Received: from mnm [127.0.0.1] by localhost with POP3 (fetchmail-5.9.0) for akpm@localhost (single-drop); Fri, 27 Jun 2003 23:17:43 -0700 (PDT) Received: by mangalore (mbox akpm) (with Cubic Circle's cucipop (v1.31 1998/05/13) Sat Jun 28 16:17:30 2003) X-From_: haveblue@us.ibm.com Sat Jun 28 16:14:22 2003 Received: from e4.ny.us.ibm.com (e4.ny.us.ibm.com [32.97.182.104]) by mangalore.zipworld.com.au (8.12.3/8.12.3/Debian-6.4) with ESMTP id h5S6EJNN005239 for ; Sat, 28 Jun 2003 16:14:20 +1000 Received: from northrelay04.pok.ibm.com (northrelay04.pok.ibm.com [9.56.224.206]) by e4.ny.us.ibm.com (8.12.9/8.12.2) with ESMTP id h5S6EGi8154664; Sat, 28 Jun 2003 02:14:16 -0400 Received: from nighthawk.sr71.net (d01av02.pok.ibm.com [9.56.224.216]) by northrelay04.pok.ibm.com (8.12.9/NCO/VER6.5) with ESMTP id h5S6EEQV030176; Sat, 28 Jun 2003 02:14:14 -0400 Subject: Re: 2.5.73-mm2 From: Dave Hansen To: akpm@zip.com.au Cc: "Martin J. Bligh" In-Reply-To: <20030627202130.066c183b.akpm@digeo.com> References: <20030627202130.066c183b.akpm@digeo.com> Content-Type: multipart/mixed; boundary="=-0X7S2SAZNKi17LAVnQ0Z" Organization: Message-Id: <1056780851.19849.235.camel@nighthawk> Mime-Version: 1.0 X-Mailer: Ximian Evolution 1.2.4 Date: 27 Jun 2003 23:14:11 -0700 X-Spam-Status: No, hits=-20.0 required=6.0 tests=BAYES_01,IN_REP_TO,REFERENCES,USER_AGENT_XIMIAN autolearn=ham version=2.53 X-Spam-Level: X-Spam-Checker-Version: SpamAssassin 2.53 (1.174.2.15-2003-03-30-exp) --=-0X7S2SAZNKi17LAVnQ0Z Content-Type: text/plain Content-Transfer-Encoding: 7bit Damn. Something in the digeo firewall does *not* like when I send patches in attachments to you. > +numa-memory-reporting-fix.patch > > Fix NUMA memory reporting (needs more work) I sent Martin a newer version of this patch which was a lot more comprehensive. You'll need to back out the previous one to apply this. I fear that the current one has broken other NUMA architectures than the i386 ones. The current numa meminfo code exports (via sysfs) pgdat->node_size, as totalram. This variable is consistently used elsewhere to mean "the number of physical pages that this particular node spans". This is _not_ what we want to see from meminfo, which is: "how much actual memory does this node have?" The following patch removes pgdat->node_size, and replaces it with ->node_spanned_pages. This is to avoid confusion with a new variable, node_present_pages, which is the _actual_ value that we want to export in meminfo. Most of the patch is a simple s/node_size/node_spanned_pages/. The node_size() macro is also removed, and replaced with new ones for node_{spanned,present}_pages() to avoid confusion. We were bitten by this problem in this bug: http://bugme.osdl.org/show_bug.cgi?id=818 Compiled and tested on NUMA-Q. -- Dave Hansen haveblue@us.ibm.com --=-0X7S2SAZNKi17LAVnQ0Z Content-Disposition: attachment; filename=node_spanned_pages-2.5.73-mm1-0.patch Content-Type: text/x-patch; NAME=node_spanned_pages-2.5.73-mm1-0.patch; CHARSET=ANSI_X3.4-1968 Content-Transfer-Encoding: 7bit diff -pur linux-2.5.73-mm1/arch/alpha/mm/numa.c linux-2.5.73-mm1-work/arch/alpha/mm/numa.c --- linux-2.5.73-mm1/arch/alpha/mm/numa.c Thu Jun 26 16:04:18 2003 +++ linux-2.5.73-mm1-work/arch/alpha/mm/numa.c Thu Jun 26 16:38:11 2003 @@ -338,7 +338,7 @@ void __init mem_init(void) lmem_map = node_mem_map(nid); pfn = NODE_DATA(nid)->node_start_pfn; - for (i = 0; i < node_size(nid); i++, pfn++) + for (i = 0; i < node_spanned_pages(nid); i++, pfn++) if (page_is_ram(pfn) && PageReserved(lmem_map+i)) reservedpages++; } @@ -372,7 +372,7 @@ show_mem(void) printk("Free swap: %6dkB\n",nr_swap_pages<<(PAGE_SHIFT-10)); for (nid = 0; nid < numnodes; nid++) { struct page * lmem_map = node_mem_map(nid); - i = node_size(nid); + i = node_spanned_pages(nid); while (i-- > 0) { total++; if (PageReserved(lmem_map+i)) diff -pur linux-2.5.73-mm1/arch/arm/mm/init.c linux-2.5.73-mm1-work/arch/arm/mm/init.c --- linux-2.5.73-mm1/arch/arm/mm/init.c Thu Jun 26 16:04:23 2003 +++ linux-2.5.73-mm1-work/arch/arm/mm/init.c Thu Jun 26 17:13:47 2003 @@ -79,7 +79,7 @@ void show_mem(void) struct page *page, *end; page = NODE_MEM_MAP(node); - end = page + NODE_DATA(node)->node_size; + end = page + NODE_DATA(node)->node_spanned_pages; do { total++; @@ -576,7 +576,7 @@ void __init mem_init(void) for (node = 0; node < numnodes; node++) { pg_data_t *pgdat = NODE_DATA(node); - if (pgdat->node_size != 0) + if (pgdat->node_spanned_pages != 0) totalram_pages += free_all_bootmem_node(pgdat); } diff -pur linux-2.5.73-mm1/arch/arm26/mm/init.c linux-2.5.73-mm1-work/arch/arm26/mm/init.c --- linux-2.5.73-mm1/arch/arm26/mm/init.c Thu Jun 26 16:04:13 2003 +++ linux-2.5.73-mm1-work/arch/arm26/mm/init.c Thu Jun 26 16:39:05 2003 @@ -68,7 +68,7 @@ void show_mem(void) page = NODE_MEM_MAP(0); - end = page + NODE_DATA(0)->node_size; + end = page + NODE_DATA(0)->node_spanned_pages; do { total++; @@ -353,7 +353,7 @@ void __init mem_init(void) max_mapnr = virt_to_page(high_memory) - mem_map; /* this will put all unused low memory onto the freelists */ - if (pgdat->node_size != 0) + if (pgdat->node_spanned_pages != 0) totalram_pages += free_all_bootmem_node(pgdat); printk(KERN_INFO "Memory:"); diff -pur linux-2.5.73-mm1/arch/i386/mm/pgtable.c linux-2.5.73-mm1-work/arch/i386/mm/pgtable.c --- linux-2.5.73-mm1/arch/i386/mm/pgtable.c Thu Jun 26 16:07:19 2003 +++ linux-2.5.73-mm1-work/arch/i386/mm/pgtable.c Thu Jun 26 17:09:22 2003 @@ -35,7 +35,7 @@ void show_mem(void) show_free_areas(); printk("Free swap: %6dkB\n",nr_swap_pages<<(PAGE_SHIFT-10)); for_each_pgdat(pgdat) { - for (i = 0; i < pgdat->node_size; ++i) { + for (i = 0; i < pgdat->node_spanned_pages; ++i) { page = pgdat->node_mem_map + i; total++; if (PageHighMem(page)) diff -pur linux-2.5.73-mm1/arch/ia64/mm/init.c linux-2.5.73-mm1-work/arch/ia64/mm/init.c --- linux-2.5.73-mm1/arch/ia64/mm/init.c Thu Jun 26 16:04:17 2003 +++ linux-2.5.73-mm1-work/arch/ia64/mm/init.c Thu Jun 26 16:39:51 2003 @@ -232,7 +232,7 @@ show_mem(void) printk("Free swap: %6dkB\n", nr_swap_pages<<(PAGE_SHIFT-10)); for_each_pgdat(pgdat) { printk("Node ID: %d\n", pgdat->node_id); - for(i = 0; i < pgdat->node_size; i++) { + for(i = 0; i < pgdat->node_spanned_pages; i++) { if (PageReserved(pgdat->node_mem_map+i)) reserved++; else if (PageSwapCache(pgdat->node_mem_map+i)) @@ -240,7 +240,7 @@ show_mem(void) else if (page_count(pgdat->node_mem_map + i)) shared += page_count(pgdat->node_mem_map + i) - 1; } - printk("\t%d pages of RAM\n", pgdat->node_size); + printk("\t%d pages of RAM\n", pgdat->node_spanned_pages); printk("\t%d reserved pages\n", reserved); printk("\t%d pages shared\n", shared); printk("\t%d pages swap cached\n", cached); diff -pur linux-2.5.73-mm1/arch/ppc64/mm/init.c linux-2.5.73-mm1-work/arch/ppc64/mm/init.c --- linux-2.5.73-mm1/arch/ppc64/mm/init.c Thu Jun 26 16:04:22 2003 +++ linux-2.5.73-mm1-work/arch/ppc64/mm/init.c Thu Jun 26 16:40:10 2003 @@ -109,7 +109,7 @@ void show_mem(void) show_free_areas(); printk("Free swap: %6dkB\n",nr_swap_pages<<(PAGE_SHIFT-10)); for_each_pgdat(pgdat) { - for (i = 0; i < pgdat->node_size; i++) { + for (i = 0; i < pgdat->node_spanned_pages; i++) { page = pgdat->node_mem_map + i; total++; if (PageReserved(page)) @@ -564,7 +564,7 @@ void __init mem_init(void) int nid; for (nid = 0; nid < numnodes; nid++) { - if (node_data[nid].node_size != 0) { + if (node_data[nid].node_spanned_pages != 0) { printk("freeing bootmem node %x\n", nid); totalram_pages += free_all_bootmem_node(NODE_DATA(nid)); diff -pur linux-2.5.73-mm1/arch/ppc64/mm/numa.c linux-2.5.73-mm1-work/arch/ppc64/mm/numa.c --- linux-2.5.73-mm1/arch/ppc64/mm/numa.c Thu Jun 26 16:04:22 2003 +++ linux-2.5.73-mm1-work/arch/ppc64/mm/numa.c Thu Jun 26 16:41:27 2003 @@ -160,21 +160,21 @@ new_range: * this simple case and complain if there is a gap in * memory */ - if (node_data[numa_domain].node_size) { + if (node_data[numa_domain].node_spanned_pages) { unsigned long shouldstart = node_data[numa_domain].node_start_pfn + - node_data[numa_domain].node_size; + node_data[numa_domain].node_spanned_pages; if (shouldstart != (start / PAGE_SIZE)) { printk(KERN_ERR "Hole in node, disabling " "region start %lx length %lx\n", start, size); continue; } - node_data[numa_domain].node_size += size / PAGE_SIZE; + node_data[numa_domain].node_spanned_pages += size / PAGE_SIZE; } else { node_data[numa_domain].node_start_pfn = start / PAGE_SIZE; - node_data[numa_domain].node_size = size / PAGE_SIZE; + node_data[numa_domain].node_spanned_pages = size / PAGE_SIZE; } for (i = start ; i < (start+size); i += MEMORY_INCREMENT) @@ -202,7 +202,7 @@ void setup_nonnuma(void) map_cpu_to_node(i, 0); node_data[0].node_start_pfn = 0; - node_data[0].node_size = lmb_end_of_DRAM() / PAGE_SIZE; + node_data[0].node_spanned_pages = lmb_end_of_DRAM() / PAGE_SIZE; for (i = 0 ; i < lmb_end_of_DRAM(); i += MEMORY_INCREMENT) numa_memory_lookup_table[i >> MEMORY_INCREMENT_SHIFT] = 0; @@ -224,12 +224,12 @@ void __init do_init_bootmem(void) unsigned long bootmem_paddr; unsigned long bootmap_pages; - if (node_data[nid].node_size == 0) + if (node_data[nid].node_spanned_pages == 0) continue; start_paddr = node_data[nid].node_start_pfn * PAGE_SIZE; end_paddr = start_paddr + - (node_data[nid].node_size * PAGE_SIZE); + (node_data[nid].node_spanned_pages * PAGE_SIZE); dbg("node %d\n", nid); dbg("start_paddr = %lx\n", start_paddr); @@ -311,7 +311,7 @@ void __init paging_init(void) unsigned long start_pfn; unsigned long end_pfn; - if (node_data[nid].node_size == 0) + if (node_data[nid].node_spanned_pages == 0) continue; start_pfn = plat_node_bdata[nid].node_boot_start >> PAGE_SHIFT; diff -pur linux-2.5.73-mm1/arch/x86_64/mm/init.c linux-2.5.73-mm1-work/arch/x86_64/mm/init.c --- linux-2.5.73-mm1/arch/x86_64/mm/init.c Thu Jun 26 16:04:24 2003 +++ linux-2.5.73-mm1-work/arch/x86_64/mm/init.c Thu Jun 26 16:41:37 2003 @@ -64,7 +64,7 @@ void show_mem(void) printk("Free swap: %6dkB\n",nr_swap_pages<<(PAGE_SHIFT-10)); for_each_pgdat(pgdat) { - for (i = 0; i < pgdat->node_size; ++i) { + for (i = 0; i < pgdat->node_spanned_pages; ++i) { page = pgdat->node_mem_map + i; total++; if (PageReserved(page)) diff -pur linux-2.5.73-mm1/arch/x86_64/mm/numa.c linux-2.5.73-mm1-work/arch/x86_64/mm/numa.c --- linux-2.5.73-mm1/arch/x86_64/mm/numa.c Thu Jun 26 16:04:24 2003 +++ linux-2.5.73-mm1-work/arch/x86_64/mm/numa.c Thu Jun 26 16:42:20 2003 @@ -86,7 +86,7 @@ void __init setup_node_bootmem(int nodei memset(NODE_DATA(nodeid), 0, sizeof(pg_data_t)); NODE_DATA(nodeid)->bdata = &plat_node_bdata[nodeid]; NODE_DATA(nodeid)->node_start_pfn = start_pfn; - NODE_DATA(nodeid)->node_size = end_pfn - start_pfn; + NODE_DATA(nodeid)->node_spanned_pages = end_pfn - start_pfn; /* Find a place for the bootmem map */ bootmap_pages = bootmem_bootmap_pages(end_pfn - start_pfn); diff -pur linux-2.5.73-mm1/include/asm-alpha/mmzone.h linux-2.5.73-mm1-work/include/asm-alpha/mmzone.h --- linux-2.5.73-mm1/include/asm-alpha/mmzone.h Thu Jun 26 16:04:30 2003 +++ linux-2.5.73-mm1-work/include/asm-alpha/mmzone.h Thu Jun 26 16:42:36 2003 @@ -31,7 +31,6 @@ extern pg_data_t node_data[]; #define pa_to_nid(pa) alpha_pa_to_nid(pa) #define NODE_DATA(nid) (&node_data[(nid)]) -#define node_size(nid) (NODE_DATA(nid)->node_size) #define node_localnr(pfn, nid) ((pfn) - NODE_DATA(nid)->node_start_pfn) @@ -124,7 +123,7 @@ PLAT_NODE_DATA_LOCALNR(unsigned long p, #define pfn_to_nid(pfn) pa_to_nid(((u64)pfn << PAGE_SHIFT)) #define pfn_valid(pfn) \ (((pfn) - node_start_pfn(pfn_to_nid(pfn))) < \ - node_size(pfn_to_nid(pfn))) \ + node_spanned_pages(pfn_to_nid(pfn))) \ #define virt_addr_valid(kaddr) pfn_valid((__pa(kaddr) >> PAGE_SHIFT)) diff -pur linux-2.5.73-mm1/include/asm-i386/mmzone.h linux-2.5.73-mm1-work/include/asm-i386/mmzone.h --- linux-2.5.73-mm1/include/asm-i386/mmzone.h Thu Jun 26 16:04:31 2003 +++ linux-2.5.73-mm1-work/include/asm-i386/mmzone.h Thu Jun 26 17:14:56 2003 @@ -32,8 +32,7 @@ extern struct pglist_data *node_data[]; #define alloc_bootmem_low_pages_node(ignore, x) \ __alloc_bootmem_node(NODE_DATA(0), (x), PAGE_SIZE, 0) -#define node_size(nid) (node_data[nid]->node_size) -#define node_localnr(pfn, nid) ((pfn) - node_data[nid]->node_start_pfn) +#define node_localnr(pfn, nid) ((pfn) - node_data[nid]->node_start_pfn) /* * Following are macros that each numa implmentation must define. @@ -54,7 +53,7 @@ extern struct pglist_data *node_data[]; #define node_end_pfn(nid) \ ({ \ pg_data_t *__pgdat = NODE_DATA(nid); \ - __pgdat->node_start_pfn + __pgdat->node_size; \ + __pgdat->node_start_pfn + __pgdat->node_spanned_pages; \ }) #define local_mapnr(kvaddr) \ diff -pur linux-2.5.73-mm1/include/asm-mips64/mmzone.h linux-2.5.73-mm1-work/include/asm-mips64/mmzone.h --- linux-2.5.73-mm1/include/asm-mips64/mmzone.h Thu Jun 26 16:07:24 2003 +++ linux-2.5.73-mm1-work/include/asm-mips64/mmzone.h Thu Jun 26 16:43:03 2003 @@ -24,7 +24,7 @@ extern plat_pg_data_t *plat_node_data[]; #define PHYSADDR_TO_NID(pa) NASID_TO_COMPACT_NODEID(NASID_GET(pa)) #define PLAT_NODE_DATA(n) (plat_node_data[n]) -#define PLAT_NODE_DATA_SIZE(n) (PLAT_NODE_DATA(n)->gendata.node_size) +#define PLAT_NODE_DATA_SIZE(n) (PLAT_NODE_DATA(n)->gendata.node_spanned_pages) #define PLAT_NODE_DATA_LOCALNR(p, n) \ (((p) >> PAGE_SHIFT) - PLAT_NODE_DATA(n)->gendata.node_start_pfn) diff -pur linux-2.5.73-mm1/include/asm-ppc64/mmzone.h linux-2.5.73-mm1-work/include/asm-ppc64/mmzone.h --- linux-2.5.73-mm1/include/asm-ppc64/mmzone.h Thu Jun 26 16:04:26 2003 +++ linux-2.5.73-mm1-work/include/asm-ppc64/mmzone.h Thu Jun 26 16:37:03 2003 @@ -54,7 +54,6 @@ static inline int pa_to_nid(unsigned lon */ #define NODE_DATA(nid) (&node_data[nid]) -#define node_size(nid) (NODE_DATA(nid)->node_size) #define node_localnr(pfn, nid) ((pfn) - NODE_DATA(nid)->node_start_pfn) /* diff -pur linux-2.5.73-mm1/include/asm-x86_64/mmzone.h linux-2.5.73-mm1-work/include/asm-x86_64/mmzone.h --- linux-2.5.73-mm1/include/asm-x86_64/mmzone.h Thu Jun 26 16:04:26 2003 +++ linux-2.5.73-mm1-work/include/asm-x86_64/mmzone.h Thu Jun 26 16:43:18 2003 @@ -40,8 +40,7 @@ static inline __attribute__((pure)) int #define node_mem_map(nid) (NODE_DATA(nid)->node_mem_map) #define node_start_pfn(nid) (NODE_DATA(nid)->node_start_pfn) #define node_end_pfn(nid) (NODE_DATA(nid)->node_start_pfn + \ - NODE_DATA(nid)->node_size) -#define node_size(nid) (NODE_DATA(nid)->node_size) + NODE_DATA(nid)->node_spanned_pages) #define local_mapnr(kvaddr) \ ( (__pa(kvaddr) >> PAGE_SHIFT) - node_start_pfn(kvaddr_to_nid(kvaddr)) ) diff -pur linux-2.5.73-mm1/include/linux/mmzone.h linux-2.5.73-mm1-work/include/linux/mmzone.h --- linux-2.5.73-mm1/include/linux/mmzone.h Thu Jun 26 16:04:27 2003 +++ linux-2.5.73-mm1-work/include/linux/mmzone.h Thu Jun 26 16:37:59 2003 @@ -184,11 +184,16 @@ typedef struct pglist_data { unsigned long *valid_addr_bitmap; struct bootmem_data *bdata; unsigned long node_start_pfn; - unsigned long node_size; + unsigned long node_present_pages; /* total number of physical pages */ + unsigned long node_spanned_pages; /* total size of physical page + range, including holes */ int node_id; struct pglist_data *pgdat_next; wait_queue_head_t kswapd_wait; } pg_data_t; + +#define node_present_pages(nid) (NODE_DATA(nid)->node_present_pages) +#define node_spanned_pages(nid) (NODE_DATA(nid)->node_spanned_pages) extern int numnodes; extern struct pglist_data *pgdat_list; diff -pur linux-2.5.73-mm1/mm/page_alloc.c linux-2.5.73-mm1-work/mm/page_alloc.c --- linux-2.5.73-mm1/mm/page_alloc.c Thu Jun 26 16:07:26 2003 +++ linux-2.5.73-mm1-work/mm/page_alloc.c Thu Jun 26 16:50:45 2003 @@ -903,7 +903,7 @@ void si_meminfo_node(struct sysinfo *val { pg_data_t *pgdat = NODE_DATA(nid); - val->totalram = pgdat->node_size; + val->totalram = pgdat->node_present_pages; val->freeram = nr_free_pages_pgdat(pgdat); val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages; val->freehigh = pgdat->node_zones[ZONE_HIGHMEM].free_pages; @@ -1138,12 +1138,13 @@ static void __init calculate_zone_totalp for (i = 0; i < MAX_NR_ZONES; i++) totalpages += zones_size[i]; - pgdat->node_size = totalpages; + pgdat->node_spanned_pages = totalpages; realtotalpages = totalpages; if (zholes_size) for (i = 0; i < MAX_NR_ZONES; i++) realtotalpages -= zholes_size[i]; + pgdat->node_present_pages = realtotalpages; printk("On node %d totalpages: %lu\n", pgdat->node_id, realtotalpages); } @@ -1349,7 +1350,7 @@ void __init free_area_init_node(int nid, pgdat->node_start_pfn = node_start_pfn; calculate_zone_totalpages(pgdat, zones_size, zholes_size); if (!node_mem_map) { - size = (pgdat->node_size + 1) * sizeof(struct page); + size = (pgdat->node_spanned_pages + 1) * sizeof(struct page); node_mem_map = alloc_bootmem_node(pgdat, size); } pgdat->node_mem_map = node_mem_map; --=-0X7S2SAZNKi17LAVnQ0Z--