Skip to contentSkip to conversation

arm64: make phys_to_virt() correct

PATCH Superseded Pingfan Liu 3 participants 14 messages Jan 18, 2022 – Jan 24, 2022

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

v4 cover arm64: make phys_to_virt() correct

Currently phys_to_virt() does not work well on 52-bits VA arm64 kernel. One issue is contributed by phys_offset not signed.
The other is contributed by wrong page_offset.

v3 -> v4:
address the date type in [1/4] and [2/4]

v2 -> v3:
Discussed with Kairui off-list, 48-bits VA kernel can not handle flipped mm yet. So introducing [4/4], which judges whether the kernel is with flipped mm layout and adopt different formula accordingly.

As for [1-3/4], they are the same as [1-3/3] in V2.

v1 -> v2
Fix broken patch [2/3] in v1
Move arch_scan_vmcoreinfo declaration to util_lib/include/elf_info.h Using UINT64_MAX instread of 0xffffffffffffffff

Cc: Kairui Song <[email protected]>
Cc: Simon Horman <[email protected]>
Cc: Philipp Rudo <[email protected]>
To: [email protected]

*** BLURB HERE ***

Kairui Song (1):
arm64: fix PAGE_OFFSET calc for flipped mm

Pingfan Liu (3):
arm64: make phys_offset signed arm64/crashdump: unify routine to get page_offset
arm64: read VA_BITS from kcore for 52-bits VA kernel

 kexec/arch/arm64/crashdump-arm64.c | 23 +----------
 kexec/arch/arm64/kexec-arm64.c     | 64 ++++++++++++++++++++++++------
 kexec/arch/arm64/kexec-arm64.h     |  3 +-
 util_lib/elf_info.c                |  7 +++-
 util_lib/include/elf_info.h        |  3 +-
 5 files changed, 62 insertions(+), 38 deletions(-)
Signature · 1 line
2.31.1
#1 lore raw

PATCH v4 1/4 arm64: make phys_offset signed

After kernel commit 7bc1a0f9e176 ("arm64: mm: use single quantity to represent the PA to VA translation"), phys_offset can be negative if running 52-bits kernel on 48-bits hardware.

So changing phys_offset from unsigned to signed.

Cc Kairui Song <[email protected]>
Cc Simon Horman <[email protected]>
Cc Philipp Rudo <[email protected]>

To: [email protected]

---
 kexec/arch/arm64/kexec-arm64.c | 12 ++++++------
 kexec/arch/arm64/kexec-arm64.h |  2 +-
 util_lib/elf_info.c            |  2 +-
 util_lib/include/elf_info.h    |  2 +-
 4 files changed, 9 insertions(+), 9 deletions(-)
kexec/arch/arm64/kexec-arm64.c +6 -6
index 6f572ed..c6c67e8 100644
@@ -859,7 +859,7 @@ void add_segment(struct kexec_info *info, const void *buf, size_t bufsz,
 	add_segment_phys_virt(info, buf, bufsz, base, memsz, 1);
 }
 
-static inline void set_phys_offset(uint64_t v, char *set_method)
+static inline void set_phys_offset(int64_t v, char *set_method)
 {
 	if (arm64_mem.phys_offset == arm64_mem_ngv
 		|| v < arm64_mem.phys_offset) {
@@ -928,7 +928,7 @@ static int get_page_offset(void)
  * from VMCOREINFO note inside 'kcore'.
  */
 
-static int get_phys_offset_from_vmcoreinfo_pt_note(unsigned long *phys_offset)
+static int get_phys_offset_from_vmcoreinfo_pt_note(long *phys_offset)
 {
 	int fd, ret = 0;
 
@@ -948,7 +948,7 @@ static int get_phys_offset_from_vmcoreinfo_pt_note(unsigned long *phys_offset)
  * from PT_LOADs inside 'kcore'.
  */
 
-int get_phys_base_from_pt_load(unsigned long *phys_offset)
+int get_phys_base_from_pt_load(long *phys_offset)
 {
 	int i, fd, ret;
 	unsigned long long phys_start;
@@ -997,7 +997,7 @@ static bool to_be_excluded(char *str)
 int get_memory_ranges(struct memory_range **range, int *ranges,
 	unsigned long kexec_flags)
 {
-	unsigned long phys_offset = UINT64_MAX;
+	long phys_offset = -1;
 	FILE *fp;
 	const char *iomem = proc_iomem();
 	char line[MAX_LINE], *str;
@@ -1019,7 +1019,7 @@ int get_memory_ranges(struct memory_range **range, int *ranges,
 		 */
 		ret = get_phys_offset_from_vmcoreinfo_pt_note(&phys_offset);
 		if (!ret) {
-			if (phys_offset != UINT64_MAX)
+			if (phys_offset != -1)
 				set_phys_offset(phys_offset,
 						"vmcoreinfo pt_note");
 		} else {
@@ -1031,7 +1031,7 @@ int get_memory_ranges(struct memory_range **range, int *ranges,
 			 */
 			ret = get_phys_base_from_pt_load(&phys_offset);
 			if (!ret)
-				if (phys_offset != UINT64_MAX)
+				if (phys_offset != -1)
 					set_phys_offset(phys_offset,
 							"pt_load");
 		}
kexec/arch/arm64/kexec-arm64.h +1 -1
index ed447ac..1844b67 100644
@@ -58,7 +58,7 @@ extern off_t initrd_size;
  */
 
 struct arm64_mem {
-	uint64_t phys_offset;
+	long phys_offset;
 	uint64_t text_offset;
 	uint64_t image_size;
 	uint64_t vp_offset;
util_lib/elf_info.c +1 -1
index 51d8b92..5574c7f 100644
@@ -1236,7 +1236,7 @@ int read_elf(int fd)
 	return 0;
 }
 
-int read_phys_offset_elf_kcore(int fd, unsigned long *phys_off)
+int read_phys_offset_elf_kcore(int fd, long *phys_off)
 {
 	int ret;
 
util_lib/include/elf_info.h +1 -1
index 4bc9279..f550d86 100644
@@ -28,7 +28,7 @@ int get_pt_load(int idx,
 	unsigned long long *phys_end,
 	unsigned long long *virt_start,
 	unsigned long long *virt_end);
-int read_phys_offset_elf_kcore(int fd, unsigned long *phys_off);
+int read_phys_offset_elf_kcore(int fd, long *phys_off);
 int read_elf(int fd);
 void dump_dmesg(int fd, void (*handler)(char*, unsigned int));
 
Signature · 1 line
2.31.1
#2 lore raw

PATCH v4 1/4 arm64: make phys_offset signed

Hi Pingfan,

Quoted 18 lines · On Tue, 18 Jan 2022 15:48:09 +0800 Pingfan Liu <[email protected]> wrote:

After kernel commit 7bc1a0f9e176 ("arm64: mm: use single quantity to represent the PA to VA translation"), phys_offset can be negative if running 52-bits kernel on 48-bits hardware.

So changing phys_offset from unsigned to signed.

Cc Kairui Song <[email protected]>
Cc Simon Horman <[email protected]>
Cc Philipp Rudo <[email protected]>

To: kexec at lists.infradead.org

---
 kexec/arch/arm64/kexec-arm64.c | 12 ++++++------
 kexec/arch/arm64/kexec-arm64.h |  2 +-
 util_lib/elf_info.c            |  2 +-
 util_lib/include/elf_info.h    |  2 +-
 4 files changed, 9 insertions(+), 9 deletions(-)

[...]

kexec/arch/arm64/kexec-arm64.h extern off_t initrd_size; · line 61

Quoted 10 lines
kexec/arch/arm64/kexec-arm64.h +1 -1
index ed447ac..1844b67 100644
@@ -58,7 +58,7 @@ extern off_t initrd_size;
  */
 
 struct arm64_mem {
-	uint64_t phys_offset;
+	long phys_offset;

I think this one should be int64_t as well.

Other than that

Quoted 3 lines
kexec/arch/arm64/kexec-arm64.h +0 -0
 	uint64_t text_offset;
 	uint64_t image_size;
 	uint64_t vp_offset;
#3 lore raw

PATCH v4 1/4 arm64: make phys_offset signed

Quoted 34 lines · Philipp Rudo

Hi Pingfan,

Quoted 18 lines · On Tue, 18 Jan 2022 15:48:09 +0800 Pingfan Liu <[email protected]> wrote:

After kernel commit 7bc1a0f9e176 ("arm64: mm: use single quantity to represent the PA to VA translation"), phys_offset can be negative if running 52-bits kernel on 48-bits hardware.

So changing phys_offset from unsigned to signed.

Cc Kairui Song <[email protected]>
Cc Simon Horman <[email protected]>
Cc Philipp Rudo <[email protected]>

To: kexec at lists.infradead.org

---
 kexec/arch/arm64/kexec-arm64.c | 12 ++++++------
 kexec/arch/arm64/kexec-arm64.h |  2 +-
 util_lib/elf_info.c            |  2 +-
 util_lib/include/elf_info.h    |  2 +-
 4 files changed, 9 insertions(+), 9 deletions(-)

[...]

Quoted 7 lines
kexec/arch/arm64/kexec-arm64.h +0 -0
index ed447ac..1844b67 100644
@@ -58,7 +58,7 @@ extern off_t initrd_size;
  */
 
Quoted 6 lines

kexec/arch/arm64/kexec-arm64.h extern off_t initrd_size; · line 61

Quoted 3 lines
kexec/arch/arm64/kexec-arm64.h +1 -1
 struct arm64_mem {
-     uint64_t phys_offset;
+     long phys_offset;

I think this one should be int64_t as well.

Yes, you are right. Thanks for your careful review.

@Simon, could you help to correct it or prefer my V5 to fix it.

Thanks,

PIngfan

#4 lore raw

PATCH v4 1/4 arm64: make phys_offset signed

Quoted 41 lines · Pingfan Liu
Quoted 34 lines · Philipp Rudo

Hi Pingfan,

Quoted 18 lines · On Tue, 18 Jan 2022 15:48:09 +0800 Pingfan Liu <[email protected]> wrote:

After kernel commit 7bc1a0f9e176 ("arm64: mm: use single quantity to represent the PA to VA translation"), phys_offset can be negative if running 52-bits kernel on 48-bits hardware.

So changing phys_offset from unsigned to signed.

Cc Kairui Song <[email protected]>
Cc Simon Horman <[email protected]>
Cc Philipp Rudo <[email protected]>

To: kexec at lists.infradead.org

---
 kexec/arch/arm64/kexec-arm64.c | 12 ++++++------
 kexec/arch/arm64/kexec-arm64.h |  2 +-
 util_lib/elf_info.c            |  2 +-
 util_lib/include/elf_info.h    |  2 +-
 4 files changed, 9 insertions(+), 9 deletions(-)

[...]

Quoted 7 lines
kexec/arch/arm64/kexec-arm64.h +0 -0
index ed447ac..1844b67 100644
@@ -58,7 +58,7 @@ extern off_t initrd_size;
  */
 
Quoted 6 lines

kexec/arch/arm64/kexec-arm64.h extern off_t initrd_size; · line 61

Quoted 3 lines
kexec/arch/arm64/kexec-arm64.h +1 -1
 struct arm64_mem {
-     uint64_t phys_offset;
+     long phys_offset;

I think this one should be int64_t as well.

Quoted 3 lines

Yes, you are right. Thanks for your careful review.

@Simon, could you help to correct it or prefer my V5 to fix it.

I think I can fix this while applying this patchset :)

#5 lore raw

PATCH v4 2/4 arm64/crashdump: unify routine to get page_offset

In reply to Pingfan Liu

There are two funcs to get page_offset:
get_kernel_page_offset()
get_page_offset()

Since get_kernel_page_offset() does not observe the kernel formula, and remove it. Unify them in order to introduce 52-bits VA kernel more easily in the coming patch.

Cc Kairui Song <[email protected]>
Cc Simon Horman <[email protected]>
Cc Philipp Rudo <[email protected]>

To: [email protected]

---
 kexec/arch/arm64/crashdump-arm64.c | 23 +----------------------
 kexec/arch/arm64/kexec-arm64.c     |  8 ++++----
 kexec/arch/arm64/kexec-arm64.h     |  1 +
 3 files changed, 6 insertions(+), 26 deletions(-)
kexec/arch/arm64/crashdump-arm64.c +1 -22
index a02019a..0a8d44c 100644
@@ -46,27 +46,6 @@ static struct crash_elf_info elf_info = {
 	.machine	= EM_AARCH64,
 };
 
-/*
- * Note: The returned value is correct only if !CONFIG_RANDOMIZE_BASE.
- */
-static uint64_t get_kernel_page_offset(void)
-{
-	int i;
-
-	if (elf_info.kern_vaddr_start == UINT64_MAX)
-		return UINT64_MAX;
-
-	/* Current max virtual memory range is 48-bits. */
-	for (i = 48; i > 0; i--)
-		if (!(elf_info.kern_vaddr_start & (1UL << i)))
-			break;
-
-	if (i <= 0)
-		return UINT64_MAX;
-	else
-		return UINT64_MAX << i;
-}
-
 /*
  * iomem_range_callback() - callback called for each iomem region
  * @data: not used
@@ -198,7 +177,7 @@ int load_crashdump_segments(struct kexec_info *info)
 	if (err)
 		return EFAILED;
 
-	elf_info.page_offset = get_kernel_page_offset();
+	get_page_offset(&elf_info.page_offset);
 	dbgprintf("%s: page_offset:   %016llx\n", __func__,
 			elf_info.page_offset);
 
kexec/arch/arm64/kexec-arm64.c +4 -4
index c6c67e8..33cc258 100644
@@ -909,7 +909,7 @@ static int get_va_bits(void)
  * get_page_offset - Helper for getting PAGE_OFFSET
  */
 
-static int get_page_offset(void)
+int get_page_offset(unsigned long *page_offset)
 {
 	int ret;
 
@@ -917,8 +917,8 @@ static int get_page_offset(void)
 	if (ret < 0)
 		return ret;
 
-	page_offset = (0xffffffffffffffffUL) << (va_bits - 1);
-	dbgprintf("page_offset : %lx\n", page_offset);
+	*page_offset = UINT64_MAX << (va_bits - 1);
+	dbgprintf("page_offset : %lx\n", *page_offset);
 
 	return 0;
 }
@@ -954,7 +954,7 @@ int get_phys_base_from_pt_load(long *phys_offset)
 	unsigned long long phys_start;
 	unsigned long long virt_start;
 
-	ret = get_page_offset();
+	ret = get_page_offset(&page_offset);
 	if (ret < 0)
 		return ret;
 
kexec/arch/arm64/kexec-arm64.h +1 -0
index 1844b67..ed99d9d 100644
@@ -69,6 +69,7 @@ extern struct arm64_mem arm64_mem;
 
 uint64_t get_phys_offset(void);
 uint64_t get_vp_offset(void);
+int get_page_offset(unsigned long *offset);
 
 static inline void reset_vp_offset(void)
 {
Signature · 1 line
2.31.1
#6 lore raw

PATCH v4 2/4 arm64/crashdump: unify routine to get page_offset

Hi Pingfan,

Quoted 8 lines · On Tue, 18 Jan 2022 15:48:10 +0800 Pingfan Liu <[email protected]> wrote:

There are two funcs to get page_offset:
get_kernel_page_offset()
get_page_offset()

Since get_kernel_page_offset() does not observe the kernel formula, and remove it. Unify them in order to introduce 52-bits VA kernel more easily in the coming patch.

Quoted 5 lines
Cc Kairui Song <[email protected]>
Cc Simon Horman <[email protected]>
Cc Philipp Rudo <[email protected]>

To: kexec at lists.infradead.org

looks good

Quoted 92 lines
---
 kexec/arch/arm64/crashdump-arm64.c | 23 +----------------------
 kexec/arch/arm64/kexec-arm64.c     |  8 ++++----
 kexec/arch/arm64/kexec-arm64.h     |  1 +
 3 files changed, 6 insertions(+), 26 deletions(-)
kexec/arch/arm64/crashdump-arm64.c +1 -22
index a02019a..0a8d44c 100644
@@ -46,27 +46,6 @@ static struct crash_elf_info elf_info = {
 	.machine	= EM_AARCH64,
 };
 
-/*
- * Note: The returned value is correct only if !CONFIG_RANDOMIZE_BASE.
- */
-static uint64_t get_kernel_page_offset(void)
-{
-	int i;
-
-	if (elf_info.kern_vaddr_start == UINT64_MAX)
-		return UINT64_MAX;
-
-	/* Current max virtual memory range is 48-bits. */
-	for (i = 48; i > 0; i--)
-		if (!(elf_info.kern_vaddr_start & (1UL << i)))
-			break;
-
-	if (i <= 0)
-		return UINT64_MAX;
-	else
-		return UINT64_MAX << i;
-}
-
 /*
  * iomem_range_callback() - callback called for each iomem region
  * @data: not used
@@ -198,7 +177,7 @@ int load_crashdump_segments(struct kexec_info *info)
 	if (err)
 		return EFAILED;
 
-	elf_info.page_offset = get_kernel_page_offset();
+	get_page_offset(&elf_info.page_offset);
 	dbgprintf("%s: page_offset:   %016llx\n", __func__,
 			elf_info.page_offset);
 
kexec/arch/arm64/kexec-arm64.c +4 -4
index c6c67e8..33cc258 100644
@@ -909,7 +909,7 @@ static int get_va_bits(void)
  * get_page_offset - Helper for getting PAGE_OFFSET
  */
 
-static int get_page_offset(void)
+int get_page_offset(unsigned long *page_offset)
 {
 	int ret;
 
@@ -917,8 +917,8 @@ static int get_page_offset(void)
 	if (ret < 0)
 		return ret;
 
-	page_offset = (0xffffffffffffffffUL) << (va_bits - 1);
-	dbgprintf("page_offset : %lx\n", page_offset);
+	*page_offset = UINT64_MAX << (va_bits - 1);
+	dbgprintf("page_offset : %lx\n", *page_offset);
 
 	return 0;
 }
@@ -954,7 +954,7 @@ int get_phys_base_from_pt_load(long *phys_offset)
 	unsigned long long phys_start;
 	unsigned long long virt_start;
 
-	ret = get_page_offset();
+	ret = get_page_offset(&page_offset);
 	if (ret < 0)
 		return ret;
 
kexec/arch/arm64/kexec-arm64.h +1 -0
index 1844b67..ed99d9d 100644
@@ -69,6 +69,7 @@ extern struct arm64_mem arm64_mem;
 
 uint64_t get_phys_offset(void);
 uint64_t get_vp_offset(void);
+int get_page_offset(unsigned long *offset);
 
 static inline void reset_vp_offset(void)
 {
#7 lore raw

PATCH v4 3/4 arm64: read VA_BITS from kcore for 52-bits VA kernel

In reply to Pingfan Liu

phys_to_virt() calculates virtual address. As a important factor, page_offset is excepted to be accurate.

Since arm64 kernel exposes va_bits through vmcore, using it.

Cc Kairui Song <[email protected]>
Cc Simon Horman <[email protected]>
Cc Philipp Rudo <[email protected]>

To: [email protected]

---
 kexec/arch/arm64/kexec-arm64.c | 34 ++++++++++++++++++++++++++++++----
 util_lib/elf_info.c            |  5 +++++
 util_lib/include/elf_info.h    |  1 +
 3 files changed, 36 insertions(+), 4 deletions(-)
kexec/arch/arm64/kexec-arm64.c +30 -4
index 33cc258..793799b 100644
@@ -54,7 +54,7 @@
 static bool try_read_phys_offset_from_kcore = false;
 
 /* Machine specific details. */
-static int va_bits;
+static int va_bits = -1;
 static unsigned long page_offset;
 
 /* Global varables the core kexec routines expect. */
@@ -876,7 +876,18 @@ static inline void set_phys_offset(int64_t v, char *set_method)
 
 static int get_va_bits(void)
 {
-	unsigned long long stext_sym_addr = get_kernel_sym("_stext");
+	unsigned long long stext_sym_addr;
+
+	/*
+	 * if already got from kcore
+	 */
+	if (va_bits != -1)
+		goto out;
+
+
+	/* For kernel older than v4.19 */
+	fprintf(stderr, "Warning, can't get the VA_BITS from kcore\n");
+	stext_sym_addr = get_kernel_sym("_stext");
 
 	if (stext_sym_addr == 0) {
 		fprintf(stderr, "Can't get the symbol of _stext.\n");
@@ -900,6 +911,7 @@ static int get_va_bits(void)
 		return -1;
 	}
 
+out:
 	dbgprintf("va_bits : %d\n", va_bits);
 
 	return 0;
@@ -917,14 +929,27 @@ int get_page_offset(unsigned long *page_offset)
 	if (ret < 0)
 		return ret;
 
-	*page_offset = UINT64_MAX << (va_bits - 1);
+	if (va_bits < 52)
+		*page_offset = UINT64_MAX << (va_bits - 1);
+	else
+		*page_offset = UINT64_MAX << va_bits;
+
 	dbgprintf("page_offset : %lx\n", *page_offset);
 
 	return 0;
 }
 
+static void arm64_scan_vmcoreinfo(char *pos)
+{
+	const char *str;
+
+	str = "NUMBER(VA_BITS)=";
+	if (memcmp(str, pos, strlen(str)) == 0)
+		va_bits = strtoul(pos + strlen(str), NULL, 10);
+}
+
 /**
- * get_phys_offset_from_vmcoreinfo_pt_note - Helper for getting PHYS_OFFSET
+ * get_phys_offset_from_vmcoreinfo_pt_note - Helper for getting PHYS_OFFSET (and va_bits)
  * from VMCOREINFO note inside 'kcore'.
  */
 
@@ -937,6 +962,7 @@ static int get_phys_offset_from_vmcoreinfo_pt_note(long *phys_offset)
 		return EFAILED;
 	}
 
+	arch_scan_vmcoreinfo = arm64_scan_vmcoreinfo;
 	ret = read_phys_offset_elf_kcore(fd, phys_offset);
 
 	close(fd);
util_lib/elf_info.c +5 -0
index 5574c7f..d252eff 100644
@@ -310,6 +310,8 @@ int get_pt_load(int idx,
 
 #define NOT_FOUND_LONG_VALUE		(-1)
 
+void (*arch_scan_vmcoreinfo)(char *pos);
+
 void scan_vmcoreinfo(char *start, size_t size)
 {
 	char *last = start + size - 1;
@@ -551,6 +553,9 @@ void scan_vmcoreinfo(char *start, size_t size)
 			}
 		}
 
+		if (arch_scan_vmcoreinfo != NULL)
+			(*arch_scan_vmcoreinfo)(pos);
+
 		if (last_line)
 			break;
 	}
util_lib/include/elf_info.h +1 -0
index f550d86..fdf4c3d 100644
@@ -31,5 +31,6 @@ int get_pt_load(int idx,
 int read_phys_offset_elf_kcore(int fd, long *phys_off);
 int read_elf(int fd);
 void dump_dmesg(int fd, void (*handler)(char*, unsigned int));
+extern void (*arch_scan_vmcoreinfo)(char *pos);
 
 #endif /* ELF_INFO_H */
Signature · 1 line
2.31.1
#8 lore raw

PATCH v4 3/4 arm64: read VA_BITS from kcore for 52-bits VA kernel

Hi Pingfan,

Quoted 10 lines · On Tue, 18 Jan 2022 15:48:11 +0800 Pingfan Liu <[email protected]> wrote:

phys_to_virt() calculates virtual address. As a important factor, page_offset is excepted to be accurate.

Since arm64 kernel exposes va_bits through vmcore, using it.

Cc Kairui Song <[email protected]>
Cc Simon Horman <[email protected]>
Cc Philipp Rudo <[email protected]>

To: kexec at lists.infradead.org

looks good

Quoted 119 lines
---
 kexec/arch/arm64/kexec-arm64.c | 34 ++++++++++++++++++++++++++++++----
 util_lib/elf_info.c            |  5 +++++
 util_lib/include/elf_info.h    |  1 +
 3 files changed, 36 insertions(+), 4 deletions(-)
kexec/arch/arm64/kexec-arm64.c +30 -4
index 33cc258..793799b 100644
@@ -54,7 +54,7 @@
 static bool try_read_phys_offset_from_kcore = false;
 
 /* Machine specific details. */
-static int va_bits;
+static int va_bits = -1;
 static unsigned long page_offset;
 
 /* Global varables the core kexec routines expect. */
@@ -876,7 +876,18 @@ static inline void set_phys_offset(int64_t v, char *set_method)
 
 static int get_va_bits(void)
 {
-	unsigned long long stext_sym_addr = get_kernel_sym("_stext");
+	unsigned long long stext_sym_addr;
+
+	/*
+	 * if already got from kcore
+	 */
+	if (va_bits != -1)
+		goto out;
+
+
+	/* For kernel older than v4.19 */
+	fprintf(stderr, "Warning, can't get the VA_BITS from kcore\n");
+	stext_sym_addr = get_kernel_sym("_stext");
 
 	if (stext_sym_addr == 0) {
 		fprintf(stderr, "Can't get the symbol of _stext.\n");
@@ -900,6 +911,7 @@ static int get_va_bits(void)
 		return -1;
 	}
 
+out:
 	dbgprintf("va_bits : %d\n", va_bits);
 
 	return 0;
@@ -917,14 +929,27 @@ int get_page_offset(unsigned long *page_offset)
 	if (ret < 0)
 		return ret;
 
-	*page_offset = UINT64_MAX << (va_bits - 1);
+	if (va_bits < 52)
+		*page_offset = UINT64_MAX << (va_bits - 1);
+	else
+		*page_offset = UINT64_MAX << va_bits;
+
 	dbgprintf("page_offset : %lx\n", *page_offset);
 
 	return 0;
 }
 
+static void arm64_scan_vmcoreinfo(char *pos)
+{
+	const char *str;
+
+	str = "NUMBER(VA_BITS)=";
+	if (memcmp(str, pos, strlen(str)) == 0)
+		va_bits = strtoul(pos + strlen(str), NULL, 10);
+}
+
 /**
- * get_phys_offset_from_vmcoreinfo_pt_note - Helper for getting PHYS_OFFSET
+ * get_phys_offset_from_vmcoreinfo_pt_note - Helper for getting PHYS_OFFSET (and va_bits)
  * from VMCOREINFO note inside 'kcore'.
  */
 
@@ -937,6 +962,7 @@ static int get_phys_offset_from_vmcoreinfo_pt_note(long *phys_offset)
 		return EFAILED;
 	}
 
+	arch_scan_vmcoreinfo = arm64_scan_vmcoreinfo;
 	ret = read_phys_offset_elf_kcore(fd, phys_offset);
 
 	close(fd);
util_lib/elf_info.c +5 -0
index 5574c7f..d252eff 100644
@@ -310,6 +310,8 @@ int get_pt_load(int idx,
 
 #define NOT_FOUND_LONG_VALUE		(-1)
 
+void (*arch_scan_vmcoreinfo)(char *pos);
+
 void scan_vmcoreinfo(char *start, size_t size)
 {
 	char *last = start + size - 1;
@@ -551,6 +553,9 @@ void scan_vmcoreinfo(char *start, size_t size)
 			}
 		}
 
+		if (arch_scan_vmcoreinfo != NULL)
+			(*arch_scan_vmcoreinfo)(pos);
+
 		if (last_line)
 			break;
 	}
util_lib/include/elf_info.h +1 -0
index f550d86..fdf4c3d 100644
@@ -31,5 +31,6 @@ int get_pt_load(int idx,
 int read_phys_offset_elf_kcore(int fd, long *phys_off);
 int read_elf(int fd);
 void dump_dmesg(int fd, void (*handler)(char*, unsigned int));
+extern void (*arch_scan_vmcoreinfo)(char *pos);
 
 #endif /* ELF_INFO_H */
#9 lore raw

PATCH v4 4/4 arm64: fix PAGE_OFFSET calc for flipped mm

In reply to Pingfan Liu

From: Kairui Song <[email protected]>

Since kernel commit 14c127c957c1 ('arm64: mm: Flip kernel VA space'), the memory layout on arm64 have changed, and kexec-tools can no longer get the the right PAGE_OFFSET based on _text symbol.

Prior to that, the kimage (_text) lays above PAGE_END with this layout:

0               -> VA_START                 : Usespace
VA_START        -> VA_START + 256M          : BPF JIT, Modules

VA_START + 256M -> PAGE_OFFSET - (~GB misc) : Vmalloc (KERNEL _text HERE)
PAGE_OFFSET -> ... : * Linear map *

And here we have:
VA_START = -1UL << VA_BITS
PAGE_OFFSET = -1UL << (VA_BITS - 1)
_text < -1UL << (VA_BITS - 1)

Kernel image lays somewhere between VA_START and PAGE_OFFSET, so we just calc VA_BITS by getting the highest unset bit of _text symbol address, and shift one less bit of VA_BITS to get page offset. This works as long as KASLR don't put kernel in a too high location (which is commented inline).

And after that commit, kernel layout have changed:

0               -> PAGE_OFFSET              : Userspace
PAGE_OFFSET     -> PAGE_END                 : * Linear map *
PAGE_END        -> PAGE_END + 128M          : bpf jit region
PAGE_END + 128M -> PAGE_END + 256MB         : modules
PAGE_END + 256M -> ...                      : vmalloc (KERNEL _text HERE)

Here we have:
PAGE_OFFSET = -1UL << VA_BITS
PAGE_END = -1UL << (VA_BITS - 1)
_text > -1UL << (VA_BITS - 1)

Kernel image now lays above PAGE_END, so we have to shift one more bit to get the VA_BITS, and shift the exact VA_BITS for PAGE_OFFSET.

We can simply check if "_text > -1UL << (VA_BITS - 1)" is true to judge which layout is being used and shift the page offset occordingly.

(rebased and stripped by Pingfan )

Cc Simon Horman <[email protected]>
Cc Philipp Rudo <[email protected]>

To: [email protected]

---
 kexec/arch/arm64/kexec-arm64.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)
kexec/arch/arm64/kexec-arm64.c +13 -1
index 793799b..ce7a5bb 100644
@@ -923,13 +923,25 @@ out:
 
 int get_page_offset(unsigned long *page_offset)
 {
+	unsigned long long text_sym_addr, kernel_va_mid;
 	int ret;
 
+	text_sym_addr = get_kernel_sym("_text");
+	if (text_sym_addr == 0) {
+		fprintf(stderr, "Can't get the symbol of _text to calculate page_offset.\n");
+		return -1;
+	}
+
 	ret = get_va_bits();
 	if (ret < 0)
 		return ret;
 
-	if (va_bits < 52)
+	/* Since kernel 5.4, kernel image is put above
+	 * UINT64_MAX << (va_bits - 1)
+	 */
+	kernel_va_mid = UINT64_MAX << (va_bits - 1);
+	/* older kernel */
+	if (text_sym_addr < kernel_va_mid)
 		*page_offset = UINT64_MAX << (va_bits - 1);
 	else
 		*page_offset = UINT64_MAX << va_bits;
Signature · 1 line
2.31.1
#10 lore raw

PATCH v4 4/4 arm64: fix PAGE_OFFSET calc for flipped mm

Hi Pingfan,

Quoted 71 lines · On Tue, 18 Jan 2022 15:48:12 +0800 Pingfan Liu <[email protected]> wrote:

From: Kairui Song <[email protected]>

Since kernel commit 14c127c957c1 ('arm64: mm: Flip kernel VA space'), the memory layout on arm64 have changed, and kexec-tools can no longer get the the right PAGE_OFFSET based on _text symbol.

Prior to that, the kimage (_text) lays above PAGE_END with this layout:

0               -> VA_START                 : Usespace
VA_START        -> VA_START + 256M          : BPF JIT, Modules

VA_START + 256M -> PAGE_OFFSET - (~GB misc) : Vmalloc (KERNEL _text HERE)
PAGE_OFFSET -> ... : * Linear map *

And here we have:
VA_START = -1UL << VA_BITS
PAGE_OFFSET = -1UL << (VA_BITS - 1)
_text < -1UL << (VA_BITS - 1)

Kernel image lays somewhere between VA_START and PAGE_OFFSET, so we just calc VA_BITS by getting the highest unset bit of _text symbol address, and shift one less bit of VA_BITS to get page offset. This works as long as KASLR don't put kernel in a too high location (which is commented inline).

And after that commit, kernel layout have changed:

0               -> PAGE_OFFSET              : Userspace
PAGE_OFFSET     -> PAGE_END                 : * Linear map *
PAGE_END        -> PAGE_END + 128M          : bpf jit region
PAGE_END + 128M -> PAGE_END + 256MB         : modules
PAGE_END + 256M -> ...                      : vmalloc (KERNEL _text HERE)

Here we have:
PAGE_OFFSET = -1UL << VA_BITS
PAGE_END = -1UL << (VA_BITS - 1)
_text > -1UL << (VA_BITS - 1)

Kernel image now lays above PAGE_END, so we have to shift one more bit to get the VA_BITS, and shift the exact VA_BITS for PAGE_OFFSET.

We can simply check if "_text > -1UL << (VA_BITS - 1)" is true to judge which layout is being used and shift the page offset occordingly.

(rebased and stripped by Pingfan )

Cc Simon Horman <[email protected]>
Cc Philipp Rudo <[email protected]>

To: kexec at lists.infradead.org

---
 kexec/arch/arm64/kexec-arm64.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)
kexec/arch/arm64/kexec-arm64.c +7 -0
index 793799b..ce7a5bb 100644
@@ -923,13 +923,25 @@ out:
 
 int get_page_offset(unsigned long *page_offset)
 {
+	unsigned long long text_sym_addr, kernel_va_mid;
 	int ret;
 
+	text_sym_addr = get_kernel_sym("_text");
+	if (text_sym_addr == 0) {
+		fprintf(stderr, "Can't get the symbol of _text to calculate page_offset.\n");
+		return -1;
+	}
+
 	ret = get_va_bits();
 	if (ret < 0)
 		return ret;
 

kexec/arch/arm64/kexec-arm64.c out: · line 947

Quoted 10 lines
kexec/arch/arm64/kexec-arm64.c +6 -1
-	if (va_bits < 52)
+	/* Since kernel 5.4, kernel image is put above
+	 * UINT64_MAX << (va_bits - 1)
+	 */
+	kernel_va_mid = UINT64_MAX << (va_bits - 1);
+	/* older kernel */
+	if (text_sym_addr < kernel_va_mid)
 		*page_offset = UINT64_MAX << (va_bits - 1);
 	else
 		*page_offset = UINT64_MAX << va_bits;

I would drop the kernel_va_mid and simply use

*page_offset = UINT64_MAX << (va_bits - 1) if (*page_offset > text_sym_addr > *page_offset)
*page_offset = UINT64_MAX << va_bits

but that's more a matter of taste.

#11 lore raw

PATCH v4 4/4 arm64: fix PAGE_OFFSET calc for flipped mm

Quoted 90 lines · Philipp Rudo

Hi Pingfan,

Quoted 71 lines · On Tue, 18 Jan 2022 15:48:12 +0800 Pingfan Liu <[email protected]> wrote:

From: Kairui Song <[email protected]>

Since kernel commit 14c127c957c1 ('arm64: mm: Flip kernel VA space'), the memory layout on arm64 have changed, and kexec-tools can no longer get the the right PAGE_OFFSET based on _text symbol.

Prior to that, the kimage (_text) lays above PAGE_END with this layout:

0               -> VA_START                 : Usespace
VA_START        -> VA_START + 256M          : BPF JIT, Modules

VA_START + 256M -> PAGE_OFFSET - (~GB misc) : Vmalloc (KERNEL _text HERE)
PAGE_OFFSET -> ... : * Linear map *

And here we have:
VA_START = -1UL << VA_BITS
PAGE_OFFSET = -1UL << (VA_BITS - 1)
_text < -1UL << (VA_BITS - 1)

Kernel image lays somewhere between VA_START and PAGE_OFFSET, so we just calc VA_BITS by getting the highest unset bit of _text symbol address, and shift one less bit of VA_BITS to get page offset. This works as long as KASLR don't put kernel in a too high location (which is commented inline).

And after that commit, kernel layout have changed:

0               -> PAGE_OFFSET              : Userspace
PAGE_OFFSET     -> PAGE_END                 : * Linear map *
PAGE_END        -> PAGE_END + 128M          : bpf jit region
PAGE_END + 128M -> PAGE_END + 256MB         : modules
PAGE_END + 256M -> ...                      : vmalloc (KERNEL _text HERE)

Here we have:
PAGE_OFFSET = -1UL << VA_BITS
PAGE_END = -1UL << (VA_BITS - 1)
_text > -1UL << (VA_BITS - 1)

Kernel image now lays above PAGE_END, so we have to shift one more bit to get the VA_BITS, and shift the exact VA_BITS for PAGE_OFFSET.

We can simply check if "_text > -1UL << (VA_BITS - 1)" is true to judge which layout is being used and shift the page offset occordingly.

(rebased and stripped by Pingfan )

Cc Simon Horman <[email protected]>
Cc Philipp Rudo <[email protected]>

To: kexec at lists.infradead.org

---
 kexec/arch/arm64/kexec-arm64.c | 14 +++++++++++++-
 1 file changed, 13 insertions(+), 1 deletion(-)
kexec/arch/arm64/kexec-arm64.c +7 -0
index 793799b..ce7a5bb 100644
@@ -923,13 +923,25 @@ out:
 
 int get_page_offset(unsigned long *page_offset)
 {
+     unsigned long long text_sym_addr, kernel_va_mid;
      int ret;
 
+     text_sym_addr = get_kernel_sym("_text");
+     if (text_sym_addr == 0) {
+             fprintf(stderr, "Can't get the symbol of _text to calculate page_offset.\n");
+             return -1;
+     }
+
      ret = get_va_bits();
      if (ret < 0)
              return ret;
 

kexec/arch/arm64/kexec-arm64.c out: · line 947

Quoted 10 lines
kexec/arch/arm64/kexec-arm64.c +6 -1
-     if (va_bits < 52)
+     /* Since kernel 5.4, kernel image is put above
+      * UINT64_MAX << (va_bits - 1)
+      */
+     kernel_va_mid = UINT64_MAX << (va_bits - 1);
+     /* older kernel */
+     if (text_sym_addr < kernel_va_mid)
              *page_offset = UINT64_MAX << (va_bits - 1);
      else
              *page_offset = UINT64_MAX << va_bits;

I would drop the kernel_va_mid and simply use

Quoted 6 lines

*page_offset = UINT64_MAX << (va_bits - 1) if (*page_offset > text_sym_addr > *page_offset)
*page_offset = UINT64_MAX << va_bits

but that's more a matter of taste.

Ah, I kept kernel_va_mid dedicatedly to illustrate the purpose.

Quoted 2 lines

Thanks for your reviewing.

Regards,

Pingfan

#12 lore raw

v4 cover arm64: make phys_to_virt() correct

In reply to Pingfan Liu
Quoted 20 lines · Pingfan Liu

Currently phys_to_virt() does not work well on 52-bits VA arm64 kernel. One issue is contributed by phys_offset not signed.
The other is contributed by wrong page_offset.

v3 -> v4:
address the date type in [1/4] and [2/4]

v2 -> v3:
Discussed with Kairui off-list, 48-bits VA kernel can not handle flipped mm yet. So introducing [4/4], which judges whether the kernel is with flipped mm layout and adopt different formula accordingly.

As for [1-3/4], they are the same as [1-3/3] in V2.

v1 -> v2
Fix broken patch [2/3] in v1
Move arch_scan_vmcoreinfo declaration to util_lib/include/elf_info.h Using UINT64_MAX instread of 0xffffffffffffffff

Quoted 4 lines

Cc: Kairui Song <[email protected]>
Cc: Simon Horman <[email protected]>
Cc: Philipp Rudo <[email protected]>
To: kexec at lists.infradead.org

Thanks, series applied.

#13 lore raw

v4 cover arm64: make phys_to_virt() correct

Quoted 22 lines · Simon Horman
Quoted 20 lines · Pingfan Liu

Currently phys_to_virt() does not work well on 52-bits VA arm64 kernel. One issue is contributed by phys_offset not signed.
The other is contributed by wrong page_offset.

v3 -> v4:
address the date type in [1/4] and [2/4]

v2 -> v3:
Discussed with Kairui off-list, 48-bits VA kernel can not handle flipped mm yet. So introducing [4/4], which judges whether the kernel is with flipped mm layout and adopt different formula accordingly.

As for [1-3/4], they are the same as [1-3/3] in V2.

v1 -> v2
Fix broken patch [2/3] in v1
Move arch_scan_vmcoreinfo declaration to util_lib/include/elf_info.h Using UINT64_MAX instread of 0xffffffffffffffff

Quoted 7 lines
Quoted 4 lines

Cc: Kairui Song <[email protected]>
Cc: Simon Horman <[email protected]>
Cc: Philipp Rudo <[email protected]>
To: kexec at lists.infradead.org

Thanks, series applied.

Hi Simon and Philipp, appreciate the help and patient review.

Best Regards,

Pingfan