[PATCH v2 0/8] Introduce sv48 support

[PATCH v2 0/8] Introduce sv48 support

[Alexandre Ghiti]

This patchset implements sv48 support at runtime. The kernel will try to         
boot with 4-level page table and will fallback to 3-level if the HW does not     
support it.                                                                      
                                                                                 
The biggest advantage is that we only have one kernel for 64bit, which           
is way easier to maintain.                                                       
                                                                                 
Folding the 4th level into a 3-level page table has almost no cost at            
runtime. But as mentioned Palmer, the relocatable code generated is less         
performant.                                                                      
                                                                                 
At the moment, there is no way to build a 3-level page table non-relocatable     
64bit kernel. We agreed that distributions will use this runtime configuration   
anyway, but Palmer proposed to introduce a new Kconfig, which I will do later    
as sv48 support was asked for 5.8.                                               
                                                                                 
Finally, the user can now ask for sv39 explicitly by using the device-tree       
which will reduce memory footprint and reduce the number of memory accesses      
in case of TLB miss.

Changes in v2:
  * Move variable declarations to pgtable.h in patch 5/7 as suggested by Anup
  * Restore mmu-type properties in patch 6 as suggested by Anup
  * Fix unused variable in patch 5 that was used in patch 6
  * Fix SPARSEMEM build (patch 2 was modified so I dropped the Reviewed-by)
  * Applied various Reviewed-by

Alexandre Ghiti (8):
  riscv: Get rid of compile time logic with MAX_EARLY_MAPPING_SIZE
  riscv: Allow to dynamically define VA_BITS
  riscv: Simplify MAXPHYSMEM config
  riscv: Prepare ptdump for vm layout dynamic addresses
  riscv: Implement sv48 support
  riscv: Allow user to downgrade to sv39 when hw supports sv48
  riscv: Use pgtable_l4_enabled to output mmu type in cpuinfo
  riscv: Explicit comment about user virtual address space size

 arch/riscv/Kconfig                  |  34 ++---
 arch/riscv/include/asm/csr.h        |   3 +-
 arch/riscv/include/asm/fixmap.h     |   1 +
 arch/riscv/include/asm/page.h       |  15 +++
 arch/riscv/include/asm/pgalloc.h    |  36 ++++++
 arch/riscv/include/asm/pgtable-64.h |  97 +++++++++++++-
 arch/riscv/include/asm/pgtable.h    |  31 ++++-
 arch/riscv/include/asm/sparsemem.h  |   6 +-
 arch/riscv/kernel/cpu.c             |  23 ++--
 arch/riscv/kernel/head.S            |   3 +-
 arch/riscv/mm/context.c             |   2 +-
 arch/riscv/mm/init.c                | 194 ++++++++++++++++++++++++----
 arch/riscv/mm/ptdump.c              |  49 +++++--
 13 files changed, 412 insertions(+), 82 deletions(-)

-- 
2.20.1

[PATCH v2 1/8] riscv: Get rid of compile time logic with MAX_EARLY_MAPPING_SIZE

[Alexandre Ghiti]

There is no need to compare at compile time MAX_EARLY_MAPPING_SIZE value
with PGDIR_SIZE since MAX_EARLY_MAPPING_SIZE is set to 128MB which is less
than PGDIR_SIZE that is equal to 1GB: that allows to simplify early_pmd
definition.

Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
Reviewed-by: Anup Patel <anup@brainfault.org>
Reviewed-by: Palmer Dabbelt <palmerdabbelt@google.com>
---
 arch/riscv/mm/init.c | 16 ++++------------
 1 file changed, 4 insertions(+), 12 deletions(-)

diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
index e63ea5b6b6cf..80fd692b72d5 100644
--- a/arch/riscv/mm/init.c
+++ b/arch/riscv/mm/init.c
@@ -256,13 +256,7 @@ static void __init create_pte_mapping(pte_t *ptep,
 
 pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
 pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
-
-#if MAX_EARLY_MAPPING_SIZE < PGDIR_SIZE
-#define NUM_EARLY_PMDS		1UL
-#else
-#define NUM_EARLY_PMDS		(1UL + MAX_EARLY_MAPPING_SIZE / PGDIR_SIZE)
-#endif
-pmd_t early_pmd[PTRS_PER_PMD * NUM_EARLY_PMDS] __initdata __aligned(PAGE_SIZE);
+pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
 
 static pmd_t *__init get_pmd_virt(phys_addr_t pa)
 {
@@ -276,14 +270,12 @@ static pmd_t *__init get_pmd_virt(phys_addr_t pa)
 
 static phys_addr_t __init alloc_pmd(uintptr_t va)
 {
-	uintptr_t pmd_num;
-
 	if (mmu_enabled)
 		return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
 
-	pmd_num = (va - kernel_virt_addr) >> PGDIR_SHIFT;
-	BUG_ON(pmd_num >= NUM_EARLY_PMDS);
-	return (uintptr_t)&early_pmd[pmd_num * PTRS_PER_PMD];
+	BUG_ON((va - kernel_virt_addr) >> PGDIR_SHIFT);
+
+	return (uintptr_t)early_pmd;
 }
 
 static void __init create_pmd_mapping(pmd_t *pmdp,
-- 
2.20.1

[PATCH v2 2/8] riscv: Allow to dynamically define VA_BITS

[Alexandre Ghiti]

With 4-level page table folding at runtime, we don't know at compile time
the size of the virtual address space so we must set VA_BITS dynamically
so that sparsemem reserves the right amount of memory for struct pages.

Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
---
 arch/riscv/Kconfig                 | 10 ----------
 arch/riscv/include/asm/pgtable.h   | 11 +++++++++--
 arch/riscv/include/asm/sparsemem.h |  6 +++++-
 3 files changed, 14 insertions(+), 13 deletions(-)

diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
index 93127d5913fe..64b25a90d60f 100644
--- a/arch/riscv/Kconfig
+++ b/arch/riscv/Kconfig
@@ -101,16 +101,6 @@ config ZONE_DMA32
 	bool
 	default y if 64BIT
 
-config VA_BITS
-	int
-	default 32 if 32BIT
-	default 39 if 64BIT
-
-config PA_BITS
-	int
-	default 34 if 32BIT
-	default 56 if 64BIT
-
 config PAGE_OFFSET
 	hex
 	default 0xC0000000 if 32BIT && MAXPHYSMEM_2GB
diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h
index 94ef3b49dfb6..ec9694624f3c 100644
--- a/arch/riscv/include/asm/pgtable.h
+++ b/arch/riscv/include/asm/pgtable.h
@@ -40,8 +40,14 @@
  * struct pages to map half the virtual address space. Then
  * position vmemmap directly below the VMALLOC region.
  */
+#ifdef CONFIG_64BIT
+#define VA_BITS		39
+#else
+#define VA_BITS		32
+#endif
+
 #define VMEMMAP_SHIFT \
-	(CONFIG_VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
+	(VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT)
 #define VMEMMAP_SIZE	BIT(VMEMMAP_SHIFT)
 #define VMEMMAP_END	(VMALLOC_START - 1)
 #define VMEMMAP_START	(VMALLOC_START - VMEMMAP_SIZE)
@@ -80,6 +86,7 @@
 #endif /* CONFIG_64BIT */
 
 #ifdef CONFIG_MMU
+
 /* Number of entries in the page global directory */
 #define PTRS_PER_PGD    (PAGE_SIZE / sizeof(pgd_t))
 /* Number of entries in the page table */
@@ -466,7 +473,7 @@ static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
  * and give the kernel the other (upper) half.
  */
 #ifdef CONFIG_64BIT
-#define KERN_VIRT_START	(-(BIT(CONFIG_VA_BITS)) + TASK_SIZE)
+#define KERN_VIRT_START	(-(BIT(VA_BITS)) + TASK_SIZE)
 #else
 #define KERN_VIRT_START	FIXADDR_START
 #endif
diff --git a/arch/riscv/include/asm/sparsemem.h b/arch/riscv/include/asm/sparsemem.h
index 45a7018a8118..63acaecc3374 100644
--- a/arch/riscv/include/asm/sparsemem.h
+++ b/arch/riscv/include/asm/sparsemem.h
@@ -4,7 +4,11 @@
 #define _ASM_RISCV_SPARSEMEM_H
 
 #ifdef CONFIG_SPARSEMEM
-#define MAX_PHYSMEM_BITS	CONFIG_PA_BITS
+#ifdef CONFIG_64BIT
+#define MAX_PHYSMEM_BITS	56
+#else
+#define MAX_PHYSMEM_BITS	34
+#endif /* CONFIG_64BIT */
 #define SECTION_SIZE_BITS	27
 #endif /* CONFIG_SPARSEMEM */
 
-- 
2.20.1

[PATCH v2 3/8] riscv: Simplify MAXPHYSMEM config

[Alexandre Ghiti]

Either the user specifies maximum physical memory size of 2GB or the
user lives with the system constraint which is 1/4th of maximum
addressable memory in Sv39 MMU mode (i.e. 128GB) for now.

Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
Reviewed-by: Anup Patel <anup@brainfault.org>
---
 arch/riscv/Kconfig | 20 ++++++--------------
 1 file changed, 6 insertions(+), 14 deletions(-)

diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
index 64b25a90d60f..e167f16131f4 100644
--- a/arch/riscv/Kconfig
+++ b/arch/riscv/Kconfig
@@ -106,7 +106,7 @@ config PAGE_OFFSET
 	default 0xC0000000 if 32BIT && MAXPHYSMEM_2GB
 	default 0x80000000 if 64BIT && !MMU
 	default 0xffffffff80000000 if 64BIT && MAXPHYSMEM_2GB
-	default 0xffffffe000000000 if 64BIT && MAXPHYSMEM_128GB
+	default 0xffffffe000000000 if 64BIT && !MAXPHYSMEM_2GB
 
 config ARCH_FLATMEM_ENABLE
 	def_bool y
@@ -223,19 +223,11 @@ config MODULE_SECTIONS
 	bool
 	select HAVE_MOD_ARCH_SPECIFIC
 
-choice
-	prompt "Maximum Physical Memory"
-	default MAXPHYSMEM_2GB if 32BIT
-	default MAXPHYSMEM_2GB if 64BIT && CMODEL_MEDLOW
-	default MAXPHYSMEM_128GB if 64BIT && CMODEL_MEDANY
-
-	config MAXPHYSMEM_2GB
-		bool "2GiB"
-	config MAXPHYSMEM_128GB
-		depends on 64BIT && CMODEL_MEDANY
-		bool "128GiB"
-endchoice
-
+config MAXPHYSMEM_2GB
+	bool "Maximum Physical Memory 2GiB"
+	default y if 32BIT
+	default y if 64BIT && CMODEL_MEDLOW
+	default n
 
 config SMP
 	bool "Symmetric Multi-Processing"
-- 
2.20.1

[PATCH v2 4/8] riscv: Prepare ptdump for vm layout dynamic addresses

[Alexandre Ghiti]

This is a preparatory patch for sv48 support that will introduce
dynamic PAGE_OFFSET.

Dynamic PAGE_OFFSET implies that all zones (vmalloc, vmemmap, fixaddr...)
whose addresses depend on PAGE_OFFSET become dynamic and can't be used
to statically initialize the array used by ptdump to identify the
different zones of the vm layout.

Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
Reviewed-by: Anup Patel <anup@brainfault.org>
---
 arch/riscv/mm/ptdump.c | 49 ++++++++++++++++++++++++++++++++++--------
 1 file changed, 40 insertions(+), 9 deletions(-)

diff --git a/arch/riscv/mm/ptdump.c b/arch/riscv/mm/ptdump.c
index 7eab76a93106..7d9386a7f5c2 100644
--- a/arch/riscv/mm/ptdump.c
+++ b/arch/riscv/mm/ptdump.c
@@ -49,22 +49,41 @@ struct addr_marker {
 	const char *name;
 };
 
+enum address_markers_idx {
+#ifdef CONFIG_KASAN
+	KASAN_SHADOW_START_NR,
+	KASAN_SHADOW_END_NR,
+#endif
+	FIXMAP_START_NR,
+	FIXMAP_END_NR,
+	PCI_IO_START_NR,
+	PCI_IO_END_NR,
+#ifdef CONFIG_SPARSEMEM_VMEMMAP
+	VMEMMAP_START_NR,
+	VMEMMAP_END_NR,
+#endif
+	VMALLOC_START_NR,
+	VMALLOC_END_NR,
+	PAGE_OFFSET_NR,
+	END_OF_SPACE_NR
+};
+
 static struct addr_marker address_markers[] = {
 #ifdef CONFIG_KASAN
 	{KASAN_SHADOW_START,	"Kasan shadow start"},
 	{KASAN_SHADOW_END,	"Kasan shadow end"},
 #endif
-	{FIXADDR_START,		"Fixmap start"},
-	{FIXADDR_TOP,		"Fixmap end"},
-	{PCI_IO_START,		"PCI I/O start"},
-	{PCI_IO_END,		"PCI I/O end"},
+	{0,			"Fixmap start"},
+	{0,			"Fixmap end"},
+	{0,			"PCI I/O start"},
+	{0,			"PCI I/O end"},
 #ifdef CONFIG_SPARSEMEM_VMEMMAP
-	{VMEMMAP_START,		"vmemmap start"},
-	{VMEMMAP_END,		"vmemmap end"},
+	{0,			"vmemmap start"},
+	{0,			"vmemmap end"},
 #endif
-	{VMALLOC_START,		"vmalloc() area"},
-	{VMALLOC_END,		"vmalloc() end"},
-	{PAGE_OFFSET,		"Linear mapping"},
+	{0,			"vmalloc() area"},
+	{0,			"vmalloc() end"},
+	{0,			"Linear mapping"},
 	{-1, NULL},
 };
 
@@ -304,6 +323,18 @@ static int ptdump_init(void)
 {
 	unsigned int i, j;
 
+	address_markers[FIXMAP_START_NR].start_address = FIXADDR_START;
+	address_markers[FIXMAP_END_NR].start_address = FIXADDR_TOP;
+	address_markers[PCI_IO_START_NR].start_address = PCI_IO_START;
+	address_markers[PCI_IO_END_NR].start_address = PCI_IO_END;
+#ifdef CONFIG_SPARSEMEM_VMEMMAP
+	address_markers[VMEMMAP_START_NR].start_address = VMEMMAP_START;
+	address_markers[VMEMMAP_END_NR].start_address = VMEMMAP_END;
+#endif
+	address_markers[VMALLOC_START_NR].start_address = VMALLOC_START;
+	address_markers[VMALLOC_END_NR].start_address = VMALLOC_END;
+	address_markers[PAGE_OFFSET_NR].start_address = PAGE_OFFSET;
+
 	for (i = 0; i < ARRAY_SIZE(pg_level); i++)
 		for (j = 0; j < ARRAY_SIZE(pte_bits); j++)
 			pg_level[i].mask |= pte_bits[j].mask;
-- 
2.20.1

[PATCH v2 5/8] riscv: Implement sv48 support

[Alexandre Ghiti]

By adding a new 4th level of page table, give the possibility to 64bit
kernel to address 2^48 bytes of virtual address: in practice, that roughly
offers ~160TB of virtual address space to userspace and allows up to 64TB
of physical memory.

If the underlying hardware does not support sv48, we will automatically
fallback to a standard 3-level page table by folding the new PUD level into
PGDIR level. In order to detect HW capabilities at runtime, we
use SATP feature that ignores writes with an unsupported mode.

Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
Reviewed-by: Anup Patel <anup@brainfault.org>
---
 arch/riscv/Kconfig                  |   6 +-
 arch/riscv/include/asm/csr.h        |   3 +-
 arch/riscv/include/asm/fixmap.h     |   1 +
 arch/riscv/include/asm/page.h       |  15 +++
 arch/riscv/include/asm/pgalloc.h    |  36 +++++++
 arch/riscv/include/asm/pgtable-64.h |  97 ++++++++++++++++-
 arch/riscv/include/asm/pgtable.h    |  10 +-
 arch/riscv/kernel/head.S            |   3 +-
 arch/riscv/mm/context.c             |   2 +-
 arch/riscv/mm/init.c                | 158 +++++++++++++++++++++++++---
 10 files changed, 307 insertions(+), 24 deletions(-)

diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
index e167f16131f4..3f73f60e9732 100644
--- a/arch/riscv/Kconfig
+++ b/arch/riscv/Kconfig
@@ -68,6 +68,7 @@ config RISCV
 	select ARCH_HAS_GCOV_PROFILE_ALL
 	select HAVE_COPY_THREAD_TLS
 	select HAVE_ARCH_KASAN if MMU && 64BIT
+	select RELOCATABLE if 64BIT
 
 config ARCH_MMAP_RND_BITS_MIN
 	default 18 if 64BIT
@@ -106,7 +107,7 @@ config PAGE_OFFSET
 	default 0xC0000000 if 32BIT && MAXPHYSMEM_2GB
 	default 0x80000000 if 64BIT && !MMU
 	default 0xffffffff80000000 if 64BIT && MAXPHYSMEM_2GB
-	default 0xffffffe000000000 if 64BIT && !MAXPHYSMEM_2GB
+	default 0xffffc00000000000 if 64BIT && !MAXPHYSMEM_2GB
 
 config ARCH_FLATMEM_ENABLE
 	def_bool y
@@ -155,8 +156,11 @@ config GENERIC_HWEIGHT
 config FIX_EARLYCON_MEM
 	def_bool MMU
 
+# On a 64BIT relocatable kernel, the 4-level page table is at runtime folded
+# on a 3-level page table when sv48 is not supported.
 config PGTABLE_LEVELS
 	int
+	default 4 if 64BIT && RELOCATABLE
 	default 3 if 64BIT
 	default 2
 
diff --git a/arch/riscv/include/asm/csr.h b/arch/riscv/include/asm/csr.h
index cec462e198ce..d41536c3f8d4 100644
--- a/arch/riscv/include/asm/csr.h
+++ b/arch/riscv/include/asm/csr.h
@@ -40,11 +40,10 @@
 #ifndef CONFIG_64BIT
 #define SATP_PPN	_AC(0x003FFFFF, UL)
 #define SATP_MODE_32	_AC(0x80000000, UL)
-#define SATP_MODE	SATP_MODE_32
 #else
 #define SATP_PPN	_AC(0x00000FFFFFFFFFFF, UL)
 #define SATP_MODE_39	_AC(0x8000000000000000, UL)
-#define SATP_MODE	SATP_MODE_39
+#define SATP_MODE_48	_AC(0x9000000000000000, UL)
 #endif
 
 /* Exception cause high bit - is an interrupt if set */
diff --git a/arch/riscv/include/asm/fixmap.h b/arch/riscv/include/asm/fixmap.h
index 2368d49eb4ef..d891cf9c73c5 100644
--- a/arch/riscv/include/asm/fixmap.h
+++ b/arch/riscv/include/asm/fixmap.h
@@ -27,6 +27,7 @@ enum fixed_addresses {
 	FIX_FDT = FIX_FDT_END + FIX_FDT_SIZE / PAGE_SIZE - 1,
 	FIX_PTE,
 	FIX_PMD,
+	FIX_PUD,
 	FIX_TEXT_POKE1,
 	FIX_TEXT_POKE0,
 	FIX_EARLYCON_MEM_BASE,
diff --git a/arch/riscv/include/asm/page.h b/arch/riscv/include/asm/page.h
index 48bb09b6a9b7..5e77fe7f0d6d 100644
--- a/arch/riscv/include/asm/page.h
+++ b/arch/riscv/include/asm/page.h
@@ -31,7 +31,19 @@
  * When not using MMU this corresponds to the first free page in
  * physical memory (aligned on a page boundary).
  */
+#ifdef CONFIG_RELOCATABLE
+#define PAGE_OFFSET		__page_offset
+
+#ifdef CONFIG_64BIT
+/*
+ * By default, CONFIG_PAGE_OFFSET value corresponds to SV48 address space so
+ * define the PAGE_OFFSET value for SV39.
+ */
+#define PAGE_OFFSET_L3		0xffffffe000000000
+#endif /* CONFIG_64BIT */
+#else
 #define PAGE_OFFSET		_AC(CONFIG_PAGE_OFFSET, UL)
+#endif /* CONFIG_RELOCATABLE */
 
 #define KERN_VIRT_SIZE (-PAGE_OFFSET)
 
@@ -102,6 +114,9 @@ extern unsigned long pfn_base;
 extern unsigned long max_low_pfn;
 extern unsigned long min_low_pfn;
 extern unsigned long kernel_virt_addr;
+#ifdef CONFIG_RELOCATABLE
+extern unsigned long __page_offset;
+#endif
 
 #define __pa_to_va_nodebug(x)	((void *)((unsigned long) (x) + va_pa_offset))
 #define linear_mapping_va_to_pa(x)	((unsigned long)(x) - va_pa_offset)
diff --git a/arch/riscv/include/asm/pgalloc.h b/arch/riscv/include/asm/pgalloc.h
index 3f601ee8233f..540eaa5a8658 100644
--- a/arch/riscv/include/asm/pgalloc.h
+++ b/arch/riscv/include/asm/pgalloc.h
@@ -36,6 +36,42 @@ static inline void pud_populate(struct mm_struct *mm, pud_t *pud, pmd_t *pmd)
 
 	set_pud(pud, __pud((pfn << _PAGE_PFN_SHIFT) | _PAGE_TABLE));
 }
+
+static inline void p4d_populate(struct mm_struct *mm, p4d_t *p4d, pud_t *pud)
+{
+	if (pgtable_l4_enabled) {
+		unsigned long pfn = virt_to_pfn(pud);
+
+		set_p4d(p4d, __p4d((pfn << _PAGE_PFN_SHIFT) | _PAGE_TABLE));
+	}
+}
+
+static inline void p4d_populate_safe(struct mm_struct *mm, p4d_t *p4d,
+				     pud_t *pud)
+{
+	if (pgtable_l4_enabled) {
+		unsigned long pfn = virt_to_pfn(pud);
+
+		set_p4d_safe(p4d,
+			     __p4d((pfn << _PAGE_PFN_SHIFT) | _PAGE_TABLE));
+	}
+}
+
+static inline pud_t *pud_alloc_one(struct mm_struct *mm, unsigned long addr)
+{
+	if (pgtable_l4_enabled)
+		return (pud_t *)__get_free_page(
+				GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_ZERO);
+	return NULL;
+}
+
+static inline void pud_free(struct mm_struct *mm, pud_t *pud)
+{
+	if (pgtable_l4_enabled)
+		free_page((unsigned long)pud);
+}
+
+#define __pud_free_tlb(tlb, pud, addr)  pud_free((tlb)->mm, pud)
 #endif /* __PAGETABLE_PMD_FOLDED */
 
 #define pmd_pgtable(pmd)	pmd_page(pmd)
diff --git a/arch/riscv/include/asm/pgtable-64.h b/arch/riscv/include/asm/pgtable-64.h
index b15f70a1fdfa..c84c31fbf8da 100644
--- a/arch/riscv/include/asm/pgtable-64.h
+++ b/arch/riscv/include/asm/pgtable-64.h
@@ -8,16 +8,32 @@
 
 #include <linux/const.h>
 
-#define PGDIR_SHIFT     30
+extern bool pgtable_l4_enabled;
+
+#define PGDIR_SHIFT     (pgtable_l4_enabled ? 39 : 30)
 /* Size of region mapped by a page global directory */
 #define PGDIR_SIZE      (_AC(1, UL) << PGDIR_SHIFT)
 #define PGDIR_MASK      (~(PGDIR_SIZE - 1))
 
+/* pud is folded into pgd in case of 3-level page table */
+#define PUD_SHIFT	30
+#define PUD_SIZE	(_AC(1, UL) << PUD_SHIFT)
+#define PUD_MASK	(~(PUD_SIZE - 1))
+
 #define PMD_SHIFT       21
 /* Size of region mapped by a page middle directory */
 #define PMD_SIZE        (_AC(1, UL) << PMD_SHIFT)
 #define PMD_MASK        (~(PMD_SIZE - 1))
 
+/* Page Upper Directory entry */
+typedef struct {
+	unsigned long pud;
+} pud_t;
+
+#define pud_val(x)      ((x).pud)
+#define __pud(x)        ((pud_t) { (x) })
+#define PTRS_PER_PUD    (PAGE_SIZE / sizeof(pud_t))
+
 /* Page Middle Directory entry */
 typedef struct {
 	unsigned long pmd;
@@ -60,6 +76,16 @@ static inline void pud_clear(pud_t *pudp)
 	set_pud(pudp, __pud(0));
 }
 
+static inline pud_t pfn_pud(unsigned long pfn, pgprot_t prot)
+{
+	return __pud((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
+}
+
+static inline unsigned long _pud_pfn(pud_t pud)
+{
+	return pud_val(pud) >> _PAGE_PFN_SHIFT;
+}
+
 static inline unsigned long pud_page_vaddr(pud_t pud)
 {
 	return (unsigned long)pfn_to_virt(pud_val(pud) >> _PAGE_PFN_SHIFT);
@@ -70,6 +96,15 @@ static inline struct page *pud_page(pud_t pud)
 	return pfn_to_page(pud_val(pud) >> _PAGE_PFN_SHIFT);
 }
 
+#define mm_pud_folded	mm_pud_folded
+static inline bool mm_pud_folded(struct mm_struct *mm)
+{
+	if (pgtable_l4_enabled)
+		return false;
+
+	return true;
+}
+
 #define pmd_index(addr) (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1))
 
 static inline pmd_t *pmd_offset(pud_t *pud, unsigned long addr)
@@ -90,4 +125,64 @@ static inline unsigned long _pmd_pfn(pmd_t pmd)
 #define pmd_ERROR(e) \
 	pr_err("%s:%d: bad pmd %016lx.\n", __FILE__, __LINE__, pmd_val(e))
 
+#define pud_ERROR(e)	\
+	pr_err("%s:%d: bad pud %016lx.\n", __FILE__, __LINE__, pud_val(e))
+
+static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
+{
+	if (pgtable_l4_enabled)
+		*p4dp = p4d;
+	else
+		set_pud((pud_t *)p4dp, (pud_t){ p4d_val(p4d) });
+}
+
+static inline int p4d_none(p4d_t p4d)
+{
+	if (pgtable_l4_enabled)
+		return (p4d_val(p4d) == 0);
+
+	return 0;
+}
+
+static inline int p4d_present(p4d_t p4d)
+{
+	if (pgtable_l4_enabled)
+		return (p4d_val(p4d) & _PAGE_PRESENT);
+
+	return 1;
+}
+
+static inline int p4d_bad(p4d_t p4d)
+{
+	if (pgtable_l4_enabled)
+		return !p4d_present(p4d);
+
+	return 0;
+}
+
+static inline void p4d_clear(p4d_t *p4d)
+{
+	if (pgtable_l4_enabled)
+		set_p4d(p4d, __p4d(0));
+}
+
+static inline unsigned long p4d_page_vaddr(p4d_t p4d)
+{
+	if (pgtable_l4_enabled)
+		return (unsigned long)pfn_to_virt(
+				p4d_val(p4d) >> _PAGE_PFN_SHIFT);
+
+	return pud_page_vaddr((pud_t) { p4d_val(p4d) });
+}
+
+#define pud_index(addr) (((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
+
+static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
+{
+	if (pgtable_l4_enabled)
+		return (pud_t *)p4d_page_vaddr(*p4d) + pud_index(address);
+
+	return (pud_t *)p4d;
+}
+
 #endif /* _ASM_RISCV_PGTABLE_64_H */
diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h
index ec9694624f3c..b4b532525fee 100644
--- a/arch/riscv/include/asm/pgtable.h
+++ b/arch/riscv/include/asm/pgtable.h
@@ -20,12 +20,14 @@
  * the kernel.
  */
 #define KERNEL_VIRT_ADDR	(VMALLOC_END - SZ_2G + 1)
-#define KERNEL_LINK_ADDR	KERNEL_VIRT_ADDR
+#define KERNEL_LINK_ADDR	(VMALLOC_LINK_END - SZ_2G + 1)
 
 #define VMALLOC_SIZE     (KERN_VIRT_SIZE >> 1)
 #define VMALLOC_END      (PAGE_OFFSET - 1)
 #define VMALLOC_START    (PAGE_OFFSET - VMALLOC_SIZE)
 
+#define VMALLOC_LINK_END	(_AC(CONFIG_PAGE_OFFSET, UL) - 1)
+
 #define BPF_JIT_REGION_SIZE	(SZ_128M)
 #define BPF_JIT_REGION_START	PFN_ALIGN((unsigned long)&_end)
 #define BPF_JIT_REGION_END	(BPF_JIT_REGION_START + BPF_JIT_REGION_SIZE)
@@ -41,7 +43,7 @@
  * position vmemmap directly below the VMALLOC region.
  */
 #ifdef CONFIG_64BIT
-#define VA_BITS		39
+#define VA_BITS		(pgtable_l4_enabled ? 48 : 39)
 #else
 #define VA_BITS		32
 #endif
@@ -73,8 +75,7 @@
 
 #ifndef __ASSEMBLY__
 
-/* Page Upper Directory not used in RISC-V */
-#include <asm-generic/pgtable-nopud.h>
+#include <asm-generic/pgtable-nop4d.h>
 #include <asm/page.h>
 #include <asm/tlbflush.h>
 #include <linux/mm_types.h>
@@ -505,6 +506,7 @@ static inline void __kernel_map_pages(struct page *page, int numpages, int enabl
 
 extern char _start[];
 extern void *dtb_early_va;
+extern u64 satp_mode;
 void setup_bootmem(void);
 void paging_init(void);
 
diff --git a/arch/riscv/kernel/head.S b/arch/riscv/kernel/head.S
index 8f5bb7731327..0632c4834c68 100644
--- a/arch/riscv/kernel/head.S
+++ b/arch/riscv/kernel/head.S
@@ -62,7 +62,8 @@ relocate:
 
 	/* Compute satp for kernel page tables, but don't load it yet */
 	srl a2, a0, PAGE_SHIFT
-	li a1, SATP_MODE
+	la a1, satp_mode
+	REG_L a1, 0(a1)
 	or a2, a2, a1
 
 	/*
diff --git a/arch/riscv/mm/context.c b/arch/riscv/mm/context.c
index 613ec81a8979..0393ca1b4416 100644
--- a/arch/riscv/mm/context.c
+++ b/arch/riscv/mm/context.c
@@ -59,7 +59,7 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,
 	cpumask_set_cpu(cpu, mm_cpumask(next));
 
 #ifdef CONFIG_MMU
-	csr_write(CSR_SATP, virt_to_pfn(next->pgd) | SATP_MODE);
+	csr_write(CSR_SATP, virt_to_pfn(next->pgd) | satp_mode);
 	local_flush_tlb_all();
 #endif
 
diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
index 80fd692b72d5..a937173af13d 100644
--- a/arch/riscv/mm/init.c
+++ b/arch/riscv/mm/init.c
@@ -25,8 +25,23 @@
 
 #include "../kernel/head.h"
 
-unsigned long kernel_virt_addr = KERNEL_VIRT_ADDR;
+#ifdef CONFIG_64BIT
+u64 satp_mode = IS_ENABLED(CONFIG_MAXPHYSMEM_2GB) ?
+				SATP_MODE_39 : SATP_MODE_48;
+bool pgtable_l4_enabled = IS_ENABLED(CONFIG_MAXPHYSMEM_2GB) ? false : true;
+#else
+u64 satp_mode = SATP_MODE_32;
+bool pgtable_l4_enabled;
+#endif
+EXPORT_SYMBOL(pgtable_l4_enabled);
+EXPORT_SYMBOL(satp_mode);
+
+unsigned long kernel_virt_addr;
 EXPORT_SYMBOL(kernel_virt_addr);
+#ifdef CONFIG_RELOCATABLE
+unsigned long __page_offset = _AC(CONFIG_PAGE_OFFSET, UL);
+EXPORT_SYMBOL(__page_offset);
+#endif
 
 unsigned long empty_zero_page[PAGE_SIZE / sizeof(unsigned long)]
 							__page_aligned_bss;
@@ -254,9 +269,12 @@ static void __init create_pte_mapping(pte_t *ptep,
 
 #ifndef __PAGETABLE_PMD_FOLDED
 
+pud_t trampoline_pud[PTRS_PER_PUD] __page_aligned_bss;
 pmd_t trampoline_pmd[PTRS_PER_PMD] __page_aligned_bss;
+pud_t fixmap_pud[PTRS_PER_PUD] __page_aligned_bss;
 pmd_t fixmap_pmd[PTRS_PER_PMD] __page_aligned_bss;
 pmd_t early_pmd[PTRS_PER_PMD] __initdata __aligned(PAGE_SIZE);
+pud_t early_pud[PTRS_PER_PUD] __initdata __aligned(PAGE_SIZE);
 
 static pmd_t *__init get_pmd_virt(phys_addr_t pa)
 {
@@ -273,7 +291,8 @@ static phys_addr_t __init alloc_pmd(uintptr_t va)
 	if (mmu_enabled)
 		return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
 
-	BUG_ON((va - kernel_virt_addr) >> PGDIR_SHIFT);
+	/* Only one PMD is available for early mapping */
+	BUG_ON((va - kernel_virt_addr) >> PUD_SHIFT);
 
 	return (uintptr_t)early_pmd;
 }
@@ -305,19 +324,70 @@ static void __init create_pmd_mapping(pmd_t *pmdp,
 	create_pte_mapping(ptep, va, pa, sz, prot);
 }
 
-#define pgd_next_t		pmd_t
-#define alloc_pgd_next(__va)	alloc_pmd(__va)
-#define get_pgd_next_virt(__pa)	get_pmd_virt(__pa)
+static pud_t *__init get_pud_virt(phys_addr_t pa)
+{
+	if (mmu_enabled) {
+		clear_fixmap(FIX_PUD);
+		return (pud_t *)set_fixmap_offset(FIX_PUD, pa);
+	} else {
+		return (pud_t *)((uintptr_t)pa);
+	}
+}
+
+static phys_addr_t __init alloc_pud(uintptr_t va)
+{
+	if (mmu_enabled)
+		return memblock_phys_alloc(PAGE_SIZE, PAGE_SIZE);
+
+	/* Only one PUD is available for early mapping */
+	BUG_ON((va - kernel_virt_addr) >> PGDIR_SHIFT);
+
+	return (uintptr_t)early_pud;
+}
+
+static void __init create_pud_mapping(pud_t *pudp,
+				      uintptr_t va, phys_addr_t pa,
+				      phys_addr_t sz, pgprot_t prot)
+{
+	pmd_t *nextp;
+	phys_addr_t next_phys;
+	uintptr_t pud_index = pud_index(va);
+
+	if (sz == PUD_SIZE) {
+		if (pud_val(pudp[pud_index]) == 0)
+			pudp[pud_index] = pfn_pud(PFN_DOWN(pa), prot);
+		return;
+	}
+
+	if (pud_val(pudp[pud_index]) == 0) {
+		next_phys = alloc_pmd(va);
+		pudp[pud_index] = pfn_pud(PFN_DOWN(next_phys), PAGE_TABLE);
+		nextp = get_pmd_virt(next_phys);
+		memset(nextp, 0, PAGE_SIZE);
+	} else {
+		next_phys = PFN_PHYS(_pud_pfn(pudp[pud_index]));
+		nextp = get_pmd_virt(next_phys);
+	}
+
+	create_pmd_mapping(nextp, va, pa, sz, prot);
+}
+
+#define pgd_next_t		pud_t
+#define alloc_pgd_next(__va)	alloc_pud(__va)
+#define get_pgd_next_virt(__pa)	get_pud_virt(__pa)
 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
-	create_pmd_mapping(__nextp, __va, __pa, __sz, __prot)
-#define fixmap_pgd_next		fixmap_pmd
+	create_pud_mapping(__nextp, __va, __pa, __sz, __prot)
+#define fixmap_pgd_next		(pgtable_l4_enabled ?			\
+			(uintptr_t)fixmap_pud : (uintptr_t)fixmap_pmd)
+#define trampoline_pgd_next	(pgtable_l4_enabled ?			\
+			(uintptr_t)trampoline_pud : (uintptr_t)trampoline_pmd)
 #else
 #define pgd_next_t		pte_t
 #define alloc_pgd_next(__va)	alloc_pte(__va)
 #define get_pgd_next_virt(__pa)	get_pte_virt(__pa)
 #define create_pgd_next_mapping(__nextp, __va, __pa, __sz, __prot)	\
 	create_pte_mapping(__nextp, __va, __pa, __sz, __prot)
-#define fixmap_pgd_next		fixmap_pte
+#define fixmap_pgd_next		((uintptr_t)fixmap_pte)
 #endif
 
 static void __init create_pgd_mapping(pgd_t *pgdp,
@@ -328,6 +398,13 @@ static void __init create_pgd_mapping(pgd_t *pgdp,
 	phys_addr_t next_phys;
 	uintptr_t pgd_index = pgd_index(va);
 
+#ifndef __PAGETABLE_PMD_FOLDED
+	if (!pgtable_l4_enabled) {
+		create_pud_mapping((pud_t *)pgdp, va, pa, sz, prot);
+		return;
+	}
+#endif
+
 	if (sz == PGDIR_SIZE) {
 		if (pgd_val(pgdp[pgd_index]) == 0)
 			pgdp[pgd_index] = pfn_pgd(PFN_DOWN(pa), prot);
@@ -419,6 +496,46 @@ void __init relocate_kernel(uintptr_t load_pa)
 	}
 }
 
+#if defined(CONFIG_64BIT) && !defined(CONFIG_MAXPHYSMEM_2GB)
+void disable_pgtable_l4(void)
+{
+	pgtable_l4_enabled = false;
+	__page_offset = PAGE_OFFSET_L3;
+	satp_mode = SATP_MODE_39;
+}
+
+/* There is a simple way to determine if 4-level is supported by the
+ * underlying hardware: establish 1:1 mapping in 4-level page table mode
+ * then read SATP to see if the configuration was taken into account
+ * meaning sv48 is supported.
+ */
+asmlinkage __init void set_satp_mode(uintptr_t load_pa)
+{
+	u64 identity_satp, hw_satp;
+
+	create_pgd_mapping(early_pg_dir, load_pa, (uintptr_t)early_pud,
+			   PGDIR_SIZE, PAGE_TABLE);
+	create_pud_mapping(early_pud, load_pa, (uintptr_t)early_pmd,
+			   PUD_SIZE, PAGE_TABLE);
+	create_pmd_mapping(early_pmd, load_pa, load_pa,
+			   PMD_SIZE, PAGE_KERNEL_EXEC);
+
+	identity_satp = PFN_DOWN((uintptr_t)&early_pg_dir) | satp_mode;
+	local_flush_tlb_all();
+	csr_write(CSR_SATP, identity_satp);
+
+	hw_satp = csr_read(CSR_SATP);
+	csr_write(CSR_SATP, 0ULL);
+	local_flush_tlb_all();
+
+	if (hw_satp != identity_satp)
+		disable_pgtable_l4();
+
+	memset(early_pg_dir, 0, PAGE_SIZE);
+	memset(early_pud, 0, PAGE_SIZE);
+	memset(early_pmd, 0, PAGE_SIZE);
+}
+#endif
 #endif
 
 static uintptr_t load_pa, load_sz;
@@ -442,9 +559,14 @@ asmlinkage void __init setup_vm(uintptr_t dtb_pa)
 	load_pa = (uintptr_t)(&_start);
 	load_sz = (uintptr_t)(&_end) - load_pa;
 
+#if defined(CONFIG_64BIT) && !defined(CONFIG_MAXPHYSMEM_2GB)
+	set_satp_mode(load_pa);
+#endif
+
+	kernel_virt_addr = KERNEL_VIRT_ADDR;
+
 	va_pa_offset = PAGE_OFFSET - load_pa;
 	va_kernel_pa_offset = kernel_virt_addr - load_pa;
-
 	pfn_base = PFN_DOWN(load_pa);
 
 #ifdef CONFIG_RELOCATABLE
@@ -473,15 +595,22 @@ asmlinkage void __init setup_vm(uintptr_t dtb_pa)
 
 	/* Setup early PGD for fixmap */
 	create_pgd_mapping(early_pg_dir, FIXADDR_START,
-			   (uintptr_t)fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
+			   fixmap_pgd_next, PGDIR_SIZE, PAGE_TABLE);
 
 #ifndef __PAGETABLE_PMD_FOLDED
-	/* Setup fixmap PMD */
+	/* Setup fixmap PUD and PMD */
+	if (pgtable_l4_enabled)
+		create_pud_mapping(fixmap_pud, FIXADDR_START,
+			   (uintptr_t)fixmap_pmd, PUD_SIZE, PAGE_TABLE);
 	create_pmd_mapping(fixmap_pmd, FIXADDR_START,
 			   (uintptr_t)fixmap_pte, PMD_SIZE, PAGE_TABLE);
+
 	/* Setup trampoline PGD and PMD */
 	create_pgd_mapping(trampoline_pg_dir, kernel_virt_addr,
-			   (uintptr_t)trampoline_pmd, PGDIR_SIZE, PAGE_TABLE);
+			   trampoline_pgd_next, PGDIR_SIZE, PAGE_TABLE);
+	if (pgtable_l4_enabled)
+		create_pud_mapping(trampoline_pud, kernel_virt_addr,
+			   (uintptr_t)trampoline_pmd, PUD_SIZE, PAGE_TABLE);
 	create_pmd_mapping(trampoline_pmd, kernel_virt_addr,
 			   load_pa, PMD_SIZE, PAGE_KERNEL_EXEC);
 #else
@@ -558,12 +687,13 @@ static void __init setup_vm_final(void)
 
 	vm_area_add_early(&vm_kernel);
 
-	/* Clear fixmap PTE and PMD mappings */
+	/* Clear fixmap page table mappings */
 	clear_fixmap(FIX_PTE);
 	clear_fixmap(FIX_PMD);
+	clear_fixmap(FIX_PUD);
 
 	/* Move to swapper page table */
-	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | SATP_MODE);
+	csr_write(CSR_SATP, PFN_DOWN(__pa_symbol(swapper_pg_dir)) | satp_mode);
 	local_flush_tlb_all();
 }
 
-- 
2.20.1

[PATCH v2 6/8] riscv: Allow user to downgrade to sv39 when hw supports sv48

[Alexandre Ghiti]

This is made possible by using the mmu-type property of the cpu node of
the device tree.

By default, the kernel will boot with 4-level page table if the hw supports
it but it can be interesting for the user to select 3-level page table as
it is less memory consuming and faster since it requires less memory
accesses in case of a TLB miss.

Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
Reviewed-by: Anup Patel <anup@brainfault.org>
---
 arch/riscv/mm/init.c | 26 ++++++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)

diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
index a937173af13d..4064639b24e4 100644
--- a/arch/riscv/mm/init.c
+++ b/arch/riscv/mm/init.c
@@ -509,10 +509,32 @@ void disable_pgtable_l4(void)
  * then read SATP to see if the configuration was taken into account
  * meaning sv48 is supported.
  */
-asmlinkage __init void set_satp_mode(uintptr_t load_pa)
+asmlinkage __init void set_satp_mode(uintptr_t load_pa, uintptr_t dtb_pa)
 {
 	u64 identity_satp, hw_satp;
+	int cpus_node;
 
+	/* 1/ Check if the user asked for sv39 explicitly in the device tree */
+	cpus_node = fdt_path_offset((void *)dtb_pa, "/cpus");
+	if (cpus_node >= 0) {
+		int node;
+
+		fdt_for_each_subnode(node, (void *)dtb_pa, cpus_node) {
+			const char *mmu_type = fdt_getprop((void *)dtb_pa, node,
+							"mmu-type", NULL);
+			if (!mmu_type)
+				continue;
+
+			if (!strcmp(mmu_type, "riscv,sv39")) {
+				disable_pgtable_l4();
+				return;
+			}
+
+			break;
+		}
+	}
+
+	/* 2/ Determine if the HW supports sv48: if not, fallback to sv39 */
 	create_pgd_mapping(early_pg_dir, load_pa, (uintptr_t)early_pud,
 			   PGDIR_SIZE, PAGE_TABLE);
 	create_pud_mapping(early_pud, load_pa, (uintptr_t)early_pmd,
@@ -560,7 +582,7 @@ asmlinkage void __init setup_vm(uintptr_t dtb_pa)
 	load_sz = (uintptr_t)(&_end) - load_pa;
 
 #if defined(CONFIG_64BIT) && !defined(CONFIG_MAXPHYSMEM_2GB)
-	set_satp_mode(load_pa);
+	set_satp_mode(load_pa, dtb_pa);
 #endif
 
 	kernel_virt_addr = KERNEL_VIRT_ADDR;
-- 
2.20.1

[PATCH v2 7/8] riscv: Use pgtable_l4_enabled to output mmu type in cpuinfo

[Alexandre Ghiti]

Now that the mmu type is determined at runtime using SATP
characteristic, use the global variable pgtable_l4_enabled to output
mmu type of the processor through /proc/cpuinfo instead of relying on
device tree infos.

Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
Reviewed-by: Anup Patel <anup@brainfault.org>
Reviewed-by: Palmer Dabbelt <palmerdabbelt@google.com>
---
 arch/riscv/include/asm/pgtable.h |  1 +
 arch/riscv/kernel/cpu.c          | 23 ++++++++++++-----------
 2 files changed, 13 insertions(+), 11 deletions(-)

diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h
index b4b532525fee..cb8c6863266b 100644
--- a/arch/riscv/include/asm/pgtable.h
+++ b/arch/riscv/include/asm/pgtable.h
@@ -507,6 +507,7 @@ static inline void __kernel_map_pages(struct page *page, int numpages, int enabl
 extern char _start[];
 extern void *dtb_early_va;
 extern u64 satp_mode;
+extern bool pgtable_l4_enabled;
 void setup_bootmem(void);
 void paging_init(void);
 
diff --git a/arch/riscv/kernel/cpu.c b/arch/riscv/kernel/cpu.c
index 40a3c442ac5f..4661b6669edb 100644
--- a/arch/riscv/kernel/cpu.c
+++ b/arch/riscv/kernel/cpu.c
@@ -7,6 +7,7 @@
 #include <linux/seq_file.h>
 #include <linux/of.h>
 #include <asm/smp.h>
+#include <asm/pgtable.h>
 
 /*
  * Returns the hart ID of the given device tree node, or -ENODEV if the node
@@ -54,18 +55,19 @@ static void print_isa(struct seq_file *f, const char *isa)
 	seq_puts(f, "\n");
 }
 
-static void print_mmu(struct seq_file *f, const char *mmu_type)
+static void print_mmu(struct seq_file *f)
 {
+	char sv_type[16];
+
 #if defined(CONFIG_32BIT)
-	if (strcmp(mmu_type, "riscv,sv32") != 0)
-		return;
+	strncpy(sv_type, "sv32", 5);
 #elif defined(CONFIG_64BIT)
-	if (strcmp(mmu_type, "riscv,sv39") != 0 &&
-	    strcmp(mmu_type, "riscv,sv48") != 0)
-		return;
+	if (pgtable_l4_enabled)
+		strncpy(sv_type, "sv48", 5);
+	else
+		strncpy(sv_type, "sv39", 5);
 #endif
-
-	seq_printf(f, "mmu\t\t: %s\n", mmu_type+6);
+	seq_printf(f, "mmu\t\t: %s\n", sv_type);
 }
 
 static void *c_start(struct seq_file *m, loff_t *pos)
@@ -90,14 +92,13 @@ static int c_show(struct seq_file *m, void *v)
 {
 	unsigned long cpu_id = (unsigned long)v - 1;
 	struct device_node *node = of_get_cpu_node(cpu_id, NULL);
-	const char *compat, *isa, *mmu;
+	const char *compat, *isa;
 
 	seq_printf(m, "processor\t: %lu\n", cpu_id);
 	seq_printf(m, "hart\t\t: %lu\n", cpuid_to_hartid_map(cpu_id));
 	if (!of_property_read_string(node, "riscv,isa", &isa))
 		print_isa(m, isa);
-	if (!of_property_read_string(node, "mmu-type", &mmu))
-		print_mmu(m, mmu);
+	print_mmu(m);
 	if (!of_property_read_string(node, "compatible", &compat)
 	    && strcmp(compat, "riscv"))
 		seq_printf(m, "uarch\t\t: %s\n", compat);
-- 
2.20.1

[PATCH v2 8/8] riscv: Explicit comment about user virtual address space size

[Alexandre Ghiti]

Define precisely the size of the user accessible virtual space size
for sv32/39/48 mmu types and explain why the whole virtual address
space is split into 2 equal chunks between kernel and user space.

Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
Reviewed-by: Anup Patel <anup@brainfault.org>
Reviewed-by: Palmer Dabbelt <palmerdabbelt@google.com>
---
 arch/riscv/include/asm/pgtable.h | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h
index cb8c6863266b..86bbc2ed1cdd 100644
--- a/arch/riscv/include/asm/pgtable.h
+++ b/arch/riscv/include/asm/pgtable.h
@@ -480,8 +480,15 @@ static inline int ptep_clear_flush_young(struct vm_area_struct *vma,
 #endif
 
 /*
- * Task size is 0x4000000000 for RV64 or 0x9fc00000 for RV32.
- * Note that PGDIR_SIZE must evenly divide TASK_SIZE.
+ * Task size is:
+ * -     0x9fc00000 (~2.5GB) for RV32.
+ * -   0x4000000000 ( 256GB) for RV64 using SV39 mmu
+ * - 0x800000000000 ( 128TB) for RV64 using SV48 mmu
+ *
+ * Note that PGDIR_SIZE must evenly divide TASK_SIZE since "RISC-V
+ * Instruction Set Manual Volume II: Privileged Architecture" states that
+ * "load and store effective addresses, which are 64bits, must have bits
+ * 63–48 all equal to bit 47, or else a page-fault exception will occur."
  */
 #ifdef CONFIG_64BIT
 #define TASK_SIZE (PGDIR_SIZE * PTRS_PER_PGD / 2)
-- 
2.20.1

Re: [PATCH v2 0/8] Introduce sv48 support

[Alex Ghiti]

Hi Palmer,

Le 6/3/20 à 4:10 AM, Alexandre Ghiti a écrit :
> This patchset implements sv48 support at runtime. The kernel will try to
> boot with 4-level page table and will fallback to 3-level if the HW does not
> support it.
>                                                                                   
> The biggest advantage is that we only have one kernel for 64bit, which
> is way easier to maintain.
>                                                                                   
> Folding the 4th level into a 3-level page table has almost no cost at
> runtime. But as mentioned Palmer, the relocatable code generated is less
> performant.
>                                                                                   
> At the moment, there is no way to build a 3-level page table non-relocatable
> 64bit kernel. We agreed that distributions will use this runtime configuration
> anyway, but Palmer proposed to introduce a new Kconfig, which I will do later
> as sv48 support was asked for 5.8.
>                                                                                   
> Finally, the user can now ask for sv39 explicitly by using the device-tree
> which will reduce memory footprint and reduce the number of memory accesses
> in case of TLB miss.
>
> Changes in v2:
>    * Move variable declarations to pgtable.h in patch 5/7 as suggested by Anup
>    * Restore mmu-type properties in patch 6 as suggested by Anup
>    * Fix unused variable in patch 5 that was used in patch 6
>    * Fix SPARSEMEM build (patch 2 was modified so I dropped the Reviewed-by)
>    * Applied various Reviewed-by
>
> Alexandre Ghiti (8):
>    riscv: Get rid of compile time logic with MAX_EARLY_MAPPING_SIZE
>    riscv: Allow to dynamically define VA_BITS
>    riscv: Simplify MAXPHYSMEM config
>    riscv: Prepare ptdump for vm layout dynamic addresses
>    riscv: Implement sv48 support
>    riscv: Allow user to downgrade to sv39 when hw supports sv48
>    riscv: Use pgtable_l4_enabled to output mmu type in cpuinfo
>    riscv: Explicit comment about user virtual address space size
>
>   arch/riscv/Kconfig                  |  34 ++---
>   arch/riscv/include/asm/csr.h        |   3 +-
>   arch/riscv/include/asm/fixmap.h     |   1 +
>   arch/riscv/include/asm/page.h       |  15 +++
>   arch/riscv/include/asm/pgalloc.h    |  36 ++++++
>   arch/riscv/include/asm/pgtable-64.h |  97 +++++++++++++-
>   arch/riscv/include/asm/pgtable.h    |  31 ++++-
>   arch/riscv/include/asm/sparsemem.h  |   6 +-
>   arch/riscv/kernel/cpu.c             |  23 ++--
>   arch/riscv/kernel/head.S            |   3 +-
>   arch/riscv/mm/context.c             |   2 +-
>   arch/riscv/mm/init.c                | 194 ++++++++++++++++++++++++----
>   arch/riscv/mm/ptdump.c              |  49 +++++--
>   13 files changed, 412 insertions(+), 82 deletions(-)
>

Do you any remark regarding this patchset and the others ?

Thanks,

Alex

Re: [PATCH v2 0/8] Introduce sv48 support

[Palmer Dabbelt]

On Wed, 03 Jun 2020 01:10:56 PDT (-0700), alex@ghiti.fr wrote:
> This patchset implements sv48 support at runtime. The kernel will try to
> boot with 4-level page table and will fallback to 3-level if the HW does not
> support it.
>
> The biggest advantage is that we only have one kernel for 64bit, which
> is way easier to maintain.
>
> Folding the 4th level into a 3-level page table has almost no cost at
> runtime. But as mentioned Palmer, the relocatable code generated is less
> performant.
>
> At the moment, there is no way to build a 3-level page table non-relocatable
> 64bit kernel. We agreed that distributions will use this runtime configuration
> anyway, but Palmer proposed to introduce a new Kconfig, which I will do later
> as sv48 support was asked for 5.8.
>
> Finally, the user can now ask for sv39 explicitly by using the device-tree
> which will reduce memory footprint and reduce the number of memory accesses
> in case of TLB miss.
>
> Changes in v2:
>   * Move variable declarations to pgtable.h in patch 5/7 as suggested by Anup
>   * Restore mmu-type properties in patch 6 as suggested by Anup
>   * Fix unused variable in patch 5 that was used in patch 6
>   * Fix SPARSEMEM build (patch 2 was modified so I dropped the Reviewed-by)
>   * Applied various Reviewed-by
>
> Alexandre Ghiti (8):
>   riscv: Get rid of compile time logic with MAX_EARLY_MAPPING_SIZE
>   riscv: Allow to dynamically define VA_BITS
>   riscv: Simplify MAXPHYSMEM config
>   riscv: Prepare ptdump for vm layout dynamic addresses
>   riscv: Implement sv48 support
>   riscv: Allow user to downgrade to sv39 when hw supports sv48
>   riscv: Use pgtable_l4_enabled to output mmu type in cpuinfo
>   riscv: Explicit comment about user virtual address space size
>
>  arch/riscv/Kconfig                  |  34 ++---
>  arch/riscv/include/asm/csr.h        |   3 +-
>  arch/riscv/include/asm/fixmap.h     |   1 +
>  arch/riscv/include/asm/page.h       |  15 +++
>  arch/riscv/include/asm/pgalloc.h    |  36 ++++++
>  arch/riscv/include/asm/pgtable-64.h |  97 +++++++++++++-
>  arch/riscv/include/asm/pgtable.h    |  31 ++++-
>  arch/riscv/include/asm/sparsemem.h  |   6 +-
>  arch/riscv/kernel/cpu.c             |  23 ++--
>  arch/riscv/kernel/head.S            |   3 +-
>  arch/riscv/mm/context.c             |   2 +-
>  arch/riscv/mm/init.c                | 194 ++++++++++++++++++++++++----
>  arch/riscv/mm/ptdump.c              |  49 +++++--
>  13 files changed, 412 insertions(+), 82 deletions(-)

Sorry I haven't had time to look at your patch sets for a bit, with the merge
window everything got a bit busy.  I'm collecting the rc1 fixes now, as we have
some major issues, but with any luck I'll have time to start looking at larger
stuff for for-next next week.

Thanks!

Re: [PATCH v2 5/8] riscv: Implement sv48 support

[Nick Kossifidis]

Στις 2020-06-03 11:11, Alexandre Ghiti έγραψε:
> By adding a new 4th level of page table, give the possibility to 64bit
> kernel to address 2^48 bytes of virtual address: in practice, that 
> roughly
> offers ~160TB of virtual address space to userspace and allows up to 
> 64TB
> of physical memory.
> 
> If the underlying hardware does not support sv48, we will automatically
> fallback to a standard 3-level page table by folding the new PUD level 
> into
> PGDIR level. In order to detect HW capabilities at runtime, we
> use SATP feature that ignores writes with an unsupported mode.
> 
> Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
> Reviewed-by: Anup Patel <anup@brainfault.org>
> ---
>  arch/riscv/Kconfig                  |   6 +-
>  arch/riscv/include/asm/csr.h        |   3 +-
>  arch/riscv/include/asm/fixmap.h     |   1 +
>  arch/riscv/include/asm/page.h       |  15 +++
>  arch/riscv/include/asm/pgalloc.h    |  36 +++++++
>  arch/riscv/include/asm/pgtable-64.h |  97 ++++++++++++++++-
>  arch/riscv/include/asm/pgtable.h    |  10 +-
>  arch/riscv/kernel/head.S            |   3 +-
>  arch/riscv/mm/context.c             |   2 +-
>  arch/riscv/mm/init.c                | 158 +++++++++++++++++++++++++---
>  10 files changed, 307 insertions(+), 24 deletions(-)
> 
> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
> index e167f16131f4..3f73f60e9732 100644
> --- a/arch/riscv/Kconfig
> +++ b/arch/riscv/Kconfig
> @@ -68,6 +68,7 @@ config RISCV
>  	select ARCH_HAS_GCOV_PROFILE_ALL
>  	select HAVE_COPY_THREAD_TLS
>  	select HAVE_ARCH_KASAN if MMU && 64BIT
> +	select RELOCATABLE if 64BIT
> 
>  config ARCH_MMAP_RND_BITS_MIN
>  	default 18 if 64BIT
> @@ -106,7 +107,7 @@ config PAGE_OFFSET
>  	default 0xC0000000 if 32BIT && MAXPHYSMEM_2GB
>  	default 0x80000000 if 64BIT && !MMU
>  	default 0xffffffff80000000 if 64BIT && MAXPHYSMEM_2GB
> -	default 0xffffffe000000000 if 64BIT && !MAXPHYSMEM_2GB
> +	default 0xffffc00000000000 if 64BIT && !MAXPHYSMEM_2GB
> 
>  config ARCH_FLATMEM_ENABLE
>  	def_bool y
> @@ -155,8 +156,11 @@ config GENERIC_HWEIGHT
>  config FIX_EARLYCON_MEM
>  	def_bool MMU
> 
> +# On a 64BIT relocatable kernel, the 4-level page table is at runtime 
> folded
> +# on a 3-level page table when sv48 is not supported.
>  config PGTABLE_LEVELS
>  	int
> +	default 4 if 64BIT && RELOCATABLE
>  	default 3 if 64BIT
>  	default 2
> 
> diff --git a/arch/riscv/include/asm/csr.h 
> b/arch/riscv/include/asm/csr.h
> index cec462e198ce..d41536c3f8d4 100644
> --- a/arch/riscv/include/asm/csr.h
> +++ b/arch/riscv/include/asm/csr.h
> @@ -40,11 +40,10 @@
>  #ifndef CONFIG_64BIT
>  #define SATP_PPN	_AC(0x003FFFFF, UL)
>  #define SATP_MODE_32	_AC(0x80000000, UL)
> -#define SATP_MODE	SATP_MODE_32
>  #else
>  #define SATP_PPN	_AC(0x00000FFFFFFFFFFF, UL)
>  #define SATP_MODE_39	_AC(0x8000000000000000, UL)
> -#define SATP_MODE	SATP_MODE_39
> +#define SATP_MODE_48	_AC(0x9000000000000000, UL)
>  #endif
> 
>  /* Exception cause high bit - is an interrupt if set */
> diff --git a/arch/riscv/include/asm/fixmap.h 
> b/arch/riscv/include/asm/fixmap.h
> index 2368d49eb4ef..d891cf9c73c5 100644
> --- a/arch/riscv/include/asm/fixmap.h
> +++ b/arch/riscv/include/asm/fixmap.h
> @@ -27,6 +27,7 @@ enum fixed_addresses {
>  	FIX_FDT = FIX_FDT_END + FIX_FDT_SIZE / PAGE_SIZE - 1,
>  	FIX_PTE,
>  	FIX_PMD,
> +	FIX_PUD,
>  	FIX_TEXT_POKE1,
>  	FIX_TEXT_POKE0,
>  	FIX_EARLYCON_MEM_BASE,
> diff --git a/arch/riscv/include/asm/page.h 
> b/arch/riscv/include/asm/page.h
> index 48bb09b6a9b7..5e77fe7f0d6d 100644
> --- a/arch/riscv/include/asm/page.h
> +++ b/arch/riscv/include/asm/page.h
> @@ -31,7 +31,19 @@
>   * When not using MMU this corresponds to the first free page in
>   * physical memory (aligned on a page boundary).
>   */
> +#ifdef CONFIG_RELOCATABLE
> +#define PAGE_OFFSET		__page_offset
> +
> +#ifdef CONFIG_64BIT
> +/*
> + * By default, CONFIG_PAGE_OFFSET value corresponds to SV48 address 
> space so
> + * define the PAGE_OFFSET value for SV39.
> + */
> +#define PAGE_OFFSET_L3		0xffffffe000000000
> +#endif /* CONFIG_64BIT */
> +#else
>  #define PAGE_OFFSET		_AC(CONFIG_PAGE_OFFSET, UL)
> +#endif /* CONFIG_RELOCATABLE */
> 
>  #define KERN_VIRT_SIZE (-PAGE_OFFSET)
> 
> @@ -102,6 +114,9 @@ extern unsigned long pfn_base;
>  extern unsigned long max_low_pfn;
>  extern unsigned long min_low_pfn;
>  extern unsigned long kernel_virt_addr;
> +#ifdef CONFIG_RELOCATABLE
> +extern unsigned long __page_offset;
> +#endif
> 
>  #define __pa_to_va_nodebug(x)	((void *)((unsigned long) (x) + 
> va_pa_offset))
>  #define linear_mapping_va_to_pa(x)	((unsigned long)(x) - va_pa_offset)
> diff --git a/arch/riscv/include/asm/pgalloc.h 
> b/arch/riscv/include/asm/pgalloc.h
> index 3f601ee8233f..540eaa5a8658 100644
> --- a/arch/riscv/include/asm/pgalloc.h
> +++ b/arch/riscv/include/asm/pgalloc.h
> @@ -36,6 +36,42 @@ static inline void pud_populate(struct mm_struct
> *mm, pud_t *pud, pmd_t *pmd)
> 
>  	set_pud(pud, __pud((pfn << _PAGE_PFN_SHIFT) | _PAGE_TABLE));
>  }
> +
> +static inline void p4d_populate(struct mm_struct *mm, p4d_t *p4d, 
> pud_t *pud)
> +{
> +	if (pgtable_l4_enabled) {
> +		unsigned long pfn = virt_to_pfn(pud);
> +
> +		set_p4d(p4d, __p4d((pfn << _PAGE_PFN_SHIFT) | _PAGE_TABLE));
> +	}
> +}
> +
> +static inline void p4d_populate_safe(struct mm_struct *mm, p4d_t *p4d,
> +				     pud_t *pud)
> +{
> +	if (pgtable_l4_enabled) {
> +		unsigned long pfn = virt_to_pfn(pud);
> +
> +		set_p4d_safe(p4d,
> +			     __p4d((pfn << _PAGE_PFN_SHIFT) | _PAGE_TABLE));
> +	}
> +}
> +
> +static inline pud_t *pud_alloc_one(struct mm_struct *mm, unsigned long 
> addr)
> +{
> +	if (pgtable_l4_enabled)
> +		return (pud_t *)__get_free_page(
> +				GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_ZERO);
> +	return NULL;
> +}
> +
> +static inline void pud_free(struct mm_struct *mm, pud_t *pud)
> +{
> +	if (pgtable_l4_enabled)
> +		free_page((unsigned long)pud);
> +}
> +
> +#define __pud_free_tlb(tlb, pud, addr)  pud_free((tlb)->mm, pud)
>  #endif /* __PAGETABLE_PMD_FOLDED */
> 
>  #define pmd_pgtable(pmd)	pmd_page(pmd)
> diff --git a/arch/riscv/include/asm/pgtable-64.h
> b/arch/riscv/include/asm/pgtable-64.h
> index b15f70a1fdfa..c84c31fbf8da 100644
> --- a/arch/riscv/include/asm/pgtable-64.h
> +++ b/arch/riscv/include/asm/pgtable-64.h
> @@ -8,16 +8,32 @@
> 
>  #include <linux/const.h>
> 
> -#define PGDIR_SHIFT     30
> +extern bool pgtable_l4_enabled;
> +
> +#define PGDIR_SHIFT     (pgtable_l4_enabled ? 39 : 30)
>  /* Size of region mapped by a page global directory */
>  #define PGDIR_SIZE      (_AC(1, UL) << PGDIR_SHIFT)
>  #define PGDIR_MASK      (~(PGDIR_SIZE - 1))
> 
> +/* pud is folded into pgd in case of 3-level page table */
> +#define PUD_SHIFT	30
> +#define PUD_SIZE	(_AC(1, UL) << PUD_SHIFT)
> +#define PUD_MASK	(~(PUD_SIZE - 1))
> +
>  #define PMD_SHIFT       21
>  /* Size of region mapped by a page middle directory */
>  #define PMD_SIZE        (_AC(1, UL) << PMD_SHIFT)
>  #define PMD_MASK        (~(PMD_SIZE - 1))
> 
> +/* Page Upper Directory entry */
> +typedef struct {
> +	unsigned long pud;
> +} pud_t;
> +
> +#define pud_val(x)      ((x).pud)
> +#define __pud(x)        ((pud_t) { (x) })
> +#define PTRS_PER_PUD    (PAGE_SIZE / sizeof(pud_t))
> +
>  /* Page Middle Directory entry */
>  typedef struct {
>  	unsigned long pmd;
> @@ -60,6 +76,16 @@ static inline void pud_clear(pud_t *pudp)
>  	set_pud(pudp, __pud(0));
>  }
> 
> +static inline pud_t pfn_pud(unsigned long pfn, pgprot_t prot)
> +{
> +	return __pud((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
> +}
> +
> +static inline unsigned long _pud_pfn(pud_t pud)
> +{
> +	return pud_val(pud) >> _PAGE_PFN_SHIFT;
> +}
> +
>  static inline unsigned long pud_page_vaddr(pud_t pud)
>  {
>  	return (unsigned long)pfn_to_virt(pud_val(pud) >> _PAGE_PFN_SHIFT);
> @@ -70,6 +96,15 @@ static inline struct page *pud_page(pud_t pud)
>  	return pfn_to_page(pud_val(pud) >> _PAGE_PFN_SHIFT);
>  }
> 
> +#define mm_pud_folded	mm_pud_folded
> +static inline bool mm_pud_folded(struct mm_struct *mm)
> +{
> +	if (pgtable_l4_enabled)
> +		return false;
> +
> +	return true;
> +}
> +
>  #define pmd_index(addr) (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1))
> 
>  static inline pmd_t *pmd_offset(pud_t *pud, unsigned long addr)
> @@ -90,4 +125,64 @@ static inline unsigned long _pmd_pfn(pmd_t pmd)
>  #define pmd_ERROR(e) \
>  	pr_err("%s:%d: bad pmd %016lx.\n", __FILE__, __LINE__, pmd_val(e))
> 
> +#define pud_ERROR(e)	\
> +	pr_err("%s:%d: bad pud %016lx.\n", __FILE__, __LINE__, pud_val(e))
> +
> +static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
> +{
> +	if (pgtable_l4_enabled)
> +		*p4dp = p4d;
> +	else
> +		set_pud((pud_t *)p4dp, (pud_t){ p4d_val(p4d) });
> +}
> +
> +static inline int p4d_none(p4d_t p4d)
> +{
> +	if (pgtable_l4_enabled)
> +		return (p4d_val(p4d) == 0);
> +
> +	return 0;
> +}
> +
> +static inline int p4d_present(p4d_t p4d)
> +{
> +	if (pgtable_l4_enabled)
> +		return (p4d_val(p4d) & _PAGE_PRESENT);
> +
> +	return 1;
> +}
> +
> +static inline int p4d_bad(p4d_t p4d)
> +{
> +	if (pgtable_l4_enabled)
> +		return !p4d_present(p4d);
> +
> +	return 0;
> +}
> +
> +static inline void p4d_clear(p4d_t *p4d)
> +{
> +	if (pgtable_l4_enabled)
> +		set_p4d(p4d, __p4d(0));
> +}
> +
> +static inline unsigned long p4d_page_vaddr(p4d_t p4d)
> +{
> +	if (pgtable_l4_enabled)
> +		return (unsigned long)pfn_to_virt(
> +				p4d_val(p4d) >> _PAGE_PFN_SHIFT);
> +
> +	return pud_page_vaddr((pud_t) { p4d_val(p4d) });
> +}
> +
> +#define pud_index(addr) (((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
> +
> +static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
> +{
> +	if (pgtable_l4_enabled)
> +		return (pud_t *)p4d_page_vaddr(*p4d) + pud_index(address);
> +
> +	return (pud_t *)p4d;
> +}
> +

In my test I had to put
#define pud_offset pud_offset
here or else I got a compilation error due to pud_offset being redefined 
on include/linux/pgtable.h:

#ifndef pud_offset
static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
{
         return (pud_t *)p4d_page_vaddr(*p4d) + pud_index(address);
}
#define pud_offset pud_offset
#endif

Re: [PATCH v2 5/8] riscv: Implement sv48 support

[Alex Ghiti]

Hi Nick,

Le 6/27/20 à 8:30 AM, Nick Kossifidis a écrit :
> Στις 2020-06-03 11:11, Alexandre Ghiti έγραψε:
>> By adding a new 4th level of page table, give the possibility to 64bit
>> kernel to address 2^48 bytes of virtual address: in practice, that 
>> roughly
>> offers ~160TB of virtual address space to userspace and allows up to 64TB
>> of physical memory.
>>
>> If the underlying hardware does not support sv48, we will automatically
>> fallback to a standard 3-level page table by folding the new PUD level 
>> into
>> PGDIR level. In order to detect HW capabilities at runtime, we
>> use SATP feature that ignores writes with an unsupported mode.
>>
>> Signed-off-by: Alexandre Ghiti <alex@ghiti.fr>
>> Reviewed-by: Anup Patel <anup@brainfault.org>
>> ---
>>  arch/riscv/Kconfig                  |   6 +-
>>  arch/riscv/include/asm/csr.h        |   3 +-
>>  arch/riscv/include/asm/fixmap.h     |   1 +
>>  arch/riscv/include/asm/page.h       |  15 +++
>>  arch/riscv/include/asm/pgalloc.h    |  36 +++++++
>>  arch/riscv/include/asm/pgtable-64.h |  97 ++++++++++++++++-
>>  arch/riscv/include/asm/pgtable.h    |  10 +-
>>  arch/riscv/kernel/head.S            |   3 +-
>>  arch/riscv/mm/context.c             |   2 +-
>>  arch/riscv/mm/init.c                | 158 +++++++++++++++++++++++++---
>>  10 files changed, 307 insertions(+), 24 deletions(-)
>>
>> diff --git a/arch/riscv/Kconfig b/arch/riscv/Kconfig
>> index e167f16131f4..3f73f60e9732 100644
>> --- a/arch/riscv/Kconfig
>> +++ b/arch/riscv/Kconfig
>> @@ -68,6 +68,7 @@ config RISCV
>>      select ARCH_HAS_GCOV_PROFILE_ALL
>>      select HAVE_COPY_THREAD_TLS
>>      select HAVE_ARCH_KASAN if MMU && 64BIT
>> +    select RELOCATABLE if 64BIT
>>
>>  config ARCH_MMAP_RND_BITS_MIN
>>      default 18 if 64BIT
>> @@ -106,7 +107,7 @@ config PAGE_OFFSET
>>      default 0xC0000000 if 32BIT && MAXPHYSMEM_2GB
>>      default 0x80000000 if 64BIT && !MMU
>>      default 0xffffffff80000000 if 64BIT && MAXPHYSMEM_2GB
>> -    default 0xffffffe000000000 if 64BIT && !MAXPHYSMEM_2GB
>> +    default 0xffffc00000000000 if 64BIT && !MAXPHYSMEM_2GB
>>
>>  config ARCH_FLATMEM_ENABLE
>>      def_bool y
>> @@ -155,8 +156,11 @@ config GENERIC_HWEIGHT
>>  config FIX_EARLYCON_MEM
>>      def_bool MMU
>>
>> +# On a 64BIT relocatable kernel, the 4-level page table is at runtime 
>> folded
>> +# on a 3-level page table when sv48 is not supported.
>>  config PGTABLE_LEVELS
>>      int
>> +    default 4 if 64BIT && RELOCATABLE
>>      default 3 if 64BIT
>>      default 2
>>
>> diff --git a/arch/riscv/include/asm/csr.h b/arch/riscv/include/asm/csr.h
>> index cec462e198ce..d41536c3f8d4 100644
>> --- a/arch/riscv/include/asm/csr.h
>> +++ b/arch/riscv/include/asm/csr.h
>> @@ -40,11 +40,10 @@
>>  #ifndef CONFIG_64BIT
>>  #define SATP_PPN    _AC(0x003FFFFF, UL)
>>  #define SATP_MODE_32    _AC(0x80000000, UL)
>> -#define SATP_MODE    SATP_MODE_32
>>  #else
>>  #define SATP_PPN    _AC(0x00000FFFFFFFFFFF, UL)
>>  #define SATP_MODE_39    _AC(0x8000000000000000, UL)
>> -#define SATP_MODE    SATP_MODE_39
>> +#define SATP_MODE_48    _AC(0x9000000000000000, UL)
>>  #endif
>>
>>  /* Exception cause high bit - is an interrupt if set */
>> diff --git a/arch/riscv/include/asm/fixmap.h 
>> b/arch/riscv/include/asm/fixmap.h
>> index 2368d49eb4ef..d891cf9c73c5 100644
>> --- a/arch/riscv/include/asm/fixmap.h
>> +++ b/arch/riscv/include/asm/fixmap.h
>> @@ -27,6 +27,7 @@ enum fixed_addresses {
>>      FIX_FDT = FIX_FDT_END + FIX_FDT_SIZE / PAGE_SIZE - 1,
>>      FIX_PTE,
>>      FIX_PMD,
>> +    FIX_PUD,
>>      FIX_TEXT_POKE1,
>>      FIX_TEXT_POKE0,
>>      FIX_EARLYCON_MEM_BASE,
>> diff --git a/arch/riscv/include/asm/page.h 
>> b/arch/riscv/include/asm/page.h
>> index 48bb09b6a9b7..5e77fe7f0d6d 100644
>> --- a/arch/riscv/include/asm/page.h
>> +++ b/arch/riscv/include/asm/page.h
>> @@ -31,7 +31,19 @@
>>   * When not using MMU this corresponds to the first free page in
>>   * physical memory (aligned on a page boundary).
>>   */
>> +#ifdef CONFIG_RELOCATABLE
>> +#define PAGE_OFFSET        __page_offset
>> +
>> +#ifdef CONFIG_64BIT
>> +/*
>> + * By default, CONFIG_PAGE_OFFSET value corresponds to SV48 address 
>> space so
>> + * define the PAGE_OFFSET value for SV39.
>> + */
>> +#define PAGE_OFFSET_L3        0xffffffe000000000
>> +#endif /* CONFIG_64BIT */
>> +#else
>>  #define PAGE_OFFSET        _AC(CONFIG_PAGE_OFFSET, UL)
>> +#endif /* CONFIG_RELOCATABLE */
>>
>>  #define KERN_VIRT_SIZE (-PAGE_OFFSET)
>>
>> @@ -102,6 +114,9 @@ extern unsigned long pfn_base;
>>  extern unsigned long max_low_pfn;
>>  extern unsigned long min_low_pfn;
>>  extern unsigned long kernel_virt_addr;
>> +#ifdef CONFIG_RELOCATABLE
>> +extern unsigned long __page_offset;
>> +#endif
>>
>>  #define __pa_to_va_nodebug(x)    ((void *)((unsigned long) (x) + 
>> va_pa_offset))
>>  #define linear_mapping_va_to_pa(x)    ((unsigned long)(x) - 
>> va_pa_offset)
>> diff --git a/arch/riscv/include/asm/pgalloc.h 
>> b/arch/riscv/include/asm/pgalloc.h
>> index 3f601ee8233f..540eaa5a8658 100644
>> --- a/arch/riscv/include/asm/pgalloc.h
>> +++ b/arch/riscv/include/asm/pgalloc.h
>> @@ -36,6 +36,42 @@ static inline void pud_populate(struct mm_struct
>> *mm, pud_t *pud, pmd_t *pmd)
>>
>>      set_pud(pud, __pud((pfn << _PAGE_PFN_SHIFT) | _PAGE_TABLE));
>>  }
>> +
>> +static inline void p4d_populate(struct mm_struct *mm, p4d_t *p4d, 
>> pud_t *pud)
>> +{
>> +    if (pgtable_l4_enabled) {
>> +        unsigned long pfn = virt_to_pfn(pud);
>> +
>> +        set_p4d(p4d, __p4d((pfn << _PAGE_PFN_SHIFT) | _PAGE_TABLE));
>> +    }
>> +}
>> +
>> +static inline void p4d_populate_safe(struct mm_struct *mm, p4d_t *p4d,
>> +                     pud_t *pud)
>> +{
>> +    if (pgtable_l4_enabled) {
>> +        unsigned long pfn = virt_to_pfn(pud);
>> +
>> +        set_p4d_safe(p4d,
>> +                 __p4d((pfn << _PAGE_PFN_SHIFT) | _PAGE_TABLE));
>> +    }
>> +}
>> +
>> +static inline pud_t *pud_alloc_one(struct mm_struct *mm, unsigned 
>> long addr)
>> +{
>> +    if (pgtable_l4_enabled)
>> +        return (pud_t *)__get_free_page(
>> +                GFP_KERNEL | __GFP_RETRY_MAYFAIL | __GFP_ZERO);
>> +    return NULL;
>> +}
>> +
>> +static inline void pud_free(struct mm_struct *mm, pud_t *pud)
>> +{
>> +    if (pgtable_l4_enabled)
>> +        free_page((unsigned long)pud);
>> +}
>> +
>> +#define __pud_free_tlb(tlb, pud, addr)  pud_free((tlb)->mm, pud)
>>  #endif /* __PAGETABLE_PMD_FOLDED */
>>
>>  #define pmd_pgtable(pmd)    pmd_page(pmd)
>> diff --git a/arch/riscv/include/asm/pgtable-64.h
>> b/arch/riscv/include/asm/pgtable-64.h
>> index b15f70a1fdfa..c84c31fbf8da 100644
>> --- a/arch/riscv/include/asm/pgtable-64.h
>> +++ b/arch/riscv/include/asm/pgtable-64.h
>> @@ -8,16 +8,32 @@
>>
>>  #include <linux/const.h>
>>
>> -#define PGDIR_SHIFT     30
>> +extern bool pgtable_l4_enabled;
>> +
>> +#define PGDIR_SHIFT     (pgtable_l4_enabled ? 39 : 30)
>>  /* Size of region mapped by a page global directory */
>>  #define PGDIR_SIZE      (_AC(1, UL) << PGDIR_SHIFT)
>>  #define PGDIR_MASK      (~(PGDIR_SIZE - 1))
>>
>> +/* pud is folded into pgd in case of 3-level page table */
>> +#define PUD_SHIFT    30
>> +#define PUD_SIZE    (_AC(1, UL) << PUD_SHIFT)
>> +#define PUD_MASK    (~(PUD_SIZE - 1))
>> +
>>  #define PMD_SHIFT       21
>>  /* Size of region mapped by a page middle directory */
>>  #define PMD_SIZE        (_AC(1, UL) << PMD_SHIFT)
>>  #define PMD_MASK        (~(PMD_SIZE - 1))
>>
>> +/* Page Upper Directory entry */
>> +typedef struct {
>> +    unsigned long pud;
>> +} pud_t;
>> +
>> +#define pud_val(x)      ((x).pud)
>> +#define __pud(x)        ((pud_t) { (x) })
>> +#define PTRS_PER_PUD    (PAGE_SIZE / sizeof(pud_t))
>> +
>>  /* Page Middle Directory entry */
>>  typedef struct {
>>      unsigned long pmd;
>> @@ -60,6 +76,16 @@ static inline void pud_clear(pud_t *pudp)
>>      set_pud(pudp, __pud(0));
>>  }
>>
>> +static inline pud_t pfn_pud(unsigned long pfn, pgprot_t prot)
>> +{
>> +    return __pud((pfn << _PAGE_PFN_SHIFT) | pgprot_val(prot));
>> +}
>> +
>> +static inline unsigned long _pud_pfn(pud_t pud)
>> +{
>> +    return pud_val(pud) >> _PAGE_PFN_SHIFT;
>> +}
>> +
>>  static inline unsigned long pud_page_vaddr(pud_t pud)
>>  {
>>      return (unsigned long)pfn_to_virt(pud_val(pud) >> _PAGE_PFN_SHIFT);
>> @@ -70,6 +96,15 @@ static inline struct page *pud_page(pud_t pud)
>>      return pfn_to_page(pud_val(pud) >> _PAGE_PFN_SHIFT);
>>  }
>>
>> +#define mm_pud_folded    mm_pud_folded
>> +static inline bool mm_pud_folded(struct mm_struct *mm)
>> +{
>> +    if (pgtable_l4_enabled)
>> +        return false;
>> +
>> +    return true;
>> +}
>> +
>>  #define pmd_index(addr) (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1))
>>
>>  static inline pmd_t *pmd_offset(pud_t *pud, unsigned long addr)
>> @@ -90,4 +125,64 @@ static inline unsigned long _pmd_pfn(pmd_t pmd)
>>  #define pmd_ERROR(e) \
>>      pr_err("%s:%d: bad pmd %016lx.\n", __FILE__, __LINE__, pmd_val(e))
>>
>> +#define pud_ERROR(e)    \
>> +    pr_err("%s:%d: bad pud %016lx.\n", __FILE__, __LINE__, pud_val(e))
>> +
>> +static inline void set_p4d(p4d_t *p4dp, p4d_t p4d)
>> +{
>> +    if (pgtable_l4_enabled)
>> +        *p4dp = p4d;
>> +    else
>> +        set_pud((pud_t *)p4dp, (pud_t){ p4d_val(p4d) });
>> +}
>> +
>> +static inline int p4d_none(p4d_t p4d)
>> +{
>> +    if (pgtable_l4_enabled)
>> +        return (p4d_val(p4d) == 0);
>> +
>> +    return 0;
>> +}
>> +
>> +static inline int p4d_present(p4d_t p4d)
>> +{
>> +    if (pgtable_l4_enabled)
>> +        return (p4d_val(p4d) & _PAGE_PRESENT);
>> +
>> +    return 1;
>> +}
>> +
>> +static inline int p4d_bad(p4d_t p4d)
>> +{
>> +    if (pgtable_l4_enabled)
>> +        return !p4d_present(p4d);
>> +
>> +    return 0;
>> +}
>> +
>> +static inline void p4d_clear(p4d_t *p4d)
>> +{
>> +    if (pgtable_l4_enabled)
>> +        set_p4d(p4d, __p4d(0));
>> +}
>> +
>> +static inline unsigned long p4d_page_vaddr(p4d_t p4d)
>> +{
>> +    if (pgtable_l4_enabled)
>> +        return (unsigned long)pfn_to_virt(
>> +                p4d_val(p4d) >> _PAGE_PFN_SHIFT);
>> +
>> +    return pud_page_vaddr((pud_t) { p4d_val(p4d) });
>> +}
>> +
>> +#define pud_index(addr) (((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1))
>> +
>> +static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
>> +{
>> +    if (pgtable_l4_enabled)
>> +        return (pud_t *)p4d_page_vaddr(*p4d) + pud_index(address);
>> +
>> +    return (pud_t *)p4d;
>> +}
>> +
> 
> In my test I had to put
> #define pud_offset pud_offset
> here or else I got a compilation error due to pud_offset being redefined 
> on include/linux/pgtable.h:
> 
> #ifndef pud_offset
> static inline pud_t *pud_offset(p4d_t *p4d, unsigned long address)
> {
>          return (pud_t *)p4d_page_vaddr(*p4d) + pud_index(address);
> }
> #define pud_offset pud_offset
> #endif

Yes, the rebase on 5.8-rc2 requires that and removing pmd_offset definition.

Alex

Re: [PATCH v2 0/8] Introduce sv48 support

[Palmer Dabbelt]

On Wed, 03 Jun 2020 01:10:56 PDT (-0700), alex@ghiti.fr wrote:
> This patchset implements sv48 support at runtime. The kernel will try to
> boot with 4-level page table and will fallback to 3-level if the HW does not
> support it.
>
> The biggest advantage is that we only have one kernel for 64bit, which
> is way easier to maintain.
>
> Folding the 4th level into a 3-level page table has almost no cost at
> runtime. But as mentioned Palmer, the relocatable code generated is less
> performant.
>
> At the moment, there is no way to build a 3-level page table non-relocatable
> 64bit kernel. We agreed that distributions will use this runtime configuration
> anyway, but Palmer proposed to introduce a new Kconfig, which I will do later
> as sv48 support was asked for 5.8.

Sorry I wasn't clear last time, but this still has the same fundamental issue:
it forces 64-bit kernels to be relocatable, which imposes a performance
penalty.  We don't have any hardware that can actually take advantage of sv48,
so I don't want to take anything that penalizes what people are actually using
in order to add a feature people can't use.

I'd be OK taking this if sv48 support simply depended on a relocatable kernel,
as then users who want the faster kernel could still build one.  I don't want
to take something that forces all 64-bit kernels to be relocatable.

> Finally, the user can now ask for sv39 explicitly by using the device-tree
> which will reduce memory footprint and reduce the number of memory accesses
> in case of TLB miss.
>
> Changes in v2:
>   * Move variable declarations to pgtable.h in patch 5/7 as suggested by Anup
>   * Restore mmu-type properties in patch 6 as suggested by Anup
>   * Fix unused variable in patch 5 that was used in patch 6
>   * Fix SPARSEMEM build (patch 2 was modified so I dropped the Reviewed-by)
>   * Applied various Reviewed-by
>
> Alexandre Ghiti (8):
>   riscv: Get rid of compile time logic with MAX_EARLY_MAPPING_SIZE
>   riscv: Allow to dynamically define VA_BITS
>   riscv: Simplify MAXPHYSMEM config
>   riscv: Prepare ptdump for vm layout dynamic addresses
>   riscv: Implement sv48 support
>   riscv: Allow user to downgrade to sv39 when hw supports sv48
>   riscv: Use pgtable_l4_enabled to output mmu type in cpuinfo
>   riscv: Explicit comment about user virtual address space size
>
>  arch/riscv/Kconfig                  |  34 ++---
>  arch/riscv/include/asm/csr.h        |   3 +-
>  arch/riscv/include/asm/fixmap.h     |   1 +
>  arch/riscv/include/asm/page.h       |  15 +++
>  arch/riscv/include/asm/pgalloc.h    |  36 ++++++
>  arch/riscv/include/asm/pgtable-64.h |  97 +++++++++++++-
>  arch/riscv/include/asm/pgtable.h    |  31 ++++-
>  arch/riscv/include/asm/sparsemem.h  |   6 +-
>  arch/riscv/kernel/cpu.c             |  23 ++--
>  arch/riscv/kernel/head.S            |   3 +-
>  arch/riscv/mm/context.c             |   2 +-
>  arch/riscv/mm/init.c                | 194 ++++++++++++++++++++++++----
>  arch/riscv/mm/ptdump.c              |  49 +++++--
>  13 files changed, 412 insertions(+), 82 deletions(-)

Re: [PATCH v2 0/8] Introduce sv48 support

[Alex Ghiti]

Hi Palmer,

Le 7/1/20 à 2:27 PM, Palmer Dabbelt a écrit :
> On Wed, 03 Jun 2020 01:10:56 PDT (-0700), alex@ghiti.fr wrote:
>> This patchset implements sv48 support at runtime. The kernel will try to
>> boot with 4-level page table and will fallback to 3-level if the HW 
>> does not
>> support it.
>>
>> The biggest advantage is that we only have one kernel for 64bit, which
>> is way easier to maintain.
>>
>> Folding the 4th level into a 3-level page table has almost no cost at
>> runtime. But as mentioned Palmer, the relocatable code generated is less
>> performant.
>>
>> At the moment, there is no way to build a 3-level page table 
>> non-relocatable
>> 64bit kernel. We agreed that distributions will use this runtime 
>> configuration
>> anyway, but Palmer proposed to introduce a new Kconfig, which I will 
>> do later
>> as sv48 support was asked for 5.8.
> 
> Sorry I wasn't clear last time, but this still has the same fundamental 
> issue:
> it forces 64-bit kernels to be relocatable, which imposes a performance
> penalty.  We don't have any hardware that can actually take advantage of 
> sv48,
> so I don't want to take anything that penalizes what people are actually 
> using
> in order to add a feature people can't use.
> 
> I'd be OK taking this if sv48 support simply depended on a relocatable 
> kernel,
> as then users who want the faster kernel could still build one.  I don't 
> want
> to take something that forces all 64-bit kernels to be relocatable.

Indeed, I had not understood that this was a requirement. I will add a 
patch on top of this one introducing a new config, I have to think about it.

But even if I understand that the new level of indirection coming with 
PIE will be slower, is this new config worth it ? Can we benchmark 
somehow the performance loss ? IMHO I think that this config will get 
broken over time by lack of testing because I believe distributions will 
go for KASLR kernel which requires the relocatability property anyway.

Alex

> 
>> Finally, the user can now ask for sv39 explicitly by using the 
>> device-tree
>> which will reduce memory footprint and reduce the number of memory 
>> accesses
>> in case of TLB miss.
>>
>> Changes in v2:
>>   * Move variable declarations to pgtable.h in patch 5/7 as suggested 
>> by Anup
>>   * Restore mmu-type properties in patch 6 as suggested by Anup
>>   * Fix unused variable in patch 5 that was used in patch 6
>>   * Fix SPARSEMEM build (patch 2 was modified so I dropped the 
>> Reviewed-by)
>>   * Applied various Reviewed-by
>>
>> Alexandre Ghiti (8):
>>   riscv: Get rid of compile time logic with MAX_EARLY_MAPPING_SIZE
>>   riscv: Allow to dynamically define VA_BITS
>>   riscv: Simplify MAXPHYSMEM config
>>   riscv: Prepare ptdump for vm layout dynamic addresses
>>   riscv: Implement sv48 support
>>   riscv: Allow user to downgrade to sv39 when hw supports sv48
>>   riscv: Use pgtable_l4_enabled to output mmu type in cpuinfo
>>   riscv: Explicit comment about user virtual address space size
>>
>>  arch/riscv/Kconfig                  |  34 ++---
>>  arch/riscv/include/asm/csr.h        |   3 +-
>>  arch/riscv/include/asm/fixmap.h     |   1 +
>>  arch/riscv/include/asm/page.h       |  15 +++
>>  arch/riscv/include/asm/pgalloc.h    |  36 ++++++
>>  arch/riscv/include/asm/pgtable-64.h |  97 +++++++++++++-
>>  arch/riscv/include/asm/pgtable.h    |  31 ++++-
>>  arch/riscv/include/asm/sparsemem.h  |   6 +-
>>  arch/riscv/kernel/cpu.c             |  23 ++--
>>  arch/riscv/kernel/head.S            |   3 +-
>>  arch/riscv/mm/context.c             |   2 +-
>>  arch/riscv/mm/init.c                | 194 ++++++++++++++++++++++++----
>>  arch/riscv/mm/ptdump.c              |  49 +++++--
>>  13 files changed, 412 insertions(+), 82 deletions(-)