[PATCH V4 0/3] arm64: Enable vmemmap mapping from device memory

[PATCH V4 0/3] arm64: Enable vmemmap mapping from device memory

[Anshuman Khandual]

This series enables vmemmap backing memory allocation from device memory
ranges on arm64. But before that, it enables vmemmap_populate_basepages()
and vmemmap_alloc_block_buf() to accommodate struct vmem_altmap based
alocation requests.

This series applies on 5.8-rc4.

Changes in V4:

- Dropped 'fallback' from vmemmap_alloc_block_buf() per Catalin

Changes in V3: (https://patchwork.kernel.org/project/linux-mm/list/?series04707)

- Dropped comment from free_hotplug_page_range() per Robin
- Modified comment in unmap_hotplug_range() per Robin
- Enabled altmap support in vmemmap_alloc_block_buf() per Robin

Changes in V2: (https://lkml.org/lkml/2020/3/4/475)

- Rebased on latest hot-remove series (v14) adding P4D page table support

Changes in V1: (https://lkml.org/lkml/2020/1/23/12)

- Added an WARN_ON() in unmap_hotplug_range() when altmap is
  provided without the page table backing memory being freed

Changes in RFC V2: (https://lkml.org/lkml/2019/10/21/11)

- Changed the commit message on 1/2 patch per Will
- Changed the commit message on 2/2 patch as well
- Rebased on arm64 memory hot remove series (v10)

RFC V1: (https://lkml.org/lkml/2019/6/28/32)

Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: linux-arm-kernel@lists.infradead.org
Cc: linux-ia64@vger.kernel.org
Cc: linux-riscv@lists.infradead.org
Cc: x86@kernel.org
Cc: linuxppc-dev@lists.ozlabs.org
Cc: linux-mm@kvack.org
Cc: linux-kernel@vger.kernel.org


Anshuman Khandual (3):
  mm/sparsemem: Enable vmem_altmap support in vmemmap_populate_basepages()
  mm/sparsemem: Enable vmem_altmap support in vmemmap_alloc_block_buf()
  arm64/mm: Enable vmem_altmap support for vmemmap mappings

 Documentation/vm/memory-model.rst |  2 +-
 arch/arm64/mm/mmu.c               | 58 ++++++++++++++++++++-----------
 arch/ia64/mm/discontig.c          |  2 +-
 arch/powerpc/mm/init_64.c         |  4 +--
 arch/riscv/mm/init.c              |  2 +-
 arch/x86/mm/init_64.c             | 11 +++---
 include/linux/mm.h                |  9 ++---
 mm/sparse-vmemmap.c               | 36 ++++++++++---------
 8 files changed, 72 insertions(+), 52 deletions(-)

-- 
2.20.1

[PATCH V4 1/3] mm/sparsemem: Enable vmem_altmap support in vmemmap_populate_basepages()

[Anshuman Khandual]

vmemmap_populate_basepages() is used across platforms to allocate backing
memory for vmemmap mapping. This is used as a standard default choice or
as a fallback when intended huge pages allocation fails. This just creates
entire vmemmap mapping with base pages (PAGE_SIZE).

On arm64 platforms, vmemmap_populate_basepages() is called instead of the
platform specific vmemmap_populate() when ARM64_SWAPPER_USES_SECTION_MAPS
is not enabled as in case for ARM64_16K_PAGES and ARM64_64K_PAGES configs.

At present vmemmap_populate_basepages() does not support allocating from
driver defined struct vmem_altmap while trying to create vmemmap mapping
for a device memory range. It prevents ARM64_16K_PAGES and ARM64_64K_PAGES
configs on arm64 from supporting device memory with vmemap_altmap request.

This enables vmem_altmap support in vmemmap_populate_basepages() unlocking
device memory allocation for vmemap mapping on arm64 platforms with 16K or
64K base page configs.

Each architecture should evaluate and decide on subscribing device memory
based base page allocation through vmemmap_populate_basepages(). Hence lets
keep it disabled on all archs in order to preserve the existing semantics.
A subsequent patch enables it on arm64.

Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: linux-arm-kernel@lists.infradead.org
Cc: linux-ia64@vger.kernel.org
Cc: linux-riscv@lists.infradead.org
Cc: x86@kernel.org
Cc: linux-kernel@vger.kernel.org
Tested-by: Jia He <justin.he@arm.com>
Acked-by: Will Deacon <will@kernel.org>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Anshuman Khandual <anshuman.khandual@arm.com>
---
 arch/arm64/mm/mmu.c      |  2 +-
 arch/ia64/mm/discontig.c |  2 +-
 arch/riscv/mm/init.c     |  2 +-
 arch/x86/mm/init_64.c    |  6 +++---
 include/linux/mm.h       |  5 +++--
 mm/sparse-vmemmap.c      | 16 +++++++++++-----
 6 files changed, 20 insertions(+), 13 deletions(-)

diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
index 1df25f26571d..63b74fd56cd8 100644
--- a/arch/arm64/mm/mmu.c
+++ b/arch/arm64/mm/mmu.c
@@ -1069,7 +1069,7 @@ static void free_empty_tables(unsigned long addr, unsigned long end,
 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
 		struct vmem_altmap *altmap)
 {
-	return vmemmap_populate_basepages(start, end, node);
+	return vmemmap_populate_basepages(start, end, node, NULL);
 }
 #else	/* !ARM64_SWAPPER_USES_SECTION_MAPS */
 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
diff --git a/arch/ia64/mm/discontig.c b/arch/ia64/mm/discontig.c
index dd8284bcbf16..4d23c81f96d8 100644
--- a/arch/ia64/mm/discontig.c
+++ b/arch/ia64/mm/discontig.c
@@ -656,7 +656,7 @@ void arch_refresh_nodedata(int update_node, pg_data_t *update_pgdat)
 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
 		struct vmem_altmap *altmap)
 {
-	return vmemmap_populate_basepages(start, end, node);
+	return vmemmap_populate_basepages(start, end, node, NULL);
 }
 
 void vmemmap_free(unsigned long start, unsigned long end,
diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
index f4adb3684f3d..8101170f54ac 100644
--- a/arch/riscv/mm/init.c
+++ b/arch/riscv/mm/init.c
@@ -530,6 +530,6 @@ void __init paging_init(void)
 int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
 			       struct vmem_altmap *altmap)
 {
-	return vmemmap_populate_basepages(start, end, node);
+	return vmemmap_populate_basepages(start, end, node, NULL);
 }
 #endif
diff --git a/arch/x86/mm/init_64.c b/arch/x86/mm/init_64.c
index dbae185511cd..19c0ed3271a3 100644
--- a/arch/x86/mm/init_64.c
+++ b/arch/x86/mm/init_64.c
@@ -1493,7 +1493,7 @@ static int __meminit vmemmap_populate_hugepages(unsigned long start,
 			vmemmap_verify((pte_t *)pmd, node, addr, next);
 			continue;
 		}
-		if (vmemmap_populate_basepages(addr, next, node))
+		if (vmemmap_populate_basepages(addr, next, node, NULL))
 			return -ENOMEM;
 	}
 	return 0;
@@ -1505,7 +1505,7 @@ int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
 	int err;
 
 	if (end - start < PAGES_PER_SECTION * sizeof(struct page))
-		err = vmemmap_populate_basepages(start, end, node);
+		err = vmemmap_populate_basepages(start, end, node, NULL);
 	else if (boot_cpu_has(X86_FEATURE_PSE))
 		err = vmemmap_populate_hugepages(start, end, node, altmap);
 	else if (altmap) {
@@ -1513,7 +1513,7 @@ int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
 				__func__);
 		err = -ENOMEM;
 	} else
-		err = vmemmap_populate_basepages(start, end, node);
+		err = vmemmap_populate_basepages(start, end, node, NULL);
 	if (!err)
 		sync_global_pgds(start, end - 1);
 	return err;
diff --git a/include/linux/mm.h b/include/linux/mm.h
index dc7b87310c10..e40ac543d248 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -3011,14 +3011,15 @@ pgd_t *vmemmap_pgd_populate(unsigned long addr, int node);
 p4d_t *vmemmap_p4d_populate(pgd_t *pgd, unsigned long addr, int node);
 pud_t *vmemmap_pud_populate(p4d_t *p4d, unsigned long addr, int node);
 pmd_t *vmemmap_pmd_populate(pud_t *pud, unsigned long addr, int node);
-pte_t *vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node);
+pte_t *vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
+			    struct vmem_altmap *altmap);
 void *vmemmap_alloc_block(unsigned long size, int node);
 struct vmem_altmap;
 void *vmemmap_alloc_block_buf(unsigned long size, int node);
 void *altmap_alloc_block_buf(unsigned long size, struct vmem_altmap *altmap);
 void vmemmap_verify(pte_t *, int, unsigned long, unsigned long);
 int vmemmap_populate_basepages(unsigned long start, unsigned long end,
-			       int node);
+			       int node, struct vmem_altmap *altmap);
 int vmemmap_populate(unsigned long start, unsigned long end, int node,
 		struct vmem_altmap *altmap);
 void vmemmap_populate_print_last(void);
diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
index 0db7738d76e9..ceed10dec31e 100644
--- a/mm/sparse-vmemmap.c
+++ b/mm/sparse-vmemmap.c
@@ -139,12 +139,18 @@ void __meminit vmemmap_verify(pte_t *pte, int node,
 			start, end - 1);
 }
 
-pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
+pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
+				       struct vmem_altmap *altmap)
 {
 	pte_t *pte = pte_offset_kernel(pmd, addr);
 	if (pte_none(*pte)) {
 		pte_t entry;
-		void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
+		void *p;
+
+		if (altmap)
+			p = altmap_alloc_block_buf(PAGE_SIZE, altmap);
+		else
+			p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
 		if (!p)
 			return NULL;
 		entry = pfn_pte(__pa(p) >> PAGE_SHIFT, PAGE_KERNEL);
@@ -212,8 +218,8 @@ pgd_t * __meminit vmemmap_pgd_populate(unsigned long addr, int node)
 	return pgd;
 }
 
-int __meminit vmemmap_populate_basepages(unsigned long start,
-					 unsigned long end, int node)
+int __meminit vmemmap_populate_basepages(unsigned long start, unsigned long end,
+					 int node, struct vmem_altmap *altmap)
 {
 	unsigned long addr = start;
 	pgd_t *pgd;
@@ -235,7 +241,7 @@ int __meminit vmemmap_populate_basepages(unsigned long start,
 		pmd = vmemmap_pmd_populate(pud, addr, node);
 		if (!pmd)
 			return -ENOMEM;
-		pte = vmemmap_pte_populate(pmd, addr, node);
+		pte = vmemmap_pte_populate(pmd, addr, node, altmap);
 		if (!pte)
 			return -ENOMEM;
 		vmemmap_verify(pte, node, addr, addr + PAGE_SIZE);
-- 
2.20.1

Re: [PATCH V4 1/3] mm/sparsemem: Enable vmem_altmap support in vmemmap_populate_basepages()

[David Hildenbrand]

>  	return 0;
> @@ -1505,7 +1505,7 @@ int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
>  	int err;
>  
>  	if (end - start < PAGES_PER_SECTION * sizeof(struct page))
> -		err = vmemmap_populate_basepages(start, end, node);
> +		err = vmemmap_populate_basepages(start, end, node, NULL);
>  	else if (boot_cpu_has(X86_FEATURE_PSE))
>  		err = vmemmap_populate_hugepages(start, end, node, altmap);
>  	else if (altmap) {

It's somewhat weird that we don't allocate basepages from altmap on x86
(both for sub-sections and without PSE). I wonder if we can simply
unlock that with your change. Especially, also handle the
!X86_FEATURE_PSE case below properly with an altmap.

a) all hw with PMEM has PSE - except special QEMU setups, so nobody
cared to implement. For the sub-section special case, nobody cared about
a handfull of memmap not ending up on the altmap. (but it's still wasted
system memory IIRC).

b) the pagetable overhead for small pages is not-neglectable and might
result in similar issues as solved by the switch to altmap on very huge
PMEM (with small amount of system RAM).

I guess it is due to a).

[...]

>  
> -pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
> +pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
> +				       struct vmem_altmap *altmap)
>  {
>  	pte_t *pte = pte_offset_kernel(pmd, addr);
>  	if (pte_none(*pte)) {
>  		pte_t entry;
> -		void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
> +		void *p;
> +
> +		if (altmap)
> +			p = altmap_alloc_block_buf(PAGE_SIZE, altmap);
> +		else
> +			p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
>  		if (!p)
>  			return NULL;

I was wondering if

if (altmap)
	p = altmap_alloc_block_buf(PAGE_SIZE, altmap);
if (!p)
	p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
if (!p)
	return NULL

Would make sense. But I guess this isn't really relevant in practice,
because the altmap is usually sized properly.

In general, LGTM.

-- 
Thanks,

David / dhildenb

Re: [PATCH V4 1/3] mm/sparsemem: Enable vmem_altmap support in vmemmap_populate_basepages()

[Anshuman Khandual]

On 07/06/2020 02:33 PM, David Hildenbrand wrote:
>>  	return 0;
>> @@ -1505,7 +1505,7 @@ int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
>>  	int err;
>>  
>>  	if (end - start < PAGES_PER_SECTION * sizeof(struct page))
>> -		err = vmemmap_populate_basepages(start, end, node);
>> +		err = vmemmap_populate_basepages(start, end, node, NULL);
>>  	else if (boot_cpu_has(X86_FEATURE_PSE))
>>  		err = vmemmap_populate_hugepages(start, end, node, altmap);
>>  	else if (altmap) {
> 
> It's somewhat weird that we don't allocate basepages from altmap on x86
> (both for sub-sections and without PSE). I wonder if we can simply
> unlock that with your change. Especially, also handle the
> !X86_FEATURE_PSE case below properly with an altmap.
> 
> a) all hw with PMEM has PSE - except special QEMU setups, so nobody
> cared to implement. For the sub-section special case, nobody cared about
> a handfull of memmap not ending up on the altmap. (but it's still wasted
> system memory IIRC).
> 
> b) the pagetable overhead for small pages is not-neglectable and might
> result in similar issues as solved by the switch to altmap on very huge
> PMEM (with small amount of system RAM).
> 
> I guess it is due to a).

Hmm, I assume these are some decisions that x86 platform will have to
make going forward in a subsequent patch as the third patch does for
the arm64 platform. But it is clearly beyond the scope of this patch
which never intended to change existing behavior on a given platform.

> 
> [...]
> 
>>  
>> -pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
>> +pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
>> +				       struct vmem_altmap *altmap)
>>  {
>>  	pte_t *pte = pte_offset_kernel(pmd, addr);
>>  	if (pte_none(*pte)) {
>>  		pte_t entry;
>> -		void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
>> +		void *p;
>> +
>> +		if (altmap)
>> +			p = altmap_alloc_block_buf(PAGE_SIZE, altmap);
>> +		else
>> +			p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
>>  		if (!p)
>>  			return NULL;
> 
> I was wondering if
> 
> if (altmap)
> 	p = altmap_alloc_block_buf(PAGE_SIZE, altmap);
> if (!p)
> 	p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
> if (!p)
> 	return NULL
> 
> Would make sense. But I guess this isn't really relevant in practice,
> because the altmap is usually sized properly.
> 
> In general, LGTM.

Okay, I assume that no further changes are required here.

Re: [PATCH V4 1/3] mm/sparsemem: Enable vmem_altmap support in vmemmap_populate_basepages()

[David Hildenbrand]

> Hmm, I assume these are some decisions that x86 platform will have to
> make going forward in a subsequent patch as the third patch does for
> the arm64 platform. But it is clearly beyond the scope of this patch
> which never intended to change existing behavior on a given platform.
> 

Yeah, I would be curious if my assumption is correct.

>>
>> [...]
>>
>>>  
>>> -pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node)
>>> +pte_t * __meminit vmemmap_pte_populate(pmd_t *pmd, unsigned long addr, int node,
>>> +				       struct vmem_altmap *altmap)
>>>  {
>>>  	pte_t *pte = pte_offset_kernel(pmd, addr);
>>>  	if (pte_none(*pte)) {
>>>  		pte_t entry;
>>> -		void *p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
>>> +		void *p;
>>> +
>>> +		if (altmap)
>>> +			p = altmap_alloc_block_buf(PAGE_SIZE, altmap);
>>> +		else
>>> +			p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
>>>  		if (!p)
>>>  			return NULL;
>>
>> I was wondering if
>>
>> if (altmap)
>> 	p = altmap_alloc_block_buf(PAGE_SIZE, altmap);
>> if (!p)
>> 	p = vmemmap_alloc_block_buf(PAGE_SIZE, node);
>> if (!p)
>> 	return NULL
>>
>> Would make sense. But I guess this isn't really relevant in practice,
>> because the altmap is usually sized properly.
>>
>> In general, LGTM.
> 
> Okay, I assume that no further changes are required here.
> 

Jep,

Reviewed-by: David Hildenbrand <david@redhat.com>

-- 
Thanks,

David / dhildenb

Re: [PATCH V4 0/3] arm64: Enable vmemmap mapping from device memory

[Anshuman Khandual]

On 07/06/2020 08:26 AM, Anshuman Khandual wrote:
> This series enables vmemmap backing memory allocation from device memory
> ranges on arm64. But before that, it enables vmemmap_populate_basepages()
> and vmemmap_alloc_block_buf() to accommodate struct vmem_altmap based
> alocation requests.
> 
> This series applies on 5.8-rc4.
> 
> Changes in V4:
> 
> - Dropped 'fallback' from vmemmap_alloc_block_buf() per Catalin

Hello Andrew,

This series has been a long running one :) Now that all the three patches
here have been reviewed, could you please consider this series for merging
towards 5.9-rc1. Catalin had suggested earlier [1] that it should go via
the MM tree instead, as it touches multiple architecture. Thank you.

[1] https://patchwork.kernel.org/patch/11611103/

- Anshuman