linux/arch/riscv/mm/tlbflush.c

206 lines
5 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
#include <linux/mm.h>
#include <linux/smp.h>
#include <linux/sched.h>
#include <linux/hugetlb.h>
#include <asm/sbi.h>
#include <asm/mmu_context.h>
/*
* Flush entire TLB if number of entries to be flushed is greater
* than the threshold below.
*/
unsigned long tlb_flush_all_threshold __read_mostly = 64;
static void local_flush_tlb_range_threshold_asid(unsigned long start,
unsigned long size,
unsigned long stride,
unsigned long asid)
{
unsigned long nr_ptes_in_range = DIV_ROUND_UP(size, stride);
int i;
if (nr_ptes_in_range > tlb_flush_all_threshold) {
local_flush_tlb_all_asid(asid);
return;
}
for (i = 0; i < nr_ptes_in_range; ++i) {
local_flush_tlb_page_asid(start, asid);
start += stride;
}
}
static inline void local_flush_tlb_range_asid(unsigned long start,
unsigned long size, unsigned long stride, unsigned long asid)
{
if (size <= stride)
local_flush_tlb_page_asid(start, asid);
else if (size == FLUSH_TLB_MAX_SIZE)
local_flush_tlb_all_asid(asid);
else
local_flush_tlb_range_threshold_asid(start, size, stride, asid);
}
/* Flush a range of kernel pages without broadcasting */
void local_flush_tlb_kernel_range(unsigned long start, unsigned long end)
{
local_flush_tlb_range_asid(start, end - start, PAGE_SIZE, FLUSH_TLB_NO_ASID);
}
static void __ipi_flush_tlb_all(void *info)
{
local_flush_tlb_all();
}
void flush_tlb_all(void)
{
if (num_online_cpus() < 2)
local_flush_tlb_all();
else if (riscv_use_sbi_for_rfence())
sbi_remote_sfence_vma_asid(NULL, 0, FLUSH_TLB_MAX_SIZE, FLUSH_TLB_NO_ASID);
riscv: Use IPIs for remote cache/TLB flushes by default An IPI backend is always required in an SMP configuration, but an SBI implementation is not. For example, SBI will be unavailable when the kernel runs in M mode. For this reason, consider IPI delivery of cache and TLB flushes to be the base case, and any other implementation (such as the SBI remote fence extension) to be an optimization. Generally, if IPIs can be delivered without firmware assistance, they are assumed to be faster than SBI calls due to the SBI context switch overhead. However, when SBI is used as the IPI backend, then the context switch cost must be paid anyway, and performing the cache/TLB flush directly in the SBI implementation is more efficient than injecting an interrupt to S-mode. This is the only existing scenario where riscv_ipi_set_virq_range() is called with use_for_rfence set to false. sbi_ipi_init() already checks riscv_ipi_have_virq_range(), so it only calls riscv_ipi_set_virq_range() when no other IPI device is available. This allows moving the static key and dropping the use_for_rfence parameter. This decouples the static key from the irqchip driver probe order. Furthermore, the static branch only makes sense when CONFIG_RISCV_SBI is enabled. Optherwise, IPIs must be used. Add a fallback definition of riscv_use_sbi_for_rfence() which handles this case and removes the need to check CONFIG_RISCV_SBI elsewhere, such as in cacheflush.c. Reviewed-by: Anup Patel <anup@brainfault.org> Signed-off-by: Samuel Holland <samuel.holland@sifive.com> Reviewed-by: Alexandre Ghiti <alexghiti@rivosinc.com> Link: https://lore.kernel.org/r/20240327045035.368512-4-samuel.holland@sifive.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-03-26 21:49:44 -07:00
else
on_each_cpu(__ipi_flush_tlb_all, NULL, 1);
}
struct flush_tlb_range_data {
unsigned long asid;
unsigned long start;
unsigned long size;
unsigned long stride;
};
static void __ipi_flush_tlb_range_asid(void *info)
{
struct flush_tlb_range_data *d = info;
local_flush_tlb_range_asid(d->start, d->size, d->stride, d->asid);
}
static void __flush_tlb_range(const struct cpumask *cmask, unsigned long asid,
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
unsigned long start, unsigned long size,
unsigned long stride)
{
unsigned int cpu;
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
if (cpumask_empty(cmask))
return;
cpu = get_cpu();
/* Check if the TLB flush needs to be sent to other CPUs. */
if (cpumask_any_but(cmask, cpu) >= nr_cpu_ids) {
local_flush_tlb_range_asid(start, size, stride, asid);
riscv: Use IPIs for remote cache/TLB flushes by default An IPI backend is always required in an SMP configuration, but an SBI implementation is not. For example, SBI will be unavailable when the kernel runs in M mode. For this reason, consider IPI delivery of cache and TLB flushes to be the base case, and any other implementation (such as the SBI remote fence extension) to be an optimization. Generally, if IPIs can be delivered without firmware assistance, they are assumed to be faster than SBI calls due to the SBI context switch overhead. However, when SBI is used as the IPI backend, then the context switch cost must be paid anyway, and performing the cache/TLB flush directly in the SBI implementation is more efficient than injecting an interrupt to S-mode. This is the only existing scenario where riscv_ipi_set_virq_range() is called with use_for_rfence set to false. sbi_ipi_init() already checks riscv_ipi_have_virq_range(), so it only calls riscv_ipi_set_virq_range() when no other IPI device is available. This allows moving the static key and dropping the use_for_rfence parameter. This decouples the static key from the irqchip driver probe order. Furthermore, the static branch only makes sense when CONFIG_RISCV_SBI is enabled. Optherwise, IPIs must be used. Add a fallback definition of riscv_use_sbi_for_rfence() which handles this case and removes the need to check CONFIG_RISCV_SBI elsewhere, such as in cacheflush.c. Reviewed-by: Anup Patel <anup@brainfault.org> Signed-off-by: Samuel Holland <samuel.holland@sifive.com> Reviewed-by: Alexandre Ghiti <alexghiti@rivosinc.com> Link: https://lore.kernel.org/r/20240327045035.368512-4-samuel.holland@sifive.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-03-26 21:49:44 -07:00
} else if (riscv_use_sbi_for_rfence()) {
sbi_remote_sfence_vma_asid(cmask, start, size, asid);
} else {
riscv: Use IPIs for remote cache/TLB flushes by default An IPI backend is always required in an SMP configuration, but an SBI implementation is not. For example, SBI will be unavailable when the kernel runs in M mode. For this reason, consider IPI delivery of cache and TLB flushes to be the base case, and any other implementation (such as the SBI remote fence extension) to be an optimization. Generally, if IPIs can be delivered without firmware assistance, they are assumed to be faster than SBI calls due to the SBI context switch overhead. However, when SBI is used as the IPI backend, then the context switch cost must be paid anyway, and performing the cache/TLB flush directly in the SBI implementation is more efficient than injecting an interrupt to S-mode. This is the only existing scenario where riscv_ipi_set_virq_range() is called with use_for_rfence set to false. sbi_ipi_init() already checks riscv_ipi_have_virq_range(), so it only calls riscv_ipi_set_virq_range() when no other IPI device is available. This allows moving the static key and dropping the use_for_rfence parameter. This decouples the static key from the irqchip driver probe order. Furthermore, the static branch only makes sense when CONFIG_RISCV_SBI is enabled. Optherwise, IPIs must be used. Add a fallback definition of riscv_use_sbi_for_rfence() which handles this case and removes the need to check CONFIG_RISCV_SBI elsewhere, such as in cacheflush.c. Reviewed-by: Anup Patel <anup@brainfault.org> Signed-off-by: Samuel Holland <samuel.holland@sifive.com> Reviewed-by: Alexandre Ghiti <alexghiti@rivosinc.com> Link: https://lore.kernel.org/r/20240327045035.368512-4-samuel.holland@sifive.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-03-26 21:49:44 -07:00
struct flush_tlb_range_data ftd;
riscv: Use IPIs for remote cache/TLB flushes by default An IPI backend is always required in an SMP configuration, but an SBI implementation is not. For example, SBI will be unavailable when the kernel runs in M mode. For this reason, consider IPI delivery of cache and TLB flushes to be the base case, and any other implementation (such as the SBI remote fence extension) to be an optimization. Generally, if IPIs can be delivered without firmware assistance, they are assumed to be faster than SBI calls due to the SBI context switch overhead. However, when SBI is used as the IPI backend, then the context switch cost must be paid anyway, and performing the cache/TLB flush directly in the SBI implementation is more efficient than injecting an interrupt to S-mode. This is the only existing scenario where riscv_ipi_set_virq_range() is called with use_for_rfence set to false. sbi_ipi_init() already checks riscv_ipi_have_virq_range(), so it only calls riscv_ipi_set_virq_range() when no other IPI device is available. This allows moving the static key and dropping the use_for_rfence parameter. This decouples the static key from the irqchip driver probe order. Furthermore, the static branch only makes sense when CONFIG_RISCV_SBI is enabled. Optherwise, IPIs must be used. Add a fallback definition of riscv_use_sbi_for_rfence() which handles this case and removes the need to check CONFIG_RISCV_SBI elsewhere, such as in cacheflush.c. Reviewed-by: Anup Patel <anup@brainfault.org> Signed-off-by: Samuel Holland <samuel.holland@sifive.com> Reviewed-by: Alexandre Ghiti <alexghiti@rivosinc.com> Link: https://lore.kernel.org/r/20240327045035.368512-4-samuel.holland@sifive.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-03-26 21:49:44 -07:00
ftd.asid = asid;
ftd.start = start;
ftd.size = size;
ftd.stride = stride;
on_each_cpu_mask(cmask, __ipi_flush_tlb_range_asid, &ftd, 1);
}
put_cpu();
}
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
static inline unsigned long get_mm_asid(struct mm_struct *mm)
{
return cntx2asid(atomic_long_read(&mm->context.id));
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
}
void flush_tlb_mm(struct mm_struct *mm)
{
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
__flush_tlb_range(mm_cpumask(mm), get_mm_asid(mm),
0, FLUSH_TLB_MAX_SIZE, PAGE_SIZE);
}
void flush_tlb_mm_range(struct mm_struct *mm,
unsigned long start, unsigned long end,
unsigned int page_size)
{
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
__flush_tlb_range(mm_cpumask(mm), get_mm_asid(mm),
start, end - start, page_size);
}
void flush_tlb_page(struct vm_area_struct *vma, unsigned long addr)
{
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
__flush_tlb_range(mm_cpumask(vma->vm_mm), get_mm_asid(vma->vm_mm),
addr, PAGE_SIZE, PAGE_SIZE);
}
void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
unsigned long end)
{
unsigned long stride_size;
if (!is_vm_hugetlb_page(vma)) {
stride_size = PAGE_SIZE;
} else {
stride_size = huge_page_size(hstate_vma(vma));
/*
* As stated in the privileged specification, every PTE in a
* NAPOT region must be invalidated, so reset the stride in that
* case.
*/
if (has_svnapot()) {
if (stride_size >= PGDIR_SIZE)
stride_size = PGDIR_SIZE;
else if (stride_size >= P4D_SIZE)
stride_size = P4D_SIZE;
else if (stride_size >= PUD_SIZE)
stride_size = PUD_SIZE;
else if (stride_size >= PMD_SIZE)
stride_size = PMD_SIZE;
else
stride_size = PAGE_SIZE;
}
}
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
__flush_tlb_range(mm_cpumask(vma->vm_mm), get_mm_asid(vma->vm_mm),
start, end - start, stride_size);
}
void flush_tlb_kernel_range(unsigned long start, unsigned long end)
{
__flush_tlb_range(cpu_online_mask, FLUSH_TLB_NO_ASID,
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
start, end - start, PAGE_SIZE);
}
#ifdef CONFIG_TRANSPARENT_HUGEPAGE
void flush_pmd_tlb_range(struct vm_area_struct *vma, unsigned long start,
unsigned long end)
{
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
__flush_tlb_range(mm_cpumask(vma->vm_mm), get_mm_asid(vma->vm_mm),
start, end - start, PMD_SIZE);
}
#endif
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
bool arch_tlbbatch_should_defer(struct mm_struct *mm)
{
return true;
}
void arch_tlbbatch_add_pending(struct arch_tlbflush_unmap_batch *batch,
struct mm_struct *mm,
unsigned long uaddr)
{
cpumask_or(&batch->cpumask, &batch->cpumask, mm_cpumask(mm));
}
void arch_flush_tlb_batched_pending(struct mm_struct *mm)
{
flush_tlb_mm(mm);
}
void arch_tlbbatch_flush(struct arch_tlbflush_unmap_batch *batch)
{
__flush_tlb_range(&batch->cpumask, FLUSH_TLB_NO_ASID, 0,
FLUSH_TLB_MAX_SIZE, PAGE_SIZE);
cpumask_clear(&batch->cpumask);
riscv: Add support for BATCHED_UNMAP_TLB_FLUSH Allow to defer the flushing of the TLB when unmapping pages, which allows to reduce the numbers of IPI and the number of sfence.vma. The ubenchmarch used in commit 43b3dfdd0455 ("arm64: support batched/deferred tlb shootdown during page reclamation/migration") that was multithreaded to force the usage of IPI shows good performance improvement on all platforms: * Unmatched: ~34% * TH1520 : ~78% * Qemu : ~81% In addition, perf on qemu reports an important decrease in time spent dealing with IPIs: Before: 68.17% main [kernel.kallsyms] [k] __sbi_rfence_v02_call After : 8.64% main [kernel.kallsyms] [k] __sbi_rfence_v02_call * Benchmark: int stick_this_thread_to_core(int core_id) { int num_cores = sysconf(_SC_NPROCESSORS_ONLN); if (core_id < 0 || core_id >= num_cores) return EINVAL; cpu_set_t cpuset; CPU_ZERO(&cpuset); CPU_SET(core_id, &cpuset); pthread_t current_thread = pthread_self(); return pthread_setaffinity_np(current_thread, sizeof(cpu_set_t), &cpuset); } static void *fn_thread (void *p_data) { int ret; pthread_t thread; stick_this_thread_to_core((int)p_data); while (1) { sleep(1); } return NULL; } int main() { volatile unsigned char *p = mmap(NULL, SIZE, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_ANONYMOUS, -1, 0); pthread_t threads[4]; int ret; for (int i = 0; i < 4; ++i) { ret = pthread_create(&threads[i], NULL, fn_thread, (void *)i); if (ret) { printf("%s", strerror (ret)); } } memset(p, 0x88, SIZE); for (int k = 0; k < 10000; k++) { /* swap in */ for (int i = 0; i < SIZE; i += 4096) { (void)p[i]; } /* swap out */ madvise(p, SIZE, MADV_PAGEOUT); } for (int i = 0; i < 4; i++) { pthread_cancel(threads[i]); } for (int i = 0; i < 4; i++) { pthread_join(threads[i], NULL); } return 0; } Signed-off-by: Alexandre Ghiti <alexghiti@rivosinc.com> Reviewed-by: Jisheng Zhang <jszhang@kernel.org> Tested-by: Jisheng Zhang <jszhang@kernel.org> # Tested on TH1520 Tested-by: Nam Cao <namcao@linutronix.de> Link: https://lore.kernel.org/r/20240108193640.344929-1-alexghiti@rivosinc.com Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>
2024-01-08 20:36:40 +01:00
}