* [RFC][PATCH] domheap optimization for NUMA
@ 2008-03-31 10:40 Zhai, Edwin
2008-03-31 10:47 ` Keir Fraser
0 siblings, 1 reply; 14+ messages in thread
From: Zhai, Edwin @ 2008-03-31 10:40 UTC (permalink / raw)
To: Keir, ryanh; +Cc: xen-devel, Zhai, Edwin
[-- Attachment #1: Type: text/plain, Size: 597 bytes --]
keir & harper,
You know, currently alloc_domheap_pages try to alloc mem from the node where
current LP sit, which is not always right. E.g some hypercall issued from dom0's
LP allocate memory for other domains.
This patch intend to resolve this issue on NUMA system by locating node from the
served domain.
A new interface alloc_domheap_pages_on_node is introduced instead of changing
current implementation, which is invasive.
There are still many places left for the new interface, but we can change them
in a incremental way if needed.
Appreciate your comments.
--
best rgds,
edwin
[-- Attachment #2: domheap_numa_2.patch --]
[-- Type: text/plain, Size: 6290 bytes --]
diff -r 5655c27da930 xen/arch/x86/domain.c
--- a/xen/arch/x86/domain.c Wed Mar 26 16:14:39 2008 +0800
+++ b/xen/arch/x86/domain.c Fri Mar 28 16:18:29 2008 +0800
@@ -477,7 +477,7 @@ int arch_domain_create(struct domain *d,
#else /* __x86_64__ */
- if ( (pg = alloc_domheap_page(NULL)) == NULL )
+ if ( (pg = alloc_domheap_page_on_node(NULL, d->vcpu[0])) == NULL )
goto fail;
d->arch.mm_perdomain_l2 = page_to_virt(pg);
clear_page(d->arch.mm_perdomain_l2);
@@ -486,7 +486,7 @@ int arch_domain_create(struct domain *d,
l2e_from_page(virt_to_page(d->arch.mm_perdomain_pt)+i,
__PAGE_HYPERVISOR);
- if ( (pg = alloc_domheap_page(NULL)) == NULL )
+ if ( (pg = alloc_domheap_page_on_node(NULL, d->vcpu[0])) == NULL )
goto fail;
d->arch.mm_perdomain_l3 = page_to_virt(pg);
clear_page(d->arch.mm_perdomain_l3);
diff -r 5655c27da930 xen/arch/x86/hvm/stdvga.c
--- a/xen/arch/x86/hvm/stdvga.c Wed Mar 26 16:14:39 2008 +0800
+++ b/xen/arch/x86/hvm/stdvga.c Fri Mar 28 16:19:03 2008 +0800
@@ -513,7 +513,7 @@ void stdvga_init(struct domain *d)
for ( i = 0; i != ARRAY_SIZE(s->vram_page); i++ )
{
- if ( (pg = alloc_domheap_page(NULL)) == NULL )
+ if ( (pg = alloc_domheap_page_on_node(NULL, d->vcpu[0])) == NULL )
break;
s->vram_page[i] = pg;
p = map_domain_page(page_to_mfn(pg));
diff -r 5655c27da930 xen/arch/x86/hvm/vlapic.c
--- a/xen/arch/x86/hvm/vlapic.c Wed Mar 26 16:14:39 2008 +0800
+++ b/xen/arch/x86/hvm/vlapic.c Thu Mar 27 17:06:53 2008 +0800
@@ -928,7 +928,7 @@ int vlapic_init(struct vcpu *v)
memflags = MEMF_bits(32);
#endif
- vlapic->regs_page = alloc_domheap_pages(NULL, 0, memflags);
+ vlapic->regs_page = alloc_domheap_pages_on_node(NULL, 0, memflags, v);
if ( vlapic->regs_page == NULL )
{
dprintk(XENLOG_ERR, "alloc vlapic regs error: %d/%d\n",
diff -r 5655c27da930 xen/arch/x86/mm/hap/hap.c
--- a/xen/arch/x86/mm/hap/hap.c Wed Mar 26 16:14:39 2008 +0800
+++ b/xen/arch/x86/mm/hap/hap.c Fri Mar 28 16:19:22 2008 +0800
@@ -135,7 +135,7 @@ static struct page_info *hap_alloc_p2m_p
&& mfn_x(page_to_mfn(pg)) >= (1UL << (32 - PAGE_SHIFT)) )
{
free_domheap_page(pg);
- pg = alloc_domheap_pages(NULL, 0, MEMF_bits(32));
+ pg = alloc_domheap_pages_on_node(NULL, 0, MEMF_bits(32), d->vcpu[0]);
if ( likely(pg != NULL) )
{
void *p = hap_map_domain_page(page_to_mfn(pg));
@@ -199,7 +199,7 @@ hap_set_allocation(struct domain *d, uns
if ( d->arch.paging.hap.total_pages < pages )
{
/* Need to allocate more memory from domheap */
- pg = alloc_domheap_page(NULL);
+ pg = alloc_domheap_page_on_node(NULL, d->vcpu[0]);
if ( pg == NULL )
{
HAP_PRINTK("failed to allocate hap pages.\n");
diff -r 5655c27da930 xen/arch/x86/mm/paging.c
--- a/xen/arch/x86/mm/paging.c Wed Mar 26 16:14:39 2008 +0800
+++ b/xen/arch/x86/mm/paging.c Fri Mar 28 16:19:39 2008 +0800
@@ -99,7 +99,7 @@ static mfn_t paging_new_log_dirty_page(s
static mfn_t paging_new_log_dirty_page(struct domain *d, void **mapping_p)
{
mfn_t mfn;
- struct page_info *page = alloc_domheap_page(NULL);
+ struct page_info *page = alloc_domheap_page_on_node(NULL, d->vcpu[0]);
if ( unlikely(page == NULL) )
{
diff -r 5655c27da930 xen/arch/x86/mm/shadow/common.c
--- a/xen/arch/x86/mm/shadow/common.c Wed Mar 26 16:14:39 2008 +0800
+++ b/xen/arch/x86/mm/shadow/common.c Fri Mar 28 16:25:56 2008 +0800
@@ -1249,7 +1249,7 @@ static unsigned int sh_set_allocation(st
{
/* Need to allocate more memory from domheap */
sp = (struct shadow_page_info *)
- alloc_domheap_pages(NULL, order, 0);
+ alloc_domheap_pages_on_node(NULL, order, 0, d->vcpu[0]);
if ( sp == NULL )
{
SHADOW_PRINTK("failed to allocate shadow pages.\n");
diff -r 5655c27da930 xen/common/page_alloc.c
--- a/xen/common/page_alloc.c Wed Mar 26 16:14:39 2008 +0800
+++ b/xen/common/page_alloc.c Thu Mar 27 17:00:47 2008 +0800
@@ -829,6 +829,13 @@ struct page_info *alloc_domheap_pages(
return __alloc_domheap_pages(d, smp_processor_id(), order, flags);
}
+struct page_info *alloc_domheap_pages_on_node(
+ struct domain *d, unsigned int order, unsigned int flags, struct vcpu *v)
+{
+ unsigned int cpu = v ? v->processor : smp_processor_id();
+ return __alloc_domheap_pages(d, cpu, order, flags);
+}
+
void free_domheap_pages(struct page_info *pg, unsigned int order)
{
int i, drop_dom_ref;
diff -r 5655c27da930 xen/drivers/passthrough/vtd/iommu.c
--- a/xen/drivers/passthrough/vtd/iommu.c Wed Mar 26 16:14:39 2008 +0800
+++ b/xen/drivers/passthrough/vtd/iommu.c Fri Mar 28 16:25:18 2008 +0800
@@ -269,7 +269,7 @@ static struct page_info *addr_to_dma_pag
if ( dma_pte_addr(*pte) == 0 )
{
- pg = alloc_domheap_page(NULL);
+ pg = alloc_domheap_page_on_node(NULL, domain->vcpu[0]);
vaddr = map_domain_page(page_to_mfn(pg));
if ( !vaddr )
{
diff -r 5655c27da930 xen/include/xen/mm.h
--- a/xen/include/xen/mm.h Wed Mar 26 16:14:39 2008 +0800
+++ b/xen/include/xen/mm.h Thu Mar 27 17:10:33 2008 +0800
@@ -54,6 +54,8 @@ void init_domheap_pages(paddr_t ps, padd
void init_domheap_pages(paddr_t ps, paddr_t pe);
struct page_info *alloc_domheap_pages(
struct domain *d, unsigned int order, unsigned int memflags);
+struct page_info *alloc_domheap_pages_on_node(
+ struct domain *d, unsigned int order, unsigned int memflags, struct vcpu *v);
struct page_info *__alloc_domheap_pages(
struct domain *d, unsigned int cpu, unsigned int order,
unsigned int memflags);
@@ -62,6 +64,7 @@ unsigned long avail_domheap_pages_region
unsigned int node, unsigned int min_width, unsigned int max_width);
unsigned long avail_domheap_pages(void);
#define alloc_domheap_page(d) (alloc_domheap_pages(d,0,0))
+#define alloc_domheap_page_on_node(d, v) (alloc_domheap_pages_on_node(d,0,0,v))
#define free_domheap_page(p) (free_domheap_pages(p,0))
void scrub_heap_pages(void);
[-- Attachment #3: Type: text/plain, Size: 138 bytes --]
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel
^ permalink raw reply [flat|nested] 14+ messages in thread* Re: [RFC][PATCH] domheap optimization for NUMA 2008-03-31 10:40 [RFC][PATCH] domheap optimization for NUMA Zhai, Edwin @ 2008-03-31 10:47 ` Keir Fraser 2008-04-02 13:06 ` Zhai, Edwin 0 siblings, 1 reply; 14+ messages in thread From: Keir Fraser @ 2008-03-31 10:47 UTC (permalink / raw) To: Zhai, Edwin, ryanh; +Cc: xen-devel On 31/3/08 11:40, "Zhai, Edwin" <edwin.zhai@intel.com> wrote: > A new interface alloc_domheap_pages_on_node is introduced instead of changing > current implementation, which is invasive. > > There are still many places left for the new interface, but we can change them > in a incremental way if needed. A function called alloc_domheap_pages_on_node() should take a *node* argument not a vcpu. Perhaps have a helper domain_default_node() which returns node for d->vcpu[0] if d->vcpu[0] is non-NULL, else returns some value meaning 'any'. This will probably require some changes to page_alloc.c. That file currently likes to pass cpu ids around rather than node ids, but it would really be cleaner to pass around the latter. -- Keir ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-03-31 10:47 ` Keir Fraser @ 2008-04-02 13:06 ` Zhai, Edwin 2008-04-02 13:27 ` Keir Fraser 0 siblings, 1 reply; 14+ messages in thread From: Zhai, Edwin @ 2008-04-02 13:06 UTC (permalink / raw) To: Keir Fraser; +Cc: ryanh, xen-devel, Zhai, Edwin On Mon, Mar 31, 2008 at 11:47:36AM +0100, Keir Fraser wrote: > On 31/3/08 11:40, "Zhai, Edwin" <edwin.zhai@intel.com> wrote: > > > A new interface alloc_domheap_pages_on_node is introduced instead of changing > > current implementation, which is invasive. > > > > There are still many places left for the new interface, but we can change them > > in a incremental way if needed. > > A function called alloc_domheap_pages_on_node() should take a *node* > argument not a vcpu. Perhaps have a helper domain_default_node() which > returns node for d->vcpu[0] if d->vcpu[0] is non-NULL, else returns some > value meaning 'any'. > > This will probably require some changes to page_alloc.c. That file currently > likes to pass cpu ids around rather than node ids, but it would really be > cleaner to pass around the latter. The issue is alloc_domheap_pages take domain* as parameter to indicate if need account pages for the domain, sometimes it's NULL. In this case, we can't deduct the node from domain. I believe it's why use cpu here as getting cpuid is easier. Another option, always use domain* to locate node(not allowed NULL) and add a new flag _MEMF_assign to indicate the assignment, which changes the interface and is invasive. > > -- Keir > > > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel > -- best rgds, edwin ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-04-02 13:06 ` Zhai, Edwin @ 2008-04-02 13:27 ` Keir Fraser 2008-04-02 22:49 ` Andre Przywara 0 siblings, 1 reply; 14+ messages in thread From: Keir Fraser @ 2008-04-02 13:27 UTC (permalink / raw) To: Zhai, Edwin; +Cc: ryanh, xen-devel On 2/4/08 14:06, "Zhai, Edwin" <edwin.zhai@intel.com> wrote: > The issue is alloc_domheap_pages take domain* as parameter to indicate if need > account pages for the domain, sometimes it's NULL. In this case, we can't > deduct > the node from domain. I believe it's why use cpu here as getting cpuid is > easier. Yes, but it's a bad interface, particularlty when the function is called alloc_domheap_pages_on_node(). Pass in a nodeid. Write a helper function to work out the nodeid from the domain*. > Another option, always use domain* to locate node(not allowed NULL) and add a > new flag _MEMF_assign to indicate the assignment, which changes the interface > and is invasive. Yes, that's a bad idea. -- Keir ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-04-02 13:27 ` Keir Fraser @ 2008-04-02 22:49 ` Andre Przywara 2008-04-02 23:21 ` Keir Fraser 0 siblings, 1 reply; 14+ messages in thread From: Andre Przywara @ 2008-04-02 22:49 UTC (permalink / raw) To: Keir Fraser; +Cc: xen-devel, edwin.zhai [-- Attachment #1: Type: text/plain, Size: 1827 bytes --] Keir Fraser wrote: > On 2/4/08 14:06, "Zhai, Edwin" <edwin.zhai@intel.com> wrote: > >> The issue is alloc_domheap_pages take domain* as parameter to indicate if need >> account pages for the domain, sometimes it's NULL. In this case, we can't >> deduct >> the node from domain. I believe it's why use cpu here as getting cpuid is >> easier. > > Yes, but it's a bad interface, particularlty when the function is called > alloc_domheap_pages_on_node(). Pass in a nodeid. Write a helper function to > work out the nodeid from the domain*. I was just looking at this code, too, so I fixed this. Eventually alloc_heap_pages is called, which deals with nodes only, so I replaced cpu with node everywhere else, too. Now __alloc_domheap_pages and alloc_domheap_pages_on_node are almost the same (except parameter ordering), so I removed the first one, since the naming of the latter is better. Passing node numbers instead of cpu numbers needs cpu_to_node and asm/numa.h, if you think there is a better way, I am all ears. > >> Another option, always use domain* to locate node(not allowed NULL) and add a >> new flag _MEMF_assign to indicate the assignment, which changes the interface >> and is invasive. > > Yes, that's a bad idea. > > -- Keir The first diff is against Edwin's patch, the second includes it. Signed-off-by: Andre Przywara <andre.przywara@amd.com> Regards, Andre. -- Andre Przywara AMD-Operating System Research Center (OSRC), Dresden, Germany Tel: +49 351 277-84917 ----to satisfy European Law for business letters: AMD Saxony Limited Liability Company & Co. KG, Wilschdorfer Landstr. 101, 01109 Dresden, Germany Register Court Dresden: HRA 4896, General Partner authorized to represent: AMD Saxony LLC (Wilmington, Delaware, US) General Manager of AMD Saxony LLC: Dr. Hans-R. Deppe, Thomas McCoy [-- Attachment #2: alloc_domheap_node_diff.patch --] [-- Type: text/plain, Size: 14257 bytes --] diff -r b1cd7c0749f4 xen/arch/x86/domain.c --- a/xen/arch/x86/domain.c Wed Apr 02 16:07:24 2008 +0200 +++ b/xen/arch/x86/domain.c Thu Apr 03 00:38:53 2008 +0200 @@ -46,6 +46,7 @@ #include <asm/debugreg.h> #include <asm/msr.h> #include <asm/nmi.h> +#include <asm/numa.h> #include <xen/iommu.h> #ifdef CONFIG_COMPAT #include <compat/vcpu.h> @@ -477,8 +478,9 @@ int arch_domain_create(struct domain *d, #else /* __x86_64__ */ - if ( (pg = alloc_domheap_page_on_node(NULL, d->vcpu[0])) == NULL ) - goto fail; + if ( (pg = alloc_domheap_page_on_node(NULL, + cpu_to_node(d->vcpu[0]->processor))) == NULL ) + goto fail; d->arch.mm_perdomain_l2 = page_to_virt(pg); clear_page(d->arch.mm_perdomain_l2); for ( i = 0; i < (1 << pdpt_order); i++ ) @@ -486,8 +488,9 @@ int arch_domain_create(struct domain *d, l2e_from_page(virt_to_page(d->arch.mm_perdomain_pt)+i, __PAGE_HYPERVISOR); - if ( (pg = alloc_domheap_page_on_node(NULL, d->vcpu[0])) == NULL ) - goto fail; + if ( (pg = alloc_domheap_page_on_node(NULL, + cpu_to_node(d->vcpu[0]->processor))) == NULL ) + goto fail; d->arch.mm_perdomain_l3 = page_to_virt(pg); clear_page(d->arch.mm_perdomain_l3); d->arch.mm_perdomain_l3[l3_table_offset(PERDOMAIN_VIRT_START)] = diff -r b1cd7c0749f4 xen/arch/x86/hvm/stdvga.c --- a/xen/arch/x86/hvm/stdvga.c Wed Apr 02 16:07:24 2008 +0200 +++ b/xen/arch/x86/hvm/stdvga.c Thu Apr 03 00:38:53 2008 +0200 @@ -32,6 +32,7 @@ #include <xen/sched.h> #include <xen/domain_page.h> #include <asm/hvm/support.h> +#include <asm/numa.h> #define PAT(x) (x) static const uint32_t mask16[16] = { @@ -513,8 +514,9 @@ void stdvga_init(struct domain *d) for ( i = 0; i != ARRAY_SIZE(s->vram_page); i++ ) { - if ( (pg = alloc_domheap_page_on_node(NULL, d->vcpu[0])) == NULL ) - break; + if ( (pg = alloc_domheap_page_on_node(NULL, + cpu_to_node(d->vcpu[0]->processor))) == NULL ) + break; s->vram_page[i] = pg; p = map_domain_page(page_to_mfn(pg)); clear_page(p); diff -r b1cd7c0749f4 xen/arch/x86/hvm/vlapic.c --- a/xen/arch/x86/hvm/vlapic.c Wed Apr 02 16:07:24 2008 +0200 +++ b/xen/arch/x86/hvm/vlapic.c Thu Apr 03 00:38:53 2008 +0200 @@ -33,6 +33,7 @@ #include <xen/sched.h> #include <asm/current.h> #include <asm/hvm/vmx/vmx.h> +#include <asm/numa.h> #include <public/hvm/ioreq.h> #include <public/hvm/params.h> @@ -928,7 +929,8 @@ int vlapic_init(struct vcpu *v) memflags = MEMF_bits(32); #endif - vlapic->regs_page = alloc_domheap_pages_on_node(NULL, 0, memflags, v); + vlapic->regs_page = alloc_domheap_pages_on_node(NULL, 0, memflags, + cpu_to_node(v->processor)); if ( vlapic->regs_page == NULL ) { dprintk(XENLOG_ERR, "alloc vlapic regs error: %d/%d\n", diff -r b1cd7c0749f4 xen/arch/x86/mm/hap/hap.c --- a/xen/arch/x86/mm/hap/hap.c Wed Apr 02 16:07:24 2008 +0200 +++ b/xen/arch/x86/mm/hap/hap.c Thu Apr 03 00:38:53 2008 +0200 @@ -38,6 +38,7 @@ #include <asm/hap.h> #include <asm/paging.h> #include <asm/domain.h> +#include <asm/numa.h> #include "private.h" @@ -135,7 +136,8 @@ static struct page_info *hap_alloc_p2m_p && mfn_x(page_to_mfn(pg)) >= (1UL << (32 - PAGE_SHIFT)) ) { free_domheap_page(pg); - pg = alloc_domheap_pages_on_node(NULL, 0, MEMF_bits(32), d->vcpu[0]); + pg = alloc_domheap_pages_on_node(NULL, 0, MEMF_bits(32), + cpu_to_node(d->vcpu[0]->processor)); if ( likely(pg != NULL) ) { void *p = hap_map_domain_page(page_to_mfn(pg)); @@ -199,7 +201,8 @@ hap_set_allocation(struct domain *d, uns if ( d->arch.paging.hap.total_pages < pages ) { /* Need to allocate more memory from domheap */ - pg = alloc_domheap_page_on_node(NULL, d->vcpu[0]); + pg = alloc_domheap_page_on_node(NULL, + cpu_to_node(d->vcpu[0]->processor)); if ( pg == NULL ) { HAP_PRINTK("failed to allocate hap pages.\n"); diff -r b1cd7c0749f4 xen/arch/x86/mm/paging.c --- a/xen/arch/x86/mm/paging.c Wed Apr 02 16:07:24 2008 +0200 +++ b/xen/arch/x86/mm/paging.c Thu Apr 03 00:38:53 2008 +0200 @@ -26,6 +26,7 @@ #include <asm/p2m.h> #include <asm/hap.h> #include <asm/guest_access.h> +#include <asm/numa.h> #include <xsm/xsm.h> #define hap_enabled(d) (is_hvm_domain(d) && (d)->arch.hvm_domain.hap_enabled) @@ -99,7 +100,8 @@ static mfn_t paging_new_log_dirty_page(s static mfn_t paging_new_log_dirty_page(struct domain *d, void **mapping_p) { mfn_t mfn; - struct page_info *page = alloc_domheap_page_on_node(NULL, d->vcpu[0]); + struct page_info *page = alloc_domheap_page_on_node(NULL, + cpu_to_node(d->vcpu[0]->processor)); if ( unlikely(page == NULL) ) { diff -r b1cd7c0749f4 xen/arch/x86/mm/shadow/common.c --- a/xen/arch/x86/mm/shadow/common.c Wed Apr 02 16:07:24 2008 +0200 +++ b/xen/arch/x86/mm/shadow/common.c Thu Apr 03 00:38:53 2008 +0200 @@ -36,6 +36,7 @@ #include <asm/current.h> #include <asm/flushtlb.h> #include <asm/shadow.h> +#include <asm/numa.h> #include "private.h" @@ -1249,7 +1250,8 @@ static unsigned int sh_set_allocation(st { /* Need to allocate more memory from domheap */ sp = (struct shadow_page_info *) - alloc_domheap_pages_on_node(NULL, order, 0, d->vcpu[0]); + alloc_domheap_pages_on_node(NULL, order, 0, + cpu_to_node(d->vcpu[0]->processor)); if ( sp == NULL ) { SHADOW_PRINTK("failed to allocate shadow pages.\n"); diff -r b1cd7c0749f4 xen/common/memory.c --- a/xen/common/memory.c Wed Apr 02 16:07:24 2008 +0200 +++ b/xen/common/memory.c Thu Apr 03 00:38:53 2008 +0200 @@ -21,6 +21,7 @@ #include <xen/errno.h> #include <asm/current.h> #include <asm/hardirq.h> +#include <asm/numa.h> #include <public/memory.h> #include <xsm/xsm.h> @@ -37,10 +38,10 @@ struct memop_args { int preempted; /* Was the hypercall preempted? */ }; -static unsigned int select_local_cpu(struct domain *d) +static unsigned int select_local_node(struct domain *d) { struct vcpu *v = d->vcpu[0]; - return (v ? v->processor : 0); + return (v ? cpu_to_node(v->processor) : 0); } static void increase_reservation(struct memop_args *a) @@ -49,7 +50,7 @@ static void increase_reservation(struct unsigned long i; xen_pfn_t mfn; struct domain *d = a->domain; - unsigned int cpu = select_local_cpu(d); + unsigned int node = select_local_node(d); if ( !guest_handle_is_null(a->extent_list) && !guest_handle_okay(a->extent_list, a->nr_extents) ) @@ -67,7 +68,8 @@ static void increase_reservation(struct goto out; } - page = __alloc_domheap_pages(d, cpu, a->extent_order, a->memflags); + page = alloc_domheap_pages_on_node ( + d, a->extent_order, a->memflags, node); if ( unlikely(page == NULL) ) { gdprintk(XENLOG_INFO, "Could not allocate order=%d extent: " @@ -96,7 +98,7 @@ static void populate_physmap(struct memo unsigned long i, j; xen_pfn_t gpfn, mfn; struct domain *d = a->domain; - unsigned int cpu = select_local_cpu(d); + unsigned int node = select_local_node(d); if ( !guest_handle_okay(a->extent_list, a->nr_extents) ) return; @@ -116,7 +118,8 @@ static void populate_physmap(struct memo if ( unlikely(__copy_from_guest_offset(&gpfn, a->extent_list, i, 1)) ) goto out; - page = __alloc_domheap_pages(d, cpu, a->extent_order, a->memflags); + page = alloc_domheap_pages_on_node ( + d, a->extent_order, a->memflags, node); if ( unlikely(page == NULL) ) { gdprintk(XENLOG_INFO, "Could not allocate order=%d extent: " @@ -296,7 +299,7 @@ static long memory_exchange(XEN_GUEST_HA unsigned long in_chunk_order, out_chunk_order; xen_pfn_t gpfn, gmfn, mfn; unsigned long i, j, k; - unsigned int memflags = 0, cpu; + unsigned int memflags = 0, node; long rc = 0; struct domain *d; struct page_info *page; @@ -352,7 +355,7 @@ static long memory_exchange(XEN_GUEST_HA memflags |= MEMF_bits(domain_clamp_alloc_bitsize( d, exch.out.address_bits ? : (BITS_PER_LONG+PAGE_SHIFT))); - cpu = select_local_cpu(d); + node = select_local_node(d); for ( i = (exch.nr_exchanged >> in_chunk_order); i < (exch.in.nr_extents >> in_chunk_order); @@ -401,8 +404,8 @@ static long memory_exchange(XEN_GUEST_HA /* Allocate a chunk's worth of anonymous output pages. */ for ( j = 0; j < (1UL << out_chunk_order); j++ ) { - page = __alloc_domheap_pages( - NULL, cpu, exch.out.extent_order, memflags); + page = alloc_domheap_pages_on_node( + NULL, exch.out.extent_order, memflags, node); if ( unlikely(page == NULL) ) { rc = -ENOMEM; diff -r b1cd7c0749f4 xen/common/page_alloc.c --- a/xen/common/page_alloc.c Wed Apr 02 16:07:24 2008 +0200 +++ b/xen/common/page_alloc.c Thu Apr 03 00:38:53 2008 +0200 @@ -36,6 +36,7 @@ #include <xen/numa.h> #include <xen/nodemask.h> #include <asm/page.h> +#include <asm/numa.h> #include <asm/flushtlb.h> /* @@ -328,10 +329,10 @@ static void init_node_heap(int node) /* Allocate 2^@order contiguous pages. */ static struct page_info *alloc_heap_pages( unsigned int zone_lo, unsigned int zone_hi, - unsigned int cpu, unsigned int order) + unsigned int node, unsigned int order) { unsigned int i, j, zone; - unsigned int node = cpu_to_node(cpu), num_nodes = num_online_nodes(); + unsigned int num_nodes = num_online_nodes(); unsigned long request = 1UL << order; cpumask_t extra_cpus_mask, mask; struct page_info *pg; @@ -670,7 +671,8 @@ void *alloc_xenheap_pages(unsigned int o ASSERT(!in_irq()); - pg = alloc_heap_pages(MEMZONE_XEN, MEMZONE_XEN, smp_processor_id(), order); + pg = alloc_heap_pages(MEMZONE_XEN, MEMZONE_XEN, + cpu_to_node(smp_processor_id()), order); if ( unlikely(pg == NULL) ) goto no_memory; @@ -778,9 +780,9 @@ int assign_pages( } -struct page_info *__alloc_domheap_pages( - struct domain *d, unsigned int cpu, unsigned int order, - unsigned int memflags) +struct page_info *alloc_domheap_pages_on_node( + struct domain *d, unsigned int order, unsigned int memflags, + unsigned int node) { struct page_info *pg = NULL; unsigned int bits = memflags >> _MEMF_bits, zone_hi = NR_ZONES - 1; @@ -797,7 +799,7 @@ struct page_info *__alloc_domheap_pages( if ( (zone_hi + PAGE_SHIFT) >= dma_bitsize ) { - pg = alloc_heap_pages(dma_bitsize - PAGE_SHIFT, zone_hi, cpu, order); + pg = alloc_heap_pages(dma_bitsize - PAGE_SHIFT, zone_hi, node, order); /* Failure? Then check if we can fall back to the DMA pool. */ if ( unlikely(pg == NULL) && @@ -811,7 +813,7 @@ struct page_info *__alloc_domheap_pages( if ( (pg == NULL) && ((pg = alloc_heap_pages(MEMZONE_XEN + 1, zone_hi, - cpu, order)) == NULL) ) + node, order)) == NULL) ) return NULL; if ( (d != NULL) && assign_pages(d, pg, order, memflags) ) @@ -826,14 +828,8 @@ struct page_info *alloc_domheap_pages( struct page_info *alloc_domheap_pages( struct domain *d, unsigned int order, unsigned int flags) { - return __alloc_domheap_pages(d, smp_processor_id(), order, flags); -} - -struct page_info *alloc_domheap_pages_on_node( - struct domain *d, unsigned int order, unsigned int flags, struct vcpu *v) -{ - unsigned int cpu = v ? v->processor : smp_processor_id(); - return __alloc_domheap_pages(d, cpu, order, flags); + return alloc_domheap_pages_on_node (d, order, flags, + cpu_to_node (smp_processor_id()); } void free_domheap_pages(struct page_info *pg, unsigned int order) diff -r b1cd7c0749f4 xen/drivers/passthrough/vtd/iommu.c --- a/xen/drivers/passthrough/vtd/iommu.c Wed Apr 02 16:07:24 2008 +0200 +++ b/xen/drivers/passthrough/vtd/iommu.c Thu Apr 03 00:38:53 2008 +0200 @@ -24,6 +24,7 @@ #include <xen/xmalloc.h> #include <xen/domain_page.h> #include <xen/iommu.h> +#include <asm/numa.h> #include "iommu.h" #include "dmar.h" #include "../pci-direct.h" @@ -269,7 +270,8 @@ static struct page_info *addr_to_dma_pag if ( dma_pte_addr(*pte) == 0 ) { - pg = alloc_domheap_page_on_node(NULL, domain->vcpu[0]); + pg = alloc_domheap_page_on_node(NULL, + cpu_to_node(domain->vcpu[0]->processor)); vaddr = map_domain_page(page_to_mfn(pg)); if ( !vaddr ) { diff -r b1cd7c0749f4 xen/include/xen/mm.h --- a/xen/include/xen/mm.h Wed Apr 02 16:07:24 2008 +0200 +++ b/xen/include/xen/mm.h Thu Apr 03 00:38:53 2008 +0200 @@ -55,16 +55,14 @@ struct page_info *alloc_domheap_pages( struct page_info *alloc_domheap_pages( struct domain *d, unsigned int order, unsigned int memflags); struct page_info *alloc_domheap_pages_on_node( - struct domain *d, unsigned int order, unsigned int memflags, struct vcpu *v); -struct page_info *__alloc_domheap_pages( - struct domain *d, unsigned int cpu, unsigned int order, - unsigned int memflags); + struct domain *d, unsigned int order, unsigned int memflags, + unsigned int node_id); void free_domheap_pages(struct page_info *pg, unsigned int order); unsigned long avail_domheap_pages_region( unsigned int node, unsigned int min_width, unsigned int max_width); unsigned long avail_domheap_pages(void); #define alloc_domheap_page(d) (alloc_domheap_pages(d,0,0)) -#define alloc_domheap_page_on_node(d, v) (alloc_domheap_pages_on_node(d,0,0,v)) +#define alloc_domheap_page_on_node(d, n) (alloc_domheap_pages_on_node(d,0,0,n)) #define free_domheap_page(p) (free_domheap_pages(p,0)) void scrub_heap_pages(void); [-- Attachment #3: alloc_domheap_node.patch --] [-- Type: text/plain, Size: 13713 bytes --] diff -r db943e8d1051 xen/arch/x86/domain.c --- a/xen/arch/x86/domain.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/domain.c Thu Apr 03 00:38:44 2008 +0200 @@ -46,6 +46,7 @@ #include <asm/debugreg.h> #include <asm/msr.h> #include <asm/nmi.h> +#include <asm/numa.h> #include <xen/iommu.h> #ifdef CONFIG_COMPAT #include <compat/vcpu.h> @@ -477,8 +478,9 @@ int arch_domain_create(struct domain *d, #else /* __x86_64__ */ - if ( (pg = alloc_domheap_page(NULL)) == NULL ) - goto fail; + if ( (pg = alloc_domheap_page_on_node(NULL, + cpu_to_node(d->vcpu[0]->processor))) == NULL ) + goto fail; d->arch.mm_perdomain_l2 = page_to_virt(pg); clear_page(d->arch.mm_perdomain_l2); for ( i = 0; i < (1 << pdpt_order); i++ ) @@ -486,8 +488,9 @@ int arch_domain_create(struct domain *d, l2e_from_page(virt_to_page(d->arch.mm_perdomain_pt)+i, __PAGE_HYPERVISOR); - if ( (pg = alloc_domheap_page(NULL)) == NULL ) - goto fail; + if ( (pg = alloc_domheap_page_on_node(NULL, + cpu_to_node(d->vcpu[0]->processor))) == NULL ) + goto fail; d->arch.mm_perdomain_l3 = page_to_virt(pg); clear_page(d->arch.mm_perdomain_l3); d->arch.mm_perdomain_l3[l3_table_offset(PERDOMAIN_VIRT_START)] = diff -r db943e8d1051 xen/arch/x86/hvm/stdvga.c --- a/xen/arch/x86/hvm/stdvga.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/hvm/stdvga.c Thu Apr 03 00:38:44 2008 +0200 @@ -32,6 +32,7 @@ #include <xen/sched.h> #include <xen/domain_page.h> #include <asm/hvm/support.h> +#include <asm/numa.h> #define PAT(x) (x) static const uint32_t mask16[16] = { @@ -513,8 +514,9 @@ void stdvga_init(struct domain *d) for ( i = 0; i != ARRAY_SIZE(s->vram_page); i++ ) { - if ( (pg = alloc_domheap_page(NULL)) == NULL ) - break; + if ( (pg = alloc_domheap_page_on_node(NULL, + cpu_to_node(d->vcpu[0]->processor))) == NULL ) + break; s->vram_page[i] = pg; p = map_domain_page(page_to_mfn(pg)); clear_page(p); diff -r db943e8d1051 xen/arch/x86/hvm/vlapic.c --- a/xen/arch/x86/hvm/vlapic.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/hvm/vlapic.c Thu Apr 03 00:38:45 2008 +0200 @@ -33,6 +33,7 @@ #include <xen/sched.h> #include <asm/current.h> #include <asm/hvm/vmx/vmx.h> +#include <asm/numa.h> #include <public/hvm/ioreq.h> #include <public/hvm/params.h> @@ -928,7 +929,8 @@ int vlapic_init(struct vcpu *v) memflags = MEMF_bits(32); #endif - vlapic->regs_page = alloc_domheap_pages(NULL, 0, memflags); + vlapic->regs_page = alloc_domheap_pages_on_node(NULL, 0, memflags, + cpu_to_node(v->processor)); if ( vlapic->regs_page == NULL ) { dprintk(XENLOG_ERR, "alloc vlapic regs error: %d/%d\n", diff -r db943e8d1051 xen/arch/x86/mm/hap/hap.c --- a/xen/arch/x86/mm/hap/hap.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/mm/hap/hap.c Thu Apr 03 00:38:45 2008 +0200 @@ -38,6 +38,7 @@ #include <asm/hap.h> #include <asm/paging.h> #include <asm/domain.h> +#include <asm/numa.h> #include "private.h" @@ -135,7 +136,8 @@ static struct page_info *hap_alloc_p2m_p && mfn_x(page_to_mfn(pg)) >= (1UL << (32 - PAGE_SHIFT)) ) { free_domheap_page(pg); - pg = alloc_domheap_pages(NULL, 0, MEMF_bits(32)); + pg = alloc_domheap_pages_on_node(NULL, 0, MEMF_bits(32), + cpu_to_node(d->vcpu[0]->processor)); if ( likely(pg != NULL) ) { void *p = hap_map_domain_page(page_to_mfn(pg)); @@ -199,7 +201,8 @@ hap_set_allocation(struct domain *d, uns if ( d->arch.paging.hap.total_pages < pages ) { /* Need to allocate more memory from domheap */ - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page_on_node(NULL, + cpu_to_node(d->vcpu[0]->processor)); if ( pg == NULL ) { HAP_PRINTK("failed to allocate hap pages.\n"); diff -r db943e8d1051 xen/arch/x86/mm/paging.c --- a/xen/arch/x86/mm/paging.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/mm/paging.c Thu Apr 03 00:38:45 2008 +0200 @@ -26,6 +26,7 @@ #include <asm/p2m.h> #include <asm/hap.h> #include <asm/guest_access.h> +#include <asm/numa.h> #include <xsm/xsm.h> #define hap_enabled(d) (is_hvm_domain(d) && (d)->arch.hvm_domain.hap_enabled) @@ -99,7 +100,8 @@ static mfn_t paging_new_log_dirty_page(s static mfn_t paging_new_log_dirty_page(struct domain *d, void **mapping_p) { mfn_t mfn; - struct page_info *page = alloc_domheap_page(NULL); + struct page_info *page = alloc_domheap_page_on_node(NULL, + cpu_to_node(d->vcpu[0]->processor)); if ( unlikely(page == NULL) ) { diff -r db943e8d1051 xen/arch/x86/mm/shadow/common.c --- a/xen/arch/x86/mm/shadow/common.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/mm/shadow/common.c Thu Apr 03 00:38:45 2008 +0200 @@ -36,6 +36,7 @@ #include <asm/current.h> #include <asm/flushtlb.h> #include <asm/shadow.h> +#include <asm/numa.h> #include "private.h" @@ -1249,7 +1250,8 @@ static unsigned int sh_set_allocation(st { /* Need to allocate more memory from domheap */ sp = (struct shadow_page_info *) - alloc_domheap_pages(NULL, order, 0); + alloc_domheap_pages_on_node(NULL, order, 0, + cpu_to_node(d->vcpu[0]->processor)); if ( sp == NULL ) { SHADOW_PRINTK("failed to allocate shadow pages.\n"); diff -r db943e8d1051 xen/common/memory.c --- a/xen/common/memory.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/common/memory.c Thu Apr 03 00:38:45 2008 +0200 @@ -21,6 +21,7 @@ #include <xen/errno.h> #include <asm/current.h> #include <asm/hardirq.h> +#include <asm/numa.h> #include <public/memory.h> #include <xsm/xsm.h> @@ -37,10 +38,10 @@ struct memop_args { int preempted; /* Was the hypercall preempted? */ }; -static unsigned int select_local_cpu(struct domain *d) +static unsigned int select_local_node(struct domain *d) { struct vcpu *v = d->vcpu[0]; - return (v ? v->processor : 0); + return (v ? cpu_to_node(v->processor) : 0); } static void increase_reservation(struct memop_args *a) @@ -49,7 +50,7 @@ static void increase_reservation(struct unsigned long i; xen_pfn_t mfn; struct domain *d = a->domain; - unsigned int cpu = select_local_cpu(d); + unsigned int node = select_local_node(d); if ( !guest_handle_is_null(a->extent_list) && !guest_handle_okay(a->extent_list, a->nr_extents) ) @@ -67,7 +68,8 @@ static void increase_reservation(struct goto out; } - page = __alloc_domheap_pages(d, cpu, a->extent_order, a->memflags); + page = alloc_domheap_pages_on_node ( + d, a->extent_order, a->memflags, node); if ( unlikely(page == NULL) ) { gdprintk(XENLOG_INFO, "Could not allocate order=%d extent: " @@ -96,7 +98,7 @@ static void populate_physmap(struct memo unsigned long i, j; xen_pfn_t gpfn, mfn; struct domain *d = a->domain; - unsigned int cpu = select_local_cpu(d); + unsigned int node = select_local_node(d); if ( !guest_handle_okay(a->extent_list, a->nr_extents) ) return; @@ -116,7 +118,8 @@ static void populate_physmap(struct memo if ( unlikely(__copy_from_guest_offset(&gpfn, a->extent_list, i, 1)) ) goto out; - page = __alloc_domheap_pages(d, cpu, a->extent_order, a->memflags); + page = alloc_domheap_pages_on_node ( + d, a->extent_order, a->memflags, node); if ( unlikely(page == NULL) ) { gdprintk(XENLOG_INFO, "Could not allocate order=%d extent: " @@ -296,7 +299,7 @@ static long memory_exchange(XEN_GUEST_HA unsigned long in_chunk_order, out_chunk_order; xen_pfn_t gpfn, gmfn, mfn; unsigned long i, j, k; - unsigned int memflags = 0, cpu; + unsigned int memflags = 0, node; long rc = 0; struct domain *d; struct page_info *page; @@ -352,7 +355,7 @@ static long memory_exchange(XEN_GUEST_HA memflags |= MEMF_bits(domain_clamp_alloc_bitsize( d, exch.out.address_bits ? : (BITS_PER_LONG+PAGE_SHIFT))); - cpu = select_local_cpu(d); + node = select_local_node(d); for ( i = (exch.nr_exchanged >> in_chunk_order); i < (exch.in.nr_extents >> in_chunk_order); @@ -401,8 +404,8 @@ static long memory_exchange(XEN_GUEST_HA /* Allocate a chunk's worth of anonymous output pages. */ for ( j = 0; j < (1UL << out_chunk_order); j++ ) { - page = __alloc_domheap_pages( - NULL, cpu, exch.out.extent_order, memflags); + page = alloc_domheap_pages_on_node( + NULL, exch.out.extent_order, memflags, node); if ( unlikely(page == NULL) ) { rc = -ENOMEM; diff -r db943e8d1051 xen/common/page_alloc.c --- a/xen/common/page_alloc.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/common/page_alloc.c Thu Apr 03 00:38:45 2008 +0200 @@ -36,6 +36,7 @@ #include <xen/numa.h> #include <xen/nodemask.h> #include <asm/page.h> +#include <asm/numa.h> #include <asm/flushtlb.h> /* @@ -328,10 +329,10 @@ static void init_node_heap(int node) /* Allocate 2^@order contiguous pages. */ static struct page_info *alloc_heap_pages( unsigned int zone_lo, unsigned int zone_hi, - unsigned int cpu, unsigned int order) + unsigned int node, unsigned int order) { unsigned int i, j, zone; - unsigned int node = cpu_to_node(cpu), num_nodes = num_online_nodes(); + unsigned int num_nodes = num_online_nodes(); unsigned long request = 1UL << order; cpumask_t extra_cpus_mask, mask; struct page_info *pg; @@ -670,7 +671,8 @@ void *alloc_xenheap_pages(unsigned int o ASSERT(!in_irq()); - pg = alloc_heap_pages(MEMZONE_XEN, MEMZONE_XEN, smp_processor_id(), order); + pg = alloc_heap_pages(MEMZONE_XEN, MEMZONE_XEN, + cpu_to_node(smp_processor_id()), order); if ( unlikely(pg == NULL) ) goto no_memory; @@ -778,9 +780,9 @@ int assign_pages( } -struct page_info *__alloc_domheap_pages( - struct domain *d, unsigned int cpu, unsigned int order, - unsigned int memflags) +struct page_info *alloc_domheap_pages_on_node( + struct domain *d, unsigned int order, unsigned int memflags, + unsigned int node) { struct page_info *pg = NULL; unsigned int bits = memflags >> _MEMF_bits, zone_hi = NR_ZONES - 1; @@ -797,7 +799,7 @@ struct page_info *__alloc_domheap_pages( if ( (zone_hi + PAGE_SHIFT) >= dma_bitsize ) { - pg = alloc_heap_pages(dma_bitsize - PAGE_SHIFT, zone_hi, cpu, order); + pg = alloc_heap_pages(dma_bitsize - PAGE_SHIFT, zone_hi, node, order); /* Failure? Then check if we can fall back to the DMA pool. */ if ( unlikely(pg == NULL) && @@ -811,7 +813,7 @@ struct page_info *__alloc_domheap_pages( if ( (pg == NULL) && ((pg = alloc_heap_pages(MEMZONE_XEN + 1, zone_hi, - cpu, order)) == NULL) ) + node, order)) == NULL) ) return NULL; if ( (d != NULL) && assign_pages(d, pg, order, memflags) ) @@ -826,7 +828,8 @@ struct page_info *alloc_domheap_pages( struct page_info *alloc_domheap_pages( struct domain *d, unsigned int order, unsigned int flags) { - return __alloc_domheap_pages(d, smp_processor_id(), order, flags); + return alloc_domheap_pages_on_node (d, order, flags, + cpu_to_node (smp_processor_id()); } void free_domheap_pages(struct page_info *pg, unsigned int order) diff -r db943e8d1051 xen/drivers/passthrough/vtd/iommu.c --- a/xen/drivers/passthrough/vtd/iommu.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/drivers/passthrough/vtd/iommu.c Thu Apr 03 00:38:45 2008 +0200 @@ -24,6 +24,7 @@ #include <xen/xmalloc.h> #include <xen/domain_page.h> #include <xen/iommu.h> +#include <asm/numa.h> #include "iommu.h" #include "dmar.h" #include "../pci-direct.h" @@ -269,7 +270,8 @@ static struct page_info *addr_to_dma_pag if ( dma_pte_addr(*pte) == 0 ) { - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page_on_node(NULL, + cpu_to_node(domain->vcpu[0]->processor)); vaddr = map_domain_page(page_to_mfn(pg)); if ( !vaddr ) { diff -r db943e8d1051 xen/include/xen/mm.h --- a/xen/include/xen/mm.h Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/include/xen/mm.h Thu Apr 03 00:38:45 2008 +0200 @@ -54,14 +54,15 @@ void init_domheap_pages(paddr_t ps, padd void init_domheap_pages(paddr_t ps, paddr_t pe); struct page_info *alloc_domheap_pages( struct domain *d, unsigned int order, unsigned int memflags); -struct page_info *__alloc_domheap_pages( - struct domain *d, unsigned int cpu, unsigned int order, - unsigned int memflags); +struct page_info *alloc_domheap_pages_on_node( + struct domain *d, unsigned int order, unsigned int memflags, + unsigned int node_id); void free_domheap_pages(struct page_info *pg, unsigned int order); unsigned long avail_domheap_pages_region( unsigned int node, unsigned int min_width, unsigned int max_width); unsigned long avail_domheap_pages(void); #define alloc_domheap_page(d) (alloc_domheap_pages(d,0,0)) +#define alloc_domheap_page_on_node(d, n) (alloc_domheap_pages_on_node(d,0,0,n)) #define free_domheap_page(p) (free_domheap_pages(p,0)) void scrub_heap_pages(void); [-- Attachment #4: Type: text/plain, Size: 138 bytes --] _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-04-02 22:49 ` Andre Przywara @ 2008-04-02 23:21 ` Keir Fraser 2008-04-03 10:39 ` Andre Przywara 2008-04-07 13:25 ` Zhai, Edwin 0 siblings, 2 replies; 14+ messages in thread From: Keir Fraser @ 2008-04-02 23:21 UTC (permalink / raw) To: Andre Przywara; +Cc: xen-devel, edwin.zhai On 2/4/08 23:49, "Andre Przywara" <andre.przywara@amd.com> wrote: >> Yes, but it's a bad interface, particularlty when the function is called >> alloc_domheap_pages_on_node(). Pass in a nodeid. Write a helper function to >> work out the nodeid from the domain*. > I was just looking at this code, too, so I fixed this. Eventually > alloc_heap_pages is called, which deals with nodes only, so I replaced > cpu with node everywhere else, too. Now __alloc_domheap_pages and > alloc_domheap_pages_on_node are almost the same (except parameter > ordering), so I removed the first one, since the naming of the latter is > better. Passing node numbers instead of cpu numbers needs cpu_to_node > and asm/numa.h, if you think there is a better way, I am all ears. That's fine. If you reference numa stuff then you need numa.h. But vcpu_to_node and domain_to_node as well as cpu_to_node, please. There's no need to be open-coding v->processor everywhere. Also in future we might care to pick node based on v's affinity map rather than just current processor value. And usage of d->vcpu[0] without checking for != NULL is asking to introduce edge-case bugs. We can easily do that NULL check in one place if we implement domain_to_node(). And, while I'm thinking about the interfaces, let's just stick to alloc_domheap_page() and alloc_domheap_pages(). Let's add a flags parameter to the former (so it matches the latter in that respect) and let's add a MEMF_node() flag subtype (similar to MEMF_bits). Semantics will be that if MEMF_node(node) is provided then we try to allocate memory from node; else we try to allocate memory from a node local to specified domain; else if domain is NULL then we ignore locality. Since zero is probably a valid numa nodeid we can define MEMF_node() as something like ((((node)+1)&0xff)<<8). Then since NUMA_NO_NODE==0xff everything works nicely: MEMF_node(NUMA_NO_NODE) is equivalent to not specifying MEMF_node() at all, which is what we would logically expect. NUMA_NO_NODE probably needs to be pulled out of asm-x86/numa.h and made the official arch-neutral way to specify 'don't care' for numa nodes. -- Keir ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-04-02 23:21 ` Keir Fraser @ 2008-04-03 10:39 ` Andre Przywara 2008-04-03 10:58 ` Keir Fraser 2008-04-07 13:25 ` Zhai, Edwin 1 sibling, 1 reply; 14+ messages in thread From: Andre Przywara @ 2008-04-03 10:39 UTC (permalink / raw) To: Keir Fraser; +Cc: xen-devel, Edwin.Zhai [-- Attachment #1: Type: text/plain, Size: 3673 bytes --] Keir, >>> Yes, but it's a bad interface, particularlty when the function is called >>> alloc_domheap_pages_on_node(). Pass in a nodeid. Write a helper function to >>> work out the nodeid from the domain*. >> I was just looking at this code, too, so I fixed this. Eventually >> alloc_heap_pages is called, which deals with nodes only, so I replaced >> cpu with node everywhere else, too. Now __alloc_domheap_pages and >> alloc_domheap_pages_on_node are almost the same (except parameter >> ordering), so I removed the first one, since the naming of the latter is >> better. Passing node numbers instead of cpu numbers needs cpu_to_node >> and asm/numa.h, if you think there is a better way, I am all ears. > > That's fine. If you reference numa stuff then you need numa.h. > > But vcpu_to_node and domain_to_node as well as cpu_to_node, please. There's > no need to be open-coding v->processor everywhere. Also in future we might > care to pick node based on v's affinity map rather than just current > processor value. And usage of d->vcpu[0] without checking for != NULL is > asking to introduce edge-case bugs. We can easily do that NULL check in one > place if we implement domain_to_node(). Ok, I did this. I provided NUMA_NO_NODE in the case d->vcpu[0] is NULL, this will be resolved to the current node in alloc_heap_pages (at least for now). By the way, can we solve the DMA_BITSIZE problem (your mail from 28th Feb) with this? If no node is specified, use the current behaviour of preferring non DMA zones, else stick to the given node. If you agree, I will implement this. > And, while I'm thinking about the interfaces, let's just stick to > alloc_domheap_page() and alloc_domheap_pages(). Let's add a flags parameter > to the former (so it matches the latter in that respect) and let's add a > MEMF_node() flag subtype (similar to MEMF_bits). Semantics will be that if > MEMF_node(node) is provided then we try to allocate memory from node; else > we try to allocate memory from a node local to specified domain; else if > domain is NULL then we ignore locality. Sounds reasonable. I changed this, too. If domain is NULL, domain_to_node will return NUMA_NO_NODE, which will eventually ignore locality (in alloc_heap_pages). > > Since zero is probably a valid numa nodeid we can define MEMF_node() as > something like ((((node)+1)&0xff)<<8). Then since NUMA_NO_NODE==0xff > everything works nicely: MEMF_node(NUMA_NO_NODE) is equivalent to not > specifying MEMF_node() at all, which is what we would logically expect. Good idea. > NUMA_NO_NODE probably needs to be pulled out of asm-x86/numa.h and made the > official arch-neutral way to specify 'don't care' for numa nodes. Is this really needed? I provided memflags=0 is all don't care cases, this should work and is more compatible. But beware that this silently assumes in page_alloc.c#alloc_domheap_pages that NUMA_NO_NODE is 0xFF, otherwise this trick will not work. Attached again a diff against my last version and the full patch (for some reason a missing bracket slipped through my last one, sorry for that). This is only quick-tested (booted and created a guest on each node). Signed-off-by: Andre Przywara <andre.przywara@amd.com> Regards, Andre. -- Andre Przywara AMD-Operating System Research Center (OSRC), Dresden, Germany Tel: +49 351 277-84917 ----to satisfy European Law for business letters: AMD Saxony Limited Liability Company & Co. KG, Wilschdorfer Landstr. 101, 01109 Dresden, Germany Register Court Dresden: HRA 4896, General Partner authorized to represent: AMD Saxony LLC (Wilmington, Delaware, US) General Manager of AMD Saxony LLC: Dr. Hans-R. Deppe, Thomas McCoy [-- Attachment #2: alloc_domheap_node_v2_diff.patch --] [-- Type: text/plain, Size: 18897 bytes --] diff -r 848a36114bb9 xen/arch/ia64/xen/mm.c --- a/xen/arch/ia64/xen/mm.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/arch/ia64/xen/mm.c Thu Apr 03 12:25:31 2008 +0200 @@ -822,7 +822,7 @@ __assign_new_domain_page(struct domain * BUG_ON(!pte_none(*pte)); - p = alloc_domheap_page(d); + p = alloc_domheap_page(d, 0); if (unlikely(!p)) { printk("assign_new_domain_page: Can't alloc!!!! Aaaargh!\n"); return(p); @@ -2316,7 +2316,7 @@ steal_page(struct domain *d, struct page unsigned long new_mfn; int ret; - new = alloc_domheap_page(d); + new = alloc_domheap_page(d, 0); if (new == NULL) { gdprintk(XENLOG_INFO, "alloc_domheap_page() failed\n"); return -1; @@ -2603,7 +2603,7 @@ void *pgtable_quicklist_alloc(void) BUG_ON(dom_p2m == NULL); if (!opt_p2m_xenheap) { - struct page_info *page = alloc_domheap_page(dom_p2m); + struct page_info *page = alloc_domheap_page(dom_p2m, 0); if (page == NULL) return NULL; p = page_to_virt(page); diff -r 848a36114bb9 xen/arch/ia64/xen/tlb_track.c --- a/xen/arch/ia64/xen/tlb_track.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/arch/ia64/xen/tlb_track.c Thu Apr 03 12:25:31 2008 +0200 @@ -48,7 +48,7 @@ tlb_track_allocate_entries(struct tlb_tr __func__, tlb_track->num_entries, tlb_track->limit); return -ENOMEM; } - entry_page = alloc_domheap_page(NULL); + entry_page = alloc_domheap_page(NULL, 0); if (entry_page == NULL) { dprintk(XENLOG_WARNING, "%s: domheap page failed. num_entries %d limit %d\n", @@ -84,7 +84,7 @@ tlb_track_create(struct domain* d) if (tlb_track == NULL) goto out; - hash_page = alloc_domheap_page(NULL); + hash_page = alloc_domheap_page(NULL, 0); if (hash_page == NULL) goto out; diff -r 848a36114bb9 xen/arch/x86/domain.c --- a/xen/arch/x86/domain.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/arch/x86/domain.c Thu Apr 03 12:25:31 2008 +0200 @@ -172,7 +172,7 @@ int setup_arg_xlat_area(struct vcpu *v, if ( !d->arch.mm_arg_xlat_l3 ) { - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page(NULL, 0); if ( !pg ) return -ENOMEM; d->arch.mm_arg_xlat_l3 = page_to_virt(pg); @@ -190,7 +190,7 @@ int setup_arg_xlat_area(struct vcpu *v, if ( !l3e_get_intpte(d->arch.mm_arg_xlat_l3[l3_table_offset(va)]) ) { - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page(NULL, 0); if ( !pg ) return -ENOMEM; clear_page(page_to_virt(pg)); @@ -199,7 +199,7 @@ int setup_arg_xlat_area(struct vcpu *v, l2tab = l3e_to_l2e(d->arch.mm_arg_xlat_l3[l3_table_offset(va)]); if ( !l2e_get_intpte(l2tab[l2_table_offset(va)]) ) { - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page(NULL, 0); if ( !pg ) return -ENOMEM; clear_page(page_to_virt(pg)); @@ -207,7 +207,7 @@ int setup_arg_xlat_area(struct vcpu *v, } l1tab = l2e_to_l1e(l2tab[l2_table_offset(va)]); BUG_ON(l1e_get_intpte(l1tab[l1_table_offset(va)])); - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page(NULL, 0); if ( !pg ) return -ENOMEM; l1tab[l1_table_offset(va)] = l1e_from_page(pg, PAGE_HYPERVISOR); @@ -253,7 +253,7 @@ static void release_arg_xlat_area(struct static int setup_compat_l4(struct vcpu *v) { - struct page_info *pg = alloc_domheap_page(NULL); + struct page_info *pg = alloc_domheap_page(NULL, 0); l4_pgentry_t *l4tab; int rc; @@ -478,8 +478,7 @@ int arch_domain_create(struct domain *d, #else /* __x86_64__ */ - if ( (pg = alloc_domheap_page_on_node(NULL, - cpu_to_node(d->vcpu[0]->processor))) == NULL ) + if ( (pg = alloc_domheap_page(NULL, MEMF_node(domain_to_node(d)))) == NULL ) goto fail; d->arch.mm_perdomain_l2 = page_to_virt(pg); clear_page(d->arch.mm_perdomain_l2); @@ -488,9 +487,8 @@ int arch_domain_create(struct domain *d, l2e_from_page(virt_to_page(d->arch.mm_perdomain_pt)+i, __PAGE_HYPERVISOR); - if ( (pg = alloc_domheap_page_on_node(NULL, - cpu_to_node(d->vcpu[0]->processor))) == NULL ) - goto fail; + if ( (pg = alloc_domheap_page(NULL, MEMF_node(domain_to_node(d)))) == NULL ) + goto fail; d->arch.mm_perdomain_l3 = page_to_virt(pg); clear_page(d->arch.mm_perdomain_l3); d->arch.mm_perdomain_l3[l3_table_offset(PERDOMAIN_VIRT_START)] = diff -r 848a36114bb9 xen/arch/x86/domain_build.c --- a/xen/arch/x86/domain_build.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/arch/x86/domain_build.c Thu Apr 03 12:25:31 2008 +0200 @@ -630,7 +630,7 @@ int __init construct_dom0( } else { - page = alloc_domheap_page(NULL); + page = alloc_domheap_page(NULL, 0); if ( !page ) panic("Not enough RAM for domain 0 PML4.\n"); l4start = l4tab = page_to_virt(page); diff -r 848a36114bb9 xen/arch/x86/hvm/stdvga.c --- a/xen/arch/x86/hvm/stdvga.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/arch/x86/hvm/stdvga.c Thu Apr 03 12:25:31 2008 +0200 @@ -514,8 +514,8 @@ void stdvga_init(struct domain *d) for ( i = 0; i != ARRAY_SIZE(s->vram_page); i++ ) { - if ( (pg = alloc_domheap_page_on_node(NULL, - cpu_to_node(d->vcpu[0]->processor))) == NULL ) + if ( (pg = alloc_domheap_page(NULL, + MEMF_node(domain_to_node(d)))) == NULL ) break; s->vram_page[i] = pg; p = map_domain_page(page_to_mfn(pg)); diff -r 848a36114bb9 xen/arch/x86/hvm/vlapic.c --- a/xen/arch/x86/hvm/vlapic.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/arch/x86/hvm/vlapic.c Thu Apr 03 12:25:31 2008 +0200 @@ -917,7 +917,7 @@ int vlapic_init(struct vcpu *v) int vlapic_init(struct vcpu *v) { struct vlapic *vlapic = vcpu_vlapic(v); - unsigned int memflags = 0; + unsigned int memflags = MEMF_node (vcpu_to_node(v)); HVM_DBG_LOG(DBG_LEVEL_VLAPIC, "%d", v->vcpu_id); @@ -926,11 +926,10 @@ int vlapic_init(struct vcpu *v) #ifdef __i386__ /* 32-bit VMX may be limited to 32-bit physical addresses. */ if ( boot_cpu_data.x86_vendor == X86_VENDOR_INTEL ) - memflags = MEMF_bits(32); + memflags |= MEMF_bits(32); #endif - vlapic->regs_page = alloc_domheap_pages_on_node(NULL, 0, memflags, - cpu_to_node(v->processor)); + vlapic->regs_page = alloc_domheap_page(NULL, memflags); if ( vlapic->regs_page == NULL ) { dprintk(XENLOG_ERR, "alloc vlapic regs error: %d/%d\n", diff -r 848a36114bb9 xen/arch/x86/mm/hap/hap.c --- a/xen/arch/x86/mm/hap/hap.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/arch/x86/mm/hap/hap.c Thu Apr 03 12:25:31 2008 +0200 @@ -136,8 +136,8 @@ static struct page_info *hap_alloc_p2m_p && mfn_x(page_to_mfn(pg)) >= (1UL << (32 - PAGE_SHIFT)) ) { free_domheap_page(pg); - pg = alloc_domheap_pages_on_node(NULL, 0, MEMF_bits(32), - cpu_to_node(d->vcpu[0]->processor)); + pg = alloc_domheap_page(NULL, MEMF_bits(32) | + MEMF_node(domain_to_node(d))); if ( likely(pg != NULL) ) { void *p = hap_map_domain_page(page_to_mfn(pg)); @@ -201,8 +201,7 @@ hap_set_allocation(struct domain *d, uns if ( d->arch.paging.hap.total_pages < pages ) { /* Need to allocate more memory from domheap */ - pg = alloc_domheap_page_on_node(NULL, - cpu_to_node(d->vcpu[0]->processor)); + pg = alloc_domheap_page(NULL, MEMF_node(domain_to_node(d))); if ( pg == NULL ) { HAP_PRINTK("failed to allocate hap pages.\n"); diff -r 848a36114bb9 xen/arch/x86/mm/paging.c --- a/xen/arch/x86/mm/paging.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/arch/x86/mm/paging.c Thu Apr 03 12:25:31 2008 +0200 @@ -100,8 +100,8 @@ static mfn_t paging_new_log_dirty_page(s static mfn_t paging_new_log_dirty_page(struct domain *d, void **mapping_p) { mfn_t mfn; - struct page_info *page = alloc_domheap_page_on_node(NULL, - cpu_to_node(d->vcpu[0]->processor)); + struct page_info *page = alloc_domheap_page(NULL, + MEMF_node(domain_to_node(d))); if ( unlikely(page == NULL) ) { diff -r 848a36114bb9 xen/arch/x86/mm/shadow/common.c --- a/xen/arch/x86/mm/shadow/common.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/arch/x86/mm/shadow/common.c Thu Apr 03 12:25:31 2008 +0200 @@ -1250,8 +1250,7 @@ static unsigned int sh_set_allocation(st { /* Need to allocate more memory from domheap */ sp = (struct shadow_page_info *) - alloc_domheap_pages_on_node(NULL, order, 0, - cpu_to_node(d->vcpu[0]->processor)); + alloc_domheap_pages(NULL, order, MEMF_node(domain_to_node(d))); if ( sp == NULL ) { SHADOW_PRINTK("failed to allocate shadow pages.\n"); diff -r 848a36114bb9 xen/arch/x86/x86_64/mm.c --- a/xen/arch/x86/x86_64/mm.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/arch/x86/x86_64/mm.c Thu Apr 03 12:25:31 2008 +0200 @@ -59,7 +59,7 @@ void *alloc_xen_pagetable(void) if ( !early_boot ) { - struct page_info *pg = alloc_domheap_page(NULL); + struct page_info *pg = alloc_domheap_page(NULL, 0); BUG_ON(pg == NULL); return page_to_virt(pg); } @@ -108,7 +108,7 @@ void __init paging_init(void) struct page_info *l1_pg, *l2_pg, *l3_pg; /* Create user-accessible L2 directory to map the MPT for guests. */ - if ( (l3_pg = alloc_domheap_page(NULL)) == NULL ) + if ( (l3_pg = alloc_domheap_page(NULL, 0)) == NULL ) goto nomem; l3_ro_mpt = page_to_virt(l3_pg); clear_page(l3_ro_mpt); @@ -134,7 +134,7 @@ void __init paging_init(void) 1UL << L2_PAGETABLE_SHIFT); if ( !((unsigned long)l2_ro_mpt & ~PAGE_MASK) ) { - if ( (l2_pg = alloc_domheap_page(NULL)) == NULL ) + if ( (l2_pg = alloc_domheap_page(NULL, 0)) == NULL ) goto nomem; va = RO_MPT_VIRT_START + (i << L2_PAGETABLE_SHIFT); l2_ro_mpt = page_to_virt(l2_pg); @@ -154,7 +154,7 @@ void __init paging_init(void) l4_table_offset(HIRO_COMPAT_MPT_VIRT_START)); l3_ro_mpt = l4e_to_l3e(idle_pg_table[l4_table_offset( HIRO_COMPAT_MPT_VIRT_START)]); - if ( (l2_pg = alloc_domheap_page(NULL)) == NULL ) + if ( (l2_pg = alloc_domheap_page(NULL, 0)) == NULL ) goto nomem; compat_idle_pg_table_l2 = l2_ro_mpt = page_to_virt(l2_pg); clear_page(l2_ro_mpt); diff -r 848a36114bb9 xen/common/grant_table.c --- a/xen/common/grant_table.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/common/grant_table.c Thu Apr 03 12:25:31 2008 +0200 @@ -1102,7 +1102,7 @@ gnttab_transfer( struct page_info *new_page; void *sp, *dp; - new_page = alloc_domheap_pages(NULL, 0, MEMF_bits(max_bitsize)); + new_page = alloc_domheap_page(NULL, MEMF_bits(max_bitsize)); if ( new_page == NULL ) { gop.status = GNTST_address_too_big; diff -r 848a36114bb9 xen/common/memory.c --- a/xen/common/memory.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/common/memory.c Thu Apr 03 12:25:32 2008 +0200 @@ -38,19 +38,13 @@ struct memop_args { int preempted; /* Was the hypercall preempted? */ }; -static unsigned int select_local_node(struct domain *d) -{ - struct vcpu *v = d->vcpu[0]; - return (v ? cpu_to_node(v->processor) : 0); -} - static void increase_reservation(struct memop_args *a) { struct page_info *page; unsigned long i; xen_pfn_t mfn; struct domain *d = a->domain; - unsigned int node = select_local_node(d); + unsigned int node = domain_to_node (d); if ( !guest_handle_is_null(a->extent_list) && !guest_handle_okay(a->extent_list, a->nr_extents) ) @@ -68,8 +62,8 @@ static void increase_reservation(struct goto out; } - page = alloc_domheap_pages_on_node ( - d, a->extent_order, a->memflags, node); + page = alloc_domheap_pages ( + d, a->extent_order, a->memflags | MEMF_node(node)); if ( unlikely(page == NULL) ) { gdprintk(XENLOG_INFO, "Could not allocate order=%d extent: " @@ -98,7 +92,7 @@ static void populate_physmap(struct memo unsigned long i, j; xen_pfn_t gpfn, mfn; struct domain *d = a->domain; - unsigned int node = select_local_node(d); + unsigned int node = domain_to_node(d); if ( !guest_handle_okay(a->extent_list, a->nr_extents) ) return; @@ -118,8 +112,8 @@ static void populate_physmap(struct memo if ( unlikely(__copy_from_guest_offset(&gpfn, a->extent_list, i, 1)) ) goto out; - page = alloc_domheap_pages_on_node ( - d, a->extent_order, a->memflags, node); + page = alloc_domheap_pages ( + d, a->extent_order, a->memflags | MEMF_node(node)); if ( unlikely(page == NULL) ) { gdprintk(XENLOG_INFO, "Could not allocate order=%d extent: " @@ -299,7 +293,7 @@ static long memory_exchange(XEN_GUEST_HA unsigned long in_chunk_order, out_chunk_order; xen_pfn_t gpfn, gmfn, mfn; unsigned long i, j, k; - unsigned int memflags = 0, node; + unsigned int memflags = 0; long rc = 0; struct domain *d; struct page_info *page; @@ -355,7 +349,7 @@ static long memory_exchange(XEN_GUEST_HA memflags |= MEMF_bits(domain_clamp_alloc_bitsize( d, exch.out.address_bits ? : (BITS_PER_LONG+PAGE_SHIFT))); - node = select_local_node(d); + memflags |= MEMF_node (domain_to_node(d)); for ( i = (exch.nr_exchanged >> in_chunk_order); i < (exch.in.nr_extents >> in_chunk_order); @@ -404,8 +398,8 @@ static long memory_exchange(XEN_GUEST_HA /* Allocate a chunk's worth of anonymous output pages. */ for ( j = 0; j < (1UL << out_chunk_order); j++ ) { - page = alloc_domheap_pages_on_node( - NULL, exch.out.extent_order, memflags, node); + page = alloc_domheap_pages( + NULL, exch.out.extent_order, memflags); if ( unlikely(page == NULL) ) { rc = -ENOMEM; diff -r 848a36114bb9 xen/common/page_alloc.c --- a/xen/common/page_alloc.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/common/page_alloc.c Thu Apr 03 12:25:32 2008 +0200 @@ -337,6 +337,7 @@ static struct page_info *alloc_heap_page cpumask_t extra_cpus_mask, mask; struct page_info *pg; + if ( node == NUMA_NO_NODE ) node = cpu_to_node(smp_processor_id()); ASSERT(node >= 0); ASSERT(node < num_nodes); ASSERT(zone_lo <= zone_hi); @@ -780,12 +781,12 @@ int assign_pages( } -struct page_info *alloc_domheap_pages_on_node( - struct domain *d, unsigned int order, unsigned int memflags, - unsigned int node) +struct page_info *alloc_domheap_pages( + struct domain *d, unsigned int order, unsigned int memflags) { struct page_info *pg = NULL; unsigned int bits = memflags >> _MEMF_bits, zone_hi = NR_ZONES - 1; + unsigned int node = (((memflags >> _MEMF_node)&0xFF) - 1 ) &0xFF; ASSERT(!in_irq()); @@ -823,13 +824,6 @@ struct page_info *alloc_domheap_pages_on } return pg; -} - -struct page_info *alloc_domheap_pages( - struct domain *d, unsigned int order, unsigned int flags) -{ - return alloc_domheap_pages_on_node (d, order, flags, - cpu_to_node (smp_processor_id()); } void free_domheap_pages(struct page_info *pg, unsigned int order) diff -r 848a36114bb9 xen/drivers/passthrough/vtd/iommu.c --- a/xen/drivers/passthrough/vtd/iommu.c Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/drivers/passthrough/vtd/iommu.c Thu Apr 03 12:25:32 2008 +0200 @@ -270,8 +270,8 @@ static struct page_info *addr_to_dma_pag if ( dma_pte_addr(*pte) == 0 ) { - pg = alloc_domheap_page_on_node(NULL, - cpu_to_node(domain->vcpu[0]->processor)); + pg = alloc_domheap_page(NULL, + MEMF_node(domain_to_node(domain))); vaddr = map_domain_page(page_to_mfn(pg)); if ( !vaddr ) { diff -r 848a36114bb9 xen/include/asm-x86/numa.h --- a/xen/include/asm-x86/numa.h Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/include/asm-x86/numa.h Thu Apr 03 12:25:32 2008 +0200 @@ -4,11 +4,16 @@ #include <xen/cpumask.h> #define NODES_SHIFT 6 +#define NUMA_NO_NODE 0xff extern unsigned char cpu_to_node[]; extern cpumask_t node_to_cpumask[]; #define cpu_to_node(cpu) (cpu_to_node[cpu]) +#define domain_to_node(domain) ((domain!=NULL && domain->vcpu[0]!=NULL)?\ + cpu_to_node[domain->vcpu[0]->processor]:\ + NUMA_NO_NODE) +#define vcpu_to_node(vcpu) (cpu_to_node[v->processor]) #define parent_node(node) (node) #define node_to_first_cpu(node) (__ffs(node_to_cpumask[node])) #define node_to_cpumask(node) (node_to_cpumask[node]) @@ -73,6 +78,5 @@ static inline __attribute__((pure)) int #define clear_node_cpumask(cpu) do {} while (0) #endif -#define NUMA_NO_NODE 0xff #endif diff -r 848a36114bb9 xen/include/xen/mm.h --- a/xen/include/xen/mm.h Thu Apr 03 00:37:33 2008 +0200 +++ b/xen/include/xen/mm.h Thu Apr 03 12:25:32 2008 +0200 @@ -54,15 +54,11 @@ void init_domheap_pages(paddr_t ps, padd void init_domheap_pages(paddr_t ps, paddr_t pe); struct page_info *alloc_domheap_pages( struct domain *d, unsigned int order, unsigned int memflags); -struct page_info *alloc_domheap_pages_on_node( - struct domain *d, unsigned int order, unsigned int memflags, - unsigned int node_id); void free_domheap_pages(struct page_info *pg, unsigned int order); unsigned long avail_domheap_pages_region( unsigned int node, unsigned int min_width, unsigned int max_width); unsigned long avail_domheap_pages(void); -#define alloc_domheap_page(d) (alloc_domheap_pages(d,0,0)) -#define alloc_domheap_page_on_node(d, n) (alloc_domheap_pages_on_node(d,0,0,n)) +#define alloc_domheap_page(d,f) (alloc_domheap_pages(d,0,f)) #define free_domheap_page(p) (free_domheap_pages(p,0)) void scrub_heap_pages(void); @@ -76,6 +72,8 @@ int assign_pages( /* memflags: */ #define _MEMF_no_refcount 0 #define MEMF_no_refcount (1U<<_MEMF_no_refcount) +#define _MEMF_node 8 +#define MEMF_node(n) ((((n)+1)&0xff)<<_MEMF_node) #define _MEMF_bits 24 #define MEMF_bits(n) ((n)<<_MEMF_bits) [-- Attachment #3: alloc_domheap_node_v2.patch --] [-- Type: text/plain, Size: 21271 bytes --] diff -r db943e8d1051 xen/arch/ia64/xen/mm.c --- a/xen/arch/ia64/xen/mm.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/ia64/xen/mm.c Thu Apr 03 12:25:50 2008 +0200 @@ -822,7 +822,7 @@ __assign_new_domain_page(struct domain * BUG_ON(!pte_none(*pte)); - p = alloc_domheap_page(d); + p = alloc_domheap_page(d, 0); if (unlikely(!p)) { printk("assign_new_domain_page: Can't alloc!!!! Aaaargh!\n"); return(p); @@ -2316,7 +2316,7 @@ steal_page(struct domain *d, struct page unsigned long new_mfn; int ret; - new = alloc_domheap_page(d); + new = alloc_domheap_page(d, 0); if (new == NULL) { gdprintk(XENLOG_INFO, "alloc_domheap_page() failed\n"); return -1; @@ -2603,7 +2603,7 @@ void *pgtable_quicklist_alloc(void) BUG_ON(dom_p2m == NULL); if (!opt_p2m_xenheap) { - struct page_info *page = alloc_domheap_page(dom_p2m); + struct page_info *page = alloc_domheap_page(dom_p2m, 0); if (page == NULL) return NULL; p = page_to_virt(page); diff -r db943e8d1051 xen/arch/ia64/xen/tlb_track.c --- a/xen/arch/ia64/xen/tlb_track.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/ia64/xen/tlb_track.c Thu Apr 03 12:25:50 2008 +0200 @@ -48,7 +48,7 @@ tlb_track_allocate_entries(struct tlb_tr __func__, tlb_track->num_entries, tlb_track->limit); return -ENOMEM; } - entry_page = alloc_domheap_page(NULL); + entry_page = alloc_domheap_page(NULL, 0); if (entry_page == NULL) { dprintk(XENLOG_WARNING, "%s: domheap page failed. num_entries %d limit %d\n", @@ -84,7 +84,7 @@ tlb_track_create(struct domain* d) if (tlb_track == NULL) goto out; - hash_page = alloc_domheap_page(NULL); + hash_page = alloc_domheap_page(NULL, 0); if (hash_page == NULL) goto out; diff -r db943e8d1051 xen/arch/x86/domain.c --- a/xen/arch/x86/domain.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/domain.c Thu Apr 03 12:25:50 2008 +0200 @@ -46,6 +46,7 @@ #include <asm/debugreg.h> #include <asm/msr.h> #include <asm/nmi.h> +#include <asm/numa.h> #include <xen/iommu.h> #ifdef CONFIG_COMPAT #include <compat/vcpu.h> @@ -171,7 +172,7 @@ int setup_arg_xlat_area(struct vcpu *v, if ( !d->arch.mm_arg_xlat_l3 ) { - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page(NULL, 0); if ( !pg ) return -ENOMEM; d->arch.mm_arg_xlat_l3 = page_to_virt(pg); @@ -189,7 +190,7 @@ int setup_arg_xlat_area(struct vcpu *v, if ( !l3e_get_intpte(d->arch.mm_arg_xlat_l3[l3_table_offset(va)]) ) { - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page(NULL, 0); if ( !pg ) return -ENOMEM; clear_page(page_to_virt(pg)); @@ -198,7 +199,7 @@ int setup_arg_xlat_area(struct vcpu *v, l2tab = l3e_to_l2e(d->arch.mm_arg_xlat_l3[l3_table_offset(va)]); if ( !l2e_get_intpte(l2tab[l2_table_offset(va)]) ) { - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page(NULL, 0); if ( !pg ) return -ENOMEM; clear_page(page_to_virt(pg)); @@ -206,7 +207,7 @@ int setup_arg_xlat_area(struct vcpu *v, } l1tab = l2e_to_l1e(l2tab[l2_table_offset(va)]); BUG_ON(l1e_get_intpte(l1tab[l1_table_offset(va)])); - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page(NULL, 0); if ( !pg ) return -ENOMEM; l1tab[l1_table_offset(va)] = l1e_from_page(pg, PAGE_HYPERVISOR); @@ -252,7 +253,7 @@ static void release_arg_xlat_area(struct static int setup_compat_l4(struct vcpu *v) { - struct page_info *pg = alloc_domheap_page(NULL); + struct page_info *pg = alloc_domheap_page(NULL, 0); l4_pgentry_t *l4tab; int rc; @@ -477,8 +478,8 @@ int arch_domain_create(struct domain *d, #else /* __x86_64__ */ - if ( (pg = alloc_domheap_page(NULL)) == NULL ) - goto fail; + if ( (pg = alloc_domheap_page(NULL, MEMF_node(domain_to_node(d)))) == NULL ) + goto fail; d->arch.mm_perdomain_l2 = page_to_virt(pg); clear_page(d->arch.mm_perdomain_l2); for ( i = 0; i < (1 << pdpt_order); i++ ) @@ -486,7 +487,7 @@ int arch_domain_create(struct domain *d, l2e_from_page(virt_to_page(d->arch.mm_perdomain_pt)+i, __PAGE_HYPERVISOR); - if ( (pg = alloc_domheap_page(NULL)) == NULL ) + if ( (pg = alloc_domheap_page(NULL, MEMF_node(domain_to_node(d)))) == NULL ) goto fail; d->arch.mm_perdomain_l3 = page_to_virt(pg); clear_page(d->arch.mm_perdomain_l3); diff -r db943e8d1051 xen/arch/x86/domain_build.c --- a/xen/arch/x86/domain_build.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/domain_build.c Thu Apr 03 12:25:50 2008 +0200 @@ -630,7 +630,7 @@ int __init construct_dom0( } else { - page = alloc_domheap_page(NULL); + page = alloc_domheap_page(NULL, 0); if ( !page ) panic("Not enough RAM for domain 0 PML4.\n"); l4start = l4tab = page_to_virt(page); diff -r db943e8d1051 xen/arch/x86/hvm/stdvga.c --- a/xen/arch/x86/hvm/stdvga.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/hvm/stdvga.c Thu Apr 03 12:25:50 2008 +0200 @@ -32,6 +32,7 @@ #include <xen/sched.h> #include <xen/domain_page.h> #include <asm/hvm/support.h> +#include <asm/numa.h> #define PAT(x) (x) static const uint32_t mask16[16] = { @@ -513,8 +514,9 @@ void stdvga_init(struct domain *d) for ( i = 0; i != ARRAY_SIZE(s->vram_page); i++ ) { - if ( (pg = alloc_domheap_page(NULL)) == NULL ) - break; + if ( (pg = alloc_domheap_page(NULL, + MEMF_node(domain_to_node(d)))) == NULL ) + break; s->vram_page[i] = pg; p = map_domain_page(page_to_mfn(pg)); clear_page(p); diff -r db943e8d1051 xen/arch/x86/hvm/vlapic.c --- a/xen/arch/x86/hvm/vlapic.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/hvm/vlapic.c Thu Apr 03 12:25:50 2008 +0200 @@ -33,6 +33,7 @@ #include <xen/sched.h> #include <asm/current.h> #include <asm/hvm/vmx/vmx.h> +#include <asm/numa.h> #include <public/hvm/ioreq.h> #include <public/hvm/params.h> @@ -916,7 +917,7 @@ int vlapic_init(struct vcpu *v) int vlapic_init(struct vcpu *v) { struct vlapic *vlapic = vcpu_vlapic(v); - unsigned int memflags = 0; + unsigned int memflags = MEMF_node (vcpu_to_node(v)); HVM_DBG_LOG(DBG_LEVEL_VLAPIC, "%d", v->vcpu_id); @@ -925,10 +926,10 @@ int vlapic_init(struct vcpu *v) #ifdef __i386__ /* 32-bit VMX may be limited to 32-bit physical addresses. */ if ( boot_cpu_data.x86_vendor == X86_VENDOR_INTEL ) - memflags = MEMF_bits(32); + memflags |= MEMF_bits(32); #endif - vlapic->regs_page = alloc_domheap_pages(NULL, 0, memflags); + vlapic->regs_page = alloc_domheap_page(NULL, memflags); if ( vlapic->regs_page == NULL ) { dprintk(XENLOG_ERR, "alloc vlapic regs error: %d/%d\n", diff -r db943e8d1051 xen/arch/x86/mm/hap/hap.c --- a/xen/arch/x86/mm/hap/hap.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/mm/hap/hap.c Thu Apr 03 12:25:50 2008 +0200 @@ -38,6 +38,7 @@ #include <asm/hap.h> #include <asm/paging.h> #include <asm/domain.h> +#include <asm/numa.h> #include "private.h" @@ -135,7 +136,8 @@ static struct page_info *hap_alloc_p2m_p && mfn_x(page_to_mfn(pg)) >= (1UL << (32 - PAGE_SHIFT)) ) { free_domheap_page(pg); - pg = alloc_domheap_pages(NULL, 0, MEMF_bits(32)); + pg = alloc_domheap_page(NULL, MEMF_bits(32) | + MEMF_node(domain_to_node(d))); if ( likely(pg != NULL) ) { void *p = hap_map_domain_page(page_to_mfn(pg)); @@ -199,7 +201,7 @@ hap_set_allocation(struct domain *d, uns if ( d->arch.paging.hap.total_pages < pages ) { /* Need to allocate more memory from domheap */ - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page(NULL, MEMF_node(domain_to_node(d))); if ( pg == NULL ) { HAP_PRINTK("failed to allocate hap pages.\n"); diff -r db943e8d1051 xen/arch/x86/mm/paging.c --- a/xen/arch/x86/mm/paging.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/mm/paging.c Thu Apr 03 12:25:50 2008 +0200 @@ -26,6 +26,7 @@ #include <asm/p2m.h> #include <asm/hap.h> #include <asm/guest_access.h> +#include <asm/numa.h> #include <xsm/xsm.h> #define hap_enabled(d) (is_hvm_domain(d) && (d)->arch.hvm_domain.hap_enabled) @@ -99,7 +100,8 @@ static mfn_t paging_new_log_dirty_page(s static mfn_t paging_new_log_dirty_page(struct domain *d, void **mapping_p) { mfn_t mfn; - struct page_info *page = alloc_domheap_page(NULL); + struct page_info *page = alloc_domheap_page(NULL, + MEMF_node(domain_to_node(d))); if ( unlikely(page == NULL) ) { diff -r db943e8d1051 xen/arch/x86/mm/shadow/common.c --- a/xen/arch/x86/mm/shadow/common.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/mm/shadow/common.c Thu Apr 03 12:25:50 2008 +0200 @@ -36,6 +36,7 @@ #include <asm/current.h> #include <asm/flushtlb.h> #include <asm/shadow.h> +#include <asm/numa.h> #include "private.h" @@ -1249,7 +1250,7 @@ static unsigned int sh_set_allocation(st { /* Need to allocate more memory from domheap */ sp = (struct shadow_page_info *) - alloc_domheap_pages(NULL, order, 0); + alloc_domheap_pages(NULL, order, MEMF_node(domain_to_node(d))); if ( sp == NULL ) { SHADOW_PRINTK("failed to allocate shadow pages.\n"); diff -r db943e8d1051 xen/arch/x86/x86_64/mm.c --- a/xen/arch/x86/x86_64/mm.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/arch/x86/x86_64/mm.c Thu Apr 03 12:25:50 2008 +0200 @@ -59,7 +59,7 @@ void *alloc_xen_pagetable(void) if ( !early_boot ) { - struct page_info *pg = alloc_domheap_page(NULL); + struct page_info *pg = alloc_domheap_page(NULL, 0); BUG_ON(pg == NULL); return page_to_virt(pg); } @@ -108,7 +108,7 @@ void __init paging_init(void) struct page_info *l1_pg, *l2_pg, *l3_pg; /* Create user-accessible L2 directory to map the MPT for guests. */ - if ( (l3_pg = alloc_domheap_page(NULL)) == NULL ) + if ( (l3_pg = alloc_domheap_page(NULL, 0)) == NULL ) goto nomem; l3_ro_mpt = page_to_virt(l3_pg); clear_page(l3_ro_mpt); @@ -134,7 +134,7 @@ void __init paging_init(void) 1UL << L2_PAGETABLE_SHIFT); if ( !((unsigned long)l2_ro_mpt & ~PAGE_MASK) ) { - if ( (l2_pg = alloc_domheap_page(NULL)) == NULL ) + if ( (l2_pg = alloc_domheap_page(NULL, 0)) == NULL ) goto nomem; va = RO_MPT_VIRT_START + (i << L2_PAGETABLE_SHIFT); l2_ro_mpt = page_to_virt(l2_pg); @@ -154,7 +154,7 @@ void __init paging_init(void) l4_table_offset(HIRO_COMPAT_MPT_VIRT_START)); l3_ro_mpt = l4e_to_l3e(idle_pg_table[l4_table_offset( HIRO_COMPAT_MPT_VIRT_START)]); - if ( (l2_pg = alloc_domheap_page(NULL)) == NULL ) + if ( (l2_pg = alloc_domheap_page(NULL, 0)) == NULL ) goto nomem; compat_idle_pg_table_l2 = l2_ro_mpt = page_to_virt(l2_pg); clear_page(l2_ro_mpt); diff -r db943e8d1051 xen/common/grant_table.c --- a/xen/common/grant_table.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/common/grant_table.c Thu Apr 03 12:25:50 2008 +0200 @@ -1102,7 +1102,7 @@ gnttab_transfer( struct page_info *new_page; void *sp, *dp; - new_page = alloc_domheap_pages(NULL, 0, MEMF_bits(max_bitsize)); + new_page = alloc_domheap_page(NULL, MEMF_bits(max_bitsize)); if ( new_page == NULL ) { gop.status = GNTST_address_too_big; diff -r db943e8d1051 xen/common/memory.c --- a/xen/common/memory.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/common/memory.c Thu Apr 03 12:25:50 2008 +0200 @@ -21,6 +21,7 @@ #include <xen/errno.h> #include <asm/current.h> #include <asm/hardirq.h> +#include <asm/numa.h> #include <public/memory.h> #include <xsm/xsm.h> @@ -37,19 +38,13 @@ struct memop_args { int preempted; /* Was the hypercall preempted? */ }; -static unsigned int select_local_cpu(struct domain *d) -{ - struct vcpu *v = d->vcpu[0]; - return (v ? v->processor : 0); -} - static void increase_reservation(struct memop_args *a) { struct page_info *page; unsigned long i; xen_pfn_t mfn; struct domain *d = a->domain; - unsigned int cpu = select_local_cpu(d); + unsigned int node = domain_to_node (d); if ( !guest_handle_is_null(a->extent_list) && !guest_handle_okay(a->extent_list, a->nr_extents) ) @@ -67,7 +62,8 @@ static void increase_reservation(struct goto out; } - page = __alloc_domheap_pages(d, cpu, a->extent_order, a->memflags); + page = alloc_domheap_pages ( + d, a->extent_order, a->memflags | MEMF_node(node)); if ( unlikely(page == NULL) ) { gdprintk(XENLOG_INFO, "Could not allocate order=%d extent: " @@ -96,7 +92,7 @@ static void populate_physmap(struct memo unsigned long i, j; xen_pfn_t gpfn, mfn; struct domain *d = a->domain; - unsigned int cpu = select_local_cpu(d); + unsigned int node = domain_to_node(d); if ( !guest_handle_okay(a->extent_list, a->nr_extents) ) return; @@ -116,7 +112,8 @@ static void populate_physmap(struct memo if ( unlikely(__copy_from_guest_offset(&gpfn, a->extent_list, i, 1)) ) goto out; - page = __alloc_domheap_pages(d, cpu, a->extent_order, a->memflags); + page = alloc_domheap_pages ( + d, a->extent_order, a->memflags | MEMF_node(node)); if ( unlikely(page == NULL) ) { gdprintk(XENLOG_INFO, "Could not allocate order=%d extent: " @@ -296,7 +293,7 @@ static long memory_exchange(XEN_GUEST_HA unsigned long in_chunk_order, out_chunk_order; xen_pfn_t gpfn, gmfn, mfn; unsigned long i, j, k; - unsigned int memflags = 0, cpu; + unsigned int memflags = 0; long rc = 0; struct domain *d; struct page_info *page; @@ -352,7 +349,7 @@ static long memory_exchange(XEN_GUEST_HA memflags |= MEMF_bits(domain_clamp_alloc_bitsize( d, exch.out.address_bits ? : (BITS_PER_LONG+PAGE_SHIFT))); - cpu = select_local_cpu(d); + memflags |= MEMF_node (domain_to_node(d)); for ( i = (exch.nr_exchanged >> in_chunk_order); i < (exch.in.nr_extents >> in_chunk_order); @@ -401,8 +398,8 @@ static long memory_exchange(XEN_GUEST_HA /* Allocate a chunk's worth of anonymous output pages. */ for ( j = 0; j < (1UL << out_chunk_order); j++ ) { - page = __alloc_domheap_pages( - NULL, cpu, exch.out.extent_order, memflags); + page = alloc_domheap_pages( + NULL, exch.out.extent_order, memflags); if ( unlikely(page == NULL) ) { rc = -ENOMEM; diff -r db943e8d1051 xen/common/page_alloc.c --- a/xen/common/page_alloc.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/common/page_alloc.c Thu Apr 03 12:25:50 2008 +0200 @@ -36,6 +36,7 @@ #include <xen/numa.h> #include <xen/nodemask.h> #include <asm/page.h> +#include <asm/numa.h> #include <asm/flushtlb.h> /* @@ -328,14 +329,15 @@ static void init_node_heap(int node) /* Allocate 2^@order contiguous pages. */ static struct page_info *alloc_heap_pages( unsigned int zone_lo, unsigned int zone_hi, - unsigned int cpu, unsigned int order) + unsigned int node, unsigned int order) { unsigned int i, j, zone; - unsigned int node = cpu_to_node(cpu), num_nodes = num_online_nodes(); + unsigned int num_nodes = num_online_nodes(); unsigned long request = 1UL << order; cpumask_t extra_cpus_mask, mask; struct page_info *pg; + if ( node == NUMA_NO_NODE ) node = cpu_to_node(smp_processor_id()); ASSERT(node >= 0); ASSERT(node < num_nodes); ASSERT(zone_lo <= zone_hi); @@ -670,7 +672,8 @@ void *alloc_xenheap_pages(unsigned int o ASSERT(!in_irq()); - pg = alloc_heap_pages(MEMZONE_XEN, MEMZONE_XEN, smp_processor_id(), order); + pg = alloc_heap_pages(MEMZONE_XEN, MEMZONE_XEN, + cpu_to_node(smp_processor_id()), order); if ( unlikely(pg == NULL) ) goto no_memory; @@ -778,12 +781,12 @@ int assign_pages( } -struct page_info *__alloc_domheap_pages( - struct domain *d, unsigned int cpu, unsigned int order, - unsigned int memflags) +struct page_info *alloc_domheap_pages( + struct domain *d, unsigned int order, unsigned int memflags) { struct page_info *pg = NULL; unsigned int bits = memflags >> _MEMF_bits, zone_hi = NR_ZONES - 1; + unsigned int node = (((memflags >> _MEMF_node)&0xFF) - 1 ) &0xFF; ASSERT(!in_irq()); @@ -797,7 +800,7 @@ struct page_info *__alloc_domheap_pages( if ( (zone_hi + PAGE_SHIFT) >= dma_bitsize ) { - pg = alloc_heap_pages(dma_bitsize - PAGE_SHIFT, zone_hi, cpu, order); + pg = alloc_heap_pages(dma_bitsize - PAGE_SHIFT, zone_hi, node, order); /* Failure? Then check if we can fall back to the DMA pool. */ if ( unlikely(pg == NULL) && @@ -811,7 +814,7 @@ struct page_info *__alloc_domheap_pages( if ( (pg == NULL) && ((pg = alloc_heap_pages(MEMZONE_XEN + 1, zone_hi, - cpu, order)) == NULL) ) + node, order)) == NULL) ) return NULL; if ( (d != NULL) && assign_pages(d, pg, order, memflags) ) @@ -821,12 +824,6 @@ struct page_info *__alloc_domheap_pages( } return pg; -} - -struct page_info *alloc_domheap_pages( - struct domain *d, unsigned int order, unsigned int flags) -{ - return __alloc_domheap_pages(d, smp_processor_id(), order, flags); } void free_domheap_pages(struct page_info *pg, unsigned int order) diff -r db943e8d1051 xen/drivers/passthrough/vtd/iommu.c --- a/xen/drivers/passthrough/vtd/iommu.c Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/drivers/passthrough/vtd/iommu.c Thu Apr 03 12:25:50 2008 +0200 @@ -24,6 +24,7 @@ #include <xen/xmalloc.h> #include <xen/domain_page.h> #include <xen/iommu.h> +#include <asm/numa.h> #include "iommu.h" #include "dmar.h" #include "../pci-direct.h" @@ -269,7 +270,8 @@ static struct page_info *addr_to_dma_pag if ( dma_pte_addr(*pte) == 0 ) { - pg = alloc_domheap_page(NULL); + pg = alloc_domheap_page(NULL, + MEMF_node(domain_to_node(domain))); vaddr = map_domain_page(page_to_mfn(pg)); if ( !vaddr ) { diff -r db943e8d1051 xen/include/asm-x86/numa.h --- a/xen/include/asm-x86/numa.h Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/include/asm-x86/numa.h Thu Apr 03 12:25:50 2008 +0200 @@ -4,11 +4,16 @@ #include <xen/cpumask.h> #define NODES_SHIFT 6 +#define NUMA_NO_NODE 0xff extern unsigned char cpu_to_node[]; extern cpumask_t node_to_cpumask[]; #define cpu_to_node(cpu) (cpu_to_node[cpu]) +#define domain_to_node(domain) ((domain!=NULL && domain->vcpu[0]!=NULL)?\ + cpu_to_node[domain->vcpu[0]->processor]:\ + NUMA_NO_NODE) +#define vcpu_to_node(vcpu) (cpu_to_node[v->processor]) #define parent_node(node) (node) #define node_to_first_cpu(node) (__ffs(node_to_cpumask[node])) #define node_to_cpumask(node) (node_to_cpumask[node]) @@ -73,6 +78,5 @@ static inline __attribute__((pure)) int #define clear_node_cpumask(cpu) do {} while (0) #endif -#define NUMA_NO_NODE 0xff #endif diff -r db943e8d1051 xen/include/xen/mm.h --- a/xen/include/xen/mm.h Tue Apr 01 10:09:33 2008 +0100 +++ b/xen/include/xen/mm.h Thu Apr 03 12:25:50 2008 +0200 @@ -54,14 +54,11 @@ void init_domheap_pages(paddr_t ps, padd void init_domheap_pages(paddr_t ps, paddr_t pe); struct page_info *alloc_domheap_pages( struct domain *d, unsigned int order, unsigned int memflags); -struct page_info *__alloc_domheap_pages( - struct domain *d, unsigned int cpu, unsigned int order, - unsigned int memflags); void free_domheap_pages(struct page_info *pg, unsigned int order); unsigned long avail_domheap_pages_region( unsigned int node, unsigned int min_width, unsigned int max_width); unsigned long avail_domheap_pages(void); -#define alloc_domheap_page(d) (alloc_domheap_pages(d,0,0)) +#define alloc_domheap_page(d,f) (alloc_domheap_pages(d,0,f)) #define free_domheap_page(p) (free_domheap_pages(p,0)) void scrub_heap_pages(void); @@ -75,6 +72,8 @@ int assign_pages( /* memflags: */ #define _MEMF_no_refcount 0 #define MEMF_no_refcount (1U<<_MEMF_no_refcount) +#define _MEMF_node 8 +#define MEMF_node(n) ((((n)+1)&0xff)<<_MEMF_node) #define _MEMF_bits 24 #define MEMF_bits(n) ((n)<<_MEMF_bits) [-- Attachment #4: Type: text/plain, Size: 138 bytes --] _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-04-03 10:39 ` Andre Przywara @ 2008-04-03 10:58 ` Keir Fraser 2008-04-03 13:57 ` Andre Przywara 0 siblings, 1 reply; 14+ messages in thread From: Keir Fraser @ 2008-04-03 10:58 UTC (permalink / raw) To: Andre Przywara; +Cc: xen-devel, Edwin.Zhai On 3/4/08 11:39, "Andre Przywara" <andre.przywara@amd.com> wrote: > By the way, can we solve the DMA_BITSIZE problem (your mail from 28th > Feb) with this? If no node is specified, use the current behaviour of > preferring non DMA zones, else stick to the given node. > If you agree, I will implement this. I don't think that gets us what we want. The fact is we specify NUMA node on nearly 100% of allocations (either explicitly via MEMF_node() or via passing a non-NULL domain pointer). So you would *always* prefer local DMA pages over remote non-DMA pages. That's not necessarily better than the current policy. My point in my email of Feb 28th was that we should set dma_bitsize 'appropriately' (well, according to a slightly arbitrary policy :-) so that *some* DMA memory is set aside and only used to satisfy allocations which cannot be satisfied by a remote, while *some* memory is always made available on every node for local allocations. Does that make sense? >> NUMA_NO_NODE probably needs to be pulled out of asm-x86/numa.h and made the >> official arch-neutral way to specify 'don't care' for numa nodes. > Is this really needed? I provided memflags=0 is all don't care cases, > this should work and is more compatible. But beware that this silently > assumes in page_alloc.c#alloc_domheap_pages that NUMA_NO_NODE is 0xFF, > otherwise this trick will not work. Yes it is needed if your patch is to work across all architectures, not just x86! Your current patch is broken in this respect because you quite unnecessarily define domain_to_node() and vcpu_to_node() in asm/numa.h rather than xen/numa.h. Please address architectural portability and re-send the patch. Apart from that I think it's just about ready to go in. Thanks, Keir > Attached again a diff against my last version and the full patch (for > some reason a missing bracket slipped through my last one, sorry for that). > > This is only quick-tested (booted and created a guest on each node). ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-04-03 10:58 ` Keir Fraser @ 2008-04-03 13:57 ` Andre Przywara 2008-04-03 14:49 ` Keir Fraser ` (2 more replies) 0 siblings, 3 replies; 14+ messages in thread From: Andre Przywara @ 2008-04-03 13:57 UTC (permalink / raw) To: Keir Fraser; +Cc: xen-devel [-- Attachment #1: Type: text/plain, Size: 1644 bytes --] Keir Fraser wrote: >>> NUMA_NO_NODE probably needs to be pulled out of asm-x86/numa.h and made the >>> official arch-neutral way to specify 'don't care' for numa nodes. >> Is this really needed? I provided memflags=0 is all don't care cases, >> this should work and is more compatible. But beware that this silently >> assumes in page_alloc.c#alloc_domheap_pages that NUMA_NO_NODE is 0xFF, >> otherwise this trick will not work. > > Yes it is needed if your patch is to work across all architectures, not just > x86! Your current patch is broken in this respect because you quite > unnecessarily define domain_to_node() and vcpu_to_node() in asm/numa.h > rather than xen/numa.h. Right you are. I fixed this below. While playing around with the headers, I realized that numa.h is not needed at all in common/page_alloc.c and common/sysctl.c. Can you confirm this? Or is this needed for the other architectures? Signed-off-by: Andre Przywara <andre.przywara@amd.com> > Please address architectural portability and re-send the patch. Apart from > that I think it's just about ready to go in. I am just trying to install an IA64 cross compiler, but it seems this will take some time... Regards, Andre. -- Andre Przywara AMD-Operating System Research Center (OSRC), Dresden, Germany Tel: +49 351 277-84917 ----to satisfy European Law for business letters: AMD Saxony Limited Liability Company & Co. KG, Wilschdorfer Landstr. 101, 01109 Dresden, Germany Register Court Dresden: HRA 4896, General Partner authorized to represent: AMD Saxony LLC (Wilmington, Delaware, US) General Manager of AMD Saxony LLC: Dr. Hans-R. Deppe, Thomas McCoy [-- Attachment #2: alloc_domheap_node_xen_numah.patch --] [-- Type: text/plain, Size: 4410 bytes --] diff -r bb6eac21369e xen/arch/x86/domain.c --- a/xen/arch/x86/domain.c Thu Apr 03 12:24:57 2008 +0200 +++ b/xen/arch/x86/domain.c Thu Apr 03 15:34:52 2008 +0200 @@ -46,7 +46,7 @@ #include <asm/debugreg.h> #include <asm/msr.h> #include <asm/nmi.h> -#include <asm/numa.h> +#include <xen/numa.h> #include <xen/iommu.h> #ifdef CONFIG_COMPAT #include <compat/vcpu.h> diff -r bb6eac21369e xen/arch/x86/hvm/stdvga.c --- a/xen/arch/x86/hvm/stdvga.c Thu Apr 03 12:24:57 2008 +0200 +++ b/xen/arch/x86/hvm/stdvga.c Thu Apr 03 15:34:52 2008 +0200 @@ -32,7 +32,7 @@ #include <xen/sched.h> #include <xen/domain_page.h> #include <asm/hvm/support.h> -#include <asm/numa.h> +#include <xen/numa.h> #define PAT(x) (x) static const uint32_t mask16[16] = { diff -r bb6eac21369e xen/arch/x86/hvm/vlapic.c --- a/xen/arch/x86/hvm/vlapic.c Thu Apr 03 12:24:57 2008 +0200 +++ b/xen/arch/x86/hvm/vlapic.c Thu Apr 03 15:34:52 2008 +0200 @@ -33,7 +33,7 @@ #include <xen/sched.h> #include <asm/current.h> #include <asm/hvm/vmx/vmx.h> -#include <asm/numa.h> +#include <xen/numa.h> #include <public/hvm/ioreq.h> #include <public/hvm/params.h> diff -r bb6eac21369e xen/arch/x86/mm/hap/hap.c --- a/xen/arch/x86/mm/hap/hap.c Thu Apr 03 12:24:57 2008 +0200 +++ b/xen/arch/x86/mm/hap/hap.c Thu Apr 03 15:34:52 2008 +0200 @@ -38,7 +38,7 @@ #include <asm/hap.h> #include <asm/paging.h> #include <asm/domain.h> -#include <asm/numa.h> +#include <xen/numa.h> #include "private.h" diff -r bb6eac21369e xen/arch/x86/mm/paging.c --- a/xen/arch/x86/mm/paging.c Thu Apr 03 12:24:57 2008 +0200 +++ b/xen/arch/x86/mm/paging.c Thu Apr 03 15:34:52 2008 +0200 @@ -26,7 +26,7 @@ #include <asm/p2m.h> #include <asm/hap.h> #include <asm/guest_access.h> -#include <asm/numa.h> +#include <xen/numa.h> #include <xsm/xsm.h> #define hap_enabled(d) (is_hvm_domain(d) && (d)->arch.hvm_domain.hap_enabled) diff -r bb6eac21369e xen/arch/x86/mm/shadow/common.c --- a/xen/arch/x86/mm/shadow/common.c Thu Apr 03 12:24:57 2008 +0200 +++ b/xen/arch/x86/mm/shadow/common.c Thu Apr 03 15:34:52 2008 +0200 @@ -36,7 +36,7 @@ #include <asm/current.h> #include <asm/flushtlb.h> #include <asm/shadow.h> -#include <asm/numa.h> +#include <xen/numa.h> #include "private.h" diff -r bb6eac21369e xen/common/memory.c --- a/xen/common/memory.c Thu Apr 03 12:24:57 2008 +0200 +++ b/xen/common/memory.c Thu Apr 03 15:34:52 2008 +0200 @@ -21,7 +21,7 @@ #include <xen/errno.h> #include <asm/current.h> #include <asm/hardirq.h> -#include <asm/numa.h> +#include <xen/numa.h> #include <public/memory.h> #include <xsm/xsm.h> diff -r bb6eac21369e xen/drivers/passthrough/vtd/iommu.c --- a/xen/drivers/passthrough/vtd/iommu.c Thu Apr 03 12:24:57 2008 +0200 +++ b/xen/drivers/passthrough/vtd/iommu.c Thu Apr 03 15:34:52 2008 +0200 @@ -24,7 +24,7 @@ #include <xen/xmalloc.h> #include <xen/domain_page.h> #include <xen/iommu.h> -#include <asm/numa.h> +#include <xen/numa.h> #include "iommu.h" #include "dmar.h" #include "../pci-direct.h" diff -r bb6eac21369e xen/include/asm-x86/numa.h --- a/xen/include/asm-x86/numa.h Thu Apr 03 12:24:57 2008 +0200 +++ b/xen/include/asm-x86/numa.h Thu Apr 03 15:34:52 2008 +0200 @@ -4,16 +4,11 @@ #include <xen/cpumask.h> #define NODES_SHIFT 6 -#define NUMA_NO_NODE 0xff extern unsigned char cpu_to_node[]; extern cpumask_t node_to_cpumask[]; #define cpu_to_node(cpu) (cpu_to_node[cpu]) -#define domain_to_node(domain) ((domain!=NULL && domain->vcpu[0]!=NULL)?\ - cpu_to_node[domain->vcpu[0]->processor]:\ - NUMA_NO_NODE) -#define vcpu_to_node(vcpu) (cpu_to_node[v->processor]) #define parent_node(node) (node) #define node_to_first_cpu(node) (__ffs(node_to_cpumask[node])) #define node_to_cpumask(node) (node_to_cpumask[node]) diff -r bb6eac21369e xen/include/xen/numa.h --- a/xen/include/xen/numa.h Thu Apr 03 12:24:57 2008 +0200 +++ b/xen/include/xen/numa.h Thu Apr 03 15:34:52 2008 +0200 @@ -8,6 +8,15 @@ #define NODES_SHIFT 0 #endif +#define NUMA_NO_NODE 0xFF + #define MAX_NUMNODES (1 << NODES_SHIFT) +#define domain_to_node(domain) (((domain)!=NULL && (domain)->vcpu[0]!=NULL)?\ + cpu_to_node[(domain)->vcpu[0]->processor]:\ + NUMA_NO_NODE) + +#define vcpu_to_node(v) (cpu_to_node[(v)->processor]) + + #endif /* _XEN_NUMA_H */ [-- Attachment #3: Type: text/plain, Size: 138 bytes --] _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-04-03 13:57 ` Andre Przywara @ 2008-04-03 14:49 ` Keir Fraser 2008-04-04 8:37 ` Isaku Yamahata 2008-04-04 13:22 ` Aron Griffis 2 siblings, 0 replies; 14+ messages in thread From: Keir Fraser @ 2008-04-03 14:49 UTC (permalink / raw) To: Andre Przywara; +Cc: xen-devel On 3/4/08 14:57, "Andre Przywara" <andre.przywara@amd.com> wrote: >> Yes it is needed if your patch is to work across all architectures, not just >> x86! Your current patch is broken in this respect because you quite >> unnecessarily define domain_to_node() and vcpu_to_node() in asm/numa.h >> rather than xen/numa.h. > Right you are. I fixed this below. While playing around with the > headers, I realized that numa.h is not needed at all in > common/page_alloc.c and common/sysctl.c. Can you confirm this? Or is > this needed for the other architectures? It's polite to directly include stuff you use and not depend on transitive header inclusion, as that can be fragile. > Signed-off-by: Andre Przywara <andre.przywara@amd.com> > > >> Please address architectural portability and re-send the patch. Apart from >> that I think it's just about ready to go in. > I am just trying to install an IA64 cross compiler, but it seems this > will take some time... The ia64 guys will thank you for it, even if I don't require it! -- Keir ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-04-03 13:57 ` Andre Przywara 2008-04-03 14:49 ` Keir Fraser @ 2008-04-04 8:37 ` Isaku Yamahata 2008-04-04 13:22 ` Aron Griffis 2 siblings, 0 replies; 14+ messages in thread From: Isaku Yamahata @ 2008-04-04 8:37 UTC (permalink / raw) To: Andre Przywara; +Cc: xen-devel On Thu, Apr 03, 2008 at 03:57:08PM +0200, Andre Przywara wrote: > I am just trying to install an IA64 cross compiler, but it seems this > will take some time... The following page would be helpfull. (prebuild packages for Fedora8 are also available.) http://wiki.xensource.com/xenwiki/CrossCompiling If your system is debian, packages are available from http://www.emdebian.org/index.html -- yamahata ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-04-03 13:57 ` Andre Przywara 2008-04-03 14:49 ` Keir Fraser 2008-04-04 8:37 ` Isaku Yamahata @ 2008-04-04 13:22 ` Aron Griffis 2 siblings, 0 replies; 14+ messages in thread From: Aron Griffis @ 2008-04-04 13:22 UTC (permalink / raw) To: Andre Przywara; +Cc: xen-devel, Keir Fraser Hi Andre, Andre Przywara wrote: [Thu Apr 03 2008, 09:57:08AM EDT] > I am just trying to install an IA64 cross compiler, but it > seems this will take some time... Thank you! It would help ia64 a lot if x86 committers would compile-test for ia64. It's actually very easy. As Isaku mentioned, there are instructions at http://wiki.xensource.com/xenwiki/CrossCompiling If you have any trouble, please let me know... Thanks, Aron ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-04-02 23:21 ` Keir Fraser 2008-04-03 10:39 ` Andre Przywara @ 2008-04-07 13:25 ` Zhai, Edwin 2008-04-07 13:51 ` Keir Fraser 1 sibling, 1 reply; 14+ messages in thread From: Zhai, Edwin @ 2008-04-07 13:25 UTC (permalink / raw) To: Keir Fraser; +Cc: Andre Przywara, xen-devel, edwin.zhai On Thu, Apr 03, 2008 at 12:21:26AM +0100, Keir Fraser wrote: > > And, while I'm thinking about the interfaces, let's just stick to > alloc_domheap_page() and alloc_domheap_pages(). Let's add a flags parameter > to the former (so it matches the latter in that respect) and let's add a > MEMF_node() flag subtype (similar to MEMF_bits). Semantics will be that if > MEMF_node(node) is provided then we try to allocate memory from node; else > we try to allocate memory from a node local to specified domain; else if > domain is NULL then we ignore locality. > > Since zero is probably a valid numa nodeid we can define MEMF_node() as > something like ((((node)+1)&0xff)<<8). Then since NUMA_NO_NODE==0xff > everything works nicely: MEMF_node(NUMA_NO_NODE) is equivalent to not > specifying MEMF_node() at all, which is what we would logically expect. > NUMA_NO_NODE probably needs to be pulled out of asm-x86/numa.h and made the > official arch-neutral way to specify 'don't care' for numa nodes. Keir, It's really good idea. BTW, Do you think if need make xenheap allocation also NUMA aware in some case, e.g VMCS? Thanks, > > -- Keir > > -- best rgds, edwin ^ permalink raw reply [flat|nested] 14+ messages in thread
* Re: [RFC][PATCH] domheap optimization for NUMA 2008-04-07 13:25 ` Zhai, Edwin @ 2008-04-07 13:51 ` Keir Fraser 0 siblings, 0 replies; 14+ messages in thread From: Keir Fraser @ 2008-04-07 13:51 UTC (permalink / raw) To: Zhai, Edwin; +Cc: Andre Przywara, xen-devel On 7/4/08 14:25, "Zhai, Edwin" <edwin.zhai@intel.com> wrote: > It's really good idea. And now done and checked in. > BTW, > Do you think if need make xenheap allocation also NUMA aware in some case, e.g > VMCS? We could get rid of the xenheap allocator altogether, at least on x86/64. Or, just for this case, alloc_domheap_pages() things like VMCS and then map_domain_page_global() them. -- Keir ^ permalink raw reply [flat|nested] 14+ messages in thread
end of thread, other threads:[~2008-04-07 13:51 UTC | newest] Thread overview: 14+ messages (download: mbox.gz follow: Atom feed -- links below jump to the message on this page -- 2008-03-31 10:40 [RFC][PATCH] domheap optimization for NUMA Zhai, Edwin 2008-03-31 10:47 ` Keir Fraser 2008-04-02 13:06 ` Zhai, Edwin 2008-04-02 13:27 ` Keir Fraser 2008-04-02 22:49 ` Andre Przywara 2008-04-02 23:21 ` Keir Fraser 2008-04-03 10:39 ` Andre Przywara 2008-04-03 10:58 ` Keir Fraser 2008-04-03 13:57 ` Andre Przywara 2008-04-03 14:49 ` Keir Fraser 2008-04-04 8:37 ` Isaku Yamahata 2008-04-04 13:22 ` Aron Griffis 2008-04-07 13:25 ` Zhai, Edwin 2008-04-07 13:51 ` Keir Fraser
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.