[PATCH v3 0/2] PPC64: Add virtual to physical address translation support for vmalloc region

[PATCH v3 0/2] PPC64: Add virtual to physical address translation support for vmalloc region

[Hari Bathini]

In makedumpfile tool, currently there is no support for virtual to physical
address translation of vmalloc region  on PPC64 architecture. These patches
add this support.  The patches have been tested by  erasing data in vmalloc
region, verfying for eraseinfo in the resultant ELF dump and verifiying the
resultant ELF dump for sanity using crash utility. Also, tested the patches
successfully with eppic scripts.

Changes in v3:
Added the buffer to read page tables in DumpInfo structure and freeing it
before exiting the tool

---

Hari Bathini (2):
      makedumpfile: Initialize for vmalloc address translation support in PPC64 arch
      makedumpfile: vtop address translation support for vmalloc region in PPC64 arch


 arch/ppc64.c   |  170 ++++++++++++++++++++++++++++++++++++++++++++++++++++----
 makedumpfile.c |    9 +++
 makedumpfile.h |   83 +++++++++++++++++++++++++++
 3 files changed, 249 insertions(+), 13 deletions(-)

-- 
- Hari


_______________________________________________
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec

[PATCH v3 1/2] makedumpfile: Initialize for vmalloc address translation support in PPC64 arch

[Hari Bathini]

In this patch, initializing of page table information that is to be
used for virtual to physical address translation  of vmalloc region
is added to the initialization sequence.

Signed-off-by: Hari Bathini <hbathini@linux.vnet.ibm.com>
---
 arch/ppc64.c   |   72 ++++++++++++++++++++++++++++++++++++++++++++++++++
 makedumpfile.c |    7 +++++
 makedumpfile.h |   80 +++++++++++++++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 158 insertions(+), 1 deletion(-)

diff --git a/arch/ppc64.c b/arch/ppc64.c
index 09c0eb3..8c7eaa7 100644
--- a/arch/ppc64.c
+++ b/arch/ppc64.c
@@ -149,6 +149,58 @@ ppc64_vmemmap_init(void)
 	return TRUE;
 }
 
+static int
+ppc64_vmalloc_init(void)
+{
+	if (info->page_size == 65536) {
+		/*
+		 * 64K pagesize
+		 */
+		if (info->kernel_version >= KERNEL_VERSION(3, 10, 0)) {
+			info->l1_index_size = PTE_INDEX_SIZE_L4_64K_3_10;
+			info->l2_index_size = PMD_INDEX_SIZE_L4_64K_3_10;
+			info->l3_index_size = PUD_INDEX_SIZE_L4_64K;
+		} else {
+			info->l1_index_size = PTE_INDEX_SIZE_L4_64K;
+			info->l2_index_size = PMD_INDEX_SIZE_L4_64K;
+			info->l3_index_size = PUD_INDEX_SIZE_L4_64K;
+		}
+
+		info->pte_shift = SYMBOL(demote_segment_4k) ?
+			PTE_SHIFT_L4_64K_V2 : PTE_SHIFT_L4_64K_V1;
+		info->l2_masked_bits = PMD_MASKED_BITS_64K;
+	} else {
+		/*
+		 * 4K pagesize
+		 */
+		info->l1_index_size = PTE_INDEX_SIZE_L4_4K;
+		info->l2_index_size = PMD_INDEX_SIZE_L4_4K;
+		info->l3_index_size = PUD_INDEX_SIZE_L4_4K;
+
+		info->pte_shift = PTE_SHIFT_L4_4K;
+		info->l2_masked_bits = PMD_MASKED_BITS_4K;
+	}
+
+	/*
+	 * Compute ptrs per each level
+	 */
+	info->l1_shift = info->page_shift;
+	info->ptrs_per_l1 = (1 << info->l1_index_size);
+	info->ptrs_per_l2 = (1 << info->l2_index_size);
+	info->ptrs_per_l3 = (1 << info->l3_index_size);
+
+	info->ptrs_per_pgd = info->ptrs_per_l3;
+
+	/*
+	 * Compute shifts
+	 */
+	info->l2_shift = info->l1_shift + info->l1_index_size;
+	info->l3_shift = info->l2_shift + info->l2_index_size;
+	info->l4_shift = info->l3_shift + info->l3_index_size;
+
+	return TRUE;
+}
+
 /*
  *  If the vmemmap address translation information is stored in the kernel,
  *  make the translation.
@@ -251,6 +303,15 @@ get_machdep_info_ppc64(void)
 	info->vmalloc_start = vmalloc_start;
 	DEBUG_MSG("vmalloc_start: %lx\n", vmalloc_start);
 
+	if (SYMBOL(swapper_pg_dir) != NOT_FOUND_SYMBOL) {
+		info->kernel_pgd = SYMBOL(swapper_pg_dir);
+	} else if (SYMBOL(cpu_pgd) != NOT_FOUND_SYMBOL) {
+		info->kernel_pgd = SYMBOL(cpu_pgd);
+	} else {
+		ERRMSG("No swapper_pg_dir or cpu_pgd symbols exist\n");
+		return FALSE;
+	}
+
 	if (SYMBOL(vmemmap_list) != NOT_FOUND_SYMBOL) {
 		info->vmemmap_start = VMEMMAP_REGION_ID << REGION_SHIFT;
 		info->vmemmap_end = info->vmemmap_start;
@@ -265,6 +326,17 @@ get_machdep_info_ppc64(void)
 }
 
 int
+get_versiondep_info_ppc64()
+{
+	if (ppc64_vmalloc_init() == FALSE) {
+		ERRMSG("Can't initialize for vmalloc translation\n");
+		return FALSE;
+	}
+
+	return TRUE;
+}
+
+int
 is_vmalloc_addr_ppc64(unsigned long vaddr)
 {
 	return (info->vmalloc_start && vaddr >= info->vmalloc_start);
diff --git a/makedumpfile.c b/makedumpfile.c
index 3884aa5..f54ff8f 100644
--- a/makedumpfile.c
+++ b/makedumpfile.c
@@ -1181,6 +1181,9 @@ get_symbol_info(void)
 	SYMBOL_INIT(mmu_psize_defs, "mmu_psize_defs");
 	SYMBOL_INIT(mmu_vmemmap_psize, "mmu_vmemmap_psize");
 
+	SYMBOL_INIT(cpu_pgd, "cpu_pgd");
+	SYMBOL_INIT(demote_segment_4k, "demote_segment_4k");
+
 	return TRUE;
 }
 
@@ -1694,6 +1697,8 @@ write_vmcoreinfo_data(void)
 	WRITE_SYMBOL("vmemmap_list", vmemmap_list);
 	WRITE_SYMBOL("mmu_psize_defs", mmu_psize_defs);
 	WRITE_SYMBOL("mmu_vmemmap_psize", mmu_vmemmap_psize);
+	WRITE_SYMBOL("cpu_pgd", cpu_pgd);
+	WRITE_SYMBOL("demote_segment_4k", demote_segment_4k);
 
 	/*
 	 * write the structure size of 1st kernel
@@ -2033,6 +2038,8 @@ read_vmcoreinfo(void)
 	READ_SYMBOL("vmemmap_list", vmemmap_list);
 	READ_SYMBOL("mmu_psize_defs", mmu_psize_defs);
 	READ_SYMBOL("mmu_vmemmap_psize", mmu_vmemmap_psize);
+	READ_SYMBOL("cpu_pgd", cpu_pgd);
+	READ_SYMBOL("demote_segment_4k", demote_segment_4k);
 
 	READ_STRUCTURE_SIZE("page", page);
 	READ_STRUCTURE_SIZE("mem_section", mem_section);
diff --git a/makedumpfile.h b/makedumpfile.h
index 9402f05..33794f5 100644
--- a/makedumpfile.h
+++ b/makedumpfile.h
@@ -586,6 +586,58 @@ do { \
 #define _MAX_PHYSMEM_BITS_3_7   (46)
 #define REGION_SHIFT            (60UL)
 #define VMEMMAP_REGION_ID       (0xfUL)
+
+#define PGDIR_SHIFT	\
+	(PAGESHIFT() + (PAGESHIFT() - 3) + (PAGESHIFT() - 2))
+#define PMD_SHIFT       (PAGESHIFT() + (PAGESHIFT() - 3))
+
+/* shift to put page number into pte */
+#define PTE_SHIFT 16
+
+#define PTE_INDEX_SIZE  9
+#define PMD_INDEX_SIZE  10
+#define PGD_INDEX_SIZE  10
+
+#define PTRS_PER_PTE    (1 << PTE_INDEX_SIZE)
+#define PTRS_PER_PMD    (1 << PMD_INDEX_SIZE)
+#define PTRS_PER_PGD    (1 << PGD_INDEX_SIZE)
+
+#define PGD_OFFSET(vaddr)       ((vaddr >> PGDIR_SHIFT) & 0x7ff)
+#define PMD_OFFSET(vaddr)       ((vaddr >> PMD_SHIFT) & (PTRS_PER_PMD - 1))
+
+/* 4-level page table support */
+
+/* 4K pagesize */
+#define PTE_INDEX_SIZE_L4_4K  9
+#define PMD_INDEX_SIZE_L4_4K  7
+#define PUD_INDEX_SIZE_L4_4K  7
+#define PGD_INDEX_SIZE_L4_4K  9
+#define PTE_SHIFT_L4_4K  17
+#define PMD_MASKED_BITS_4K  0
+
+/* 64K pagesize */
+#define PTE_INDEX_SIZE_L4_64K   12
+#define PMD_INDEX_SIZE_L4_64K   12
+#define PUD_INDEX_SIZE_L4_64K   0
+#define PGD_INDEX_SIZE_L4_64K   4
+#define PTE_INDEX_SIZE_L4_64K_3_10  8
+#define PMD_INDEX_SIZE_L4_64K_3_10  10
+#define PGD_INDEX_SIZE_L4_64K_3_10  12
+#define PTE_SHIFT_L4_64K_V1  32
+#define PTE_SHIFT_L4_64K_V2  30
+#define PMD_MASKED_BITS_64K  0x1ff
+
+#define L4_MASK		\
+	(info->kernel_version >= KERNEL_VERSION(3, 10, 0) ? 0xfff : 0x1ff)
+#define L4_OFFSET(vaddr)	((vaddr >> (info->l4_shift)) & L4_MASK)
+
+#define PGD_OFFSET_L4(vaddr)	\
+	((vaddr >> (info->l3_shift)) & (info->ptrs_per_l3 - 1))
+
+#define PMD_OFFSET_L4(vaddr)	\
+	((vaddr >> (info->l2_shift)) & (info->ptrs_per_l2 - 1))
+
+#define _PAGE_PRESENT		0x1UL
 #endif
 
 #ifdef __powerpc32__
@@ -731,10 +783,11 @@ unsigned long long vaddr_to_paddr_x86_64(unsigned long vaddr);
 
 #ifdef __powerpc64__ /* powerpc64 */
 int get_machdep_info_ppc64(void);
+int get_versiondep_info_ppc64(void);
 unsigned long long vaddr_to_paddr_ppc64(unsigned long vaddr);
 #define get_phys_base()		TRUE
 #define get_machdep_info()	get_machdep_info_ppc64()
-#define get_versiondep_info()	TRUE
+#define get_versiondep_info()	get_versiondep_info_ppc64()
 #define vaddr_to_paddr(X)	vaddr_to_paddr_ppc64(X)
 #endif          /* powerpc64 */
 
@@ -932,6 +985,25 @@ struct DumpInfo {
 	struct ppc64_vmemmap	*vmemmap_list;
 
 	/*
+	 * page table info for ppc64
+	 */
+	int		ptrs_per_pgd;
+	uint		l3_index_size;
+	uint		l2_index_size;
+	uint		l1_index_size;
+	uint		ptrs_per_l3;
+	uint		ptrs_per_l2;
+	uint		ptrs_per_l1;
+	uint		l4_shift;
+	uint		l3_shift;
+	uint		l2_shift;
+	uint		l1_shift;
+	uint		pte_shift;
+	uint		l2_masked_bits;
+	ulong		kernel_pgd;
+	char		*page_buf; /* Page buffer to read page tables */
+
+	/*
 	 * Filter config file containing filter commands to filter out kernel
 	 * data from vmcore.
 	 */
@@ -1192,6 +1264,12 @@ struct symbol_table {
 	unsigned long long		vmemmap_list;
 	unsigned long long		mmu_vmemmap_psize;
 	unsigned long long		mmu_psize_defs;
+
+	/*
+	 * vm related symbols for ppc64 arch
+	 */
+	unsigned long long		cpu_pgd;
+	unsigned long long		demote_segment_4k;
 };
 
 struct size_table {


_______________________________________________
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec

[PATCH v3 2/2] makedumpfile: vtop address translation support for vmalloc region in PPC64 arch

[Hari Bathini]

This patch adds virtual to physical address translation support
for vmalloc region in PPC64 architecture.

Signed-off-by: Hari Bathini <hbathini@linux.vnet.ibm.com>
---
 arch/ppc64.c   |   98 +++++++++++++++++++++++++++++++++++++++++++++++++-------
 makedumpfile.c |    2 +
 makedumpfile.h |    3 ++
 3 files changed, 91 insertions(+), 12 deletions(-)

diff --git a/arch/ppc64.c b/arch/ppc64.c
index 8c7eaa7..89a7f05 100644
--- a/arch/ppc64.c
+++ b/arch/ppc64.c
@@ -224,6 +224,83 @@ ppc64_vmemmap_to_phys(unsigned long vaddr)
 	return paddr;
 }
 
+static unsigned long long
+ppc64_vtop_level4(unsigned long vaddr)
+{
+	ulong *level4, *level4_dir;
+	ulong *page_dir, *page_middle;
+	ulong *page_table;
+	unsigned long long level4_pte, pgd_pte;
+	unsigned long long pmd_pte, pte;
+	unsigned long long paddr = NOT_PADDR;
+
+	if (info->page_buf == NULL) {
+		/*
+		 * This is the first vmalloc address translation request
+		 */
+		info->page_buf = (char *)calloc(1, PAGESIZE());
+		if (info->page_buf == NULL) {
+			ERRMSG("Can't allocate memory to read page tables. %s\n",
+			       strerror(errno));
+			return NOT_PADDR;
+		}
+	}
+
+	level4 = (ulong *)info->kernel_pgd;
+	level4_dir = (ulong *)((ulong *)level4 + L4_OFFSET(vaddr));
+	if (!readmem(VADDR, PAGEBASE(level4), info->page_buf, PAGESIZE())) {
+		ERRMSG("Can't read level4 page: 0x%llx\n", PAGEBASE(level4));
+		return NOT_PADDR;
+	}
+	level4_pte = ULONG((info->page_buf + PAGEOFFSET(level4_dir)));
+	if (!level4_pte)
+		return NOT_PADDR;
+
+	/*
+	 * Sometimes we don't have level3 pagetable entries
+	 */
+	if (info->l3_index_size != 0) {
+		page_dir = (ulong *)((ulong *)level4_pte + PGD_OFFSET_L4(vaddr));
+		if (!readmem(VADDR, PAGEBASE(level4_pte), info->page_buf, PAGESIZE())) {
+			ERRMSG("Can't read PGD page: 0x%llx\n", PAGEBASE(level4_pte));
+			return NOT_PADDR;
+		}
+		pgd_pte = ULONG((info->page_buf + PAGEOFFSET(page_dir)));
+		if (!pgd_pte)
+			return NOT_PADDR;
+	} else {
+		pgd_pte = level4_pte;
+	}
+
+	page_middle = (ulong *)((ulong *)pgd_pte + PMD_OFFSET_L4(vaddr));
+	if (!readmem(VADDR, PAGEBASE(pgd_pte), info->page_buf, PAGESIZE())) {
+		ERRMSG("Can't read PMD page: 0x%llx\n", PAGEBASE(pgd_pte));
+		return NOT_PADDR;
+	}
+	pmd_pte = ULONG((info->page_buf + PAGEOFFSET(page_middle)));
+	if (!(pmd_pte))
+		return NOT_PADDR;
+
+	page_table = (ulong *)(pmd_pte & ~(info->l2_masked_bits))
+			+ (BTOP(vaddr) & (info->ptrs_per_l1 - 1));
+	if (!readmem(VADDR, PAGEBASE(pmd_pte), info->page_buf, PAGESIZE())) {
+		ERRMSG("Can't read page table: 0x%llx\n", PAGEBASE(pmd_pte));
+		return NOT_PADDR;
+	}
+	pte = ULONG((info->page_buf + PAGEOFFSET(page_table)));
+	if (!(pte & _PAGE_PRESENT)) {
+		ERRMSG("Page not present!\n");
+		return NOT_PADDR;
+	}
+
+	if (!pte)
+		return NOT_PADDR;
+
+	paddr = PAGEBASE(PTOB(pte >> info->pte_shift)) + PAGEOFFSET(vaddr);
+
+	return paddr;
+}
+
 int
 set_ppc64_max_physmem_bits(void)
 {
@@ -347,10 +424,18 @@ vaddr_to_paddr_ppc64(unsigned long vaddr)
 {
 	unsigned long long paddr;
 
+	if ((info->flag_vmemmap)
+	    && (vaddr >= info->vmemmap_start)) {
+		return ppc64_vmemmap_to_phys(vaddr);
+	}
+
 	paddr = vaddr_to_paddr_general(vaddr);
 	if (paddr != NOT_PADDR)
 		return paddr;
 
+	if (!is_vmalloc_addr_ppc64(vaddr))
+		return (vaddr - info->kernel_start);
+
 	if ((SYMBOL(vmap_area_list) == NOT_FOUND_SYMBOL)
 	    || (OFFSET(vmap_area.va_start) == NOT_FOUND_STRUCTURE)
 	    || (OFFSET(vmap_area.list) == NOT_FOUND_STRUCTURE)) {
@@ -360,19 +445,8 @@ vaddr_to_paddr_ppc64(unsigned long vaddr)
 			return NOT_PADDR;
 		}
 	}
-	if (!is_vmalloc_addr_ppc64(vaddr))
-		return (vaddr - info->kernel_start);
 
-	if ((info->flag_vmemmap)
-	    && (vaddr >= info->vmemmap_start)) {
-		return ppc64_vmemmap_to_phys(vaddr);
-	}
-
-	/*
-	 * TODO: Support vmalloc translation.
-	 */
-	ERRMSG("This makedumpfile does not support vmalloc translation.\n");
-	return NOT_PADDR;
+	return ppc64_vtop_level4(vaddr);
 }
 
 #endif /* powerpc64 */
diff --git a/makedumpfile.c b/makedumpfile.c
index f54ff8f..11cd473 100644
--- a/makedumpfile.c
+++ b/makedumpfile.c
@@ -9297,6 +9297,8 @@ out:
 			free(info->splitting_info);
 		if (info->p2m_mfn_frame_list != NULL)
 			free(info->p2m_mfn_frame_list);
+		if (info->page_buf != NULL)
+			free(info->page_buf);
 		free(info);
 	}
 	free_elf_info();
diff --git a/makedumpfile.h b/makedumpfile.h
index 33794f5..eba9798 100644
--- a/makedumpfile.h
+++ b/makedumpfile.h
@@ -158,6 +158,9 @@ isAnon(unsigned long mapping)
 	return ((unsigned long)mapping & PAGE_MAPPING_ANON) != 0;
 }
 
+#define PTOB(X)			(((unsigned long long)(X)) << PAGESHIFT())
+#define BTOP(X)			(((unsigned long long)(X)) >> PAGESHIFT())
+
 #define PAGESIZE()		(info->page_size)
 #define PAGESHIFT()		(info->page_shift)
 #define PAGEOFFSET(X)		(((unsigned long long)(X)) & (PAGESIZE() - 1))


_______________________________________________
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec

RE: [PATCH v3 0/2] PPC64: Add virtual to physical address translation support for vmalloc region

[Atsushi Kumagai]

>In makedumpfile tool, currently there is no support for virtual to physical
>address translation of vmalloc region  on PPC64 architecture. These patches
>add this support.  The patches have been tested by  erasing data in vmalloc
>region, verfying for eraseinfo in the resultant ELF dump and verifiying the
>resultant ELF dump for sanity using crash utility. Also, tested the patches
>successfully with eppic scripts.
>
>Changes in v3:
>Added the buffer to read page tables in DumpInfo structure and freeing it
>before exiting the tool

Thanks Hari, I'll merge this version into v1.5.7.


Atsushi Kumagai

>---
>
>Hari Bathini (2):
>      makedumpfile: Initialize for vmalloc address translation support in PPC64 arch
>      makedumpfile: vtop address translation support for vmalloc region in PPC64 arch
>
>
> arch/ppc64.c   |  170 ++++++++++++++++++++++++++++++++++++++++++++++++++++----
> makedumpfile.c |    9 +++
> makedumpfile.h |   83 +++++++++++++++++++++++++++
> 3 files changed, 249 insertions(+), 13 deletions(-)
>
>--
>- Hari
_______________________________________________
kexec mailing list
kexec@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/kexec