Re: [Crash-utility] [PATCH RFC] Add "kmem -r" to display accumulated slab statistics like /proc/slabinfo

----- Original Message ----- > Nowadays, the "kmem -s" output can become very long vertically too, > due to the memcg kmem caches. It look like the longer a system has > run, the longer it becomes. > > crash> kmem -s | wc -l > 19855 > > On the other hand, since /proc/slabinfo accumulates the values of > each slab_root_caches and its children, it's still short relatively. > And I think there are many cases that support folks want to see the > accumulated values like /proc/slabinfo from vmcore, in order to > grasp the overview of slab activity quickly. > > We can use something like the attached script to accumulate them, > but I believe it would be more useful to implement it in crash. > > This patch introduces the "kmem -r" option to imitate /proc/slabinfo, > but it is limited to CONFIG_SLUB for now. And it looks like it's limited to Linux 4.11 and later, correct?

Thanks, Dave > > I tested this patch with the kmem-s2r.awk script: > > crash> kmem -s | awk -f kmem-s2r.awk > kmem-s2r.txt > crash> kmem -r > kmem-r.txt > > # diff -u kmem-s2r.txt kmem-r.txt > > Supported: > crash> kmem -r > crash> kmem -r list > crash> kmem -r <slab name> > > Signed-off-by: Kazuhito Hagio <k-hagio(a)ab.jp.nec.com&gt; > --- > defs.h | 5 ++ > help.c | 10 +-- > memory.c | 219 > ++++++++++++++++++++++++++++++++++++++++++++++++++++++++------ > symbols.c | 9 +++ > 4 files changed, 220 insertions(+), 23 deletions(-) > > diff --git a/defs.h b/defs.h > index 6fdb478..8687ff1 100644 > --- a/defs.h > +++ b/defs.h > @@ -2032,6 +2032,10 @@ struct offset_table { /* stash of > commonly-used offsets */ > long bpf_prog_aux_user; > long user_struct_uid; > long idr_cur; > + long kmem_cache_memcg_params; > + long memcg_cache_params___root_caches_node; > + long memcg_cache_params_children; > + long memcg_cache_params_children_node; > }; > > struct size_table { /* stash of commonly-used sizes */ > @@ -2438,6 +2442,7 @@ struct vm_table { /* kernel VM-related > data */ > #define PAGEFLAGS (0x4000000) > #define SLAB_OVERLOAD_PAGE (0x8000000) > #define SLAB_CPU_CACHE (0x10000000) > +#define SLAB_ROOT_CACHES (0x20000000) > > #define IS_FLATMEM() (vt->flags & FLATMEM) > #define IS_DISCONTIGMEM() (vt->flags & DISCONTIGMEM) > diff --git a/help.c b/help.c > index aeeb056..ee8b999 100644 > --- a/help.c > +++ b/help.c > @@ -6448,7 +6448,7 @@ char *help_kmem[] = { > "kmem", > "kernel memory", > "[-f|-F|-c|-C|-i|-v|-V|-n|-z|-o|-h] [-p | -m member[,member]]\n" > -" [[-s|-S] [slab] [-I slab[,slab]]] [-g [flags]] [[-P] address]]", > +" [[-s|-S|-r] [slab] [-I slab[,slab]]] [-g [flags]] [[-P] address]]", > " This command displays information about the use of kernel memory.\n", > " -f displays the contents of the system free memory headers.", > " also verifies that the page count equals nr_free_pages.", > @@ -6490,10 +6490,12 @@ char *help_kmem[] = { > " slab data for each per-cpu slab is displayed, along with the", > " address of each kmem_cache_node, its count of full and > partial", > " slabs, and a list of all tracked slabs.", > -" slab when used with -s or -S, limits the command to only the slab > cache", > -" of name \"slab\". If the slab argument is \"list\", then", > +" -r displays accumulated kmalloc() slab data of each > slab_root_caches", > +" and its children. Available only if CONFIG_SLUB for now.", > +" slab when used with -s, -S or -r, limits the command to only the > slab", > +" cache of name \"slab\". If the slab argument is \"list\", > then", > " all slab cache names and addresses are listed.", > -" -I slab when used with -s or -S, one or more slab cache names in a", > +" -I slab when used with -s, -S or -r, one or more slab cache names in > a", > " comma-separated list may be specified as slab caches to > ignore.", > " -g displays the enumerator value of all bits in the page > structure's", > " \"flags\" field.", > diff --git a/memory.c b/memory.c > index e02ba68..1501b21 100644 > --- a/memory.c > +++ b/memory.c > @@ -167,12 +167,12 @@ static int kmem_cache_downsize(void); > static int ignore_cache(struct meminfo *, char *); > static char *is_kmem_cache_addr(ulong, char *); > static char *is_kmem_cache_addr_common(ulong, char *); > -static void kmem_cache_list(void); > +static void kmem_cache_list(struct meminfo *); > static void dump_kmem_cache(struct meminfo *); > static void dump_kmem_cache_percpu_v1(struct meminfo *); > static void dump_kmem_cache_percpu_v2(struct meminfo *); > static void dump_kmem_cache_slub(struct meminfo *); > -static void kmem_cache_list_common(void); > +static void kmem_cache_list_common(struct meminfo *); > static ulong get_cpu_slab_ptr(struct meminfo *, int, ulong *); > static unsigned int oo_order(ulong); > static unsigned int oo_objects(ulong); > @@ -276,6 +276,8 @@ static int generic_read_dumpfile(ulonglong, void *, long, > char *, ulong); > static int generic_write_dumpfile(ulonglong, void *, long, char *, ulong); > static int page_to_nid(ulong); > static int get_kmem_cache_list(ulong **); > +static int get_kmem_cache_root_list(ulong **); > +static int get_kmem_cache_child_list(ulong **, ulong); > static int get_kmem_cache_slub_data(long, struct meminfo *); > static ulong compound_head(ulong); > static long count_partial(ulong, struct meminfo *, ulong *); > @@ -815,6 +817,23 @@ vm_init(void) > "kmem_slab_s", "s_magic"); > } > > + if (kernel_symbol_exists("slab_root_caches")) { > + MEMBER_OFFSET_INIT(kmem_cache_memcg_params, > + "kmem_cache", "memcg_params"); > + MEMBER_OFFSET_INIT(memcg_cache_params___root_caches_node, > + "memcg_cache_params", "__root_caches_node"); > + MEMBER_OFFSET_INIT(memcg_cache_params_children, > + "memcg_cache_params", "children"); > + MEMBER_OFFSET_INIT(memcg_cache_params_children_node, > + "memcg_cache_params", "children_node"); > + > + if (VALID_MEMBER(kmem_cache_memcg_params) > + && VALID_MEMBER(memcg_cache_params___root_caches_node) > + && VALID_MEMBER(memcg_cache_params_children) > + && VALID_MEMBER(memcg_cache_params_children_node)) > + vt->flags |= SLAB_ROOT_CACHES; > + } > + > if (!kt->kernel_NR_CPUS) { > if (enumerator_value("WORK_CPU_UNBOUND", (long *)&value1)) > kt->kernel_NR_CPUS = (int)value1; > @@ -4713,6 +4732,7 @@ get_task_mem_usage(ulong task, struct task_mem_usage > *tm) > #define SLAB_OVERLOAD_PAGE_PTR (ADDRESS_SPECIFIED << 24) > #define SLAB_BITFIELD (ADDRESS_SPECIFIED << 25) > #define SLAB_GATHER_FAILURE (ADDRESS_SPECIFIED << 26) > +#define GET_SLAB_ROOT_CACHES (ADDRESS_SPECIFIED << 27) > > #define GET_ALL \ > (GET_SHARED_PAGES|GET_TOTALRAM_PAGES|GET_BUFFERS_PAGES|GET_SLAB_PAGES) > @@ -4724,6 +4744,7 @@ cmd_kmem(void) > int c; > int sflag, Sflag, pflag, fflag, Fflag, vflag, zflag, oflag, gflag; > int nflag, cflag, Cflag, iflag, lflag, Lflag, Pflag, Vflag, hflag; > + int rflag; > struct meminfo meminfo; > ulonglong value[MAXARGS]; > char buf[BUFSIZE]; > @@ -4733,13 +4754,13 @@ cmd_kmem(void) > spec_addr = 0; > sflag = Sflag = pflag = fflag = Fflag = Pflag = zflag = oflag = 0; > vflag = Cflag = cflag = iflag = nflag = lflag = Lflag = Vflag = 0; > - gflag = hflag = 0; > + gflag = hflag = rflag = 0; > escape = FALSE; > BZERO(&meminfo, sizeof(struct meminfo)); > BZERO(&value[0], sizeof(ulonglong)*MAXARGS); > pc->curcmd_flags &= ~HEADER_PRINTED; > > - while ((c = getopt(argcnt, args, "gI:sSFfm:pvczCinl:L:PVoh")) != > EOF) { > + while ((c = getopt(argcnt, args, "gI:sSrFfm:pvczCinl:L:PVoh")) != > EOF) { > switch(c) > { > case 'V': > @@ -4775,11 +4796,15 @@ cmd_kmem(void) > break; > > case 's': > - sflag = 1; Sflag = 0; > + sflag = 1; Sflag = rflag = 0; > break; > > case 'S': > - Sflag = 1; sflag = 0; > + Sflag = 1; sflag = rflag = 0; > + break; > + > + case 'r': > + rflag = 1; sflag = Sflag = 0; > break; > > case 'F': > @@ -4859,12 +4884,13 @@ cmd_kmem(void) > cmd_usage(pc->curcmd, SYNOPSIS); > > if ((sflag + Sflag + pflag + fflag + Fflag + Vflag + oflag + > - vflag + Cflag + cflag + iflag + lflag + Lflag + gflag + hflag) > > 1) { > + vflag + Cflag + cflag + iflag + lflag + Lflag + gflag + > + hflag + rflag) > 1) { > error(INFO, "only one flag allowed!\n"); > cmd_usage(pc->curcmd, SYNOPSIS); > } > > - if (sflag || Sflag || !(vt->flags & KMEM_CACHE_INIT)) > + if (sflag || Sflag || rflag || !(vt->flags & KMEM_CACHE_INIT)) > kmem_cache_init(); > > while (args[optind]) { > @@ -4881,7 +4907,7 @@ cmd_kmem(void) > escape = TRUE; > } else > meminfo.reqname = args[optind]; > - if (!sflag && !Sflag) > + if (!sflag && !Sflag && !rflag) > cmd_usage(pc->curcmd, SYNOPSIS); > } > > @@ -4994,7 +5020,7 @@ cmd_kmem(void) > * no value arguments allowed! > */ > if (zflag || nflag || iflag || Fflag || Cflag || Lflag || > - Vflag || oflag || hflag) { > + Vflag || oflag || hflag || rflag) { > error(INFO, > "no address arguments allowed with this option\n"); > cmd_usage(pc->curcmd, SYNOPSIS); > @@ -5030,9 +5056,17 @@ cmd_kmem(void) > if (hflag == 1) > dump_hstates(); > > - if (sflag == 1) { > + if (sflag == 1 || rflag == 1) { > + if (rflag) { > + if (!((vt->flags & KMALLOC_SLUB) > + && (vt->flags & SLAB_ROOT_CACHES))) > + error(FATAL, > + "-r option doesn't support this kernel\n"); > + > + meminfo.flags = GET_SLAB_ROOT_CACHES; > + } > if (!escape && STREQ(meminfo.reqname, "list")) > - kmem_cache_list(); > + kmem_cache_list(&meminfo); > else if (vt->flags & KMEM_CACHE_UNAVAIL) > error(FATAL, > "kmem cache slab subsystem not available\n"); > @@ -5042,7 +5076,7 @@ cmd_kmem(void) > > if (Sflag == 1) { > if (STREQ(meminfo.reqname, "list")) > - kmem_cache_list(); > + kmem_cache_list(&meminfo); > else if (vt->flags & KMEM_CACHE_UNAVAIL) > error(FATAL, > "kmem cache slab subsystem not available\n"); > @@ -5092,7 +5126,8 @@ cmd_kmem(void) > > if (!(sflag + Sflag + pflag + fflag + Fflag + vflag + > Vflag + zflag + oflag + cflag + Cflag + iflag + > - nflag + lflag + Lflag + gflag + hflag + meminfo.calls)) > + nflag + lflag + Lflag + gflag + hflag + rflag + > + meminfo.calls)) > cmd_usage(pc->curcmd, SYNOPSIS); > > } > @@ -9117,7 +9152,7 @@ is_kmem_cache_addr(ulong vaddr, char *kbuf) > * dumps all slab cache names and their addresses. > */ > static void > -kmem_cache_list(void) > +kmem_cache_list(struct meminfo *mi) > { > ulong cache, cache_cache, name; > long next_offset, name_offset; > @@ -9132,7 +9167,7 @@ kmem_cache_list(void) > } > > if (vt->flags & (KMALLOC_SLUB|KMALLOC_COMMON)) { > - kmem_cache_list_common(); > + kmem_cache_list_common(mi); > return; > } > > @@ -13564,6 +13599,8 @@ dump_vm_table(int verbose) > fprintf(fp, "%sSLAB_OVERLOAD_PAGE", others++ ? "|" : "");\ > if (vt->flags & SLAB_CPU_CACHE) > fprintf(fp, "%sSLAB_CPU_CACHE", others++ ? "|" : "");\ > + if (vt->flags & SLAB_ROOT_CACHES) > + fprintf(fp, "%sSLAB_ROOT_CACHES", others++ ? "|" : "");\ > if (vt->flags & USE_VMAP_AREA) > fprintf(fp, "%sUSE_VMAP_AREA", others++ ? "|" : "");\ > if (vt->flags & CONFIG_NUMA) > @@ -18044,14 +18081,17 @@ kmem_cache_init_slub(void) > } > > static void > -kmem_cache_list_common(void) > +kmem_cache_list_common(struct meminfo *mi) > { > int i, cnt; > ulong *cache_list; > ulong name; > char buf[BUFSIZE]; > > - cnt = get_kmem_cache_list(&cache_list); > + if (mi->flags & GET_SLAB_ROOT_CACHES) > + cnt = get_kmem_cache_root_list(&cache_list); > + else > + cnt = get_kmem_cache_list(&cache_list); > > for (i = 0; i < cnt; i++) { > fprintf(fp, "%lx ", cache_list[i]); > @@ -18087,7 +18127,11 @@ dump_kmem_cache_slub(struct meminfo *si) > } > > order = objects = 0; > - si->cache_count = get_kmem_cache_list(&si->cache_list); > + if (si->flags & GET_SLAB_ROOT_CACHES) > + si->cache_count = get_kmem_cache_root_list(&si->cache_list); > + else > + si->cache_count = get_kmem_cache_list(&si->cache_list); > + > si->cache_buf = GETBUF(SIZE(kmem_cache)); > > if (VALID_MEMBER(page_objects) && > @@ -18168,6 +18212,79 @@ dump_kmem_cache_slub(struct meminfo *si) > !get_kmem_cache_slub_data(GET_SLUB_OBJECTS, si)) > si->flags |= SLAB_GATHER_FAILURE; > > + /* accumulate children's slabinfo */ > + if (si->flags & GET_SLAB_ROOT_CACHES) { > + struct meminfo *mi; > + int j; > + char buf2[BUFSIZE]; > + > + mi = (struct meminfo *)GETBUF(sizeof(struct meminfo)); > + memcpy(mi, si, sizeof(struct meminfo)); > + > + mi->cache_count = get_kmem_cache_child_list(&mi->cache_list, > + si->cache_list[i]); > + > + if (!mi->cache_count) > + goto no_children; > + > + mi->cache_buf = GETBUF(SIZE(kmem_cache)); > + > + for (j = 0; j < mi->cache_count; j++) { > + BZERO(mi->cache_buf, SIZE(kmem_cache)); > + if (!readmem(mi->cache_list[j], KVADDR, mi->cache_buf, > + SIZE(kmem_cache), "kmem_cache buffer", > + RETURN_ON_ERROR|RETURN_PARTIAL)) > + continue; > + > + name = ULONG(mi->cache_buf + OFFSET(kmem_cache_name)); > + if (!read_string(name, buf2, BUFSIZE-1)) > + sprintf(buf2, "(unknown)"); > + > + objsize = UINT(mi->cache_buf + OFFSET(kmem_cache_objsize)); > + size = UINT(mi->cache_buf + OFFSET(kmem_cache_size)); > + offset = UINT(mi->cache_buf + OFFSET(kmem_cache_offset)); > + if (VALID_MEMBER(kmem_cache_objects)) { > + objects = UINT(mi->cache_buf + > + OFFSET(kmem_cache_objects)); > + order = UINT(mi->cache_buf + OFFSET(kmem_cache_order)); > + } else if (VALID_MEMBER(kmem_cache_oo)) { > + oo = ULONG(mi->cache_buf + OFFSET(kmem_cache_oo)); > + objects = oo_objects(oo); > + order = oo_order(oo); > + } else > + error(FATAL, "cannot determine " > + "kmem_cache objects/order values\n"); > + > + mi->cache = mi->cache_list[j]; > + mi->curname = buf2; > + mi->objsize = objsize; > + mi->size = size; > + mi->objects = objects; > + mi->slabsize = (PAGESIZE() << order); > + mi->inuse = mi->num_slabs = 0; > + mi->slab_offset = offset; > + mi->random = VALID_MEMBER(kmem_cache_random) ? > + ULONG(mi->cache_buf + OFFSET(kmem_cache_random)) : 0; > + > + if (!get_kmem_cache_slub_data(GET_SLUB_SLABS, mi) || > + !get_kmem_cache_slub_data(GET_SLUB_OBJECTS, mi)) { > + si->flags |= SLAB_GATHER_FAILURE; > + continue; > + } > + > + si->inuse += mi->inuse; > + si->free += mi->free; > + si->num_slabs += mi->num_slabs; > + > + if (CRASHDEBUG(1)) > + dump_kmem_cache_info(mi); > + } > + FREEBUF(mi->cache_buf); > + FREEBUF(mi->cache_list); > +no_children: > + FREEBUF(mi); > + } > + > DUMP_KMEM_CACHE_INFO(); > > if (si->flags & SLAB_GATHER_FAILURE) { > @@ -18964,6 +19081,70 @@ get_kmem_cache_list(ulong **cache_buf) > return cnt; > } > > +static int > +get_kmem_cache_root_list(ulong **cache_buf) > +{ > + int cnt; > + ulong vaddr; > + struct list_data list_data, *ld; > + > + get_symbol_data("slab_root_caches", sizeof(void *), &vaddr); > + > + ld = &list_data; > + BZERO(ld, sizeof(struct list_data)); > + ld->flags |= LIST_ALLOCATE; > + ld->start = vaddr; > + ld->list_head_offset = OFFSET(kmem_cache_memcg_params) > + + OFFSET(memcg_cache_params___root_caches_node); > + ld->end = symbol_value("slab_root_caches"); > + if (CRASHDEBUG(3)) > + ld->flags |= VERBOSE; > + > + cnt = do_list(ld); > + *cache_buf = ld->list_ptr; > + > + return cnt; > +} > + > +static int > +get_kmem_cache_child_list(ulong **cache_buf, ulong root) > +{ > + int cnt; > + ulong vaddr, children; > + struct list_data list_data, *ld; > + > + children = root + OFFSET(kmem_cache_memcg_params) > + + OFFSET(memcg_cache_params_children); > + > + readmem(children, KVADDR, &vaddr, sizeof(ulong), > + "kmem_cache.memcg_params.children", > + FAULT_ON_ERROR); > + > + /* > + * When no children, since there is the difference of offset > + * of children list between root and child, do_list returns > + * an incorrect cache_buf[0]. So we determine wheather it has > + * children or not with the value of list_head.next. > + */ > + if (children == vaddr) > + return 0; > + > + ld = &list_data; > + BZERO(ld, sizeof(struct list_data)); > + ld->flags |= LIST_ALLOCATE; > + ld->start = vaddr; > + ld->list_head_offset = > + OFFSET(kmem_cache_memcg_params) > + + OFFSET(memcg_cache_params_children_node); > + ld->end = children; > + if (CRASHDEBUG(3)) > + ld->flags |= VERBOSE; > + > + cnt = do_list(ld); > + *cache_buf = ld->list_ptr; > + > + return cnt; > +} > > /* > * Get the address of the head page of a compound page. > diff --git a/symbols.c b/symbols.c > index bee60ba..2e6713a 100644 > --- a/symbols.c > +++ b/symbols.c > @@ -9451,6 +9451,15 @@ dump_offset_table(char *spec, ulong makestruct) > fprintf(fp, " kmem_cache_flags: %ld\n", > OFFSET(kmem_cache_flags)); > > + fprintf(fp, " kmem_cache_memcg_params: %ld\n", > + OFFSET(kmem_cache_memcg_params)); > + fprintf(fp, "memcg_cache_params___root_caches_node: %ld\n", > + OFFSET(memcg_cache_params___root_caches_node)); > + fprintf(fp, " memcg_cache_params_children: %ld\n", > + OFFSET(memcg_cache_params_children)); > + fprintf(fp, " memcg_cache_params_children_node: %ld\n", > + OFFSET(memcg_cache_params_children_node)); > + > fprintf(fp, " net_device_next: %ld\n", > OFFSET(net_device_next)); > fprintf(fp, " net_device_name: %ld\n", > -- > 1.8.3.1 > > -- > Crash-utility mailing list > Crash-utility(a)redhat.com > https://www.redhat.com/mailman/listinfo/crash-utility -- Crash-utility mailing list Crash-utility(a)redhat.com https://www.redhat.com/mailman/listinfo/crash-utility