1_nuao_reg reg, ptr, offset=0
@@ -89,8 +81,7 @@
 	.endm
 
 	.macro ldr1_nuao reg, ptr, offset=0
-	8888: ldr \reg, [\ptr, \offset]
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, ldr \reg, [\ptr, \offset])
 	.endm
 
 	.macro str1_nuao reg, ptr, offset=0
@@ -98,8 +89,7 @@
 	.endm
 
 	.macro ldp1_nuao  regA, regB, ptr, offset=0
-	8888: ldp \regA, \regB, [\ptr, \offset]
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, ldp \regA, \regB, [\ptr, \offset])
 	.endm
 
 	.macro stp1_nuao regA, regB, ptr, offset=0
@@ -107,8 +97,7 @@
 	.endm
 
 	.macro ldp1_pre_nuao regA, regB, ptr, offset
-	8888: ldp \regA, \regB, [\ptr, \offset]!
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, ldp \regA, \regB, [\ptr, \offset]!)
 	.endm
 
 	.macro stp1_pre_nuao regA, regB, ptr, offset
arch/arm64/lib/copy_in_user.S +24 -48

141 lines not on this page · raw message

arch/arm64/lib/copy_to_user.S +11 -22
index 6b4742cac083..23af6af254da 100644
@@ -24,8 +24,7 @@
 	.endm
 
 	.macro strb1 reg, ptr, offset=0
-	8888: sttrb \reg, [\ptr, \offset]
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, sttrb \reg, [\ptr, \offset])
 	.endm
 
 	.macro ldrb1_reg reg, ptr, offset
@@ -34,9 +33,8 @@
 
 	.macro strb1_reg reg, ptr, offset
 	add \ptr, \ptr, \offset
-	8888: sttrb \reg, [\ptr]
+	USER_F(9998f, sttrb \reg, [\ptr])
 	sub \ptr, \ptr, \offset
-	_asm_extable_faultaddr	8888b,9998f;
 	.endm
 
 	.macro ldr1 reg, ptr, offset=0
@@ -44,8 +42,7 @@
 	.endm
 
 	.macro str1 reg, ptr, offset=0
-	8888: sttr \reg, [\ptr, \offset]
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, sttr \reg, [\ptr, \offset])
 	.endm
 
 	.macro ldp1 regA, regB, ptr, offset=0
@@ -53,10 +50,8 @@
 	.endm
 
 	.macro stp1 regA, regB, ptr, offset=0
-	8888: sttr \regA, [\ptr, \offset]
-	8889: sttr \regB, [\ptr, \offset + 8]
-	_asm_extable_faultaddr	8888b,9998f;
-	_asm_extable_faultaddr	8889b,9998f;
+	USER_F(9998f, sttr \regA, [\ptr, \offset])
+	USER_F(9998f, sttr \regB, [\ptr, \offset + 8])
 	.endm
 
 	.macro ldp1_pre regA, regB, ptr, offset
@@ -64,11 +59,9 @@
 	.endm
 
 	.macro stp1_pre regA, regB, ptr, offset
-	8888: sttr \regA, [\ptr, \offset]
-	8889: sttr \regB, [\ptr, \offset + 8]
+	USER_F(9998f, sttr \regA, [\ptr, \offset])
+	USER_F(9998f, sttr \regB, [\ptr, \offset + 8])
 	add \ptr, \ptr, \offset
-	_asm_extable_faultaddr	8888b,9998f;
-	_asm_extable_faultaddr	8889b,9998f;
 	.endm
 
 	.macro ldrb1_nuao reg, ptr, offset=0
@@ -76,8 +69,7 @@
 	.endm
 
 	.macro strb1_nuao reg, ptr, offset=0
-	8888: strb \reg, [\ptr, \offset]
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, strb \reg, [\ptr, \offset])
 	.endm
 
 	.macro ldrb1_nuao_reg reg, ptr, offset=0
@@ -93,8 +85,7 @@
 	.endm
 
 	.macro str1_nuao reg, ptr, offset=0
-	8888: str \reg, [\ptr, \offset]
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, str \reg, [\ptr, \offset])
 	.endm
 
 	.macro ldp1_nuao  regA, regB, ptr, offset=0
@@ -106,13 +97,11 @@
 	.endm
 
 	.macro stp1_nuao regA, regB, ptr, offset=0
-	8888: stp \regA, \regB, [\ptr, \offset]
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, stp \regA, \regB, [\ptr, \offset])
 	.endm
 
 	.macro stp1_pre_nuao regA, regB, ptr, offset
-	8888: stp \regA, \regB, [\ptr, \offset]!
-	_asm_extable_faultaddr	8888b,9998f;
+	USER_F(9998f, stp \regA, \regB, [\ptr, \offset]!)
 	.endm
 
 	.macro copy_exit
Signature · 1 line
2.17.1
#10 lore raw

PATCH v4 10/14 arm64: Store the arguments to copy_*_user on the stack

In reply to Oliver Swede

This preserves the initial arguments of the user copy calls so that they can be restored by the fixup routines.

The values in the relevant three registers (x0/dstin, x1/src, x2/count) may be modified by the optimized memcpy algorithm for large copy sizes, and this stores them before it begins executing.

The stack is used instead of other general-purpose registers due to resource constraints: the algorithm is optimized with respect to the Procedure Call Standard in the Arm ABI, which assumes that x0-x17 can be used as scratch registers and utilizes all of them during copying, but leaves alone the rest that have specific uses in the broader system. As there are no more temporary registers, the stack can be used to preserve the initial arguments to provide fixup routines with more information to use in the calculation of the number of bytes that failed to copy.

The stack pointer is restored to its initial position, either from the fixup code in the case of a fault, or at the end of the copy algorithm otherwise (uaccess_finish is extended to restore the sp, and this code is also moved to copy_template_user.S as it is common to all of the copy routines that access userspace memory).

---
 arch/arm64/lib/copy_from_user.S     | 3 ---
 arch/arm64/lib/copy_in_user.S       | 3 ---
 arch/arm64/lib/copy_template_user.S | 6 ++++++
 arch/arm64/lib/copy_to_user.S       | 3 ---
 arch/arm64/lib/copy_user_fixup.S    | 1 +
 5 files changed, 7 insertions(+), 9 deletions(-)
arch/arm64/lib/copy_from_user.S +0 -3
index fa319f27a42b..cd3042e98394 100644
@@ -110,9 +110,6 @@
 
 SYM_FUNC_START(__arch_copy_from_user)
 #include "copy_template_user.S"
-.Luaccess_finish:
-	mov	x0, #0
-	ret
 SYM_FUNC_END(__arch_copy_from_user)
 EXPORT_SYMBOL(__arch_copy_from_user)
 #include "copy_user_fixup.S"
arch/arm64/lib/copy_in_user.S +0 -3
index 6b9bb6091dd8..fe035e513b34 100644
@@ -117,9 +117,6 @@
 
 SYM_FUNC_START(__arch_copy_in_user)
 #include "copy_template_user.S"
-.Luaccess_finish:
-	mov	x0, #0
-	ret
 SYM_FUNC_END(__arch_copy_in_user)
 EXPORT_SYMBOL(__arch_copy_in_user)
 #include "copy_user_fixup.S"
arch/arm64/lib/copy_template_user.S +6 -0
index 3db24dcdab05..1d13daf314b0 100644
@@ -21,4 +21,10 @@
 L(copy_non_uao):
 #undef L
 #define L(l) .Lnuao ## l
+	str     x2, [sp, #-16]!		// count
+	stp     x0, x1, [sp, #-16]!	// dstin, src
 #include "copy_template.S"
+.Luaccess_finish:
+	add	sp, sp, 32
+	mov	x0, #0
+	ret
arch/arm64/lib/copy_to_user.S +0 -3
index 23af6af254da..de0af211b3ba 100644
@@ -110,9 +110,6 @@
 
 SYM_FUNC_START(__arch_copy_to_user)
 #include "copy_template_user.S"
-.Luaccess_finish:
-	mov	x0, #0
-	ret
 SYM_FUNC_END(__arch_copy_to_user)
 EXPORT_SYMBOL(__arch_copy_to_user)
 #include "copy_user_fixup.S"
arch/arm64/lib/copy_user_fixup.S +1 -0
index 117c37598691..fe9f5ac19605 100644
@@ -5,5 +5,6 @@ addr	.req	x15
 .align	2
 9998:
 	// TODO: add accurate fixup
+	add	sp, sp, 32
 	ret
 
Signature · 1 line
2.17.1
#11 lore raw

PATCH v4 11/14 arm64: Use additional memcpy macros and fixups

In reply to Oliver Swede

The Arm-provided memcpy routine has been updated with more recent optimizations. The kernel's helper functions for copying to/from user space memory, which import this algorithm and create exception table entries for instructions that reference user space, require new recovery code to accurately determine the number of bytes that successfully copied before a page fault.

This adds new macro instantiations in the copy template, and corresponding definitions to each of the copy_*_user() functions. This allows more fixup routines to be added so that an accurate value for the number of bytes that failed to copy can be returned.

This increases the flexibility of the fixup code, as certain propertes can be encapsulated by the mapping to implicitly provide the routines with more information than would otherwise be available.

In the case of the current memcpy, the number of bytes already copied depends highly on the type of instruction that caused the fault (load vs. store), so the use of separate fixups enables specific routines for each case. This is an alternative to (for instance) using a single fixup for both the loads and stores, as this would be subject to issues relating to overlapping src and dst buffers in copy_in_user().
The outcome also depends largely on other factors, such as if the target address is specified relative to the start or end of the buffer, and whether or not the store targets are guaranteed to be 16-byte aligned.

These characteristics (obtained from analysing the copy algorithm) enable fixups to be written that are modular and accurate for each case. The fixup logic should be straightforward to modify in the future, e.g. if there are further improvements to the memcpy routine.

---
 arch/arm64/lib/copy_from_user.S     | 197 ++++++++++++++++++++++++--
 arch/arm64/lib/copy_in_user.S       | 205 ++++++++++++++++++++++++++--
 arch/arm64/lib/copy_template.S      | 126 ++++++++---------
 arch/arm64/lib/copy_template_user.S |  20 +++
 arch/arm64/lib/copy_to_user.S       | 179 +++++++++++++++++++++++-
 arch/arm64/lib/copy_user_fixup.S    |  13 +-
 arch/arm64/lib/memcpy.S             |  80 +++++++++++
 7 files changed, 728 insertions(+), 92 deletions(-)
arch/arm64/lib/copy_from_user.S +184 -13
index cd3042e98394..0056d1fc06eb 100644
@@ -19,17 +19,46 @@
  * Returns:
  *	x0 - bytes not copied
  */
+
 	.macro ldrb1 reg, ptr, offset=0
-	USER_F(9998f, ldtrb \reg, [\ptr, \offset])
+	USER_F(9993f, ldtrb \reg, [\ptr, \offset])
+	.endm
+
+	.macro ldrb2 reg, ptr, offset=0
+	USER_F(9994f, ldtrb \reg, [\ptr, \offset])
+	.endm
+
+	.macro ldrb3 reg, ptr, offset=0
+	USER_F(9995f, ldtrb \reg, [\ptr, \offset])
 	.endm
 
 	.macro strb1 reg, ptr, offset=0
 	strb \reg, [\ptr, \offset]
 	.endm
 
+	.macro strb2 reg, ptr, offset=0
+	strb \reg, [\ptr, \offset]
+	.endm
+
+	.macro strb3 reg, ptr, offset=0
+	strb \reg, [\ptr, \offset]
+	.endm
+
 	.macro ldrb1_reg reg, ptr, offset
 	add \ptr, \ptr, \offset
-	USER_F(9998f, ldtrb \reg, [\ptr])
+	USER_F(9993f, ldtrb \reg, [\ptr])
+	sub \ptr, \ptr, \offset
+	.endm
+
+	.macro ldrb2_reg reg, ptr, offset
+	add \ptr, \ptr, \offset
+	USER_F(9994f, ldtrb \reg, [\ptr])
+	sub \ptr, \ptr, \offset
+	.endm
+
+	.macro ldrb3_reg reg, ptr, offset
+	add \ptr, \ptr, \offset
+	USER_F(9995f, ldtrb \reg, [\ptr])
 	sub \ptr, \ptr, \offset
 	.endm
 
@@ -37,26 +66,80 @@
 	strb \reg, [\ptr, \offset]
 	.endm
 
+	.macro strb2_reg reg, ptr, offset
+	strb \reg, [\ptr, \offset]
+	.endm
+
+	.macro strb3_reg reg, ptr, offset
+	strb \reg, [\ptr, \offset]
+	.endm
+
 	.macro ldr1 reg, ptr, offset=0
-	USER_F(9998f, ldtr \reg, [\ptr, \offset])
+	USER_F(9993f, ldtr \reg, [\ptr, \offset])
+	.endm
+
+	.macro ldr2 reg, ptr, offset=0
+	USER_F(9994f, ldtr \reg, [\ptr, \offset])
+	.endm
+
+	.macro ldr3 reg, ptr, offset=0
+	USER_F(9995f, ldtr \reg, [\ptr, \offset])
 	.endm
 
 	.macro str1 reg, ptr, offset=0
 	str \reg, [\ptr, \offset]
 	.endm
 
+	.macro str2 reg, ptr, offset=0
+	str \reg, [\ptr, \offset]
+	.endm
+
+	.macro str3 reg, ptr, offset=0
+	str \reg, [\ptr, \offset]
+	.endm
+
 	.macro ldp1 regA, regB, ptr, offset=0
-	USER_F(9998f, ldtr \regA, [\ptr, \offset])
-	USER_F(9998f, ldtr \regB, [\ptr, \offset + 8])
+	USER_F(9993f, ldtr \regA, [\ptr, \offset])
+	USER_F(9993f, ldtr \regB, [\ptr, \offset + 8])
+	.endm
+
+	.macro ldp2 regA, regB, ptr, offset=0
+	USER_F(9994f, ldtr \regA, [\ptr, \offset])
+	USER_F(9994f, ldtr \regB, [\ptr, \offset + 8])
+	.endm
+
+	.macro ldp3 regA, regB, ptr, offset=0
+	USER_F(9995f, ldtr \regA, [\ptr, \offset])
+	USER_F(9995f, ldtr \regB, [\ptr, \offset + 8])
 	.endm
 
 	.macro stp1 regA, regB, ptr, offset=0
 	stp \regA, \regB, [\ptr, \offset]
 	.endm
 
+	.macro stp2 regA, regB, ptr, offset=0
+	stp \regA, \regB, [\ptr, \offset]
+	.endm
+
+	.macro stp3 regA, regB, ptr, offset=0
+	stp \regA, \regB, [\ptr, \offset]
+	.endm
+
 	.macro ldp1_pre regA, regB, ptr, offset
-	USER_F(9998f, ldtr \regA, [\ptr, \offset])
-	USER_F(9998f, ldtr \regB, [\ptr, \offset + 8])
+	USER_F(9993f, ldtr \regA, [\ptr, \offset])
+	USER_F(9993f, ldtr \regB, [\ptr, \offset + 8])
+	add \ptr, \ptr, \offset
+	.endm
+
+	.macro ldp2_pre regA, regB, ptr, offset
+	USER_F(9994f, ldtr \regA, [\ptr, \offset])
+	USER_F(9994f, ldtr \regB, [\ptr, \offset + 8])
+	add \ptr, \ptr, \offset
+	.endm
+
+	.macro ldp3_pre regA, regB, ptr, offset
+	USER_F(9995f, ldtr \regA, [\ptr, \offset])
+	USER_F(9995f, ldtr \regB, [\ptr, \offset + 8])
 	add \ptr, \ptr, \offset
 	.endm
 
@@ -64,46 +147,134 @@
 	stp \regA, \regB, [\ptr, \offset]!
 	.endm
 
+	.macro stp2_pre regA, regB, ptr, offset
+	stp \regA, \regB, [\ptr, \offset]!
+	.endm
+
+	.macro stp3_pre regA, regB, ptr, offset
+	stp \regA, \regB, [\ptr, \offset]!
+	.endm
+
 	.macro ldrb1_nuao reg, ptr, offset=0
-	USER_F(9998f, ldrb \reg, [\ptr, \offset])
+	USER_F(9993f, ldrb \reg, [\ptr, \offset])
+	.endm
+
+	.macro ldrb2_nuao reg, ptr, offset=0
+	USER_F(9994f, ldrb \reg, [\ptr, \offset])
+	.endm
+
+	.macro ldrb3_nuao reg, ptr, offset=0
+	USER_F(9995f, ldrb \reg, [\ptr, \offset])
 	.endm
 
 	.macro strb1_nuao reg, ptr, offset=0
 	strb \reg, [\ptr, \offset]
 	.endm
 
+	.macro strb2_nuao reg, ptr, offset=0
+	strb \reg, [\ptr, \offset]
+	.endm
+
+	.macro strb3_nuao reg, ptr, offset=0
+	strb \reg, [\ptr, \offset]
+	.endm
+
 	.macro ldrb1_nuao_reg reg, ptr, offset=0
-	USER_F(9998f, ldrb \reg, [\ptr, \offset])
+	USER_F(9993f, ldrb \reg, [\ptr, \offset])
+	.endm
+
+	.macro ldrb2_nuao_reg reg, ptr, offset=0
+	USER_F(9994f, ldrb \reg, [\ptr, \offset])
+	.endm
+
+	.macro ldrb3_nuao_reg reg, ptr, offset=0
+	USER_F(9995f, ldrb \reg, [\ptr, \offset])
 	.endm
 
 	.macro strb1_nuao_reg reg, ptr, offset=0
 	strb \reg, [\ptr, \offset]
 	.endm
 
+	.macro strb2_nuao_reg reg, ptr, offset=0
+	strb \reg, [\ptr, \offset]
+	.endm
+
+	.macro strb3_nuao_reg reg, ptr, offset=0
+	strb \reg, [\ptr, \offset]
+	.endm
+
 	.macro ldr1_nuao reg, ptr, offset=0
-	USER_F(9998f, ldr \reg, [\ptr, \offset])
+	USER_F(9993f, ldr \reg, [\ptr, \offset])
+	.endm
+
+	.macro ldr2_nuao reg, ptr, offset=0
+	USER_F(9994f, ldr \reg, [\ptr, \offset])
+	.endm
+
+	.macro ldr3_nuao reg, ptr, offset=0
+	USER_F(9995f, ldr \reg, [\ptr, \offset])
 	.endm
 
 	.macro str1_nuao reg, ptr, offset=0
 	str \reg, [\ptr, \offset]
 	.endm
 
-	.macro ldp1_nuao  regA, regB, ptr, offset=0
-	USER_F(9998f, ldp \regA, \regB, [\ptr, \offset])
+	.macro str2_nuao reg, ptr, offset=0
+	str \reg, [\ptr, \offset]
+	.endm
+
+	.macro str3_nuao reg, ptr, offset=0
+	str \reg, [\ptr, \offset]
+	.endm
+
+	.macro ldp1_nuao regA, regB, ptr, offset=0
+	USER_F(9993f, ldp \regA, \regB, [\ptr, \offset])
+	.endm
+
+	.macro ldp2_nuao regA, regB, ptr, offset=0
+	USER_F(9994f, ldp \regA, \regB, [\ptr, \offset])
+	.endm
+
+	.macro ldp3_nuao regA, regB, ptr, offset=0
+	USER_F(9995f, ldp \regA, \regB, [\ptr, \offset])
 	.endm
 
 	.macro stp1_nuao regA, regB, ptr, offset=0
 	stp \regA, \regB, [\ptr, \offset]
 	.endm
 
+	.macro stp2_nuao regA, regB, ptr, offset=0
+	stp \regA, \regB, [\ptr, \offset]
+	.endm
+
+	.macro stp3_nuao regA, regB, ptr, offset=0
+	stp \regA, \regB, [\ptr, \offset]
+	.endm
+
 	.macro ldp1_pre_nuao regA, regB, ptr, offset
-	USER_F(9998f, ldp \regA, \regB, [\ptr, \offset]!)
+	USER_F(9993f, ldp \regA, \regB, [\ptr, \offset]!)
+	.endm
+
+	.macro ldp2_pre_nuao regA, regB, ptr, offset
+	USER_F(9994f, ldp \regA, \regB, [\ptr, \offset]!)
+	.endm
+
+	.macro ldp3_pre_nuao regA, regB, ptr, offset
+	USER_F(9995f, ldp \regA, \regB, [\ptr, \offset]!)
 	.endm
 
 	.macro stp1_pre_nuao regA, regB, ptr, offset
 	stp \regA, \regB, [\ptr, \offset]!
 	.endm
 
+	.macro stp2_pre_nuao regA, regB, ptr, offset
+	stp \regA, \regB, [\ptr, \offset]!
+	.endm
+
+	.macro stp3_pre_nuao regA, regB, ptr, offset
+	stp \regA, \regB, [\ptr, \offset]!
+	.endm
+
 	.macro copy_exit
 	b	.Luaccess_finish
 	.endm
arch/arm64/lib/copy_in_user.S +193 -12

286 lines not on this page · raw message

arch/arm64/lib/copy_template.S +64 -62

216 lines not on this page · raw message

arch/arm64/lib/copy_template_user.S +20 -0
index 1d13daf314b0..f36c77738e42 100644
@@ -8,15 +8,35 @@
 #include "copy_template.S"
 
 #define ldp1 ldp1_nuao
+#define ldp2 ldp2_nuao
+#define ldp3 ldp3_nuao
 #define ldp1_pre ldp1_pre_nuao
+#define ldp2_pre ldp2_pre_nuao
+#define ldp3_pre ldp3_pre_nuao
 #define stp1 stp1_nuao
+#define stp2 stp2_nuao
+#define stp3 stp3_nuao
 #define stp1_pre stp1_pre_nuao
+#define stp2_pre stp2_pre_nuao
+#define stp3_pre stp3_pre_nuao
 #define ldr1 ldr1_nuao
+#define ldr2 ldr2_nuao
+#define ldr3 ldr3_nuao
 #define str1 str1_nuao
+#define str2 str2_nuao
+#define str3 str3_nuao
 #define ldrb1 ldrb1_nuao
+#define ldrb2 ldrb2_nuao
+#define ldrb3 ldrb3_nuao
 #define strb1 strb1_nuao
+#define strb2 strb2_nuao
+#define strb3 strb3_nuao
 #define ldrb1_reg ldrb1_nuao_reg
+#define ldrb2_reg ldrb2_nuao_reg
+#define ldrb3_reg ldrb3_nuao_reg
 #define strb1_reg strb1_nuao_reg
+#define strb2_reg strb2_nuao_reg
+#define strb3_reg strb3_nuao_reg
 
 L(copy_non_uao):
 #undef L
arch/arm64/lib/copy_to_user.S +175 -4

253 lines not on this page · raw message

arch/arm64/lib/copy_user_fixup.S +12 -1
index fe9f5ac19605..f878c8831b14 100644
@@ -3,8 +3,19 @@
 addr	.req	x15
 .section .fixup,"ax"
 .align	2
+9993:
+9994:
+9995:
+9996:
+9997:
 9998:
-	// TODO: add accurate fixup
+	/* Retrieve useful information & free the stack area */
+	ldr	count, [sp, #16]	// x2
 	add	sp, sp, 32
+	/*
+	 * Return the initial count as the (under-estimated) number
+	 * of bytes that failed to copy
+	 */
+	mov	x0, count
 	ret
 
arch/arm64/lib/memcpy.S +80 -0

122 lines not on this page · raw message

Signature · 1 line
2.17.1
#12 lore raw

PATCH v4 12/14 arm64: Add fixup routines for usercopy load exceptions

In reply to Oliver Swede

This adds the fixup routines for exceptions that occur on load operations while copying, by providing the calling code with a more accurate value for the number of bytes that failed to copy.

The three routines for load exceptions work together to analyse the position of the fault relative to the start or the end of the buffer, and backtrack from the optimized memcpy algorithm to determine if some number of bytes has already been successfully copied.

The new template uses out-of-order copying, and this fixup routine is specific to the latest memcpy implementation. It is assumed there is no requirement to follow through with the copying of data that may reside in temporary registers on a fault, as this would greatly increase the fixup's complexity.

---
 arch/arm64/lib/copy_user_fixup.S | 172 ++++++++++++++++++++++++++++++-
 1 file changed, 168 insertions(+), 4 deletions(-)
arch/arm64/lib/copy_user_fixup.S +168 -4
index f878c8831b14..ae94b492129d 100644
@@ -3,19 +3,183 @@
 addr	.req	x15
 .section .fixup,"ax"
 .align	2
+
+/*
+ * These fixup routines assume no requirements to follow through
+ * with copying any intermediate data in registers.
+ *
+ * The fixup subroutine that is directed to depends on the
+ * characteristics of the target address in the faulting instruction:
+ * if it is an address specified relative to the start or end of the
+ * buffer, and (in the case of copy sizes >128B) if the target address
+ * is 16-byte aligned.
+ */
+
+/*
+ * The following three routines are directed to from faults on load
+ * instructions. For 9993, 9994 and 9995, in each case nothing has
+ * been copied if either:
+ *	a) the copy size is less than 128 bytes - the algorithm does
+ *	   not interleave load/store instructions and starts with the
+ *	   load instructions, so for these sizes nothing has been
+ *	   copied on a load fault and the full buffer width srcend-src
+ *	   can be returned, or
+ *	b) the copy size is greater than or equal to 128 bytes and the
+ *	   src and dst buffers overlap - this would result in a
+ *	   backwards copy, however the calling code expects the return
+ *	   value, the numbwe of bytes not copied, to be from an in-order
+ *	   perspective.
+ */
 9993:
+	/*
+	 * This is reached from load instructions that are specified
+	 * relative to the start of a user space memory buffer, and
+	 * provide target addresses that are 16-byte aligned.
+	 */
+
+	/* Retrieve original params from stack */
+	ldp	dst, src, [sp], #16	// dst: x3, src: x1
+	ldr	count, [sp], #16	// count: x2
+	add	srcend, src, count
+	add	dstend, dst, count
+
+	/*
+	 * If the fault occurred with a copy size larger than 128 bytes,
+	 * and if the buffers do not overlap (where it would start copying
+	 * from the end), it still has not copied anything at this point
+	 * as the fault occurred at the start of the algorithm's subroutine
+	 * for long forward copies, and no bytes have been stored at this
+	 * point.
+	 */
+	b	L(none_copied)
 9994:
+	/*
+	 * This is reached from load instructions that are specified
+	 * relative to the end of a user space memory buffer, and
+	 * where the store targets are not guaranteed to be 16-byte
+	 * aligned.
+	 */
+
+	/* Save the current (adjusted) dst for later restoration */
+	mov	tmp1, dst
+
+	/* Retrieve original params from stack */
+	ldp	dst, src, [sp], #16	// dst: x3, src: x1
+	ldr	count, [sp], #16	// count: x2
+	add	srcend, src, count
+	add	dstend, dst, count
+
+	/* Copy size <= 128 bytes */
+	cmp	count, 128
+	b.le	L(none_copied)
+
+	/*
+	 * Overlapping buffers:
+	 * (src <= dst && dst < srcend) || (dst <= src && src < dstend)
+	 */
+	cmp	src, dst
+	ccmp	dst, srcend, #0, le
+	b.lt	L(none_copied)
+	cmp	dst, src
+	ccmp	src, dstend, #0, le
+	b.lt	L(none_copied)
+
+	/*
+	 * In a long copy over 128B, in the case of a fault at an
+	 * address relative to the end of the buffer, this will
+	 * occur in copy64_from_end() and thus the last 64 bytes.
+	 *
+	 * The algorithm first copies 16 bytes, and then aligns dst
+	 * to 16 bytes for the rest of the copy.
+	 * At the end of each loop64(), dst is incremented by 64.
+	 *
+	 * This takes the minimum of (16, max((current_dst + 16 - original_dst), 16) 
+	 */
+
+	/* already copied max((tmp1 + 16 - dst), 16) */
+	add	tmp1, tmp1, 16
+	sub	tmp1, tmp1, dst
+	mov	x0, 16
+	cmp	x0, tmp1
+	csel	tmp1, x0, tmp1, ge // threshold it has copied up to
+	sub	x0, srcend, 64 // base of final 64 bytes
+	sub	x0, addr, x0 // fault offset in final 64 bytes
+	bic	x0, x0, 15 // discard bytes between target and fault addr)
+	add	x0, x0, tmp1 // copied up to here
+	sub	x0, count, x0
+	b	L(end_fixup)
+
 9995:
+	/*
+	 * This is reached from load instructions that are specified
+	 * relative to the start of a user space memory buffer, and
+	 * where the store targets are guaranteed to be 16-byte aligned.
+	 */
+	
+	/* Save the current (adjusted) dst for later restoration */
+	mov	tmp1, dst
+
+	/* Retrieve original params from stack */
+	ldp	dst, src, [sp], #16	// dst: x3, src: x1
+	ldr	count, [sp], #16	// count: x2
+	add	srcend, src, count
+	add	dstend, dst, count
+
+	/* Copy size <= 128 bytes */
+	cmp	count, 128
+	b.le	L(none_copied)
+
+	/*
+	 * Overlapping buffers:
+	 * (src <= dst && dst < srcend) || (dst <= src && src < dstend)
+	 */
+	cmp	src, dst
+	ccmp	dst, srcend, #0, le
+	b.lt	L(none_copied)
+	cmp	dst, src
+	ccmp	src, dstend, #0, le
+	b.lt	L(none_copied)
+
+	/*
+	 * These stores provide target addresses that are 16-byte
+	 * aligned, and this allows for a mapping:
+	 */
+	and	x0, dst, 15
+	sub	x0, src, x0 // aligned src: (src - (dst & 15))
+	sub	x0, addr, src // offset of fault relative to aligned src
+	bic	x0, x0, 15 // round down to base of faulting 16B chunk
+	mov	tmp1, 16
+	cmp	x0, tmp1
+	/* fault_offset < 32 -> count */
+	b.le	L(none_copied)
+	mov	tmp1, 64
+	cmp	x0, tmp1
+	mov	tmp1, x0
+	sub	x0, count, 16
+	/* 32 <= fault_offset < 80 -> count-16 */
+	b.le	L(end_fixup)
+	mov	x0, tmp1
+	sub	x0, x0, 48 /* faulted in a loop64 subroutine */
+	and	tmp1, dst, 15
+	sub	x0, x0, tmp1
+	sub	x0, count, x0
+	/*
+	 * fault_offset >= 80 ->
+	 * count-((fault_offset&~15)-48)-(dst & 15))
+	 */
+	b	L(end_fixup)
+	
 9996:
 9997:
 9998:
-	/* Retrieve useful information & free the stack area */
-	ldr	count, [sp, #16]	// x2
+	/* Retrieve info from stack */
+	ldr	count, [sp], #16	// count: x2
 	add	sp, sp, 32
+L(none_copied):
 	/*
-	 * Return the initial count as the (under-estimated) number
+	 * Return the initial count as the number
 	 * of bytes that failed to copy
 	 */
 	mov	x0, count
+L(end_fixup):
 	ret
-
Signature · 1 line
2.17.1
#13 lore raw

PATCH v4 13/14 arm64: Add fixup routines for usercopy store exceptions

In reply to Oliver Swede

This adds the fixup routines for exceptions that occur on store operations while copying, by providing the calling code with a more accurate value for the number of bytes that failed to copy, rather than returning the full buffer width.

The three routines for store exceptions work together to analyse the position of the fault relative to the start or the end of the buffer, and backtrack from the optimized memcpy algorithm to determine if some number of bytes has already been successfully copied.

The store operations occur mostly in-order, with the exception of a few copy size ranges - this is specific to the new copy template, which uses the latest memcpy implementation.

---
 arch/arm64/lib/copy_user_fixup.S | 217 ++++++++++++++++++++++++++++++-
 1 file changed, 215 insertions(+), 2 deletions(-)
arch/arm64/lib/copy_user_fixup.S +215 -2
index ae94b492129d..37ca3d99a02a 100644
@@ -169,12 +169,225 @@ addr	.req	x15
 	 */
 	b	L(end_fixup)
 	
+/*
+ * The following three routines are directed to from faults
+ * on store instructions.
+ */
 9996:
+	/*
+	 * This routine is reached from faults on store instructions
+	 * where the target address has been specified relative to the
+	 * start of the user space memory buffer, and is also not
+	 * guaranteed to be 16-byte aligned.
+	 *
+	 * For copy sizes <= 128 bytes, the stores occur in-order,
+	 * so it has copied up to (addr-dst)&~15.
+	 * For copy sizes > 128 bytes, this should only be directed
+	 * to from a fault on the first store of the long copy, before
+	 * the algorithm aligns dst to 16B, so no bytes have copied at
+	 * this point.
+	 */
+
+	/* Retrieve original params from stack */
+	ldr	dst, [sp], #16	// dst: x3, src: x1
+	ldr	count, [sp], #16	// count: x2
+	add	srcend, src, count
+	add	dstend, dst, count
+
+	/* count <= 3 -> count - (addr-dst) */
+	cmp	count, 3
+	sub	x0, addr, dst // relative fault offset in buffer
+	sub	x0, count, x0 // bytes yet to copy
+	b.le	L(end_fixup)
+        /* 3 < count <= 32 -> count - (addr-dst) */
+	cmp	count, 32
+	b.le	L(none_copied)
+	/* 32 < count < 128 -> count - ((addr-dst)&15) */
+	cmp	count, 128
+	sub	x0, addr, dst // relative fault offset
+	bic	x0, x0, 15 // bytes already copied (steps of 16B stores)
+	sub	x0, count, x0 // bytes yet to copy
+	b.le	L(end_fixup)
+	/* 128 < count -> count */
+	b	L(none_copied)
+
 9997:
+	/*
+	 * This routine is reached from faults on store instructions
+	 * where the target address has been specified relative to
+	 * the end of the user space memory buffer and is also not
+	 * guaranteed to be 16-byte aligned.
+	 *
+	 * In this scenario, the copy is close to completion and
+	 * has occurred in-order, so it is straightforward to
+	 * calculate the remaining bytes.
+	 *
+	 * The algorithm increments dst by 64B for each loop64()
+	 * subroutine, and tmp1 stores its latest value, which
+	 * allows for the calculation of a threshold that it has
+	 * copied up to.
+	 * 
+	 * To account for faults on data that has already been copied
+	 * (which can occur due to the way the algorithm uses
+	 * overlapping copies within a buffer), this threshold can be
+	 * subtracted from the copy size and the result compared
+	 * against the remaining bytes after the fault offset in the
+	 * last 64B; the minimum of the two can then be taken for the
+	 * return value.
+	 */
+
+	/* Save the current (adjusted) dst for later restoration */
+	mov	tmp1, dst
+
+	/* Retrieve original params from stack */
+	ldp	dst, src, [sp], #16	// dst: x3, src: x1
+	ldr	count, [sp], #16	// count: x2
+	add	srcend, src, count
+	add	dstend, dst, count	
+
+	/*
+	 * Overlapping buffers:
+	 * (src <= dst && dst < srcend) || (dst <= src && src < dstend)
+	 */
+	cmp	src, dst
+	ccmp	dst, srcend, #0, le
+	b.lt	L(none_copied)
+	cmp	dst, src
+	ccmp	src, dstend, #0, le
+	b.lt	L(none_copied)
+	
+	/*
+	 * For copy size >128 bytes, select max of
+	 * { tmp1-dst+80, ((addr-dstend+64)&15) }
+	 */
+	sub	tmp1, tmp1, dst // relative position of new dst
+	add	tmp1, tmp1, 80 // copied up to here
+	sub	tmp1, count, tmp1 // remaining bytes after non-overlapping section
+	sub	x0, dstend, 64
+	sub	x0, addr, x0
+	bic	x0, x0, 15 // fault offset within dest. buffer
+	add	x0, dstend, x0
+	sub	x0, x0, 64
+	sub	x0, dstend, x0 // remaining bytes in final (overlapping) 64B
+	cmp	x0, tmp1
+	csel	x0, x0, tmp1, lt
+	cmp	count, 128
+	b.gt	L(end_fixup)
+
+	cmp	count, 2
+	b.le	L(none_copied)
+	cmp	count, 3 // one byte left
+	mov	x0, 1
+	b.eq	L(end_fixup)
+	cmp	count, 7 // four bytes left
+	sub	x0, count, 4
+	b.le	L(end_fixup)
+	cmp	count, 15 // eight bytes left
+	sub	x0, count, 8
+	b.le	L(end_fixup)
+	cmp	count, 32 // 16 bytes left
+	sub	x0, count, 16	
+	b.le	L(end_fixup)
+	/*
+	 * For copy size 33..64 select min of
+	 * {(32 - fault_offset), (count-32)}
+	 * as copy may overlap
+	 */
+	sub	tmp1, dstend, 32
+	sub	tmp1, addr, tmp1
+	bic	tmp1, tmp1, 15
+	mov	x0, 32
+	sub	tmp1, x0, tmp1
+	sub	x0, count, 32
+	cmp	x0, tmp1
+	csel	x0, x0, tmp1, lt
+	cmp	count, 64
+	b.le	L(end_fixup)
+	/* For copy size 65..96 select min of
+	 * {(count - fault_offset), (count-64)}
+	 * as copy may overlap
+	 */
+	sub	tmp1, dstend, 32
+	sub	tmp1, addr, tmp1
+	bic	tmp1, tmp1, 15
+	mov	x0, 32
+	sub	tmp1, x0, tmp1
+	sub	x0, count, 64
+	cmp	x0, tmp1
+	csel	x0, x0, tmp1, lt
+	cmp	count, 96
+	b.lt	L(end_fixup)
+	/*
+	 * For copy size 97..128 same as above, but account for
+	 * out-of-order initial stores, where no bytes have been
+	 * copied on those faults
+	 */
+	sub	tmp1, dstend, 64
+	sub	tmp1, addr, tmp1
+	bic	tmp1, tmp1, 15
+	mov	x0, 64
+	sub	tmp1, x0, tmp1
+	cmp	count, 128
+	mov	x0, 32
+	ccmp	tmp1, x0, #0, le
+	b.gt	L(none_copied) // none copied if faults in first or second chunk
+	sub	x0, count, 64
+	cmp	x0, tmp1
+	csel	x0, x0, tmp1, lt
+	cmp	count, 128
+	b.le	L(end_fixup)
+
+	b	L(none_copied)
+
 9998:
-	/* Retrieve info from stack */
+	/*
+	 * This routine is reached from faults on store instructions
+	 * where the target address has been specified relative to the
+	 * start of the user space memory buffer, and is also
+	 * guaranteed to be 16-byte aligned.
+	 *
+	 * These instrucions occur after the algorithm aligns dst
+	 * to 16B, which is after the very first store in a long copy.
+	 * It then continues copying from dst+16 onwards.
+	 *
+	 * This could result in the second store attempting to copy
+	 * data that has already been copied, as there would be an
+	 * overlap in the buffer if the original dst is not 16 bytes
+	 * aligned. In this case we report that 16 bytes has already
+	 * already been copied, so it must take the minimum of the
+	 * two values.
+	 */
+
+	/* Retrieve original params from stack */
+	ldp	dst, src, [sp], #16	// dst: x3, src: x1
 	ldr	count, [sp], #16	// count: x2
-	add	sp, sp, 32
+	add	srcend, src, count
+	add	dstend, dst, count	
+
+	/*
+	 * Overlapping buffers:
+	 * (src <= dst && dst < srcend) || (dst <= src && src < dstend)
+	 */
+	cmp	src, dst
+	ccmp	dst, srcend, #0, le
+	b.lt	L(none_copied)
+	cmp	dst, src
+	ccmp	src, dstend, #0, le
+	b.lt	L(none_copied)
+
+	/* Take the min from {16,(fault_addr&15)-(dst&15)}
+	 * and subtract from count to obtain the return value */
+	bic	tmp1, dst, 15 // aligned dst
+	bic	x0, addr, 15
+	sub	x0, x0, tmp1 // relative fault offset
+	cmp	x0, 16
+	bic	x0, addr, 15
+	sub	x0, x0, dst
+	sub	x0, count, x0
+	b.gt	L(end_fixup)
+	sub	x0, count, 16
+	b	L(end_fixup) // initial unaligned chunk copied
+
 L(none_copied):
 	/*
 	 * Return the initial count as the number
Signature · 1 line
2.17.1
#14 lore raw

PATCH v4 14/14 arm64: Improve accuracy of fixup for UAO cases

In reply to Oliver Swede

This accounts for variations in the number of bytes copied to the destination buffer that could result from the substitution of STP instructions with 2x unprivileged STTR variants if UAO is supported and enabled.

Rather than duplicating the store fixups with the modifications, the relevant alternatives are inserted in-line.

---
 arch/arm64/lib/copy_user_fixup.S | 47 ++++++++++++++++++++++++++++----
 1 file changed, 41 insertions(+), 6 deletions(-)
arch/arm64/lib/copy_user_fixup.S +41 -6
index 37ca3d99a02a..2d413f9ba5d3 100644
@@ -205,7 +205,12 @@ addr	.req	x15
 	/* 32 < count < 128 -> count - ((addr-dst)&15) */
 	cmp	count, 128
 	sub	x0, addr, dst // relative fault offset
+	/* fault offset within dest. buffer */
+	alternative_if ARM64_HAS_UAO
+	bic	x0, x0, 7 // stp subst. for 2x sttr
+	alternative_else
 	bic	x0, x0, 15 // bytes already copied (steps of 16B stores)
+	alternative_endif
 	sub	x0, count, x0 // bytes yet to copy
 	b.le	L(end_fixup)
 	/* 128 < count -> count */
@@ -265,7 +270,12 @@ addr	.req	x15
 	sub	tmp1, count, tmp1 // remaining bytes after non-overlapping section
 	sub	x0, dstend, 64
 	sub	x0, addr, x0
-	bic	x0, x0, 15 // fault offset within dest. buffer
+	/* fault offset within dest. buffer */
+	alternative_if ARM64_HAS_UAO
+	bic	x0, x0, 7 // stp subst. for 2x sttr
+	alternative_else
+	bic	x0, x0, 15 // bytes already copied (steps of 16B stores)
+	alternative_endif
 	add	x0, dstend, x0
 	sub	x0, x0, 64
 	sub	x0, dstend, x0 // remaining bytes in final (overlapping) 64B
@@ -295,7 +305,12 @@ addr	.req	x15
 	 */
 	sub	tmp1, dstend, 32
 	sub	tmp1, addr, tmp1
-	bic	tmp1, tmp1, 15
+	/* fault offset */
+	alternative_if ARM64_HAS_UAO
+	bic	tmp1, tmp1, 7 // stp subst. for 2x sttr
+	alternative_else
+	bic	tmp1, tmp1, 15 // bytes already copied (steps of 16B stores)
+	alternative_endif
 	mov	x0, 32
 	sub	tmp1, x0, tmp1
 	sub	x0, count, 32
@@ -309,7 +324,12 @@ addr	.req	x15
 	 */
 	sub	tmp1, dstend, 32
 	sub	tmp1, addr, tmp1
-	bic	tmp1, tmp1, 15
+	/* fault offset */
+	alternative_if ARM64_HAS_UAO
+	bic	tmp1, tmp1, 7 // stp subst. for 2x sttr
+	alternative_else
+	bic	tmp1, tmp1, 15 // bytes already copied (steps of 16B stores)
+	alternative_endif
 	mov	x0, 32
 	sub	tmp1, x0, tmp1
 	sub	x0, count, 64
@@ -324,7 +344,12 @@ addr	.req	x15
 	 */
 	sub	tmp1, dstend, 64
 	sub	tmp1, addr, tmp1
-	bic	tmp1, tmp1, 15
+	/* fault offset */
+	alternative_if ARM64_HAS_UAO
+	bic	tmp1, tmp1, 7 // stp subst. for 2x sttr
+	alternative_else
+	bic	tmp1, tmp1, 15 // bytes already copied (steps of 16B stores)
+	alternative_endif
 	mov	x0, 64
 	sub	tmp1, x0, tmp1
 	cmp	count, 128
@@ -378,10 +403,20 @@ addr	.req	x15
 	/* Take the min from {16,(fault_addr&15)-(dst&15)}
 	 * and subtract from count to obtain the return value */
 	bic	tmp1, dst, 15 // aligned dst
-	bic	x0, addr, 15
+	/* fault offset */
+	alternative_if ARM64_HAS_UAO
+	bic	x0, addr, 7 // stp subst. for 2x sttr
+	alternative_else
+	bic	x0, addr, 15 // bytes already copied (steps of 16B stores)
+	alternative_endif
 	sub	x0, x0, tmp1 // relative fault offset
 	cmp	x0, 16
-	bic	x0, addr, 15
+	/* fault offset */
+	alternative_if ARM64_HAS_UAO
+	bic	x0, addr, 7 // stp subst. for 2x sttr
+	alternative_else
+	bic	x0, addr, 15 // bytes already copied (steps of 16B stores)
+	alternative_endif
 	sub	x0, x0, dst
 	sub	x0, count, x0
 	b.gt	L(end_fixup)
Signature · 1 line
2.17.1
author
#15 lore raw
In reply to Oliver Swede

Hi all,

(Apologies if you have already received this follow-up, please ignore any previous duplicates; I am re-sending this with the lists in the CC as the formatting filters caught me out :) Thanks ]

Quoted 3 lines

Version 3 addressed this but I later found some issues with the fixup correctness after further testing, and have partially re-written them here, and addressed some other behaviours of the copy algorithm.

Further to the above description from the cover letter I have expanded on this below.

The approach introduced in this version redirects the faulting instruction to one of multiple fixup subroutines, depending on properties of the instance of the instruction in the copy template: if it specifies the target address relative to the start or end of the buffer, and if the algorithm has reached a stage where the destination address has been set to be 16-byte aligned (this is more relevant for longer copies).

These two properties are both in relation to the copy algorithm's aim at reducing the number of total instruction by making accesses that may overlap, and thus copy the same data multiple times. This is in contrast to the current algorithm which may need to branch to subroutines that complete an odd number of bytes that isn't divisible by the initial unit that was directed to. E.g. with the new algorithm, 17 bytes will complete with two load and two store instructions, by coping the first 16 bytes from the start of the buffer, then the last 16 bytes to the left of the end of the buffer.

There appear to be two types of overlapping copies within a source or destination buffer:
1. If src is initially non-16-byte aligned, then it is re-set to (src &~ 15), and the copying then resumes from aligned_src+16. In this scenario any data between aligned_src+16 and src+16 could repeat in the overlapping instruction.
2. If the target address is specified relative to the end of the buffer (e.g. [srcend, -16]) then an overlap is taking place, so it should check the fault address against the threshold known to have already been copied up to.

In particular, encapsulating these properties through separate fixups allows for:

a) Accurate rounding-down to the base of the block within which the fault occurred.
The assumption here is that load and and store instructions are single-copy atomic, so a fault in one instruction will not result in any of the other bytes succeeding. For instance, the copy unit size could be 1, 4, 8, or 16, depending on the instruction (ldrb/strb, ldr/str, or ldp/stp), but any threshold prior to which copying may have succeeded will be the target address, i.e. the base of this block. If the fixup routing itself implies certain information about the target address, then the scenario shouldn't occur where it faults on an address but the backtracking to the target is inaccurate (as a result of copy overlaps within the buffer).

b) All data copied up to some address to be taken into account, in the case where a page is swapped out immediately before the same address is accessed as a result of an overlap.
This is somewhat related to the above, but addresses a specific situation, where a page could be swapped out mid-way through a copy and result in the same address faulting on its second copy, even if the algorithm has succeeded up to a threshold address beyond the fault.

Within each fixup is the calculation for the number of bytes remaining is performed, with the knowledge about the stage of the algorithm that has been reached and any previous steps before the fault, and the information retrieved from the stack, which contains the original arguments to the usercopy function. It is then able to determine the accurate return value. For systems implementing UAO, due to the ldp/stp instructions being substituted for 2x ldtr/sttr instructions, a fault in the second half of the doubleword needs to to take any completion of the first half into account, and there are in-line modifications for this (but they are not necessary for the load instruction fixups).

The categorization of the fixups based on the algorithm's behaviour may also provide a general approach to replacing the fixups for future copy imports, as they are fairly modular. However, there is an open question as to whether it would be easy to replace them given this framework, as the fixup subroutines are relatively long for the store instructions. I have added comments to explain the calculations, but I am wondering if it would be possible to divert the fixup logic to make it easier to modify, if the aim is to reduce the coupling between the fixup routines and the copy algorithm.

I am waiting on access to the relevant machine before posting the benchmark results for this optimized memcpy, but Sam reported the following with the similar (but now slightly older) cortex-strings version:
* copy_from_user: 13.17%
* copy_to_user: 4.8%
* memcpy: 27.88%
* copy_in_user: Didn't appear in the test results.
This machine will also be used to check the fixups are accurate on a system with UAO - they appear to be exact on a non-UAO system with PAN that I've been working on locally.

I should also mention that the correctness of these routines were tested using a selftest test module akin to lib/test_user_copy.c (whose usercopy functionality checks these patches do pass) but which is more specific to the fixup accuracy, in that it compares the return value with the true number of bytes remaining in the destination buffer at the point of a fault.

Thanks in advance,
Oliver

Quoted 84 lines · Oliver Swede

Hi,

This contains an update to the cortex-strings patchset: the correctness of the fixup routines are improved, with the aim being to return the exact number of remaining bytes for all copy sizes. To ensure they are exact - which the current fixups are not for some copy sizes and are off by a few byes - is an extension to the original intention of fixing an issue reported by an LTP run last year, where the fixup routine in v2 of this patchset (which was importing the cortex-strings memcpy implementation) would over-report the number of bytes that successfully copied.
Version 3 addressed this but I later found some issues with the fixup correctness after further testing, and have partially re-written them here, and addressed some other behaviours of the copy algorithm.

Comments welcome,

Thanks
Oliver

v1: https://lore.kernel.org/linux-arm-kernel/[email protected]/
v2: https://lore.kernel.org/linux-arm-kernel/[email protected]/
v3: https://lore.kernel.org/linux-arm-kernel/[email protected]/

Changes since v3:
* Improves the accuracy of the fixups in response to issues that arose during futher testing
* Accounts for faults on store instructions on systems with UAO enabled
* Expands on comments detailing the implementation

Changes since v2:
* Adds Robin's separate patch that fixes a compilation issue with KProbes fixup [1]
* Imports the most recent memcpy implementation by updating Sam's patch (and moves this patch to occur after the cortex-strings importing so that it's closer to the patches containing its corresponding fixups)
* Uses the stack to preserve the initial parameters
* Replaces the usercopy fixup routine in v2 with multiple longer fixups that each make use of the fault address to return the exact number of bytes that haven't yet copied.

[1] https://lore.kernel.org/linux-arm-kernel/e70f7b9de7e601b9e4a6fedad8eaf64d304b1637.1571326276.git.robin.murphy@arm.com/

Oliver Swede (5):
arm64: Store the arguments to copy_*_user on the stack
arm64: Use additional memcpy macros and fixups
arm64: Add fixup routines for usercopy load exceptions
arm64: Add fixup routines for usercopy store exceptions
arm64: Improve accuracy of fixup for UAO cases

Robin Murphy (2):
arm64: kprobes: Drop open-coded exception fixup
arm64: Tidy up _asm_extable_faultaddr usage

Sam Tebbs (7):
arm64: Allow passing fault address to fixup handlers
arm64: Import latest version of Cortex Strings' memcmp
arm64: Import latest version of Cortex Strings' memmove
arm64: Import latest version of Cortex Strings' strcmp
arm64: Import latest version of Cortex Strings' strlen
arm64: Import latest version of Cortex Strings' strncmp
arm64: Import latest optimization of memcpy

  arch/arm64/include/asm/alternative.h |  36 ---
  arch/arm64/include/asm/assembler.h   |  13 +
  arch/arm64/include/asm/extable.h     |  10 +-
  arch/arm64/kernel/probes/kprobes.c   |   7 -
  arch/arm64/lib/copy_from_user.S      | 272 +++++++++++++++--
  arch/arm64/lib/copy_in_user.S        | 287 ++++++++++++++++--
  arch/arm64/lib/copy_template.S       | 377 +++++++++++++----------
  arch/arm64/lib/copy_template_user.S  |  50 ++++
  arch/arm64/lib/copy_to_user.S        | 273 +++++++++++++++--
  arch/arm64/lib/copy_user_fixup.S     | 433 +++++++++++++++++++++++++++
  arch/arm64/lib/memcmp.S              | 333 ++++++++------------
  arch/arm64/lib/memcpy.S              | 127 ++++++--
  arch/arm64/lib/memmove.S             | 232 +++++---------
  arch/arm64/lib/strcmp.S              | 272 +++++++----------
  arch/arm64/lib/strlen.S              | 247 ++++++++++-----
  arch/arm64/lib/strncmp.S             | 363 ++++++++++------------
  arch/arm64/mm/extable.c              |  13 +-
  arch/arm64/mm/fault.c                |   2 +-
  18 files changed, 2228 insertions(+), 1119 deletions(-)
Quoted 3 lines

create mode 100644 arch/arm64/lib/copy_template_user.S create mode 100644 arch/arm64/lib/copy_user_fixup.S

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

maintainer
#16 lore raw

Hi Oli,

Thanks for this. Just a few high-level comments below.

Quoted 3 lines · Oli Swede
Quoted 3 lines

Version 3 addressed this but I later found some issues with the fixup correctness after further testing, and have partially re-written them here, and addressed some other behaviours of the copy algorithm.

[...]

Quoted 10 lines

I am waiting on access to the relevant machine before posting the benchmark results for this optimized memcpy, but Sam reported the following with the similar (but now slightly older) cortex-strings version:
* copy_from_user: 13.17%
* copy_to_user: 4.8%
* memcpy: 27.88%
* copy_in_user: Didn't appear in the test results.
This machine will also be used to check the fixups are accurate on a system with UAO - they appear to be exact on a non-UAO system with PAN that I've been working on locally.

I'm inclined to say that cortex-strings is probably not a good basis for our uaccess routines. The code needs to be adapted in a non-straightforward way so that we lose pretty much all of the benefits we'd usually get from adopted an existing implementation; we can't pull in fixes or improvements without a lot of manual effort, we can't reuse existing testing infrastructure (see below) and we end up being a "second-class" user of the routines because of the discrepancies in implementation.

So why don't we use cortex-strings as a basis for the in-kernel routines only, preferably in a form where the code can be used directly and updated with a script (e.g. similar to how we pull in arch/arm64/crypto routines from OpenSSL). We can then roll our own uaccess routines, using a slightly more straight-forward implementation which is more amenable to handling user faults and doesn't do things like over copying.

Quoted 5 lines

I should also mention that the correctness of these routines were tested using a selftest test module akin to lib/test_user_copy.c (whose usercopy functionality checks these patches do pass) but which is more specific to the fixup accuracy, in that it compares the return value with the true number of bytes remaining in the destination buffer at the point of a fault.

Can we put this test module into the kernel source tree, please, maybe as part of lkdtm? Given the control flow of these optimised functions, I think we absolutely need targetted testing to make sure we're getting complete coverage.

Will

maintainer
#17 lore raw
Quoted 18 lines · Will Deacon
Quoted 3 lines · Oli Swede
Quoted 3 lines

Version 3 addressed this but I later found some issues with the fixup correctness after further testing, and have partially re-written them here, and addressed some other behaviours of the copy algorithm.

[...]

Quoted 10 lines

I am waiting on access to the relevant machine before posting the benchmark results for this optimized memcpy, but Sam reported the following with the similar (but now slightly older) cortex-strings version:
* copy_from_user: 13.17%
* copy_to_user: 4.8%
* memcpy: 27.88%
* copy_in_user: Didn't appear in the test results.
This machine will also be used to check the fixups are accurate on a system with UAO - they appear to be exact on a non-UAO system with PAN that I've been working on locally.

Quoted 7 lines

I'm inclined to say that cortex-strings is probably not a good basis for our uaccess routines. The code needs to be adapted in a non-straightforward way so that we lose pretty much all of the benefits we'd usually get from adopted an existing implementation; we can't pull in fixes or improvements without a lot of manual effort, we can't reuse existing testing infrastructure (see below) and we end up being a "second-class" user of the routines because of the discrepancies in implementation.

I was a bit more optimistic about this series until I saw the copy_user_fixup.S changes (patches 12 to 14). I have a suspicion it's only Oli (and maybe Robin) who understands it, so from a maintainer's perspective it doesn't really scale. It's also very fragile with any minor tweak in the actual copy routine potentially breaking the fixup code.

Quoted 6 lines

So why don't we use cortex-strings as a basis for the in-kernel routines only, preferably in a form where the code can be used directly and updated with a script (e.g. similar to how we pull in arch/arm64/crypto routines from OpenSSL). We can then roll our own uaccess routines, using a slightly more straight-forward implementation which is more amenable to handling user faults and doesn't do things like over copying.

I think that's probably the best option. I wouldn't mind replacing the in-kernel memcpy/strcpy etc. with these patches since the work was done already but definitely not for the uaccess and fixup routines (we still have the original implementation in the git log).

A script would work even better. Do we have any issue with licensing though? Cortex Strings is BSD (3-clause IIRC) and copyright owned by Linaro. I got them to officially agree with relicensing (dual license) the latest copy under GPLv2 so that we can contribute it to the kernel. But since the project license is still BSD, any future updates in there are BSD-only.

Maybe someone who understands this stuff can confirm that it's ok to regularly grab the Cortex Strings files into a GPLv2 codebase without asking for Linaro's permission.

BTW, you could pick the kprobes patch in here, that explicit fixup call is not necessary.

Signature · 6 lines
Catalin

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
author
#18 lore raw

Hi Will and Catalin,

Thank you both for having a look at this, the coupling between the usercopy routines and the fixup routines was my main concern as well and I understand the desire to take an approach that avoids this dependency when it comes to importing new copy routines.

Quoted 8 lines · Catalin Marinas

I was a bit more optimistic about this series until I saw the copy_user_fixup.S changes (patches 12 to 14). I have a suspicion it's only Oli (and maybe Robin) who understands it, so from a maintainer's perspective it doesn't really scale. It's also very fragile with any minor tweak in the actual copy routine potentially breaking the fixup code.

I was wondering if it would be possible to check that an alternative expression of the logic involved (e.g. diverted to a C function - if technically feasible - that is easier to understand and modify) would not really be desirable either?

Quoted 10 lines · Catalin Marinas
Quoted 8 lines · Will Deacon

I'm inclined to say that cortex-strings is probably not a good basis for our uaccess routines. The code needs to be adapted in a non-straightforward way so that we lose pretty much all of the benefits we'd usually get from adopted an existing implementation; we can't pull in fixes or improvements without a lot of manual effort, we can't reuse existing testing infrastructure (see below) and we end up being a "second-class" user of the routines because of the discrepancies in implementation.

Quoted 7 lines

I was a bit more optimistic about this series until I saw the copy_user_fixup.S changes (patches 12 to 14). I have a suspicion it's only Oli (and maybe Robin) who understands it, so from a maintainer's perspective it doesn't really scale. It's also very fragile with any minor tweak in the actual copy routine potentially breaking the fixup code.

An alternative approach to the fixup routines came up recently in discussions; it could potentially enable the copy template to be preserved in its current form and allow the uaccess routines to re-use a shared copy algorithm (including cortex-strings and/or future optimized out-of-order algorithms), and is based on one of Robin's suggestions from a few weeks ago (I found some time to prototype this locally the other day). The algorithm-specific backtracking would be dropped in favour of first- and second-stage fixups, with each of these being relatively short & straightforward, and (together) likely compatible with future copy routines. More specifically, within the fixup path invoked after the initial fault, it could jump back N bytes and then begin an in-order byte-wise copy from within the first fixup, and then intentionally fault for the second time at either the same or another fault address. As the intermediate scan-copy should continue right up to the 'true' fault address, and ignore any out-of-order techniques used by the optimized algorithm, the fixup for the second fault could then contain an accurate, conclusive calculation of the return value.

A technique for this which would need to be carried forward from patch 11 in v4 would be to redirect the PC to a different 'intermediate' fixup (though only two choices this time) depending on whether the faulting instruction is a load or store, so it can establish if the minimum address to jump back to should be src or buf when comparing to (faddr-N). Additionally, for this the parameters to the usercopy functions would still need to be stored on the stack (patch 10 in v4). Otherwise, it would involve a slight rewrite of v4 that could provide a more maintainable approach; the same fixups would be shared for all copy sizes & fault addresses (versus in v4) and the only variable subject to re-evaluation with new imports might be N (currently for copy sizes over 64B it is guaranteed anything below the previous 64B will have been copied). It could also remove the need to special-case UAO systems when compared to v4.

If it's possible for pages to be swapped out in the middle of a usercopy function then this method might not provide the extra implicit knowledge used by the fixups in v4 about the stage of the algorithm that has been reached (deduced from the fixup that has been routed to, implied by the fault-instruction-to-fixup mapping) to compensate for this, so it could under-report the bytes copied if it faults at an earlier address during the in-order (secondary) copy. This could be outweighed by the potential advantages mentioned above, and I mainly include this as v4 should account for this scenario (but perhaps unnecessarily if the scenario isn't possible or too unlikely :) ), but as you rightly point out it would be tiresome to re-write these routines.

Quoted 7 lines · Will Deacon

So why don't we use cortex-strings as a basis for the in-kernel routines only, preferably in a form where the code can be used directly and updated with a script (e.g. similar to how we pull in arch/arm64/crypto routines from OpenSSL). We can then roll our own uaccess routines, using a slightly more straight-forward implementation which is more amenable to handling user faults and doesn't do things like over copying.

I was wondering if the proposal described above is something that you would want to look into, or if the use of separate copy routines (script-based or otherwise) for in-kernel/uaccess is the preferred approach?

Quoted 5 lines · Will Deacon

Can we put this test module into the kernel source tree, please, maybe as part of lkdtm? Given the control flow of these optimised functions, I think we absolutely need targeted testing to make sure we're getting complete coverage.

I will post the standalone module in its current state later today/early next week, the tests are exhaustive but if it would be useful to include a mechanism for specifying certain ranges of faults/copy sizes or improve the error reporting, I'd be happy to add these features (and/or integrate it into lkdtm).

Many thanks,
Oli

Quoted 53 lines · Catalin Marinas
Quoted 18 lines · Will Deacon
Quoted 3 lines · Oli Swede
Quoted 3 lines

Version 3 addressed this but I later found some issues with the fixup correctness after further testing, and have partially re-written them here, and addressed some other behaviours of the copy algorithm.

[...]

Quoted 10 lines

I am waiting on access to the relevant machine before posting the benchmark results for this optimized memcpy, but Sam reported the following with the similar (but now slightly older) cortex-strings version:
* copy_from_user: 13.17%
* copy_to_user: 4.8%
* memcpy: 27.88%
* copy_in_user: Didn't appear in the test results.
This machine will also be used to check the fixups are accurate on a system with UAO - they appear to be exact on a non-UAO system with PAN that I've been working on locally.

Quoted 7 lines

I'm inclined to say that cortex-strings is probably not a good basis for our uaccess routines. The code needs to be adapted in a non-straightforward way so that we lose pretty much all of the benefits we'd usually get from adopted an existing implementation; we can't pull in fixes or improvements without a lot of manual effort, we can't reuse existing testing infrastructure (see below) and we end up being a "second-class" user of the routines because of the discrepancies in implementation.

I was a bit more optimistic about this series until I saw the copy_user_fixup.S changes (patches 12 to 14). I have a suspicion it's only Oli (and maybe Robin) who understands it, so from a maintainer's perspective it doesn't really scale. It's also very fragile with any minor tweak in the actual copy routine potentially breaking the fixup code.

Quoted 6 lines

So why don't we use cortex-strings as a basis for the in-kernel routines only, preferably in a form where the code can be used directly and updated with a script (e.g. similar to how we pull in arch/arm64/crypto routines from OpenSSL). We can then roll our own uaccess routines, using a slightly more straight-forward implementation which is more amenable to handling user faults and doesn't do things like over copying.

I think that's probably the best option. I wouldn't mind replacing the in-kernel memcpy/strcpy etc. with these patches since the work was done already but definitely not for the uaccess and fixup routines (we still have the original implementation in the git log).

A script would work even better. Do we have any issue with licensing though? Cortex Strings is BSD (3-clause IIRC) and copyright owned by Linaro. I got them to officially agree with relicensing (dual license) the latest copy under GPLv2 so that we can contribute it to the kernel. But since the project license is still BSD, any future updates in there are BSD-only.

Quoted 7 lines

Maybe someone who understands this stuff can confirm that it's ok to regularly grab the Cortex Strings files into a GPLv2 codebase without asking for Linaro's permission.

BTW, you could pick the kprobes patch in here, that explicit fixup call is not necessary.

_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel