Patchwork [8/10] memory-hotplug : remove page table of x86_64 architecture

login
register
mail settings
Submitter Yasuaki Ishimatsu
Date Oct. 5, 2012, 2:36 a.m.
Message ID <506E4799.30407@jp.fujitsu.com>
Download mbox | patch
Permalink /patch/189403/
State Not Applicable
Delegated to: David Miller
Headers show

Comments

Yasuaki Ishimatsu - Oct. 5, 2012, 2:36 a.m.
From: Wen Congyang <wency@cn.fujitsu.com>

For hot removing memory, we sholud remove page table about the memory.
So the patch searches a page table about the removed memory, and clear
page table.

CC: David Rientjes <rientjes@google.com>
CC: Jiang Liu <liuj97@gmail.com>
CC: Len Brown <len.brown@intel.com>
CC: Christoph Lameter <cl@linux.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
CC: Andrew Morton <akpm@linux-foundation.org>
CC: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
CC: Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com>
Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
---
 arch/x86/include/asm/pgtable_types.h |    1 
 arch/x86/mm/init_64.c                |  147 +++++++++++++++++++++++++++++++++++
 arch/x86/mm/pageattr.c               |   47 +++++------
 3 files changed, 173 insertions(+), 22 deletions(-)


--
To unsubscribe from this list: send the line "unsubscribe sparclinux" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Andi Kleen - Oct. 8, 2012, 4:37 a.m.
Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com> writes:
> +			}
> +
> +			/*
> +			 * We use 2M page, but we need to remove part of them,
> +			 * so split 2M page to 4K page.
> +			 */
> +			pte = alloc_low_page(&pte_phys);

What happens when the allocation fails?

alloc_low_page seems to be buggy there too, it would __pa a NULL 
pointer.

> +		if (pud_large(*pud)) {
> +			if ((addr & ~PUD_MASK) == 0 && next <= end) {
> +				set_pud(pud, __pud(0));
> +				pages++;
> +				continue;
> +			}
> +
> +			/*
> +			 * We use 1G page, but we need to remove part of them,
> +			 * so split 1G page to 2M page.
> +			 */
> +			pmd = alloc_low_page(&pmd_phys);

Same here

> +			__split_large_page((pte_t *)pud, addr, (pte_t *)pmd);
> +
> +			spin_lock(&init_mm.page_table_lock);
> +			pud_populate(&init_mm, pud, __va(pmd_phys));
> +			spin_unlock(&init_mm.page_table_lock);
> +		}
> +
> +		pmd = map_low_page(pmd_offset(pud, 0));
> +		phys_pmd_remove(pmd, addr, end);
> +		unmap_low_page(pmd);
> +		__flush_tlb_all();
> +	}
> +	__flush_tlb_all();

This doesn't flush the other CPUs doesn't it?

-Andi
Wen Congyang - Oct. 8, 2012, 5:23 a.m.
At 10/08/2012 12:37 PM, Andi Kleen Wrote:
> Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com> writes:
>> +			}
>> +
>> +			/*
>> +			 * We use 2M page, but we need to remove part of them,
>> +			 * so split 2M page to 4K page.
>> +			 */
>> +			pte = alloc_low_page(&pte_phys);
> 
> What happens when the allocation fails?
> 
> alloc_low_page seems to be buggy there too, it would __pa a NULL 
> pointer.

Yes, it will cause kernek panicked in __pa() if CONFI_DEBUG_VIRTUAL is set.
Otherwise, it will return a NULL pointer. I will update this patch to deal
with NULL pointer.

> 
>> +		if (pud_large(*pud)) {
>> +			if ((addr & ~PUD_MASK) == 0 && next <= end) {
>> +				set_pud(pud, __pud(0));
>> +				pages++;
>> +				continue;
>> +			}
>> +
>> +			/*
>> +			 * We use 1G page, but we need to remove part of them,
>> +			 * so split 1G page to 2M page.
>> +			 */
>> +			pmd = alloc_low_page(&pmd_phys);
> 
> Same here
> 
>> +			__split_large_page((pte_t *)pud, addr, (pte_t *)pmd);
>> +
>> +			spin_lock(&init_mm.page_table_lock);
>> +			pud_populate(&init_mm, pud, __va(pmd_phys));
>> +			spin_unlock(&init_mm.page_table_lock);
>> +		}
>> +
>> +		pmd = map_low_page(pmd_offset(pud, 0));
>> +		phys_pmd_remove(pmd, addr, end);
>> +		unmap_low_page(pmd);
>> +		__flush_tlb_all();
>> +	}
>> +	__flush_tlb_all();
> 
> This doesn't flush the other CPUs doesn't it?

How to flush the other CPU's tlb? use on_each_cpu() to run __flush_tlb_all()
on each online cpu?

Thanks
Wen Congyang

> 
> -Andi
> 

--
To unsubscribe from this list: send the line "unsubscribe sparclinux" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Ni zhan Chen - Oct. 11, 2012, 12:35 a.m.
On 10/08/2012 01:23 PM, Wen Congyang wrote:
> At 10/08/2012 12:37 PM, Andi Kleen Wrote:
>> Yasuaki Ishimatsu <isimatu.yasuaki@jp.fujitsu.com> writes:
>>> +			}
>>> +
>>> +			/*
>>> +			 * We use 2M page, but we need to remove part of them,
>>> +			 * so split 2M page to 4K page.
>>> +			 */
>>> +			pte = alloc_low_page(&pte_phys);
>> What happens when the allocation fails?
>>
>> alloc_low_page seems to be buggy there too, it would __pa a NULL
>> pointer.
> Yes, it will cause kernek panicked in __pa() if CONFI_DEBUG_VIRTUAL is set.
> Otherwise, it will return a NULL pointer. I will update this patch to deal
> with NULL pointer.
>
>>> +		if (pud_large(*pud)) {
>>> +			if ((addr & ~PUD_MASK) == 0 && next <= end) {
>>> +				set_pud(pud, __pud(0));
>>> +				pages++;
>>> +				continue;
>>> +			}
>>> +
>>> +			/*
>>> +			 * We use 1G page, but we need to remove part of them,
>>> +			 * so split 1G page to 2M page.
>>> +			 */
>>> +			pmd = alloc_low_page(&pmd_phys);
>> Same here
>>
>>> +			__split_large_page((pte_t *)pud, addr, (pte_t *)pmd);
>>> +
>>> +			spin_lock(&init_mm.page_table_lock);
>>> +			pud_populate(&init_mm, pud, __va(pmd_phys));
>>> +			spin_unlock(&init_mm.page_table_lock);
>>> +		}
>>> +
>>> +		pmd = map_low_page(pmd_offset(pud, 0));
>>> +		phys_pmd_remove(pmd, addr, end);
>>> +		unmap_low_page(pmd);
>>> +		__flush_tlb_all();
>>> +	}
>>> +	__flush_tlb_all();

Hi Congyang,

I see you call __flush_tlb_all() every pud entry(all pmd, pte related to 
it changed) modified, then how to determine the flush frequency? why not 
every pmd entry?

Regards,
Chen

>> This doesn't flush the other CPUs doesn't it?
> How to flush the other CPU's tlb? use on_each_cpu() to run __flush_tlb_all()
> on each online cpu?
>
> Thanks
> Wen Congyang
>
>> -Andi
>>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@kvack.org.  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
>

--
To unsubscribe from this list: send the line "unsubscribe sparclinux" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

Index: linux-3.6/arch/x86/mm/init_64.c
===================================================================
--- linux-3.6.orig/arch/x86/mm/init_64.c	2012-10-04 18:30:21.171698416 +0900
+++ linux-3.6/arch/x86/mm/init_64.c	2012-10-04 18:30:27.317704652 +0900
@@ -675,6 +675,151 @@  int arch_add_memory(int nid, u64 start, 
 }
 EXPORT_SYMBOL_GPL(arch_add_memory);
 
+static void __meminit
+phys_pte_remove(pte_t *pte_page, unsigned long addr, unsigned long end)
+{
+	unsigned pages = 0;
+	int i = pte_index(addr);
+
+	pte_t *pte = pte_page + pte_index(addr);
+
+	for (; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE, pte++) {
+
+		if (addr >= end)
+			break;
+
+		if (!pte_present(*pte))
+			continue;
+
+		pages++;
+		set_pte(pte, __pte(0));
+	}
+
+	update_page_count(PG_LEVEL_4K, -pages);
+}
+
+static void __meminit
+phys_pmd_remove(pmd_t *pmd_page, unsigned long addr, unsigned long end)
+{
+	unsigned long pages = 0, next;
+	int i = pmd_index(addr);
+
+	for (; i < PTRS_PER_PMD; i++, addr = next) {
+		unsigned long pte_phys;
+		pmd_t *pmd = pmd_page + pmd_index(addr);
+		pte_t *pte;
+
+		if (addr >= end)
+			break;
+
+		next = (addr & PMD_MASK) + PMD_SIZE;
+
+		if (!pmd_present(*pmd))
+			continue;
+
+		if (pmd_large(*pmd)) {
+			if ((addr & ~PMD_MASK) == 0 && next <= end) {
+				set_pmd(pmd, __pmd(0));
+				pages++;
+				continue;
+			}
+
+			/*
+			 * We use 2M page, but we need to remove part of them,
+			 * so split 2M page to 4K page.
+			 */
+			pte = alloc_low_page(&pte_phys);
+			__split_large_page((pte_t *)pmd, addr, pte);
+
+			spin_lock(&init_mm.page_table_lock);
+			pmd_populate_kernel(&init_mm, pmd, __va(pte_phys));
+			spin_unlock(&init_mm.page_table_lock);
+		}
+
+		spin_lock(&init_mm.page_table_lock);
+		pte = map_low_page((pte_t *)pmd_page_vaddr(*pmd));
+		phys_pte_remove(pte, addr, end);
+		unmap_low_page(pte);
+		spin_unlock(&init_mm.page_table_lock);
+	}
+	update_page_count(PG_LEVEL_2M, -pages);
+}
+
+static void __meminit
+phys_pud_remove(pud_t *pud_page, unsigned long addr, unsigned long end)
+{
+	unsigned long pages = 0, next;
+	int i = pud_index(addr);
+
+	for (; i < PTRS_PER_PUD; i++, addr = next) {
+		unsigned long pmd_phys;
+		pud_t *pud = pud_page + pud_index(addr);
+		pmd_t *pmd;
+
+		if (addr >= end)
+			break;
+
+		next = (addr & PUD_MASK) + PUD_SIZE;
+
+		if (!pud_present(*pud))
+			continue;
+
+		if (pud_large(*pud)) {
+			if ((addr & ~PUD_MASK) == 0 && next <= end) {
+				set_pud(pud, __pud(0));
+				pages++;
+				continue;
+			}
+
+			/*
+			 * We use 1G page, but we need to remove part of them,
+			 * so split 1G page to 2M page.
+			 */
+			pmd = alloc_low_page(&pmd_phys);
+			__split_large_page((pte_t *)pud, addr, (pte_t *)pmd);
+
+			spin_lock(&init_mm.page_table_lock);
+			pud_populate(&init_mm, pud, __va(pmd_phys));
+			spin_unlock(&init_mm.page_table_lock);
+		}
+
+		pmd = map_low_page(pmd_offset(pud, 0));
+		phys_pmd_remove(pmd, addr, end);
+		unmap_low_page(pmd);
+		__flush_tlb_all();
+	}
+	__flush_tlb_all();
+
+	update_page_count(PG_LEVEL_1G, -pages);
+}
+
+void __meminit
+kernel_physical_mapping_remove(unsigned long start, unsigned long end)
+{
+	unsigned long next;
+
+	start = (unsigned long)__va(start);
+	end = (unsigned long)__va(end);
+
+	for (; start < end; start = next) {
+		pgd_t *pgd = pgd_offset_k(start);
+		pud_t *pud;
+
+		next = (start + PGDIR_SIZE) & PGDIR_MASK;
+		if (next > end)
+			next = end;
+
+		if (!pgd_present(*pgd))
+			continue;
+
+		pud = map_low_page((pud_t *)pgd_page_vaddr(*pgd));
+		phys_pud_remove(pud, __pa(start), __pa(end));
+		unmap_low_page(pud);
+	}
+
+	__flush_tlb_all();
+}
+
 #ifdef CONFIG_MEMORY_HOTREMOVE
 int __ref arch_remove_memory(u64 start, u64 size)
 {
@@ -687,6 +832,8 @@  int __ref arch_remove_memory(u64 start, 
 	ret = __remove_pages(zone, start_pfn, nr_pages);
 	WARN_ON_ONCE(ret);
 
+	kernel_physical_mapping_remove(start, start + size);
+
 	return ret;
 }
 #endif
Index: linux-3.6/arch/x86/include/asm/pgtable_types.h
===================================================================
--- linux-3.6.orig/arch/x86/include/asm/pgtable_types.h	2012-10-04 18:26:51.925486954 +0900
+++ linux-3.6/arch/x86/include/asm/pgtable_types.h	2012-10-04 18:30:27.322704656 +0900
@@ -334,6 +334,7 @@  static inline void update_page_count(int
  * as a pte too.
  */
 extern pte_t *lookup_address(unsigned long address, unsigned int *level);
+extern int __split_large_page(pte_t *kpte, unsigned long address, pte_t *pbase);
 
 #endif	/* !__ASSEMBLY__ */
 
Index: linux-3.6/arch/x86/mm/pageattr.c
===================================================================
--- linux-3.6.orig/arch/x86/mm/pageattr.c	2012-10-04 18:26:51.923486952 +0900
+++ linux-3.6/arch/x86/mm/pageattr.c	2012-10-04 18:30:27.328704662 +0900
@@ -501,21 +501,13 @@  out_unlock:
 	return do_split;
 }
 
-static int split_large_page(pte_t *kpte, unsigned long address)
+int __split_large_page(pte_t *kpte, unsigned long address, pte_t *pbase)
 {
 	unsigned long pfn, pfninc = 1;
 	unsigned int i, level;
-	pte_t *pbase, *tmp;
+	pte_t *tmp;
 	pgprot_t ref_prot;
-	struct page *base;
-
-	if (!debug_pagealloc)
-		spin_unlock(&cpa_lock);
-	base = alloc_pages(GFP_KERNEL | __GFP_NOTRACK, 0);
-	if (!debug_pagealloc)
-		spin_lock(&cpa_lock);
-	if (!base)
-		return -ENOMEM;
+	struct page *base = virt_to_page(pbase);
 
 	spin_lock(&pgd_lock);
 	/*
@@ -523,10 +515,11 @@  static int split_large_page(pte_t *kpte,
 	 * up for us already:
 	 */
 	tmp = lookup_address(address, &level);
-	if (tmp != kpte)
-		goto out_unlock;
+	if (tmp != kpte) {
+		spin_unlock(&pgd_lock);
+		return 1;
+	}
 
-	pbase = (pte_t *)page_address(base);
 	paravirt_alloc_pte(&init_mm, page_to_pfn(base));
 	ref_prot = pte_pgprot(pte_clrhuge(*kpte));
 	/*
@@ -579,17 +572,27 @@  static int split_large_page(pte_t *kpte,
 	 * going on.
 	 */
 	__flush_tlb_all();
+	spin_unlock(&pgd_lock);
 
-	base = NULL;
+	return 0;
+}
 
-out_unlock:
-	/*
-	 * If we dropped out via the lookup_address check under
-	 * pgd_lock then stick the page back into the pool:
-	 */
-	if (base)
+static int split_large_page(pte_t *kpte, unsigned long address)
+{
+	pte_t *pbase;
+	struct page *base;
+
+	if (!debug_pagealloc)
+		spin_unlock(&cpa_lock);
+	base = alloc_pages(GFP_KERNEL | __GFP_NOTRACK, 0);
+	if (!debug_pagealloc)
+		spin_lock(&cpa_lock);
+	if (!base)
+		return -ENOMEM;
+
+	pbase = (pte_t *)page_address(base);
+	if (__split_large_page(kpte, address, pbase))
 		__free_page(base);
-	spin_unlock(&pgd_lock);
 
 	return 0;
 }