powerpc/pseries: CMO unused page hinting

Adds support for the "unused" page hint which can be used in shared memory partitions to flag pages not in use, which will then be stolen before active pages by the hypervisor when memory needs to be moved to LPARs in need of additional memory. Failure to mark pages as 'unused' makes the LPAR slower to give up unused memory to other partitions. This adds the kernel parameter 'cmo_free_hint' to disable this functionality. Signed-off-by: Brian King <brking@linux.vnet.ibm.com> Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

powerpc/pseries: CMO unused page hinting
Adds support for the "unused" page hint which can be used in shared memory partitions to flag pages not in use, which will then be stolen before active pages by the hypervisor when memory needs to be moved to LPARs in need of additional memory. Failure to mark pages as 'unused' makes the LPAR slower to give up unused memory to other partitions. This adds the kernel parameter 'cmo_free_hint' to disable this functionality. Signed-off-by: Brian King <brking@linux.vnet.ibm.com> Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Robert Jennings · Benjamin Herrenschmidt
1 parent d69a78d7da
Showing 3 changed files with 64 additions and 0 deletions Side-by-side Diff
Documentation/kernel-parameters.txt
arch/powerpc/include/asm/page.h
arch/powerpc/platforms/pseries/lpar.c
@@ -497,6 +497,13 @@
 			Also note the kernel might malfunction if you disable
 			some critical bits.
  
+	cmo_free_hint=	[PPC] Format: { yes | no }
+			Specify whether pages are marked as being inactive
+			when they are freed.  This is used in CMO environments
+			to determine OS memory pressure for page stealing by
+			a hypervisor.
+			Default: yes
+
 	code_bytes	[X86] How many bytes of object code to print
 			in an oops report.
 			Range: 0 - 8192
@@ -231,6 +231,11 @@
 		struct page *p);
 extern int page_is_ram(unsigned long pfn);
  
+#ifdef CONFIG_PPC_SMLPAR
+void arch_free_page(struct page *page, int order);
+#define HAVE_ARCH_FREE_PAGE
+#endif
+
 struct vm_area_struct;
  
 typedef struct page *pgtable_t;
@@ -609,4 +609,56 @@
 	ppc_md.flush_hash_range	= pSeries_lpar_flush_hash_range;
 	ppc_md.hpte_clear_all   = pSeries_lpar_hptab_clear;
 }
+
+#ifdef CONFIG_PPC_SMLPAR
+#define CMO_FREE_HINT_DEFAULT 1
+static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
+
+static int __init cmo_free_hint(char *str)
+{
+	char *parm;
+	parm = strstrip(str);
+
+	if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
+		printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
+		cmo_free_hint_flag = 0;
+		return 1;
+	}
+
+	cmo_free_hint_flag = 1;
+	printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
+
+	if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
+		return 1;
+
+	return 0;
+}
+
+__setup("cmo_free_hint=", cmo_free_hint);
+
+static void pSeries_set_page_state(struct page *page, int order,
+				   unsigned long state)
+{
+	int i, j;
+	unsigned long cmo_page_sz, addr;
+
+	cmo_page_sz = cmo_get_page_size();
+	addr = __pa((unsigned long)page_address(page));
+
+	for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
+		for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
+			plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
+	}
+}
+
+void arch_free_page(struct page *page, int order)
+{
+	if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
+		return;
+
+	pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
+}
+EXPORT_SYMBOL(arch_free_page);
+
+#endif
...	...	@@ -497,6 +497,13 @@
497	497	Also note the kernel might malfunction if you disable
498	498	some critical bits.
499	499
	500	+ cmo_free_hint= [PPC] Format: { yes \| no }
	501	+ Specify whether pages are marked as being inactive
	502	+ when they are freed. This is used in CMO environments
	503	+ to determine OS memory pressure for page stealing by
	504	+ a hypervisor.
	505	+ Default: yes
	506	+
500	507	code_bytes [X86] How many bytes of object code to print
501	508	in an oops report.
502	509	Range: 0 - 8192
...	...	@@ -231,6 +231,11 @@
231	231	struct page *p);
232	232	extern int page_is_ram(unsigned long pfn);
233	233
	234	+#ifdef CONFIG_PPC_SMLPAR
	235	+void arch_free_page(struct page *page, int order);
	236	+#define HAVE_ARCH_FREE_PAGE
	237	+#endif
	238	+
234	239	struct vm_area_struct;
235	240
236	241	typedef struct page *pgtable_t;
...	...	@@ -609,4 +609,56 @@
609	609	ppc_md.flush_hash_range = pSeries_lpar_flush_hash_range;
610	610	ppc_md.hpte_clear_all = pSeries_lpar_hptab_clear;
611	611	}
	612	+
	613	+#ifdef CONFIG_PPC_SMLPAR
	614	+#define CMO_FREE_HINT_DEFAULT 1
	615	+static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
	616	+
	617	+static int __init cmo_free_hint(char *str)
	618	+{
	619	+ char *parm;
	620	+ parm = strstrip(str);
	621	+
	622	+ if (strcasecmp(parm, "no") == 0 \|\| strcasecmp(parm, "off") == 0) {
	623	+ printk(KERN_INFO "cmo_free_hint: CMO free page hinting is not active.\n");
	624	+ cmo_free_hint_flag = 0;
	625	+ return 1;
	626	+ }
	627	+
	628	+ cmo_free_hint_flag = 1;
	629	+ printk(KERN_INFO "cmo_free_hint: CMO free page hinting is active.\n");
	630	+
	631	+ if (strcasecmp(parm, "yes") == 0 \|\| strcasecmp(parm, "on") == 0)
	632	+ return 1;
	633	+
	634	+ return 0;
	635	+}
	636	+
	637	+__setup("cmo_free_hint=", cmo_free_hint);
	638	+
	639	+static void pSeries_set_page_state(struct page *page, int order,
	640	+ unsigned long state)
	641	+{
	642	+ int i, j;
	643	+ unsigned long cmo_page_sz, addr;
	644	+
	645	+ cmo_page_sz = cmo_get_page_size();
	646	+ addr = __pa((unsigned long)page_address(page));
	647	+
	648	+ for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
	649	+ for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
	650	+ plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
	651	+ }
	652	+}
	653	+
	654	+void arch_free_page(struct page *page, int order)
	655	+{
	656	+ if (!cmo_free_hint_flag \|\| !firmware_has_feature(FW_FEATURE_CMO))
	657	+ return;
	658	+
	659	+ pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
	660	+}
	661	+EXPORT_SYMBOL(arch_free_page);
	662	+
	663	+#endif