[Crash-utility] [PATCH v3] Speed up "kmem -[sS]" by optimizing is_page_ptr() for x86_64
changes v2 -> v3:
- move the setting point of machdep->is_page_ptr to machdep_init(SETUP_ENV)
changes v1 -> v2:
- rewrite based on the per-architecture function call Dave provided
- remove the part which used page.flags for non-VMEMMAP kernels
- add address range/position check first
- remove/optimize the calculations of mem_map and phys address
- modify the patch description
The "kmem -[sS]" commands can take several minutes to complete with
the following conditions:
- The system has a lot of memory sections with CONFIG_SPARSEMEM, and
- The kernel uses SLUB and it has a very long partial slab list.
crash> kmem -s dentry
CACHE NAME OBJSIZE ALLOCATED TOTAL SLABS SSIZE
ffff88017fc78a00 dentry 192 9038949 10045728 239184 8k
crash> kmem -s dentry | bash -c 'cat >/dev/null ; echo $SECONDS'
133
crash> kmem -S dentry | bash -c 'cat >/dev/null ; echo $SECONDS'
656
One of the causes is that is_page_ptr() in count_partial() determines
whether a given slub page address is a page struct by searching all
mem_sections available for the one which includes it.
With CONFIG_SPARSEMEM_VMEMMAP on x86_64, we can do that by checking
its address range and whether its calculated mem_section is valid.
With this patch, the computation amount can be significantly reduced
in that case.
crash> kmem -s dentry | bash -c 'cat >/dev/null ; echo $SECONDS'
1
crash> kmem -S dentry | bash -c 'cat >/dev/null ; echo $SECONDS'
1
Signed-off-by: Kazuhito Hagio <k-hagio(a)ab.jp.nec.com>
---
defs.h | 1 +
x86_64.c | 23 +++++++++++++++++++++++
2 files changed, 24 insertions(+)
diff --git a/defs.h b/defs.h
index 9663bd8..7998ebf 100644
--- a/defs.h
+++ b/defs.h
@@ -5133,6 +5133,7 @@ int vaddr_type(ulong, struct task_context *);
char *format_stack_entry(struct bt_info *bt, char *, ulong, ulong);
int in_user_stack(ulong, ulong);
int dump_inode_page(ulong);
+ulong valid_section_nr(ulong);
/*
diff --git a/x86_64.c b/x86_64.c
index 7449571..0d5e150 100644
--- a/x86_64.c
+++ b/x86_64.c
@@ -77,6 +77,7 @@ static void x86_64_calc_phys_base(void);
static int x86_64_is_module_addr(ulong);
static int x86_64_is_kvaddr(ulong);
static int x86_64_is_uvaddr(ulong, struct task_context *);
+static int x86_64_is_page_ptr(ulong, physaddr_t *);
static ulong *x86_64_kpgd_offset(ulong, int, int);
static ulong x86_64_upgd_offset(struct task_context *, ulong, int, int);
static ulong x86_64_upgd_offset_legacy(struct task_context *, ulong, int, int);
@@ -156,6 +157,7 @@ x86_64_init(int when)
{
case SETUP_ENV:
machdep->process_elf_notes = x86_process_elf_notes;
+ machdep->is_page_ptr = x86_64_is_page_ptr;
break;
case PRE_SYMTAB:
machdep->verify_symbol = x86_64_verify_symbol;
@@ -802,6 +804,7 @@ x86_64_dump_machdep_table(ulong arg)
fprintf(fp, " get_smp_cpus: x86_64_get_smp_cpus()\n");
fprintf(fp, " is_kvaddr: x86_64_is_kvaddr()\n");
fprintf(fp, " is_uvaddr: x86_64_is_uvaddr()\n");
+ fprintf(fp, " is_page_ptr: x86_64_is_page_ptr()\n");
fprintf(fp, " verify_paddr: x86_64_verify_paddr()\n");
fprintf(fp, " get_kvaddr_ranges: x86_64_get_kvaddr_ranges()\n");
fprintf(fp, " init_kernel_pgd: x86_64_init_kernel_pgd()\n");
@@ -1594,6 +1597,26 @@ x86_64_is_uvaddr(ulong addr, struct task_context *tc)
return (addr < USERSPACE_TOP);
}
+static int
+x86_64_is_page_ptr(ulong addr, physaddr_t *phys)
+{
+ ulong pfn, nr;
+
+ if (IS_SPARSEMEM() && (machdep->flags & VMEMMAP) &&
+ (addr >= VMEMMAP_VADDR && addr <= VMEMMAP_END) &&
+ !((addr - VMEMMAP_VADDR) % SIZE(page))) {
+
+ pfn = (addr - VMEMMAP_VADDR) / SIZE(page);
+ nr = pfn_to_section_nr(pfn);
+ if (valid_section_nr(nr)) {
+ if (phys)
+ *phys = PTOB(pfn);
+ return TRUE;
+ }
+ }
+ return FALSE;
+}
+
/*
* Find the kernel pgd entry..
* pgd = pgd_offset_k(addr);
--
1.8.3.1