Excerpts from Christophe Leroy's message of May 3, 2021 3:11 am:
>
>
> Le 02/05/2021 à 13:00, Nicholas Piggin a écrit :
>> This reduces TLB misses by nearly 30x on a `git diff` workload on a
>> 2-node POWER9 (59,800 -> 2,100) and reduces CPU cycles by 0.54%, due
>> to vfs hashes being allocated with 2MB pages.
>>
>> Acked-by: Michael Ellerman <m...@ellerman.id.au>
>> Signed-off-by: Nicholas Piggin <npig...@gmail.com>
>> ---
>> Since v1:
>> - Don't define MODULES_VADDR which has some other side effect (e.g.,
>> ptdump).
>> - Fixed (hopefully) kbuild warning.
>> - Keep __vmalloc_node_range call on 3 lines.
>>
>> .../admin-guide/kernel-parameters.txt | 2 ++
>> arch/powerpc/Kconfig | 1 +
>> arch/powerpc/kernel/module.c | 18 +++++++++++++-----
>> 3 files changed, 16 insertions(+), 5 deletions(-)
>>
>> diff --git a/Documentation/admin-guide/kernel-parameters.txt
>> b/Documentation/admin-guide/kernel-parameters.txt
>> index 1c0a3cf6fcc9..1be38b25c485 100644
>> --- a/Documentation/admin-guide/kernel-parameters.txt
>> +++ b/Documentation/admin-guide/kernel-parameters.txt
>> @@ -3250,6 +3250,8 @@
>>
>> nohugeiomap [KNL,X86,PPC,ARM64] Disable kernel huge I/O mappings.
>>
>> + nohugevmalloc [PPC] Disable kernel huge vmalloc mappings.
>> +
>> nosmt [KNL,S390] Disable symmetric multithreading (SMT).
>> Equivalent to smt=1.
>>
>> diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
>> index 1e6230bea09d..c547a9d6a2dd 100644
>> --- a/arch/powerpc/Kconfig
>> +++ b/arch/powerpc/Kconfig
>> @@ -185,6 +185,7 @@ config PPC
>> select GENERIC_VDSO_TIME_NS
>> select HAVE_ARCH_AUDITSYSCALL
>> select HAVE_ARCH_HUGE_VMAP if PPC_BOOK3S_64 &&
>> PPC_RADIX_MMU
>> + select HAVE_ARCH_HUGE_VMALLOC if HAVE_ARCH_HUGE_VMAP
>> select HAVE_ARCH_JUMP_LABEL
>> select HAVE_ARCH_JUMP_LABEL_RELATIVE
>> select HAVE_ARCH_KASAN if PPC32 && PPC_PAGE_SHIFT <= 14
>> diff --git a/arch/powerpc/kernel/module.c b/arch/powerpc/kernel/module.c
>> index fab84024650c..ea1fa55a6897 100644
>> --- a/arch/powerpc/kernel/module.c
>> +++ b/arch/powerpc/kernel/module.c
>> @@ -8,6 +8,7 @@
>> #include <linux/moduleloader.h>
>> #include <linux/err.h>
>> #include <linux/vmalloc.h>
>> +#include <linux/mm.h>
>> #include <linux/bug.h>
>> #include <asm/module.h>
>> #include <linux/uaccess.h>
>> @@ -88,17 +89,22 @@ int module_finalize(const Elf_Ehdr *hdr,
>> return 0;
>> }
>>
>> -#ifdef MODULES_VADDR
>> static __always_inline void *
>> __module_alloc(unsigned long size, unsigned long start, unsigned long end)
>> {
>> - return __vmalloc_node_range(size, 1, start, end, GFP_KERNEL,
>> - PAGE_KERNEL_EXEC, VM_FLUSH_RESET_PERMS,
>> NUMA_NO_NODE,
>> - __builtin_return_address(0));
>> + /*
>> + * Don't do huge page allocations for modules yet until more testing
>> + * is done. STRICT_MODULE_RWX may require extra work to support this
>> + * too.
>> + */
>> + return __vmalloc_node_range(size, 1, start, end, GFP_KERNEL,
>> PAGE_KERNEL_EXEC,
>> + VM_FLUSH_RESET_PERMS | VM_NO_HUGE_VMAP,
>> + NUMA_NO_NODE, __builtin_return_address(0));
>> }
>>
>> void *module_alloc(unsigned long size)
>> {
>> +#ifdef CONFIG_PPC32
>
> What then happens to PPC32 platforms that doesn't define MODULES_VADDR, for
> instance 4xx or booke ?
>
> I think it should be:
>
> #ifdef MODULES_VADDR
Yes the kernel build robot agrees with you. I'll respin.
Thanks,
Nick
>
>> unsigned long limit = (unsigned long)_etext - SZ_32M;
>> void *ptr = NULL;
>>
>> @@ -112,5 +118,7 @@ void *module_alloc(unsigned long size)
>> ptr = __module_alloc(size, MODULES_VADDR, MODULES_END);
>>
>> return ptr;
>> -}
>> +#else
>> + return __module_alloc(size, VMALLOC_START, VMALLOC_END);
>> #endif
>> +}
>>
>
