varun-r-mallya/sysprof
1
0
Fork 0
mirror of https://github.com/varun-r-mallya/sysprof.git synced 2026-02-12 16:10:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add a memcpy() that uses nontemporal writes. (copy_kernel_stack): Add

Browse Source 
2008-02-16  Soren Sandmann  <sandmann@redhat.com>

	* module/sysprof-module.c (nt_memcpy): Add a memcpy() that uses
	nontemporal writes.
	(copy_kernel_stack): Add commented-out use of it here.



svn path=/trunk/; revision=396
This commit is contained in:
Soren Sandmann
2008-02-16 15:54:40 +00:00
committed by Søren Sandmann Pedersen
parent 64d220e346
commit 9d2f7a1a9d
2 changed files with 42 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
ChangeLog
View File

@ -1,3 +1,9 @@
2008-02-16 Soren Sandmann <sandmann@redhat.com>
* module/sysprof-module.c (nt_memcpy): Add a memcpy() that uses
nontemporal writes.
(copy_kernel_stack): Add commented-out use of it here.
2007-11-20 Soren Sandmann <sandmann@daimi.au.dk> 2007-11-20 Soren Sandmann <sandmann@daimi.au.dk>
* binfile.c (find_separate_debug_file): Use hash table instead of * binfile.c (find_separate_debug_file): Use hash table instead of

37
module/sysprof-module.c
View File

@ -112,6 +112,38 @@ minimum (int a, int b)
return a > b ? b : a; return a > b ? b : a;
} }
static void
nt_memcpy (void *dst, void *src, int n_bytes)
{
#if defined(CONFIG_X86_64) || defined(CONFIG_X86)
if (((unsigned long)dst & 3) ||
((unsigned long)src & 3)) {
memcpy (dst, src, n_bytes);
}
else {
int i;
if ((unsigned long)src & 7) {
*(uint32_t *)dst = *(uint32_t *)src;
src += 4;
dst += 4;
}
for (i = 0; i < n_bytes; i += 8) {
__asm__ __volatile__ (
" movq (%0, %2), %%mm0\n"
" movntq %%mm0, (%1, %2)\n"
:
: "r" (src), "r" (dst), "r" (i)
: "memory");
}
__asm__ __volatile__ (
"emms\n");
}
#else
memcpy (dst, src, n_bytes);
#endif
}
static struct pt_regs * static struct pt_regs *
copy_kernel_stack (struct pt_regs *regs, copy_kernel_stack (struct pt_regs *regs,
SysprofStackTrace *trace) SysprofStackTrace *trace)
@ -137,6 +169,9 @@ copy_kernel_stack (struct pt_regs *regs,
sizeof (trace->kernel_stack)); sizeof (trace->kernel_stack));
if (n_bytes > 0) { if (n_bytes > 0) {
#if 0
nt_memcpy (&(trace->kernel_stack[1]), esp, n_bytes);
#endif
memcpy (&(trace->kernel_stack[1]), esp, n_bytes); memcpy (&(trace->kernel_stack[1]), esp, n_bytes);
trace->n_kernel_words += (n_bytes) / sizeof (void *); trace->n_kernel_words += (n_bytes) / sizeof (void *);
@ -203,7 +238,7 @@ heuristic_trace (struct pt_regs *regs,
if (vma && vma->vm_flags & VM_EXEC) if (vma && vma->vm_flags & VM_EXEC)
if (vma->vm_start <= x && x <= vma->vm_end) if (vma->vm_start <= x && x <= vma->vm_end)
trace->addresses[j++] = x; trace->addresses[j++] = (void *)x;
} }
#if 0 #if 0