| Message ID | 20230714165508.94561-2-charlie@rivosinc.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | RISC-V: mm: Make SV48 the default address space | expand |
| Context | Check | Description |
|---|---|---|
| conchuod/cover_letter | success | Series has a cover letter |
| conchuod/tree_selection | success | Guessed tree name to be for-next at HEAD 471aba2e4760 |
| conchuod/fixes_present | success | Fixes tag not required for -next series |
| conchuod/maintainers_pattern | success | MAINTAINERS pattern errors before the patch: 4 and now 4 |
| conchuod/verify_signedoff | success | Signed-off-by tag matches author and committer |
| conchuod/kdoc | success | Errors and warnings before: 0 this patch: 0 |
| conchuod/build_rv64_clang_allmodconfig | success | Errors and warnings before: 2810 this patch: 2810 |
| conchuod/module_param | success | Was 0 now: 0 |
| conchuod/build_rv64_gcc_allmodconfig | fail | Errors and warnings before: 15876 this patch: 15877 |
| conchuod/build_rv32_defconfig | success | Build OK |
| conchuod/dtb_warn_rv64 | success | Errors and warnings before: 3 this patch: 3 |
| conchuod/header_inline | success | No static functions without inline keyword in header files |
| conchuod/checkpatch | warning | CHECK: Unnecessary parentheses around 'VA_BITS >= VA_BITS_SV57' |
| conchuod/build_rv64_nommu_k210_defconfig | success | Build OK |
| conchuod/verify_fixes | success | No Fixes tag |
| conchuod/build_rv64_nommu_virt_defconfig | success | Build OK |
On Fri, Jul 14, 2023 at 6:55 PM Charlie Jenkins <charlie@rivosinc.com> wrote: > > Make sv48 the default address space for mmap as some applications > currently depend on this assumption. A hint address passed to mmap will > cause the largest address space that fits entirely into the hint to be > used. If the hint is less than or equal to 1<<38, an sv39 address will > be used. An exception is that if the hint address is 0, then a sv48 > address will be used. After an address space is completely full, the next > smallest address space will be used. > > Signed-off-by: Charlie Jenkins <charlie@rivosinc.com> > --- > arch/riscv/include/asm/elf.h | 2 +- > arch/riscv/include/asm/pgtable.h | 12 +++++++- > arch/riscv/include/asm/processor.h | 46 +++++++++++++++++++++++++----- > 3 files changed, 51 insertions(+), 9 deletions(-) > > diff --git a/arch/riscv/include/asm/elf.h b/arch/riscv/include/asm/elf.h > index c24280774caf..5d3368d5585c 100644 > --- a/arch/riscv/include/asm/elf.h > +++ b/arch/riscv/include/asm/elf.h > @@ -49,7 +49,7 @@ extern bool compat_elf_check_arch(Elf32_Ehdr *hdr); > * the loader. We need to make sure that it is out of the way of the program > * that it will "exec", and that there is sufficient room for the brk. > */ > -#define ELF_ET_DYN_BASE ((TASK_SIZE / 3) * 2) > +#define ELF_ET_DYN_BASE ((DEFAULT_MAP_WINDOW / 3) * 2) > > #ifdef CONFIG_64BIT > #ifdef CONFIG_COMPAT > diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h > index 75970ee2bda2..e13f5872bfe9 100644 > --- a/arch/riscv/include/asm/pgtable.h > +++ b/arch/riscv/include/asm/pgtable.h > @@ -63,12 +63,22 @@ > * position vmemmap directly below the VMALLOC region. > */ > #ifdef CONFIG_64BIT > +#define VA_BITS_SV39 39 > +#define VA_BITS_SV48 48 > +#define VA_BITS_SV57 57 > + > +#define VA_USER_SV39 (UL(1) << (VA_BITS_SV39 - 1)) > +#define VA_USER_SV48 (UL(1) << (VA_BITS_SV48 - 1)) > +#define VA_USER_SV57 (UL(1) << (VA_BITS_SV57 - 1)) > + > #define VA_BITS (pgtable_l5_enabled ? \ > - 57 : (pgtable_l4_enabled ? 48 : 39)) > + VA_BITS_SV57 : (pgtable_l4_enabled ? VA_BITS_SV48 : VA_BITS_SV39)) > #else > #define VA_BITS 32 > #endif > > +#define MMAP_VA_BITS ((VA_BITS >= VA_BITS_SV48) ? VA_BITS_SV48 : VA_BITS) > + > #define VMEMMAP_SHIFT \ > (VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT) > #define VMEMMAP_SIZE BIT(VMEMMAP_SHIFT) > diff --git a/arch/riscv/include/asm/processor.h b/arch/riscv/include/asm/processor.h > index c950a8d9edef..14a5396eed3d 100644 > --- a/arch/riscv/include/asm/processor.h > +++ b/arch/riscv/include/asm/processor.h > @@ -13,20 +13,52 @@ > > #include <asm/ptrace.h> > > -/* > - * This decides where the kernel will search for a free chunk of vm > - * space during mmap's. > - */ > -#define TASK_UNMAPPED_BASE PAGE_ALIGN(TASK_SIZE / 3) > - > -#define STACK_TOP TASK_SIZE > #ifdef CONFIG_64BIT > +#define DEFAULT_MAP_WINDOW (UL(1) << (MMAP_VA_BITS - 1)) > #define STACK_TOP_MAX TASK_SIZE_64 > + > +#define arch_get_mmap_end(addr, len, flags) \ > +({ \ > + unsigned long mmap_end; \ > + if ((addr) >= VA_USER_SV57) \ > + mmap_end = STACK_TOP_MAX; \ > + else if ((((addr) >= VA_USER_SV48)) && (VA_BITS >= VA_BITS_SV48)) \ > + mmap_end = VA_USER_SV48; \ > + else if ((addr) == 0) \ > + mmap_end = DEFAULT_MAP_WINDOW; \ > + else \ > + mmap_end = VA_USER_SV39; \ > + mmap_end; \ > +}) What about the following instead: #define arch_get_mmap_end(addr, len, flags) \ ({ \ unsigned long mmap_end; \ if ((addr) >= VA_USER_SV57) \ mmap_end = STACK_TOP_MAX; \ // Maybe a comment here that says it returns the max user address of the current mode, not obvious at first sight. else \ mmap_end = DEFAULT_MAP_WINDOW; \ mmap_end; \ }) The only corner case is when sv57 is active, then only a hint greater than VA_USER_SV57 can return a sv57 user address. Otherwise, we just need to return the default mmap end right? > + > +#define arch_get_mmap_base(addr, base) \ > +({ \ > + unsigned long mmap_base; \ > + if (((addr) >= VA_USER_SV57) && (VA_BITS >= VA_BITS_SV57)) \ > + mmap_base = (base) + (VA_USER_SV57 - DEFAULT_MAP_WINDOW); \ > + else if ((((addr) >= VA_USER_SV48)) && (VA_BITS >= VA_BITS_SV48)) \ > + mmap_base = (base) + (VA_USER_SV48 - DEFAULT_MAP_WINDOW); \ > + else if ((addr) == 0) \ > + mmap_base = (base); \ > + else \ > + mmap_base = (base) + (VA_USER_SV39 - DEFAULT_MAP_WINDOW); \ > + mmap_base; \ > +}) > + From arch_pick_mmap_layout() (https://elixir.bootlin.com/linux/latest/source/mm/util.c#L433), the "base" argument is: - either STACK_TOP in top-down (more or less some random offset) - or TASK_UNMAPPED_BASE in bottom-up (more or less some random offset) When bottom-up is the current mode, we should not change the base, so adding (VA_USER_SV57 - DEFAULT_MAP_WINDOW) in the first case is not right for me. When sv48 or sv57 are the active mode, DEFAULT_MAP_WINDOW is equal to VA_USER_SV48 right? So (VA_USER_SV48 - DEFAULT_MAP_WINDOW) is 0, so not useful. And for the last case, when the user asks for a sv39 address whereas the active mode is sv48 or sv57, then (VA_USER_SV39 - DEFAULT_MAP_WINDOW) is negative and the base is smaller which is not correct. In the bottom-up case, we should preserve the base and I think that again, only sv57 is the corner case to deal with. > #else > +#define DEFAULT_MAP_WINDOW TASK_SIZE > #define STACK_TOP_MAX TASK_SIZE > #endif > #define STACK_ALIGN 16 > > +#define STACK_TOP DEFAULT_MAP_WINDOW > + > +/* > + * This decides where the kernel will search for a free chunk of vm > + * space during mmap's. > + */ > +#define TASK_UNMAPPED_BASE PAGE_ALIGN(DEFAULT_MAP_WINDOW / 3) > + > #ifndef __ASSEMBLY__ > > struct task_struct; > -- > 2.41.0 >
On Thu, Jul 20, 2023 at 10:13:54AM +0200, Alexandre Ghiti wrote: > On Fri, Jul 14, 2023 at 6:55 PM Charlie Jenkins <charlie@rivosinc.com> wrote: > > > > Make sv48 the default address space for mmap as some applications > > currently depend on this assumption. A hint address passed to mmap will > > cause the largest address space that fits entirely into the hint to be > > used. If the hint is less than or equal to 1<<38, an sv39 address will > > be used. An exception is that if the hint address is 0, then a sv48 > > address will be used. After an address space is completely full, the next > > smallest address space will be used. > > > > Signed-off-by: Charlie Jenkins <charlie@rivosinc.com> > > --- > > arch/riscv/include/asm/elf.h | 2 +- > > arch/riscv/include/asm/pgtable.h | 12 +++++++- > > arch/riscv/include/asm/processor.h | 46 +++++++++++++++++++++++++----- > > 3 files changed, 51 insertions(+), 9 deletions(-) > > > > diff --git a/arch/riscv/include/asm/elf.h b/arch/riscv/include/asm/elf.h > > index c24280774caf..5d3368d5585c 100644 > > --- a/arch/riscv/include/asm/elf.h > > +++ b/arch/riscv/include/asm/elf.h > > @@ -49,7 +49,7 @@ extern bool compat_elf_check_arch(Elf32_Ehdr *hdr); > > * the loader. We need to make sure that it is out of the way of the program > > * that it will "exec", and that there is sufficient room for the brk. > > */ > > -#define ELF_ET_DYN_BASE ((TASK_SIZE / 3) * 2) > > +#define ELF_ET_DYN_BASE ((DEFAULT_MAP_WINDOW / 3) * 2) > > > > #ifdef CONFIG_64BIT > > #ifdef CONFIG_COMPAT > > diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h > > index 75970ee2bda2..e13f5872bfe9 100644 > > --- a/arch/riscv/include/asm/pgtable.h > > +++ b/arch/riscv/include/asm/pgtable.h > > @@ -63,12 +63,22 @@ > > * position vmemmap directly below the VMALLOC region. > > */ > > #ifdef CONFIG_64BIT > > +#define VA_BITS_SV39 39 > > +#define VA_BITS_SV48 48 > > +#define VA_BITS_SV57 57 > > + > > +#define VA_USER_SV39 (UL(1) << (VA_BITS_SV39 - 1)) > > +#define VA_USER_SV48 (UL(1) << (VA_BITS_SV48 - 1)) > > +#define VA_USER_SV57 (UL(1) << (VA_BITS_SV57 - 1)) > > + > > #define VA_BITS (pgtable_l5_enabled ? \ > > - 57 : (pgtable_l4_enabled ? 48 : 39)) > > + VA_BITS_SV57 : (pgtable_l4_enabled ? VA_BITS_SV48 : VA_BITS_SV39)) > > #else > > #define VA_BITS 32 > > #endif > > > > +#define MMAP_VA_BITS ((VA_BITS >= VA_BITS_SV48) ? VA_BITS_SV48 : VA_BITS) > > + > > #define VMEMMAP_SHIFT \ > > (VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT) > > #define VMEMMAP_SIZE BIT(VMEMMAP_SHIFT) > > diff --git a/arch/riscv/include/asm/processor.h b/arch/riscv/include/asm/processor.h > > index c950a8d9edef..14a5396eed3d 100644 > > --- a/arch/riscv/include/asm/processor.h > > +++ b/arch/riscv/include/asm/processor.h > > @@ -13,20 +13,52 @@ > > > > #include <asm/ptrace.h> > > > > -/* > > - * This decides where the kernel will search for a free chunk of vm > > - * space during mmap's. > > - */ > > -#define TASK_UNMAPPED_BASE PAGE_ALIGN(TASK_SIZE / 3) > > - > > -#define STACK_TOP TASK_SIZE > > #ifdef CONFIG_64BIT > > +#define DEFAULT_MAP_WINDOW (UL(1) << (MMAP_VA_BITS - 1)) > > #define STACK_TOP_MAX TASK_SIZE_64 > > + > > +#define arch_get_mmap_end(addr, len, flags) \ > > +({ \ > > + unsigned long mmap_end; \ > > + if ((addr) >= VA_USER_SV57) \ > > + mmap_end = STACK_TOP_MAX; \ > > + else if ((((addr) >= VA_USER_SV48)) && (VA_BITS >= VA_BITS_SV48)) \ > > + mmap_end = VA_USER_SV48; \ > > + else if ((addr) == 0) \ > > + mmap_end = DEFAULT_MAP_WINDOW; \ > > + else \ > > + mmap_end = VA_USER_SV39; \ > > + mmap_end; \ > > +}) > > What about the following instead: > > #define arch_get_mmap_end(addr, len, flags) \ > ({ \ > unsigned long mmap_end; \ > if ((addr) >= VA_USER_SV57) \ > mmap_end = STACK_TOP_MAX; \ // Maybe a comment here that > says it returns the max user address of the current mode, not obvious > at first sight. > else \ > mmap_end = DEFAULT_MAP_WINDOW; \ > mmap_end; \ > }) > > The only corner case is when sv57 is active, then only a hint greater > than VA_USER_SV57 can return a sv57 user address. Otherwise, we just > need to return the default mmap end right? > > > + > > +#define arch_get_mmap_base(addr, base) \ > > +({ \ > > + unsigned long mmap_base; \ > > + if (((addr) >= VA_USER_SV57) && (VA_BITS >= VA_BITS_SV57)) \ > > + mmap_base = (base) + (VA_USER_SV57 - DEFAULT_MAP_WINDOW); \ > > + else if ((((addr) >= VA_USER_SV48)) && (VA_BITS >= VA_BITS_SV48)) \ > > + mmap_base = (base) + (VA_USER_SV48 - DEFAULT_MAP_WINDOW); \ > > + else if ((addr) == 0) \ > > + mmap_base = (base); \ > > + else \ > > + mmap_base = (base) + (VA_USER_SV39 - DEFAULT_MAP_WINDOW); \ > > + mmap_base; \ > > +}) > > + > > From arch_pick_mmap_layout() > (https://elixir.bootlin.com/linux/latest/source/mm/util.c#L433), the > "base" argument is: > > - either STACK_TOP in top-down (more or less some random offset) > - or TASK_UNMAPPED_BASE in bottom-up (more or less some random offset) > > When bottom-up is the current mode, we should not change the base, so > adding (VA_USER_SV57 - DEFAULT_MAP_WINDOW) in the first case is not > right for me. When sv48 or sv57 are the active mode, > DEFAULT_MAP_WINDOW is equal to VA_USER_SV48 right? So (VA_USER_SV48 - > DEFAULT_MAP_WINDOW) is 0, so not useful. And for the last case, when > the user asks for a sv39 address whereas the active mode is sv48 or > sv57, then (VA_USER_SV39 - DEFAULT_MAP_WINDOW) is negative and the > base is smaller which is not correct. > In the bottom-up case mm->mmap_base is directly used instead of arch_get_mmap_base so base will not be modified in that case. The math here is confusing so I can refactor it. I like your suggestion to extract out the rnd value of base with (base) - DEFAULT_MAP_WINDOW since base is defined as PAGE_ALIGN(STACK_TOP - gap - rnd) and STACK_TOP is DEFAULT_MAP_WINDOW. This (-gap-rnd) value can then directly be used instead of redoing the math in each if. > In the bottom-up case, we should preserve the base and I think that > again, only sv57 is the corner case to deal with. > > For both this comment and the one above, I think we should allow the user to default back to sv38. I talked to Palmer and he would prefer allowing selection of sv38. Since there is no overhead, there is no need to prevent the user from doing so. > > #else > > +#define DEFAULT_MAP_WINDOW TASK_SIZE > > #define STACK_TOP_MAX TASK_SIZE > > #endif > > #define STACK_ALIGN 16 > > > > +#define STACK_TOP DEFAULT_MAP_WINDOW > > + > > +/* > > + * This decides where the kernel will search for a free chunk of vm > > + * space during mmap's. > > + */ > > +#define TASK_UNMAPPED_BASE PAGE_ALIGN(DEFAULT_MAP_WINDOW / 3) > > + > > #ifndef __ASSEMBLY__ > > > > struct task_struct; > > -- > > 2.41.0 > >
diff --git a/arch/riscv/include/asm/elf.h b/arch/riscv/include/asm/elf.h index c24280774caf..5d3368d5585c 100644 --- a/arch/riscv/include/asm/elf.h +++ b/arch/riscv/include/asm/elf.h @@ -49,7 +49,7 @@ extern bool compat_elf_check_arch(Elf32_Ehdr *hdr); * the loader. We need to make sure that it is out of the way of the program * that it will "exec", and that there is sufficient room for the brk. */ -#define ELF_ET_DYN_BASE ((TASK_SIZE / 3) * 2) +#define ELF_ET_DYN_BASE ((DEFAULT_MAP_WINDOW / 3) * 2) #ifdef CONFIG_64BIT #ifdef CONFIG_COMPAT diff --git a/arch/riscv/include/asm/pgtable.h b/arch/riscv/include/asm/pgtable.h index 75970ee2bda2..e13f5872bfe9 100644 --- a/arch/riscv/include/asm/pgtable.h +++ b/arch/riscv/include/asm/pgtable.h @@ -63,12 +63,22 @@ * position vmemmap directly below the VMALLOC region. */ #ifdef CONFIG_64BIT +#define VA_BITS_SV39 39 +#define VA_BITS_SV48 48 +#define VA_BITS_SV57 57 + +#define VA_USER_SV39 (UL(1) << (VA_BITS_SV39 - 1)) +#define VA_USER_SV48 (UL(1) << (VA_BITS_SV48 - 1)) +#define VA_USER_SV57 (UL(1) << (VA_BITS_SV57 - 1)) + #define VA_BITS (pgtable_l5_enabled ? \ - 57 : (pgtable_l4_enabled ? 48 : 39)) + VA_BITS_SV57 : (pgtable_l4_enabled ? VA_BITS_SV48 : VA_BITS_SV39)) #else #define VA_BITS 32 #endif +#define MMAP_VA_BITS ((VA_BITS >= VA_BITS_SV48) ? VA_BITS_SV48 : VA_BITS) + #define VMEMMAP_SHIFT \ (VA_BITS - PAGE_SHIFT - 1 + STRUCT_PAGE_MAX_SHIFT) #define VMEMMAP_SIZE BIT(VMEMMAP_SHIFT) diff --git a/arch/riscv/include/asm/processor.h b/arch/riscv/include/asm/processor.h index c950a8d9edef..14a5396eed3d 100644 --- a/arch/riscv/include/asm/processor.h +++ b/arch/riscv/include/asm/processor.h @@ -13,20 +13,52 @@ #include <asm/ptrace.h> -/* - * This decides where the kernel will search for a free chunk of vm - * space during mmap's. - */ -#define TASK_UNMAPPED_BASE PAGE_ALIGN(TASK_SIZE / 3) - -#define STACK_TOP TASK_SIZE #ifdef CONFIG_64BIT +#define DEFAULT_MAP_WINDOW (UL(1) << (MMAP_VA_BITS - 1)) #define STACK_TOP_MAX TASK_SIZE_64 + +#define arch_get_mmap_end(addr, len, flags) \ +({ \ + unsigned long mmap_end; \ + if ((addr) >= VA_USER_SV57) \ + mmap_end = STACK_TOP_MAX; \ + else if ((((addr) >= VA_USER_SV48)) && (VA_BITS >= VA_BITS_SV48)) \ + mmap_end = VA_USER_SV48; \ + else if ((addr) == 0) \ + mmap_end = DEFAULT_MAP_WINDOW; \ + else \ + mmap_end = VA_USER_SV39; \ + mmap_end; \ +}) + +#define arch_get_mmap_base(addr, base) \ +({ \ + unsigned long mmap_base; \ + if (((addr) >= VA_USER_SV57) && (VA_BITS >= VA_BITS_SV57)) \ + mmap_base = (base) + (VA_USER_SV57 - DEFAULT_MAP_WINDOW); \ + else if ((((addr) >= VA_USER_SV48)) && (VA_BITS >= VA_BITS_SV48)) \ + mmap_base = (base) + (VA_USER_SV48 - DEFAULT_MAP_WINDOW); \ + else if ((addr) == 0) \ + mmap_base = (base); \ + else \ + mmap_base = (base) + (VA_USER_SV39 - DEFAULT_MAP_WINDOW); \ + mmap_base; \ +}) + #else +#define DEFAULT_MAP_WINDOW TASK_SIZE #define STACK_TOP_MAX TASK_SIZE #endif #define STACK_ALIGN 16 +#define STACK_TOP DEFAULT_MAP_WINDOW + +/* + * This decides where the kernel will search for a free chunk of vm + * space during mmap's. + */ +#define TASK_UNMAPPED_BASE PAGE_ALIGN(DEFAULT_MAP_WINDOW / 3) + #ifndef __ASSEMBLY__ struct task_struct;
Make sv48 the default address space for mmap as some applications currently depend on this assumption. A hint address passed to mmap will cause the largest address space that fits entirely into the hint to be used. If the hint is less than or equal to 1<<38, an sv39 address will be used. An exception is that if the hint address is 0, then a sv48 address will be used. After an address space is completely full, the next smallest address space will be used. Signed-off-by: Charlie Jenkins <charlie@rivosinc.com> --- arch/riscv/include/asm/elf.h | 2 +- arch/riscv/include/asm/pgtable.h | 12 +++++++- arch/riscv/include/asm/processor.h | 46 +++++++++++++++++++++++++----- 3 files changed, 51 insertions(+), 9 deletions(-)