在之前公司搞了5年的linux内核和驱动，打算翻翻笔记拿点东西出来写。
查内存泄露时自己想的一个办法，经过多次考验是有效的，在32位，64位的arm上都试过，检查出来过原生内核的内存泄露bug，后来在mips系统也用过一次。
简单的原理是，在内存分配的入口函数__alloc_pages_nodemask调用save_stack()把内存分配的堆栈保存下来，然后每隔一段时间比如两分钟，统计一下已分配出去的页是通过怎样的堆栈分配出去的，
比如以下例子，如果统计出来非常多的页是通过下面堆栈分配的，那就说明videobuf_alloc函数存在内存分配但不释放的情况。
[ 869.964000] [<800baf34>] slob_new_pages.clone.19+0x18/0x38
[ 869.968000] [<800bb138>] slob_alloc.clone.20+0x1e4/0x310
[ 869.976000] [<800bb3b4>] __kmalloc_node+0xbc/0xd8
[ 869.980000] [<802d7034>] videobuf_alloc+0x34/0x74
[ 869.984000] [<802d7160>] __videobuf_mmap_setup+0xec/0x154
[ 869.992000] [<802d79e4>] videobuf_reqbufs+0x190/0x22c
[ 869.996000] [<802d95fc>] gl5201vout_vidioc_reqbufs+0x94/0x114
[ 870.000000] [<802cc8c4>] __video_do_ioctl+0x1420/0x4114
[ 870.008000] [<802cf794>] video_ioctl2+0x1dc/0x558
[ 870.012000] [<800cf2dc>] vfs_ioctl+0xd0/0xec
[ 870.016000] [<800cfbf4>] do_vfs_ioctl+0x90/0x6b8
[ 870.020000] [<800d0260>] sys_ioctl+0x44/0x94
[ 870.024000] [<80018524>] stack_done+0x20/0x3c
下面是代码示例，首先是在一个初始化函数分配足够的内存用于保存堆栈，每一个page都要一块内存保存堆栈，512M内存含131072个页，需要131072*sizeof(struct my_save_condense)内存空间，
然后搞一个定期统计的work。
#include <linux/stacktrace.h>
struct my_save_condense {
struct stack_trace trace;
long entries[16];
int valid;
};
extern struct my_save_condense *my_stack_save2;
#define MAX_PAGE_COUNT 131072
static void my_work_proc(struct work_struct *ignored);
static DECLARE_DELAYED_WORK(my_work, my_work_proc);
extern void show_mem();
unsigned int pm_inited=0;
static void my_work_proc(struct work_struct *ignored)
{
show_mem();
schedule_delayed_work(&my_work, 120*HZ);
}
static int __init asoc_pm_init(void)
{
pr_info("[PM] ASOC Power Management\n");
schedule_delayed_work(&my_work, HZ*20);
my_stack_save2=(struct my_save_condense *)kmalloc(MAX_PAGE_COUNT*sizeof(struct my_save_condense));
pm_inited=1;
}
late_initcall(asoc_pm_init);
下面是在内存分配的地方，page_alloc.c的__alloc_pages_nodemask添加保存堆栈的代码，灰色高显的代码。
page_alloc.c
#include <linux/stacktrace.h>
struct my_save_condense {
struct stack_trace trace;
long entries[16];
int valid;
};
struct my_save_condense *my_stack_save2;
extern unsigned int pm_inited;
/*
* This is the 'heart' of the zoned buddy allocator.
*/
struct page *
__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
struct zonelist *zonelist, nodemask_t *nodemask)
{
enum zone_type high_zoneidx = gfp_zone(gfp_mask);
struct zone *preferred_zone;
struct page *page;
unsigned long pfn;
int migratetype = allocflags_to_migratetype(gfp_mask);
gfp_mask &= gfp_allowed_mask;
lockdep_trace_alloc(gfp_mask);
might_sleep_if(gfp_mask & __GFP_WAIT);
if (should_fail_alloc_page(gfp_mask, order))
return NULL;
/*
* Check the zones suitable for the gfp_mask contain at least one
* valid zone. It's possible to have an empty zonelist as a result
* of GFP_THISNODE and a memoryless node
*/
if (unlikely(!zonelist->_zonerefs->zone))
return NULL;
get_mems_allowed();
/* The preferred zone is used for statistics later */
first_zones_zonelist(zonelist, high_zoneidx, nodemask, &preferred_zone);
if (!preferred_zone) {
put_mems_allowed();
return NULL;
}
/* First allocation attempt */
page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET,
preferred_zone, migratetype);
if (unlikely(!page))
page = __alloc_pages_slowpath(gfp_mask, order,
zonelist, high_zoneidx, nodemask,
preferred_zone, migratetype);
put_mems_allowed();
trace_mm_page_alloc(page, order, gfp_mask, migratetype);
pfn=page_to_pfn(page);
if (pfn_valid(pfn)&&pm_inited==1){
my_stack_save2[pfn].valid = 0x12345678;
memset(&(my_stack_save2[pfn].entries[0]), 0, 64);
my_stack_save2[pfn].trace.entries = my_stack_save2[pfn].entries;
my_stack_save2[pfn].trace.max_entries = 16;
my_stack_save2[pfn].trace.skip = 2;
my_stack_save2[pfn].trace.nr_entries=0;
save_stack_trace(&my_stack_save2[pfn].trace);
}
return page;
}
定期统计的函数单独放在show_mem.c,在内核show_mem()函数进行修改。
遍历每个已分配出去的页，即page_count(page)>0的页，这里有一个compare_stack_my()函数，用来比较堆栈是否相同，相同的堆栈，页数累加，
这里的全局变量struct my_save_condense save_it_begin[250]，假定不同内存分配堆栈数量不超过250个，
最后打印有多少个堆栈，每个堆栈分配了多少个page。
show_mem.c
#include <linux/mm.h>
#include <linux/nmi.h>
#include <linux/quicklist.h>
#include <linux/stacktrace.h>
#include <linux/kallsyms.h>
struct my_save_condense {
struct stack_trace trace;
long entries[16];
int valid;
};
struct my_save_condense save_it_begin[250];
int now_pos_begin=0;
extern struct my_save_condense *my_stack_save2;
void compare_stack_my(struct my_save_condense *p_save,struct my_save_condense *save_array,int array_count, int *now_pos,int count)
{
int i,j,diff,k,index=0;
diff=0;
for (i=0;i<*now_pos;i++){
diff=0;
for(j=0;j<p_save->trace.nr_entries&&j<8;j++){
if(save_array[i].entries[j]!=p_save->entries[j]) {
diff=1;
break;
}
}
if(diff==0){
save_array[i].valid+=count;
index=i;
break;
}
}
if (diff==1 || *now_pos==0){
if(*now_pos>=array_count){
printk(KERN_ALERT"overflow\n");
} else {
*now_pos = *now_pos + 1;
memcpy(&save_array[*now_pos-1],p_save,sizeof(struct my_save_condense));
save_array[*now_pos-1].trace.entries = save_array[*now_pos-1].entries;
save_array[*now_pos-1].valid=count;
index=*now_pos-1;
}
}
}
void show_mem(void)
{
pg_data_t *pgdat;
unsigned long total = 0, reserved = 0, shared = 0,
nonshared = 0, highmem = 0,mapcount_shared=0,mapcount_nonshared=0,anon_not_mapped=0,file_not_mapped=0,my_vmalloc=0,
mapcount_nonshared_file=0,mapcount_nonshared_anon=0,
mapcount_shared_anon=0,mapcount_shared_file=0,
mapcount_shares_file=0,mapcount_shares_anon=0,my_private=0,my_slabnotfree=0,my_other=0,slab_count=0;
struct anon_vma *anon_vma;
struct anon_vma_chain *avc;
pgoff_t pgoff;
int j,k,l;
struct mm_struct *mm;
struct vm_area_struct *vma;
struct task_struct *t,*p;
char *q;
struct dentry *alias;
struct inode *ii;
pte_t *pte;
spinlock_t *ptl;
int c=1;
printk("Mem-Info:\n");
show_free_areas();
memset(save_it_begin,0,250*sizeof(struct my_save_condense));
now_pos_begin=0;
memset(save_it_end,0,250*sizeof(struct my_save_condense));
now_pos_end=0;
for_each_online_pgdat(pgdat) {
unsigned long i, flags;
pgdat_resize_lock(pgdat, &flags);
for (i = 0; i < pgdat->node_spanned_pages; i+=c) {
struct page *page;
unsigned long pfn = pgdat->node_start_pfn + i;
if (unlikely(!(i % MAX_ORDER_NR_PAGES)))
touch_nmi_watchdog();
if (!pfn_valid(pfn))
continue;
page = pfn_to_page(pfn);
if (PageHighMem(page))
highmem++;
if (PageReserved(page)){
reserved++;
c=1;
}
else if (page_count(page) == 1){
if(PageCompound(page)){
if(PageTail(page)){
//destroy_compound_page先__ClearPageHead,再__ClearPageTail,错误统计几次,无所谓
printk(KERN_ALERT"pfn %ld PageTail first_page pfn %ld\n",pfn,
page_to_pfn(page->first_page));
c=1;
} else {
struct page *pg=page+1;;
c=1<<((int)(page[1].lru.prev));
if(c>2048){
printk(KERN_ALERT"possible?compound page over 8M\n");
}
//printk(KERN_DEBUG"Compound pfn %ld count %d\n",pfn,c);
}
} else {
c=1;
}
nonshared+=c;
if(PageAnon(page)){
if(page_mapcount(page)==0){
anon_not_mapped+=c;
}else if (page_mapcount(page)>1)
{
pr_alert(".possible? anon page_count 1 mapcount %d pfn %d\n",page_mapcount(page),pfn);
}
}else if (PageSlab(page)){
slab_count++;
if (my_stack_save2[pfn].valid==0x12345678 ){
compare_stack_my(&my_stack_save2[pfn],
save_it_end,
250,
&now_pos_end,
c);
}
}else{
if(page_mapcount(page)==0){
struct address_space *mapping1 = page->mapping;
if(mapping1 == NULL){
if (my_stack_save2[pfn].valid==0x12345678 ){
compare_stack_my(&my_stack_save2[pfn],
save_it_begin,
250,
&now_pos_begin,
c);
}
if(PageHead(page)&&!(page->flags & (1 << PG_slab))){
my_slabnotfree+=c;
} else if (page->flags & (1 << PG_private)){
my_private+=c;
} else {
my_other+=c;//main thread_info,pte
}
} else {
file_not_mapped+=c;
}
}else if (page_mapcount(page)>1)
{
pr_alert(".possible? file page_count 1 mapcount %d pfn %d\n",page_mapcount(page),pfn);
}
}
}
else if (page_count(page) > 1){
if(PageCompound(page)){
if(PageTail(page)){
printk(KERN_ALERT"pfn %ld PageTail first_page pfn %ld\n",pfn,
page_to_pfn(page->first_page));
c=1;
} else {
struct page *pg=page+1;;
c=1<<((int)(page[1].lru.prev));
if(c>2048){
printk(KERN_ALERT"possible?compound page over 8M\n");
}
//printk(KERN_DEBUG"Compound pfn %ld count %d %d\n",pfn,c);
}
} else {
c=1;
}
shared+=c;
if(PageAnon(page)){
if(page_mapcount(page)==0){
pr_alert("possible? anon page_count %d mapcount 0 pfn %d\n",page_count(page),pfn);
}
}else if(PageSlab(page)){
slab_count++;
if (my_stack_save2[pfn].valid==0x12345678 ){
compare_stack_my(&my_stack_save2[pfn],
save_it_end,
250,
&now_pos_end,
c);
}
}else{
if(page_mapcount(page)==0){
struct address_space *mapping1 = page->mapping;
if(mapping1 == NULL){
if (my_stack_save2[pfn].valid==0x12345678 ){
compare_stack_my(&my_stack_save2[pfn],
save_it_begin,
250,
&now_pos_begin,
c);
}
if(PageHead(page)&&!(page->flags & (1 << PG_slab))){
my_slabnotfree+=c;
} else if (page->flags & (1 << PG_private)){
my_private+=c;
}else {
my_other+=c;//main thread_info pte
}
} else {
file_not_mapped+=c;
}
}
}
} else {
c=1;
}
total+=c;
}
pgdat_resize_unlock(pgdat, &flags);
}
printk("%lu pages RAM\n", total);
#ifdef CONFIG_HIGHMEM
printk("%lu pages HighMem\n", highmem);
#endif
printk("%lu pages reserved\n", reserved);
printk("%lu pages shared\n", shared);
printk("%lu pages non-shared\n", nonshared);
#ifdef CONFIG_QUICKLIST
printk("%lu pages in pagetable cache\n",
quicklist_total_size());
#endif
printk(KERN_DEBUG"anon_not_mapped %lu file_not_mapped %lu private %lu slabnotfree %lu other %lu slab %d\n",
anon_not_mapped,file_not_mapped,my_private,my_slabnotfree,my_other,slab_count);
printk("now_pos_begin %d\n",now_pos_begin);
for(j=0;j<now_pos_begin;j++){
if(save_it_begin[j].valid>50){
printk(KERN_DEBUG"[%d] nr_entries %d valid %d\n",j,
save_it_begin[j].trace.nr_entries,save_it_begin[j].valid);
print_stack_trace(&(save_it_begin[j].trace),0);
}
}
}
345
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
