euleros inclusion category: feature feature: a tiered memory manager that tracks page hotness and migrate them between tiers of memory bugzilla: https://gitee.com/openeuler/kernel/issues/I8T1MB?from=project-issue CVE: NA

-------------------------------------------------

This patch proposes the etmem feature to openEuler.

etmem aims at solving the DRAM shortage issue that we are facing with the emergence of memory intensive workloads.

With etmem, we can replace DRAM with less costly tiers of memory, for example, AEP, without significant performance loss. etmem tracks page hotness and migrate them between tiers of memory, so that, ideally, we can exploit fast memory tier for high performance and slower memory tier for large capacity.

etmem has two kernel space components: etmem scan and etmem swap, the former for page hotness tracking and the latter for migration.

Signed-off-by: Yuchen Tang tangyuchen5@huawei.com --- mm/Kconfig | 12 ++++++++++++ 1 file changed, 12 insertions(+)

diff --git a/mm/Kconfig b/mm/Kconfig index c277bb069ab7..7493fe4ce8da 100644 --- a/mm/Kconfig +++ b/mm/Kconfig @@ -1348,6 +1348,18 @@ config CLEAR_FREELIST_PAGE To enable this feature, kernel parameter "clear_freelist" also needs to be added.

+config ETMEM + bool "Enable etmem feature" + depends on MMU + depends on X86 || ARM64 + default n + help + etmem is a tiered memory extension technology that uses DRAM and memory + compression/high-performance storage media to form tiered memory storage. + Memory data is tiered, and cold data is migrated from memory media to + high-performance storage media to release memory space and reduce + memory costs. + source "mm/damon/Kconfig"

endmenu

euleros inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I8T1MB?from=project-issue CVE: NA

-------------------------------------------------

Export __pud_trans_huge_lock so that etmem_scan can exploit this for page table walk.

__pud_trans_huge_lock returns page table lock pointer if a given pud maps a transparent huge page, which will be useful when encountering a pud entry during the page table walk in etmem scan module.

Signed-off-by: Yuchen Tang tangyuchen5@huawei.com --- mm/huge_memory.c | 1 + 1 file changed, 1 insertion(+)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c index a13418df1115..8c8da505739f 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -1963,6 +1963,7 @@ spinlock_t *__pud_trans_huge_lock(pud_t *pud, struct vm_area_struct *vma) spin_unlock(ptl); return NULL; } +EXPORT_SYMBOL_GPL(__pud_trans_huge_lock);

#ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD int zap_huge_pud(struct mmu_gather *tlb, struct vm_area_struct *vma,

euleros inclusion category: feature feature: Classify page hotness through page table scanning bugzilla: https://gitee.com/openeuler/kernel/issues/I8T1MB?from=project-issue CVE: NA

-------------------------------------------------

This patch implements the etmem feature to openEuler.

etmem scan module communicates with the user space program through registered proc file system. It periodically scans the vma segments of the target process, by walking its page table and check access bit of each page, before reporting the scan results to user space, so that we can better classify hotness of pages and further migrate hot ones to fast memory tier and cold ones to slow memory tier.

Signed-off-by: yanxiaodan yanxiaodan@huawei.com Signed-off-by: Feilong Lin linfeilong@huawei.com Signed-off-by: geruijun geruijun@huawei.com Signed-off-by: liubo liubo254@huawei.com Signed-off-by: Yuchen Tang tangyuchen5@huawei.com --- fs/proc/Makefile | 2 + fs/proc/base.c | 6 + fs/proc/etmem_proc.c | 122 +++ fs/proc/etmem_scan.c | 1383 ++++++++++++++++++++++++ fs/proc/etmem_scan.h | 149 +++ fs/proc/internal.h | 3 + include/linux/etmem.h | 27 + include/linux/mm.h | 4 + include/linux/mm_types.h | 5 + include/uapi/asm-generic/mman-common.h | 4 + mm/madvise.c | 16 + virt/kvm/kvm_main.c | 8 + 12 files changed, 1729 insertions(+) create mode 100644 fs/proc/etmem_proc.c create mode 100644 fs/proc/etmem_scan.c create mode 100644 fs/proc/etmem_scan.h create mode 100644 include/linux/etmem.h

diff --git a/fs/proc/Makefile b/fs/proc/Makefile index bd08616ed8ba..b9a7bc7d8a75 100644 --- a/fs/proc/Makefile +++ b/fs/proc/Makefile @@ -34,3 +34,5 @@ proc-$(CONFIG_PROC_VMCORE) += vmcore.o proc-$(CONFIG_PRINTK) += kmsg.o proc-$(CONFIG_PROC_PAGE_MONITOR) += page.o proc-$(CONFIG_BOOT_CONFIG) += bootconfig.o +obj-$(CONFIG_ETMEM_SCAN) += etmem_scan.o +proc-${CONFIG_ETMEM} += etmem_proc.o diff --git a/fs/proc/base.c b/fs/proc/base.c index 243c15919e18..eb5ab83e5c22 100644 --- a/fs/proc/base.c +++ b/fs/proc/base.c @@ -3355,6 +3355,9 @@ static const struct pid_entry tgid_base_stuff[] = { REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations), REG("pagemap", S_IRUSR, proc_pagemap_operations), #endif +#ifdef CONFIG_ETMEM + REG("idle_pages", S_IRUSR|S_IWUSR, proc_mm_idle_operations), +#endif #ifdef CONFIG_SECURITY DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations), #endif @@ -3704,6 +3707,9 @@ static const struct pid_entry tid_base_stuff[] = { REG("smaps_rollup", S_IRUGO, proc_pid_smaps_rollup_operations), REG("pagemap", S_IRUSR, proc_pagemap_operations), #endif +#ifdef CONFIG_ETMEM + REG("idle_pages", S_IRUSR|S_IWUSR, proc_mm_idle_operations), +#endif #ifdef CONFIG_SECURITY DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations), #endif diff --git a/fs/proc/etmem_proc.c b/fs/proc/etmem_proc.c new file mode 100644 index 000000000000..edacb9260345 --- /dev/null +++ b/fs/proc/etmem_proc.c @@ -0,0 +1,122 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <linux/seq_file.h> +#include <linux/pagemap.h> +#include <linux/mempolicy.h> +#include <linux/swap.h> +#include <linux/sched/mm.h> +#include <linux/mmu_notifier.h> +#include <linux/page_idle.h> +#include <linux/uaccess.h> +#include <linux/pkeys.h> +#include <linux/module.h> +#include <linux/ptrace.h> +#include <linux/etmem.h> + +#include "internal.h" + +static DEFINE_SPINLOCK(scan_lock); + +static int page_scan_lock(struct file *file, int is_lock, struct file_lock *flock) +{ + if (is_lock) + spin_lock(&scan_lock); + else + spin_unlock(&scan_lock); + + return 0; +} + +/* will be filled when kvm_ept_idle module loads */ +struct file_operations proc_page_scan_operations = { + .flock = page_scan_lock, +}; +EXPORT_SYMBOL_GPL(proc_page_scan_operations); + +static ssize_t mm_idle_read(struct file *file, char __user *buf, + size_t count, loff_t *ppos) +{ + struct mm_struct *mm = file->private_data; + int ret = 0; + + if (!mm || !mmget_not_zero(mm)) { + ret = -ESRCH; + return ret; + } + if (proc_page_scan_operations.read) + ret = proc_page_scan_operations.read(file, buf, count, ppos); + + mmput(mm); + return ret; +} + +static int mm_idle_open(struct inode *inode, struct file *file) +{ + struct mm_struct *mm = NULL; + struct module *module = NULL; + int ret = -1; + + if (!file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN)) + return -EPERM; + + page_scan_lock(NULL, 1, NULL); + module = proc_page_scan_operations.owner; + if (module != NULL && try_module_get(module)) + ret = 0; + page_scan_lock(NULL, 0, NULL); + if (ret != 0) { + /* no scan ko installed, avoid to return valid file */ + return -ENODEV; + } + + mm = proc_mem_open(inode, PTRACE_MODE_READ); + if (IS_ERR(mm)) { + module_put(module); + return PTR_ERR(mm); + } + + file->private_data = mm; + + if (proc_page_scan_operations.open) + ret = proc_page_scan_operations.open(inode, file); + + if (ret != 0) + module_put(module); + + return ret; +} + +static int mm_idle_release(struct inode *inode, struct file *file) +{ + struct mm_struct *mm = file->private_data; + int ret = 0; + + if (mm) { + if (!mm_kvm(mm)) + flush_tlb_mm(mm); + mmdrop(mm); + } + + if (proc_page_scan_operations.release) + ret = proc_page_scan_operations.release(inode, file); + + if (proc_page_scan_operations.owner) + module_put(proc_page_scan_operations.owner); + + return ret; +} + +static long mm_idle_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) +{ + if (proc_page_scan_operations.unlocked_ioctl) + return proc_page_scan_operations.unlocked_ioctl(filp, cmd, arg); + + return 0; +} + +const struct file_operations proc_mm_idle_operations = { + .llseek = mem_lseek, /* borrow this */ + .read = mm_idle_read, + .open = mm_idle_open, + .release = mm_idle_release, + .unlocked_ioctl = mm_idle_ioctl, +}; diff --git a/fs/proc/etmem_scan.c b/fs/proc/etmem_scan.c new file mode 100644 index 000000000000..1b5436d50e47 --- /dev/null +++ b/fs/proc/etmem_scan.c @@ -0,0 +1,1383 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <linux/pagemap.h> +#include <linux/mm.h> +#include <linux/hugetlb.h> +#include <linux/kernel.h> +#include <linux/sched.h> +#include <linux/proc_fs.h> +#include <linux/uaccess.h> +#include <linux/kvm.h> +#include <linux/kvm_host.h> +#include <linux/bitmap.h> +#include <linux/sched/mm.h> +#include <linux/version.h> +#include <linux/module.h> +#include <linux/io.h> +#include <linux/pagewalk.h> +#include <linux/uaccess.h> +#include <asm/cacheflush.h> +#include <asm/page.h> +#include <asm/pgalloc.h> +#include <asm/pgtable.h> +#include <linux/huge_mm.h> +#include <linux/etmem.h> +#ifdef CONFIG_ARM64 +#include <asm/pgtable-types.h> +#include <asm/memory.h> +#include <asm/kvm_mmu.h> +#include <asm/kvm_arm.h> +#include <asm/stage2_pgtable.h> +#endif +#include "etmem_scan.h" +#include <linux/hugetlb_inline.h> + +#ifdef CONFIG_X86_64 +/* + * Fallback to false for kernel doens't support KVM_INVALID_SPTE + * ept_idle can sitll work in this situation but the scan accuracy may drop, + * depends on the access frequences of the workload. + */ +#ifdef KVM_INVALID_SPTE +#define KVM_CHECK_INVALID_SPTE(val) ((val) == KVM_INVALID_SPTE) +#else +#define KVM_CHECK_INVALID_SPTE(val) (0) +#endif + +# define kvm_arch_mmu_pointer(vcpu) (vcpu->arch.mmu) +# define kvm_mmu_ad_disabled(mmu) (mmu->cpu_role.base.ad_disabled) +#endif /*CONFIG_X86_64*/ + +#ifdef CONFIG_ARM64 +#define if_pmd_thp_or_huge(pmd) (if_pmd_huge(pmd) || pmd_trans_huge(pmd)) +#endif /* CONFIG_ARM64 */ + +#ifdef DEBUG + +#define debug_printk trace_printk + +#define set_restart_gpa(val, note) ({ \ + unsigned long old_val = pic->restart_gpa; \ + pic->restart_gpa = (val); \ + trace_printk("restart_gpa=%lx %luK %s %s %d\n", \ + (val), (pic->restart_gpa - old_val) >> 10, \ + note, __func__, __LINE__); \ +}) + +#define set_next_hva(val, note) ({ \ + unsigned long old_val = pic->next_hva; \ + pic->next_hva = (val); \ + trace_printk(" next_hva=%lx %luK %s %s %d\n", \ + (val), (pic->next_hva - old_val) >> 10, \ + note, __func__, __LINE__); \ +}) + +#else + +#define debug_printk(...) + +#define set_restart_gpa(val, note) ({ \ + pic->restart_gpa = (val); \ +}) + +#define set_next_hva(val, note) ({ \ + pic->next_hva = (val); \ +}) + +#endif + +#define RET_RESCAN_FLAG 0x10000 + +/* error return IDLE_PAGE_TYPE_MAX or return valid page type */ +enum ProcIdlePageType (*vm_handle_pte_hole)(unsigned long addr, + unsigned long next, int depth, struct mm_walk *walk) = NULL; +EXPORT_SYMBOL_GPL(vm_handle_pte_hole); + +static int set_walk_step(const char *val, const struct kernel_param *kp) +{ + int ret; + unsigned int n; + + ret = kstrtouint(val, 0, &n); + if (ret != 0 || n == 0) + return -EINVAL; + + return param_set_uint(val, kp); +} + +static struct kernel_param_ops walk_step_ops = { + .set = set_walk_step, + .get = param_get_uint, +}; + +static unsigned int __read_mostly walk_step = 512; // in PAGE_SIZE +module_param_cb(walk_step, &walk_step_ops, &walk_step, 0644); + +static unsigned int resched_step = 10; +module_param(resched_step, uint, 0644); + +static unsigned long pagetype_size[16] = { + [PTE_ACCESSED] = PAGE_SIZE, /* 4k page */ + [PMD_ACCESSED] = PMD_SIZE, /* 2M page */ + [PUD_PRESENT] = PUD_SIZE, /* 1G page */ + + [PTE_DIRTY_M] = PAGE_SIZE, + [PMD_DIRTY_M] = PMD_SIZE, + + [PTE_IDLE] = PAGE_SIZE, + [PMD_IDLE] = PMD_SIZE, + [PMD_IDLE_PTES] = PMD_SIZE, + + [PTE_HOLE] = PAGE_SIZE, + [PMD_HOLE] = PMD_SIZE, +}; + +static void u64_to_u8(uint64_t n, uint8_t *p) +{ + p += sizeof(uint64_t) - 1; + + *p-- = n; n >>= 8; + *p-- = n; n >>= 8; + *p-- = n; n >>= 8; + *p-- = n; n >>= 8; + + *p-- = n; n >>= 8; + *p-- = n; n >>= 8; + *p-- = n; n >>= 8; + *p = n; +} + +static void dump_pic(struct page_idle_ctrl *pic) +{ + debug_printk("page_idle_ctrl: pie_read=%d pie_read_max=%d", + pic->pie_read, + pic->pie_read_max); + debug_printk(" buf_size=%d bytes_copied=%d next_hva=%pK", + pic->buf_size, + pic->bytes_copied, + pic->next_hva); + debug_printk(" restart_gpa=%pK pa_to_hva=%pK\n", + pic->restart_gpa, + pic->gpa_to_hva); +} + +#ifdef CONFIG_ARM64 +static int if_pmd_huge(pmd_t pmd) +{ + return pmd_val(pmd) && !(pmd_val(pmd) & PMD_TABLE_BIT); +} + +static int if_pud_huge(pud_t pud) +{ +#ifndef __PAGETABLE_PMD_FOLDED + return pud_val(pud) && !(pud_val(pud) & PUD_TABLE_BIT); +#else + return 0; +#endif +} +#endif + +static void pic_report_addr(struct page_idle_ctrl *pic, unsigned long addr) +{ + unsigned long hva; + + pic->kpie[pic->pie_read++] = PIP_CMD_SET_HVA; + hva = addr; + u64_to_u8(hva, &pic->kpie[pic->pie_read]); + pic->pie_read += sizeof(uint64_t); + dump_pic(pic); +} + +static int pic_add_page(struct page_idle_ctrl *pic, + unsigned long addr, + unsigned long next, + enum ProcIdlePageType page_type) +{ + unsigned long page_size = pagetype_size[page_type]; + + dump_pic(pic); + + /* align kernel/user vision of cursor position */ + next = round_up(next, page_size); + + if (!pic->pie_read || + addr + pic->gpa_to_hva != pic->next_hva) { + /* merge hole */ + if (page_type == PTE_HOLE || + page_type == PMD_HOLE) { + set_restart_gpa(next, "PTE_HOLE|PMD_HOLE"); + return 0; + } + + if (addr + pic->gpa_to_hva < pic->next_hva) { + debug_printk("page_idle: addr moves backwards\n"); + WARN_ONCE(1, "page_idle: addr moves backwards"); + } + + if (pic->pie_read + sizeof(uint64_t) + 2 >= pic->pie_read_max) { + set_restart_gpa(addr, "PAGE_IDLE_KBUF_FULL"); + return PAGE_IDLE_KBUF_FULL; + } + + pic_report_addr(pic, round_down(addr, page_size) + + pic->gpa_to_hva); + } else { + if (PIP_TYPE(pic->kpie[pic->pie_read - 1]) == page_type && + PIP_SIZE(pic->kpie[pic->pie_read - 1]) < 0xF) { + set_next_hva(next + pic->gpa_to_hva, "IN-PLACE INC"); + set_restart_gpa(next, "IN-PLACE INC"); + pic->kpie[pic->pie_read - 1]++; + WARN_ONCE(page_size < next-addr, "next-addr too large"); + return 0; + } + if (pic->pie_read >= pic->pie_read_max) { + set_restart_gpa(addr, "PAGE_IDLE_KBUF_FULL"); + return PAGE_IDLE_KBUF_FULL; + } + } + + set_next_hva(next + pic->gpa_to_hva, "NEW-ITEM"); + set_restart_gpa(next, "NEW-ITEM"); + pic->kpie[pic->pie_read] = PIP_COMPOSE(page_type, 1); + pic->pie_read++; + + return 0; +} + +static int init_page_idle_ctrl_buffer(struct page_idle_ctrl *pic) +{ + pic->pie_read = 0; + pic->pie_read_max = min(PAGE_IDLE_KBUF_SIZE, + pic->buf_size - pic->bytes_copied); + /* reserve space for PIP_CMD_SET_HVA in the end */ + pic->pie_read_max -= sizeof(uint64_t) + 1; + + /* + * Align with PAGE_IDLE_KBUF_FULL + * logic in pic_add_page(), to avoid pic->pie_read = 0 when + * PAGE_IDLE_KBUF_FULL happened. + */ + if (pic->pie_read_max <= sizeof(uint64_t) + 2) + return PAGE_IDLE_KBUF_FULL; + + memset(pic->kpie, 0, sizeof(pic->kpie)); + return 0; +} + +static void setup_page_idle_ctrl(struct page_idle_ctrl *pic, void *buf, + int buf_size, unsigned int flags) +{ + pic->buf = buf; + pic->buf_size = buf_size; + pic->bytes_copied = 0; + pic->next_hva = 0; + pic->gpa_to_hva = 0; + pic->restart_gpa = 0; + pic->last_va = 0; + pic->flags = flags; +} + +static int page_idle_copy_user(struct page_idle_ctrl *pic, + unsigned long start, unsigned long end) +{ + int bytes_read; + int ret; + + dump_pic(pic); + + bytes_read = pic->pie_read; + if (!bytes_read) + return 0; + + ret = copy_to_user(pic->buf, pic->kpie, bytes_read); + if (ret) + return -EFAULT; + + pic->buf += bytes_read; + pic->bytes_copied += bytes_read; + if (pic->bytes_copied >= pic->buf_size) + return PAGE_IDLE_BUF_FULL; + + ret = init_page_idle_ctrl_buffer(pic); + if (ret) + return ret; + + cond_resched(); + return 0; +} + +#ifdef CONFIG_X86_64 +static int vm_walk_host_range(unsigned long long start, + unsigned long end, + struct mm_walk *walk) +{ + int ret; + struct page_idle_ctrl *pic = walk->private; + unsigned long tmp_gpa_to_hva = pic->gpa_to_hva; + + pic->gpa_to_hva = 0; + read_unlock(&pic->kvm->mmu_lock); + mmap_read_lock(walk->mm); + local_irq_disable(); + ret = walk_page_range(walk->mm, start + tmp_gpa_to_hva, end + tmp_gpa_to_hva, + walk->ops, walk->private); + local_irq_enable(); + mmap_read_unlock(walk->mm); + pic->gpa_to_hva = tmp_gpa_to_hva; + if (pic->flags & VM_SCAN_HOST) { + pic->restart_gpa -= tmp_gpa_to_hva; + pic->flags &= ~VM_SCAN_HOST; + } + if (ret != PAGE_IDLE_KBUF_FULL && end > pic->restart_gpa) + pic->restart_gpa = end; + + /* ept page table may change after spin_unlock, rescan vm from root ept */ + ret |= RET_RESCAN_FLAG; + + return ret; +} + +static int ept_pte_range(struct page_idle_ctrl *pic, + pmd_t *pmd, unsigned long addr, unsigned long end, + struct mm_walk *walk) +{ + pte_t *pte; + enum ProcIdlePageType page_type; + int err = 0; + + pte = pte_offset_kernel(pmd, addr); + do { + if (KVM_CHECK_INVALID_SPTE(pte->pte)) { + page_type = PTE_IDLE; + } else if (!ept_pte_present(*pte)) { + err = vm_walk_host_range(addr, end, walk); + goto next; + } else if (!test_and_clear_bit(_PAGE_BIT_EPT_ACCESSED, + (unsigned long *) &pte->pte)) + page_type = PTE_IDLE; + else { + page_type = PTE_ACCESSED; + if (pic->flags & SCAN_DIRTY_PAGE) { + if (test_and_clear_bit(_PAGE_BIT_EPT_DIRTY, + (unsigned long *) &pte->pte)) + page_type = PTE_DIRTY_M; + } + } + + err = pic_add_page(pic, addr, addr + PAGE_SIZE, page_type); +next: + if (err) + break; + } while (pte++, addr += PAGE_SIZE, addr != end); + + return err; +} + +static enum ProcIdlePageType ept_huge_accessed(pmd_t *pmd, unsigned long addr, + unsigned long end) +{ + int accessed = PMD_IDLE; + pte_t *pte; + + pte = pte_offset_kernel(pmd, addr); + do { + if (!KVM_CHECK_INVALID_SPTE(pte->pte)) + continue; + if (!ept_pte_present(*pte)) + continue; + if (!test_and_clear_bit(_PAGE_BIT_EPT_ACCESSED, + (unsigned long *)&pte->pte)) + continue; + accessed = PMD_ACCESSED; + } while (pte++, addr += PAGE_SIZE, addr != end); + + return accessed; +} + +static int ept_pmd_range(struct page_idle_ctrl *pic, + pud_t *pud, unsigned long addr, unsigned long end, + struct mm_walk *walk) +{ + pmd_t *pmd; + unsigned long next; + enum ProcIdlePageType page_type; + enum ProcIdlePageType pte_page_type; + int err = 0; + + if (pic->flags & SCAN_HUGE_PAGE) + pte_page_type = PMD_IDLE_PTES; + else + pte_page_type = IDLE_PAGE_TYPE_MAX; + + pmd = pmd_offset(pud, addr); + do { + next = pmd_addr_end(addr, end); + if (KVM_CHECK_INVALID_SPTE(pmd->pmd)) + page_type = PMD_IDLE; + else if (!ept_pmd_present(*pmd)) { + err = vm_walk_host_range(addr, next, walk); + goto next; + } else if (!pmd_large(*pmd)) { + if (pic->flags & SCAN_AS_HUGE) + page_type = ept_huge_accessed(pmd, addr, next); + else + page_type = pte_page_type; + } else if (!test_and_clear_bit(_PAGE_BIT_EPT_ACCESSED, + (unsigned long *)pmd)) + page_type = PMD_IDLE; + else { + page_type = PMD_ACCESSED; + if ((pic->flags & SCAN_DIRTY_PAGE) && + test_and_clear_bit(_PAGE_BIT_EPT_DIRTY, + (unsigned long *) pmd)) + page_type = PMD_DIRTY_M; + } + + if (page_type != IDLE_PAGE_TYPE_MAX) + err = pic_add_page(pic, addr, next, page_type); + else + err = ept_pte_range(pic, pmd, addr, next, walk); + +next: + if (err) + break; + } while (pmd++, addr = next, addr != end); + + return err; +} + + +static int ept_pud_range(struct page_idle_ctrl *pic, + p4d_t *p4d, unsigned long addr, unsigned long end, + struct mm_walk *walk) +{ + pud_t *pud; + unsigned long next; + int err = 0; + + pud = pud_offset(p4d, addr); + do { + next = pud_addr_end(addr, end); + + if (!ept_pud_present(*pud)) { + err = vm_walk_host_range(addr, next, walk); + goto next; + } + + if (pud_large(*pud)) + err = pic_add_page(pic, addr, next, PUD_PRESENT); + else + err = ept_pmd_range(pic, pud, addr, next, walk); + +next: + if (err) + break; + } while (pud++, addr = next, addr != end); + + return err; +} + +static int ept_p4d_range(struct page_idle_ctrl *pic, + p4d_t *p4d, unsigned long addr, unsigned long end, + struct mm_walk *walk) +{ + unsigned long next; + int err = 0; + + p4d += p4d_index(addr); + do { + next = p4d_addr_end(addr, end); + if (!ept_p4d_present(*p4d)) { + set_restart_gpa(next, "P4D_HOLE"); + continue; + } + + err = ept_pud_range(pic, p4d, addr, next, walk); + if (err) + break; + } while (p4d++, addr = next, addr != end); + + return err; +} + +static int ept_pgd_range(struct page_idle_ctrl *pic, + pgd_t *pgd, + unsigned long addr, + unsigned long end, + struct mm_walk *walk) +{ + p4d_t *p4d; + unsigned long next; + int err = 0; + + pgd = pgd_offset_pgd(pgd, addr); + do { + next = pgd_addr_end(addr, end); + if (!ept_pgd_present(*pgd)) { + set_restart_gpa(next, "PGD_HOLE"); + continue; + } + + p4d = (p4d_t *)pgd_page_vaddr(*pgd); + err = ept_p4d_range(pic, p4d, addr, next, walk); + if (err) + break; + } while (pgd++, addr = next, addr != end); + + return err; +} + +static int ept_page_range(struct page_idle_ctrl *pic, + unsigned long addr, + unsigned long end, + struct mm_walk *walk) +{ + struct kvm_vcpu *vcpu; + struct kvm_mmu *mmu; + uint64_t *ept_root; + int err = 0; + + WARN_ON(addr >= end); + + read_lock(&pic->kvm->mmu_lock); + + vcpu = kvm_get_vcpu(pic->kvm, 0); + if (!vcpu) { + pic->gpa_to_hva = 0; + set_restart_gpa(TASK_SIZE, "NO-VCPU"); + read_unlock(&pic->kvm->mmu_lock); + return -EINVAL; + } + + mmu = kvm_arch_mmu_pointer(vcpu); + if (!VALID_PAGE(mmu->root.hpa)) { + pic->gpa_to_hva = 0; + set_restart_gpa(TASK_SIZE, "NO-HPA"); + read_unlock(&pic->kvm->mmu_lock); + return -EINVAL; + } + + ept_root = __va(mmu->root.hpa); + + /* Walk start at p4d when vm has 4 level table pages */ + if (mmu->root_role.level != 4) + err = ept_pgd_range(pic, (pgd_t *)ept_root, addr, end, walk); + else + err = ept_p4d_range(pic, (p4d_t *)ept_root, addr, end, walk); + + /* mmu_lock is unlock in vm_walk_host_range which will unlock mmu_lock + * and RET_RESCAN_FLAG will be set in ret value + */ + if (!(err & RET_RESCAN_FLAG)) + read_unlock(&pic->kvm->mmu_lock); + else + err &= ~RET_RESCAN_FLAG; + + return err; +} + +static int ept_idle_supports_cpu(struct kvm *kvm) +{ + struct kvm_vcpu *vcpu; + struct kvm_mmu *mmu; + int ret; + + vcpu = kvm_get_vcpu(kvm, 0); + if (!vcpu) + return -EINVAL; + + read_lock(&kvm->mmu_lock); + mmu = kvm_arch_mmu_pointer(vcpu); + if (kvm_mmu_ad_disabled(mmu)) { + pr_notice("CPU does not support EPT A/D bits tracking\n"); + ret = -EINVAL; + } else if (mmu->root_role.level < 4 || + (mmu->root_role.level == 5 && !pgtable_l5_enabled())) { + pr_notice("Unsupported EPT level %d\n", mmu->root_role.level); + ret = -EINVAL; + } else + ret = 0; + read_unlock(&kvm->mmu_lock); + + return ret; +} + +#else +static inline phys_addr_t stage2_range_addr_end(phys_addr_t addr, phys_addr_t end) +{ + phys_addr_t size = kvm_granule_size(KVM_PGTABLE_MIN_BLOCK_LEVEL); + phys_addr_t boundary = ALIGN_DOWN(addr + size, size); + + return (boundary - 1 < end - 1) ? boundary : end; +} + +static int arm_pte_range(struct page_idle_ctrl *pic, + pmd_t *pmd, unsigned long addr, unsigned long end) +{ + pte_t *pte; + enum ProcIdlePageType page_type; + int err = 0; + + pte = pte_offset_kernel(pmd, addr); + do { + if (!pte_present(*pte)) + page_type = PTE_HOLE; + else if (!test_and_clear_bit(_PAGE_MM_BIT_ACCESSED, + (unsigned long *) &pte->pte)) + page_type = PTE_IDLE; + else + page_type = PTE_ACCESSED; + + err = pic_add_page(pic, addr, addr + PAGE_SIZE, page_type); + if (err) + break; + } while (pte++, addr += PAGE_SIZE, addr != end); + + return err; +} + +static int arm_pmd_range(struct page_idle_ctrl *pic, + pud_t *pud, unsigned long addr, unsigned long end) +{ + pmd_t *pmd; + unsigned long next; + enum ProcIdlePageType page_type; + enum ProcIdlePageType pte_page_type; + int err = 0; + + if (pic->flags & SCAN_HUGE_PAGE) + pte_page_type = PMD_IDLE_PTES; + else + pte_page_type = IDLE_PAGE_TYPE_MAX; + + pmd = pmd_offset(pud, addr); + do { + next = pmd_addr_end(addr, end); + if (!pmd_present(*pmd)) + page_type = PMD_HOLE; + else if (!if_pmd_thp_or_huge(*pmd)) + page_type = pte_page_type; + else if (!test_and_clear_bit(_PAGE_MM_BIT_ACCESSED, + (unsigned long *)pmd)) + page_type = PMD_IDLE; + else + page_type = PMD_ACCESSED; + + if (page_type != IDLE_PAGE_TYPE_MAX) + err = pic_add_page(pic, addr, next, page_type); + else + err = arm_pte_range(pic, pmd, addr, next); + if (err) + break; + } while (pmd++, addr = next, addr != end); + + return err; +} + +static int arm_pud_range(struct page_idle_ctrl *pic, + p4d_t *p4d, unsigned long addr, unsigned long end) +{ + pud_t *pud = (pud_t *)p4d; + unsigned long next; + int err = 0; + + pud += pud_index(addr); + do { + next = pud_addr_end(addr, end); + if (!pud_present(*pud)) { + set_restart_gpa(next, "PUD_HOLE"); + continue; + } + + if (if_pud_huge(*pud)) + err = pic_add_page(pic, addr, next, PUD_PRESENT); + else + err = arm_pmd_range(pic, pud, addr, next); + if (err) + break; + } while (pud++, addr = next, addr != end); + + return err; +} + +static int arm_p4d_range(struct page_idle_ctrl *pic, + pgd_t *pgd, unsigned long addr, unsigned long end) +{ + p4d_t *p4d; + unsigned long next; + int err = 0; + + p4d = p4d_offset(pgd, addr); + do { + next = p4d_addr_end(addr, end); + if (!p4d_present(*p4d)) { + set_restart_gpa(next, "P4D_HOLE"); + continue; + } + + err = arm_pud_range(pic, p4d, addr, next); + if (err) + break; + } while (p4d++, addr = next, addr != end); + + return err; +} + +static int arm_page_range(struct page_idle_ctrl *pic, + unsigned long addr, + unsigned long end) +{ + pgd_t *pgd; + unsigned long next; + struct kvm *kvm = pic->kvm; + int err = 0; + + WARN_ON(addr >= end); + + read_lock(&pic->kvm->mmu_lock); + pgd = (pgd_t *)kvm->arch.mmu.pgt->pgd + pgd_index(addr) * PTRS_PER_PTE; + read_unlock(&pic->kvm->mmu_lock); + + local_irq_disable(); + do { + next = pgd_addr_end(addr, end); + if (!pgd_present(*pgd)) { + set_restart_gpa(next, "PGD_HOLE"); + continue; + } + + err = arm_p4d_range(pic, pgd, addr, next); + if (err) + break; + } while (pgd++, addr = next, addr != end); + + local_irq_enable(); + return err; +} +#endif + +/* + * Depending on whether hva falls in a memslot: + * + * 1) found => return gpa and remaining memslot size in *addr_range + * + * |<----- addr_range --------->| + * [ mem slot ] + * ^hva + * + * 2) not found => return hole size in *addr_range + * + * |<----- addr_range --------->| + * [first mem slot above hva ] + * ^hva + * + * If hva is above all mem slots, *addr_range will be ~0UL. + * We can finish read(2). + */ +static unsigned long vm_idle_find_gpa(struct page_idle_ctrl *pic, + unsigned long hva, + unsigned long *addr_range) +{ + struct kvm *kvm = pic->kvm; + struct kvm_memslots *slots; + struct kvm_memory_slot *memslot; + unsigned long hva_end; + gfn_t gfn; + int bkt; + + *addr_range = ~0UL; + mutex_lock(&kvm->slots_lock); + slots = kvm_memslots(pic->kvm); + kvm_for_each_memslot(memslot, bkt, slots) { + hva_end = memslot->userspace_addr + + (memslot->npages << PAGE_SHIFT); + + if (hva >= memslot->userspace_addr && hva < hva_end) { + gpa_t gpa; + + gfn = hva_to_gfn_memslot(hva, memslot); + *addr_range = hva_end - hva; + gpa = gfn_to_gpa(gfn); + mutex_unlock(&kvm->slots_lock); + return gpa; + } + + if (memslot->userspace_addr > hva) + *addr_range = min(*addr_range, + memslot->userspace_addr - hva); + } + mutex_unlock(&kvm->slots_lock); + return INVALID_PAGE; +} + +static inline unsigned long mask_to_size(unsigned long mask) +{ + return ~mask + 1; +} + +static int mm_idle_hugetlb_entry(pte_t *pte, unsigned long hmask, + unsigned long addr, unsigned long next, + struct mm_walk *walk); +static int vm_idle_hugetlb_entry(pte_t *pte, unsigned long hmask, + unsigned long addr, unsigned long next, + struct mm_walk *walk) +{ + struct page_idle_ctrl *pic = walk->private; + enum ProcIdlePageType page_type; + + pic->flags |= VM_SCAN_HOST; + + /* hugetlb page table entry of vm maybe not present while page is resident + * in address_space + */ + if (mask_to_size(hmask) != PUD_SIZE && !pte_present(*pte) && + vm_handle_pte_hole != NULL) { + page_type = vm_handle_pte_hole(addr, next, -1, walk); + if (page_type < IDLE_PAGE_TYPE_MAX) + return pic_add_page(pic, addr, next, page_type); + } + + return mm_idle_hugetlb_entry(pte, hmask, addr, next, walk); +} + +static int vm_idle_pte_hole(unsigned long addr, unsigned long next, int depth, struct mm_walk *walk) +{ + struct page_idle_ctrl *pic = walk->private; + enum ProcIdlePageType pagetype; + + if (vm_handle_pte_hole == NULL) + return 0; + + pagetype = vm_handle_pte_hole(addr, next, depth, walk); + if (pagetype >= IDLE_PAGE_TYPE_MAX) + return 0; + + debug_printk("scan pte hole addr %pK type %d\n", addr, pagetype); + pic->flags |= VM_SCAN_HOST; + return pic_add_page(pic, addr, next, pagetype); +} + +static int mm_idle_pmd_entry(pmd_t *pmd, unsigned long addr, + unsigned long next, struct mm_walk *walk); +static int vm_idle_pmd_entry(pmd_t *pmd, unsigned long addr, + unsigned long next, struct mm_walk *walk) +{ + struct page_idle_ctrl *pic = walk->private; + + pic->flags |= VM_SCAN_HOST; + return mm_idle_pmd_entry(pmd, addr, next, walk); +} + +static int mm_idle_pud_entry(pud_t *pud, unsigned long addr, + unsigned long next, struct mm_walk *walk); +static int vm_idle_pud_entry(pud_t *pud, unsigned long addr, + unsigned long next, struct mm_walk *walk) +{ + struct page_idle_ctrl *pic = walk->private; + + pic->flags |= VM_SCAN_HOST; + return mm_idle_pud_entry(pud, addr, next, walk); +} + +static int vm_idle_walk_hva_range(struct page_idle_ctrl *pic, + unsigned long start, unsigned long end, + struct mm_walk *walk) +{ + unsigned long gpa_addr; + unsigned long gpa_next; + unsigned long gpa_end; + unsigned long addr_range; + unsigned long va_end; + int ret; + int steps; + +#ifdef CONFIG_X86_64 + ret = ept_idle_supports_cpu(pic->kvm); + if (ret) + return ret; +#endif + + ret = init_page_idle_ctrl_buffer(pic); + if (ret) + return ret; + + for (; start < end;) { + gpa_addr = vm_idle_find_gpa(pic, start, &addr_range); + + if (gpa_addr == INVALID_PAGE) { + pic->gpa_to_hva = 0; + if (addr_range == ~0UL) { + set_restart_gpa(TASK_SIZE, "EOF"); + va_end = end; + } else { + start += addr_range; + set_restart_gpa(start, "OUT-OF-SLOT"); + va_end = start; + } + } else { + pic->gpa_to_hva = start - gpa_addr; + gpa_end = gpa_addr + addr_range; + steps = 0; + for (; gpa_addr < gpa_end;) { + gpa_next = min(gpa_end, gpa_addr + walk_step * PAGE_SIZE); +#ifdef CONFIG_ARM64 + ret = arm_page_range(pic, gpa_addr, gpa_next); +#else + ret = ept_page_range(pic, gpa_addr, gpa_next, walk); +#endif + gpa_addr = pic->restart_gpa; + + if (ret) + break; + + if (++steps >= resched_step) { + cond_resched(); + steps = 0; + } + } + va_end = pic->gpa_to_hva + gpa_end; + } + + start = pic->restart_gpa + pic->gpa_to_hva; + ret = page_idle_copy_user(pic, start, va_end); + if (ret) + break; + } + + if (start > pic->next_hva) + set_next_hva(start, "NEXT-START"); + + if (pic->bytes_copied) + ret = 0; + return ret; +} + +static int mm_idle_test_walk(unsigned long start, unsigned long end, + struct mm_walk *walk); +static ssize_t vm_idle_read(struct file *file, char *buf, + size_t count, loff_t *ppos) +{ + struct mm_struct *mm = file->private_data; + struct mm_walk mm_walk = {}; + struct mm_walk_ops mm_walk_ops = {}; + struct page_idle_ctrl *pic; + unsigned long hva_start = *ppos; + unsigned long hva_end = hva_start + (count << (3 + PAGE_SHIFT)); + int ret; + + pic = kzalloc(sizeof(*pic), GFP_KERNEL); + if (!pic) + return -ENOMEM; + + setup_page_idle_ctrl(pic, buf, count, file->f_flags); + pic->kvm = mm_kvm(mm); + + mm_walk_ops.pmd_entry = vm_idle_pmd_entry; + mm_walk_ops.pud_entry = vm_idle_pud_entry; + mm_walk_ops.hugetlb_entry = vm_idle_hugetlb_entry; + mm_walk_ops.pte_hole = vm_idle_pte_hole; + mm_walk_ops.test_walk = mm_idle_test_walk; + + mm_walk.mm = mm; + mm_walk.ops = &mm_walk_ops; + mm_walk.private = pic; + + ret = vm_idle_walk_hva_range(pic, hva_start, hva_end, &mm_walk); + if (ret) + goto out_kvm; + + ret = pic->bytes_copied; + *ppos = pic->next_hva; +out_kvm: + kfree(pic); + return ret; + +} + +static ssize_t mm_idle_read(struct file *file, char *buf, + size_t count, loff_t *ppos); + +static ssize_t page_scan_read(struct file *file, char *buf, + size_t count, loff_t *ppos) +{ + struct mm_struct *mm = file->private_data; + unsigned long hva_start = *ppos; + unsigned long hva_end = hva_start + (count << (3 + PAGE_SHIFT)); + + if ((hva_start >= TASK_SIZE) || (hva_end >= TASK_SIZE)) { + debug_printk("page_idle_read past TASK_SIZE: %pK %pK %lx\n", + hva_start, hva_end, TASK_SIZE); + return 0; + } + if (hva_end <= hva_start) { + debug_printk("page_idle_read past EOF: %pK %pK\n", + hva_start, hva_end); + return 0; + } + if (*ppos & (PAGE_SIZE - 1)) { + debug_printk("page_idle_read unaligned ppos: %pK\n", + hva_start); + return -EINVAL; + } + if (count < PAGE_IDLE_BUF_MIN) { + debug_printk("page_idle_read small count: %lx\n", + (unsigned long)count); + return -EINVAL; + } + + if (!mm_kvm(mm)) + return mm_idle_read(file, buf, count, ppos); + + return vm_idle_read(file, buf, count, ppos); +} + +static int page_scan_open(struct inode *inode, struct file *file) +{ + if (!try_module_get(THIS_MODULE)) + return -EBUSY; + + return 0; +} + +static int page_scan_release(struct inode *inode, struct file *file) +{ + struct mm_struct *mm = file->private_data; + struct kvm *kvm; + int ret = 0; + + if (!mm) { + ret = -EBADF; + goto out; + } + + kvm = mm_kvm(mm); + if (!kvm) { + ret = -EINVAL; + goto out; + } +#ifdef CONFIG_X86_64 + write_lock(&kvm->mmu_lock); + kvm_flush_remote_tlbs(kvm); + write_unlock(&kvm->mmu_lock); +#endif + +out: + module_put(THIS_MODULE); + return ret; +} + +static int mm_idle_pmd_large(pmd_t pmd) +{ +#ifdef CONFIG_ARM64 + return if_pmd_thp_or_huge(pmd); +#else + return pmd_large(pmd); +#endif +} + +static int mm_idle_pte_range(struct page_idle_ctrl *pic, pmd_t *pmd, + unsigned long addr, unsigned long next) +{ + enum ProcIdlePageType page_type; + pte_t *pte; + int err = 0; + + pte = pte_offset_kernel(pmd, addr); + do { + if (!pte_present(*pte)) + page_type = PTE_HOLE; + else if (pic->flags & SCAN_IGN_HOST) + page_type = PTE_IDLE; + else if (!test_and_clear_bit(_PAGE_MM_BIT_ACCESSED, + (unsigned long *) &pte->pte)) + page_type = PTE_IDLE; + else { + page_type = PTE_ACCESSED; + } + + err = pic_add_page(pic, addr, addr + PAGE_SIZE, page_type); + if (err) + break; + } while (pte++, addr += PAGE_SIZE, addr != next); + + return err; +} + +static int mm_idle_hugetlb_entry(pte_t *pte, unsigned long hmask, + unsigned long addr, unsigned long next, + struct mm_walk *walk) +{ + struct page_idle_ctrl *pic = walk->private; + enum ProcIdlePageType page_type; + unsigned long start = addr & hmask; /* hugepage may be splited in vm */ + int ret; + + if (mask_to_size(hmask) == PUD_SIZE) { + page_type = PUD_PRESENT; + goto add_page; + } + + if (!pte_present(*pte)) + page_type = PMD_HOLE; + else if (pic->flags & SCAN_IGN_HOST) + page_type = PMD_IDLE; + else if (!test_and_clear_bit(_PAGE_MM_BIT_ACCESSED, (unsigned long *)pte)) + page_type = PMD_IDLE; + else + page_type = PMD_ACCESSED; + +add_page: + ret = pic_add_page(pic, start, start + pagetype_size[page_type], page_type); + return ret; +} + +static int mm_idle_pmd_entry(pmd_t *pmd, unsigned long addr, + unsigned long next, struct mm_walk *walk) +{ + struct page_idle_ctrl *pic = walk->private; + enum ProcIdlePageType page_type; + enum ProcIdlePageType pte_page_type; + int err; + + /* + * Skip duplicate PMD_IDLE_PTES: when the PMD crosses VMA boundary, + * walk_page_range() can call on the same PMD twice. + */ + if ((addr & PMD_MASK) == (pic->last_va & PMD_MASK) && (pic->flags & SCAN_HUGE_PAGE)) { + debug_printk("ignore duplicate addr %pK %pK\n", + addr, pic->last_va); + set_restart_gpa(round_up(next, PMD_SIZE), "DUP_ADDR"); + return 0; + } + pic->last_va = addr; + + if (pic->flags & SCAN_HUGE_PAGE) + pte_page_type = PMD_IDLE_PTES; + else + pte_page_type = IDLE_PAGE_TYPE_MAX; + + if (!pmd_present(*pmd)) + page_type = PMD_HOLE; + else if (!mm_idle_pmd_large(*pmd)) + page_type = pte_page_type; + else if (!test_and_clear_bit(_PAGE_MM_BIT_ACCESSED, + (unsigned long *)pmd) || + pic->flags & SCAN_IGN_HOST) + page_type = PMD_IDLE; + else + page_type = PMD_ACCESSED; + + if (page_type != IDLE_PAGE_TYPE_MAX) + err = pic_add_page(pic, addr, next, page_type); + else + err = mm_idle_pte_range(pic, pmd, addr, next); + + return err; +} + +static int mm_idle_pud_entry(pud_t *pud, unsigned long addr, + unsigned long next, struct mm_walk *walk) +{ + struct page_idle_ctrl *pic = walk->private; + + spinlock_t *ptl = pud_trans_huge_lock(pud, walk->vma); + + if (ptl) { + if ((addr & PUD_MASK) != (pic->last_va & PUD_MASK)) { + pic_add_page(pic, addr, next, PUD_PRESENT); + pic->last_va = addr; + } + spin_unlock(ptl); + return 1; + } + + return 0; +} + +static int mm_idle_test_walk(unsigned long start, unsigned long end, + struct mm_walk *walk) +{ + struct vm_area_struct *vma = walk->vma; + struct page_idle_ctrl *pic = walk->private; + + /* If the specified page swapout is set, the untagged vma is skipped. */ + if ((pic->flags & VMA_SCAN_FLAG) && !(vma->vm_flags & VM_SWAPFLAG)) + return 1; + + if (vma->vm_file) { + if (is_vm_hugetlb_page(vma)) + return 0; + if ((vma->vm_flags & (VM_WRITE|VM_MAYSHARE)) == VM_WRITE) + return 0; + return 1; + } + + return 0; +} + +static int mm_idle_walk_range(struct page_idle_ctrl *pic, + unsigned long start, + unsigned long end, + struct mm_walk *walk) +{ + struct vm_area_struct *vma; + int ret = 0; + + ret = init_page_idle_ctrl_buffer(pic); + if (ret) + return ret; + + for (; start < end;) { + mmap_read_lock(walk->mm); + vma = find_vma(walk->mm, start); + if (vma) { + if (end > vma->vm_start) { + local_irq_disable(); + ret = walk_page_range(walk->mm, start, end, + walk->ops, walk->private); + local_irq_enable(); + } else + set_restart_gpa(vma->vm_start, "VMA-HOLE"); + } else + set_restart_gpa(TASK_SIZE, "EOF"); + mmap_read_unlock(walk->mm); + WARN_ONCE(pic->gpa_to_hva, "non-zero gpa_to_hva"); + if (ret != PAGE_IDLE_KBUF_FULL && end > pic->restart_gpa) + pic->restart_gpa = end; + start = pic->restart_gpa; + ret = page_idle_copy_user(pic, start, end); + if (ret) + break; + } + + if (start > pic->next_hva) + set_next_hva(start, "NEXT-START"); + + if (pic->bytes_copied) { + if (ret != PAGE_IDLE_BUF_FULL && pic->next_hva < end) + debug_printk("partial scan: next_hva=%pK end=%pK\n", + pic->next_hva, end); + ret = 0; + } else + debug_printk("nothing read"); + return ret; +} + +static ssize_t mm_idle_read(struct file *file, char *buf, + size_t count, loff_t *ppos) +{ + struct mm_struct *mm = file->private_data; + struct mm_walk_ops *mm_walk_ops = NULL; + struct mm_walk mm_walk = {}; + struct page_idle_ctrl *pic; + unsigned long va_start = *ppos; + unsigned long va_end = va_start + (count << (3 + PAGE_SHIFT)); + int ret; + + if (va_end <= va_start) { + debug_printk("%s past EOF: %pK %pK\n", + __func__, va_start, va_end); + return 0; + } + if (*ppos & (PAGE_SIZE - 1)) { + debug_printk("%s unaligned ppos: %pK\n", + __func__, va_start); + return -EINVAL; + } + if (count < PAGE_IDLE_BUF_MIN) { + debug_printk("%s small count: %lx\n", + __func__, (unsigned long)count); + return -EINVAL; + } + + pic = kzalloc(sizeof(*pic), GFP_KERNEL); + if (!pic) + return -ENOMEM; + + mm_walk_ops = kzalloc(sizeof(struct mm_walk_ops), GFP_KERNEL); + if (!mm_walk_ops) { + kfree(pic); + return -ENOMEM; + } + + setup_page_idle_ctrl(pic, buf, count, file->f_flags); + + mm_walk_ops->pmd_entry = mm_idle_pmd_entry; + mm_walk_ops->pud_entry = mm_idle_pud_entry; + mm_walk_ops->hugetlb_entry = mm_idle_hugetlb_entry; + mm_walk_ops->test_walk = mm_idle_test_walk; + + mm_walk.mm = mm; + mm_walk.ops = mm_walk_ops; + mm_walk.private = pic; + mm_walk.pgd = NULL; + mm_walk.no_vma = false; + ret = mm_idle_walk_range(pic, va_start, va_end, &mm_walk); + if (ret) + goto out_free; + + ret = pic->bytes_copied; + *ppos = pic->next_hva; +out_free: + kfree(pic); + kfree(mm_walk_ops); + return ret; +} + +static long page_scan_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) +{ + void __user *argp = (void __user *)arg; + unsigned int flags; + + if (get_user(flags, (unsigned int __user *)argp)) + return -EFAULT; + flags &= ALL_SCAN_FLAGS; + + switch (cmd) { + case IDLE_SCAN_ADD_FLAGS: + filp->f_flags |= flags; + break; + case IDLE_SCAN_REMOVE_FLAGS: + filp->f_flags &= ~flags; + break; + case VMA_SCAN_ADD_FLAGS: + filp->f_flags |= flags; + break; + case VMA_SCAN_REMOVE_FLAGS: + filp->f_flags &= ~flags; + break; + default: + return -EOPNOTSUPP; + } + + return 0; +} + +extern struct file_operations proc_page_scan_operations; + +static int page_scan_entry(void) +{ + proc_page_scan_operations.flock(NULL, 1, NULL); + proc_page_scan_operations.owner = THIS_MODULE; + proc_page_scan_operations.read = page_scan_read; + proc_page_scan_operations.open = page_scan_open; + proc_page_scan_operations.release = page_scan_release; + proc_page_scan_operations.unlocked_ioctl = page_scan_ioctl; + proc_page_scan_operations.flock(NULL, 0, NULL); + + return 0; +} + +static void page_scan_exit(void) +{ + proc_page_scan_operations.flock(NULL, 1, NULL); + proc_page_scan_operations.owner = NULL; + proc_page_scan_operations.read = NULL; + proc_page_scan_operations.open = NULL; + proc_page_scan_operations.release = NULL; + proc_page_scan_operations.unlocked_ioctl = NULL; + proc_page_scan_operations.flock(NULL, 0, NULL); +} + +MODULE_LICENSE("GPL"); +module_init(page_scan_entry); +module_exit(page_scan_exit); diff --git a/fs/proc/etmem_scan.h b/fs/proc/etmem_scan.h new file mode 100644 index 000000000000..e109f7f350e1 --- /dev/null +++ b/fs/proc/etmem_scan.h @@ -0,0 +1,149 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _PAGE_IDLE_H +#define _PAGE_IDLE_H + +#include <linux/types.h> + +#define SCAN_HUGE_PAGE O_NONBLOCK /* only huge page */ +#define SCAN_SKIM_IDLE O_NOFOLLOW /* stop on PMD_IDLE_PTES */ +#define SCAN_DIRTY_PAGE O_NOATIME /* report pte/pmd dirty bit */ + +/* define to not used file flags */ +#define SCAN_AS_HUGE 0100000000 /* treat normal page as hugepage in vm */ +#define SCAN_IGN_HOST 0200000000 /* ignore host access when scan vm */ +#define VM_SCAN_HOST 0400000000 /* scan and add host page for vm hole(internal) */ +#define VMA_SCAN_FLAG 0x1000 /* scan the specifics vma with flag */ + +#define ALL_SCAN_FLAGS (SCAN_HUGE_PAGE | SCAN_SKIM_IDLE | SCAN_DIRTY_PAGE | \ + SCAN_AS_HUGE | SCAN_IGN_HOST | VM_SCAN_HOST | VMA_SCAN_FLAG) + +#define IDLE_SCAN_MAGIC 0x66 +#define IDLE_SCAN_ADD_FLAGS _IOW(IDLE_SCAN_MAGIC, 0x0, unsigned int) +#define IDLE_SCAN_REMOVE_FLAGS _IOW(IDLE_SCAN_MAGIC, 0x1, unsigned int) +#define VMA_SCAN_ADD_FLAGS _IOW(IDLE_SCAN_MAGIC, 0x2, unsigned int) +#define VMA_SCAN_REMOVE_FLAGS _IOW(IDLE_SCAN_MAGIC, 0x3, unsigned int) + +enum ProcIdlePageType { + PTE_ACCESSED, /* 4k page */ + PMD_ACCESSED, /* 2M page */ + PUD_PRESENT, /* 1G page */ + + PTE_DIRTY_M, + PMD_DIRTY_M, + + PTE_IDLE, + PMD_IDLE, + PMD_IDLE_PTES, /* all PTE idle */ + + PTE_HOLE, + PMD_HOLE, + + PIP_CMD, + + IDLE_PAGE_TYPE_MAX +}; + +#define PIP_TYPE(a) (0xf & (a >> 4)) +#define PIP_SIZE(a) (0xf & a) +#define PIP_COMPOSE(type, nr) ((type << 4) | nr) + +#define PIP_CMD_SET_HVA PIP_COMPOSE(PIP_CMD, 0) + +#ifndef INVALID_PAGE +#define INVALID_PAGE ~0UL +#endif + +#ifdef CONFIG_ARM64 +#define _PAGE_MM_BIT_ACCESSED 10 +#else +#define _PAGE_MM_BIT_ACCESSED _PAGE_BIT_ACCESSED +#endif + +#ifdef CONFIG_X86_64 +#define _PAGE_BIT_EPT_ACCESSED 8 +#define _PAGE_BIT_EPT_DIRTY 9 +#define _PAGE_EPT_ACCESSED (_AT(pteval_t, 1) << _PAGE_BIT_EPT_ACCESSED) +#define _PAGE_EPT_DIRTY (_AT(pteval_t, 1) << _PAGE_BIT_EPT_DIRTY) + +#define _PAGE_EPT_PRESENT (_AT(pteval_t, 7)) + +static inline int ept_pte_present(pte_t a) +{ + return pte_flags(a) & _PAGE_EPT_PRESENT; +} + +static inline int ept_pmd_present(pmd_t a) +{ + return pmd_flags(a) & _PAGE_EPT_PRESENT; +} + +static inline int ept_pud_present(pud_t a) +{ + return pud_flags(a) & _PAGE_EPT_PRESENT; +} + +static inline int ept_p4d_present(p4d_t a) +{ + return p4d_flags(a) & _PAGE_EPT_PRESENT; +} + +static inline int ept_pgd_present(pgd_t a) +{ + return pgd_flags(a) & _PAGE_EPT_PRESENT; +} + +static inline int ept_pte_accessed(pte_t a) +{ + return pte_flags(a) & _PAGE_EPT_ACCESSED; +} + +static inline int ept_pmd_accessed(pmd_t a) +{ + return pmd_flags(a) & _PAGE_EPT_ACCESSED; +} + +static inline int ept_pud_accessed(pud_t a) +{ + return pud_flags(a) & _PAGE_EPT_ACCESSED; +} + +static inline int ept_p4d_accessed(p4d_t a) +{ + return p4d_flags(a) & _PAGE_EPT_ACCESSED; +} + +static inline int ept_pgd_accessed(pgd_t a) +{ + return pgd_flags(a) & _PAGE_EPT_ACCESSED; +} +#endif + +extern struct file_operations proc_page_scan_operations; + +#define PAGE_IDLE_KBUF_FULL 1 +#define PAGE_IDLE_BUF_FULL 2 +#define PAGE_IDLE_BUF_MIN (sizeof(uint64_t) * 2 + 3) + +#define PAGE_IDLE_KBUF_SIZE 8000 + +struct page_idle_ctrl { + struct mm_struct *mm; + struct kvm *kvm; + + uint8_t kpie[PAGE_IDLE_KBUF_SIZE]; + int pie_read; + int pie_read_max; + + void __user *buf; + int buf_size; + int bytes_copied; + + unsigned long next_hva; /* GPA for EPT; VA for PT */ + unsigned long gpa_to_hva; + unsigned long restart_gpa; + unsigned long last_va; + + unsigned int flags; +}; + +#endif diff --git a/fs/proc/internal.h b/fs/proc/internal.h index 9a8f32f21ff5..a9615455b709 100644 --- a/fs/proc/internal.h +++ b/fs/proc/internal.h @@ -303,6 +303,9 @@ extern const struct file_operations proc_pid_smaps_operations; extern const struct file_operations proc_pid_smaps_rollup_operations; extern const struct file_operations proc_clear_refs_operations; extern const struct file_operations proc_pagemap_operations; +#ifdef CONFIG_ETMEM +extern const struct file_operations proc_mm_idle_operations; +#endif

extern unsigned long task_vsize(struct mm_struct *); extern unsigned long task_statm(struct mm_struct *, diff --git a/include/linux/etmem.h b/include/linux/etmem.h new file mode 100644 index 000000000000..e8a2585f3891 --- /dev/null +++ b/include/linux/etmem.h @@ -0,0 +1,27 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef __MM_ETMEM_H_ +#define __MM_ETMEM_H_ + +#include <linux/list.h> +#include <asm/page.h> +#include <linux/mmzone.h> +#include <linux/memcontrol.h> +#include <linux/page-flags.h> + +#ifdef CONFIG_ETMEM + +#if IS_ENABLED(CONFIG_KVM) +static inline struct kvm *mm_kvm(struct mm_struct *mm) +{ + return mm->kvm; +} +#else +static inline struct kvm *mm_kvm(struct mm_struct *mm) +{ + return NULL; +} +#endif + + +#endif /* #ifdef CONFIG_ETMEM */ +#endif /* define __MM_ETMEM_H_ */ diff --git a/include/linux/mm.h b/include/linux/mm.h index 80bacc4da324..a3f0556a91fc 100644 --- a/include/linux/mm.h +++ b/include/linux/mm.h @@ -308,6 +308,10 @@ extern unsigned int kobjsize(const void *objp); # define VM_SOFTDIRTY 0 #endif

+#ifdef CONFIG_ETMEM +#define VM_SWAPFLAG 0x400000000000000 /* memory swap out flag in vma */ +#endif + #define VM_MIXEDMAP 0x10000000 /* Can contain "struct page" and pure PFN pages */ #define VM_HUGEPAGE 0x20000000 /* MADV_HUGEPAGE marked this vma */ #define VM_NOHUGEPAGE 0x40000000 /* MADV_NOHUGEPAGE marked this vma */ diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h index 0bc3c7c191a5..b0447e056b20 100644 --- a/include/linux/mm_types.h +++ b/include/linux/mm_types.h @@ -31,6 +31,7 @@

struct address_space; struct mem_cgroup; +struct kvm;

/* * Each physical page in the system has a struct page associated with @@ -937,6 +938,10 @@ struct mm_struct { #ifdef CONFIG_SHARE_POOL struct sp_group_master *sp_group_master; #endif +#if IS_ENABLED(CONFIG_ETMEM) && IS_ENABLED(CONFIG_KVM) + struct kvm *kvm; +#endif + } __randomize_layout;

/* diff --git a/include/uapi/asm-generic/mman-common.h b/include/uapi/asm-generic/mman-common.h index 6ce1f1ceb432..14e5498efd7a 100644 --- a/include/uapi/asm-generic/mman-common.h +++ b/include/uapi/asm-generic/mman-common.h @@ -79,6 +79,10 @@

#define MADV_COLLAPSE 25 /* Synchronous hugepage collapse */

+#define MADV_ETMEM_BASE 0x1100 +#define MADV_SWAPFLAG MADV_ETMEM_BASE /* for memory to be swap out */ +#define MADV_SWAPFLAG_REMOVE (MADV_SWAPFLAG + 1) + /* compatibility flags */ #define MAP_FILE 0

diff --git a/mm/madvise.c b/mm/madvise.c index 4dded5d27e7e..726668a521dc 100644 --- a/mm/madvise.c +++ b/mm/madvise.c @@ -1092,6 +1092,14 @@ static int madvise_vma_behavior(struct vm_area_struct *vma, if (error) goto out; break; +#ifdef CONFIG_ETMEM + case MADV_SWAPFLAG: + new_flags |= VM_SWAPFLAG; + break; + case MADV_SWAPFLAG_REMOVE: + new_flags &= ~VM_SWAPFLAG; + break; +#endif case MADV_COLLAPSE: return madvise_collapse(vma, prev, start, end); } @@ -1196,6 +1204,10 @@ madvise_behavior_valid(int behavior) #ifdef CONFIG_MEMORY_FAILURE case MADV_SOFT_OFFLINE: case MADV_HWPOISON: +#endif +#ifdef CONFIG_ETMEM + case MADV_SWAPFLAG: + case MADV_SWAPFLAG_REMOVE: #endif return true;

@@ -1386,6 +1398,10 @@ int madvise_set_anon_name(struct mm_struct *mm, unsigned long start, * easily if memory pressure happens. * MADV_PAGEOUT - the application is not expected to use this memory soon, * page out the pages in this range immediately. + * MADV_SWAPFLAG - Used in the etmem memory extension feature, the process + * specifies the memory swap area by adding a flag to a specific + * vma address. + * MADV_SWAPFLAG_REMOVE - remove the specific vma flag * MADV_POPULATE_READ - populate (prefault) page tables readable by * triggering read faults if required * MADV_POPULATE_WRITE - populate (prefault) page tables writable by diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c index 486800a7024b..bb5f6cf4da33 100644 --- a/virt/kvm/kvm_main.c +++ b/virt/kvm/kvm_main.c @@ -1305,6 +1305,10 @@ static void kvm_destroy_vm(struct kvm *kvm)

kvm_destroy_pm_notifier(kvm); kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm); +#if IS_ENABLED(CONFIG_ETMEM) && IS_ENABLED(CONFIG_KVM) + if (mm->kvm == kvm) + mm->kvm = NULL; +#endif kvm_destroy_vm_debugfs(kvm); kvm_arch_sync_events(kvm); mutex_lock(&kvm_lock); @@ -5098,6 +5102,10 @@ static int kvm_dev_ioctl_create_vm(unsigned long type) goto put_kvm; }

+#if IS_ENABLED(CONFIG_ETMEM) && IS_ENABLED(CONFIG_KVM) + if (kvm->mm->kvm == NULL) + kvm->mm->kvm = kvm; +#endif /* * Don't call kvm_put_kvm anymore at this point; file->f_op is * already set, with ->release() being kvm_vm_release(). In error

euleros inclusion category: feature feature: Add pages to swapcache and swap them out proactively bugzilla: https://gitee.com/openeuler/kernel/issues/I8T1MB?from=project-issue CVE: NA

-------------------------------------------------

This patch implements the etmem swap feature.

etmem swap, like etmem scan, also communicated with the user space program thought a registered proc file system. It adds target pages to swap cache to be further reclaimed by kswapd, and dwell in swap space.

Signed-off-by: yanxiaodan yanxiaodan@huawei.com Signed-off-by: linmiaohe linmiaohe@huawei.com Signed-off-by: louhongxiang louhongxiang@huawei.com Signed-off-by: liubo liubo254@huawei.com Signed-off-by: geruijun geruijun@huawei.com Signed-off-by: Yuchen Tang tangyuchen5@huawei.com --- fs/proc/Makefile | 5 +- fs/proc/base.c | 2 + fs/proc/etmem_proc.c | 94 ++++++++++++++++++++++++++++++++++++ fs/proc/etmem_swap.c | 109 ++++++++++++++++++++++++++++++++++++++++++ fs/proc/internal.h | 1 + include/linux/etmem.h | 13 +++++ mm/Kconfig | 10 ++++ mm/Makefile | 1 + mm/etmem.c | 64 +++++++++++++++++++++++++ 9 files changed, 297 insertions(+), 2 deletions(-) create mode 100644 fs/proc/etmem_swap.c create mode 100644 mm/etmem.c

diff --git a/fs/proc/Makefile b/fs/proc/Makefile index b9a7bc7d8a75..fe283f354d61 100644 --- a/fs/proc/Makefile +++ b/fs/proc/Makefile @@ -34,5 +34,6 @@ proc-$(CONFIG_PROC_VMCORE) += vmcore.o proc-$(CONFIG_PRINTK) += kmsg.o proc-$(CONFIG_PROC_PAGE_MONITOR) += page.o proc-$(CONFIG_BOOT_CONFIG) += bootconfig.o -obj-$(CONFIG_ETMEM_SCAN) += etmem_scan.o -proc-${CONFIG_ETMEM} += etmem_proc.o +obj-$(CONFIG_ETMEM_SCAN) += etmem_scan.o +obj-$(CONFIG_ETMEM_SWAP) += etmem_swap.o +proc-${CONFIG_ETMEM} += etmem_proc.o diff --git a/fs/proc/base.c b/fs/proc/base.c index eb5ab83e5c22..5be0273a70c4 100644 --- a/fs/proc/base.c +++ b/fs/proc/base.c @@ -3357,6 +3357,7 @@ static const struct pid_entry tgid_base_stuff[] = { #endif #ifdef CONFIG_ETMEM REG("idle_pages", S_IRUSR|S_IWUSR, proc_mm_idle_operations), + REG("swap_pages", S_IWUSR, proc_mm_swap_operations), #endif #ifdef CONFIG_SECURITY DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations), @@ -3709,6 +3710,7 @@ static const struct pid_entry tid_base_stuff[] = { #endif #ifdef CONFIG_ETMEM REG("idle_pages", S_IRUSR|S_IWUSR, proc_mm_idle_operations), + REG("swap_pages", S_IWUSR, proc_mm_swap_operations), #endif #ifdef CONFIG_SECURITY DIR("attr", S_IRUGO|S_IXUGO, proc_attr_dir_inode_operations, proc_attr_dir_operations), diff --git a/fs/proc/etmem_proc.c b/fs/proc/etmem_proc.c index edacb9260345..2e6712cc43b2 100644 --- a/fs/proc/etmem_proc.c +++ b/fs/proc/etmem_proc.c @@ -120,3 +120,97 @@ const struct file_operations proc_mm_idle_operations = { .release = mm_idle_release, .unlocked_ioctl = mm_idle_ioctl, }; + +static DEFINE_SPINLOCK(swap_lock); + +static int page_swap_lock(struct file *file, int is_lock, struct file_lock *flock) +{ + if (is_lock) + spin_lock(&swap_lock); + else + spin_unlock(&swap_lock); + + return 0; +} +/*swap pages*/ +struct file_operations proc_swap_pages_operations = { + .flock = page_swap_lock, +}; +EXPORT_SYMBOL_GPL(proc_swap_pages_operations); + +static ssize_t mm_swap_write(struct file *file, const char __user *buf, + size_t count, loff_t *ppos) +{ + if (proc_swap_pages_operations.write) + return proc_swap_pages_operations.write(file, buf, count, ppos); + + return -1; +} + +static int mm_swap_open(struct inode *inode, struct file *file) +{ + struct mm_struct *mm = NULL; + struct module *module = NULL; + int ret = -1; + + if (!file_ns_capable(file, &init_user_ns, CAP_SYS_ADMIN)) + return -EPERM; + + page_swap_lock(NULL, 1, NULL); + module = proc_swap_pages_operations.owner; + if (module != NULL && try_module_get(module)) + ret = 0; + page_swap_lock(NULL, 0, NULL); + if (ret != 0) { + /* no swap ko installed, avoid to return valid file */ + return -ENODEV; + } + + mm = proc_mem_open(inode, PTRACE_MODE_READ); + if (IS_ERR(mm)) { + module_put(module); + return PTR_ERR(mm); + } + + file->private_data = mm; + + if (proc_swap_pages_operations.open) + ret = proc_swap_pages_operations.open(inode, file); + + if (ret != 0) + module_put(module); + + return ret; +} + +static int mm_swap_release(struct inode *inode, struct file *file) +{ + struct mm_struct *mm = file->private_data; + int ret = 0; + + if (mm) + mmdrop(mm); + + if (proc_swap_pages_operations.release) + ret = proc_swap_pages_operations.release(inode, file); + + if (proc_swap_pages_operations.owner) + module_put(proc_swap_pages_operations.owner); + + return ret; +} + +static long mm_swap_ioctl(struct file *filp, unsigned int cmd, unsigned long arg) +{ + if (proc_swap_pages_operations.unlocked_ioctl) + return proc_swap_pages_operations.unlocked_ioctl(filp, cmd, arg); + return 0; +} + +const struct file_operations proc_mm_swap_operations = { + .llseek = mem_lseek, + .write = mm_swap_write, + .open = mm_swap_open, + .release = mm_swap_release, + .unlocked_ioctl = mm_swap_ioctl, +}; diff --git a/fs/proc/etmem_swap.c b/fs/proc/etmem_swap.c new file mode 100644 index 000000000000..4aad6b9db9a6 --- /dev/null +++ b/fs/proc/etmem_swap.c @@ -0,0 +1,109 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <linux/init.h> +#include <linux/kernel.h> +#include <linux/module.h> +#include <linux/string.h> +#include <linux/proc_fs.h> +#include <linux/sched/mm.h> +#include <linux/mm.h> +#include <linux/swap.h> +#include <linux/mempolicy.h> +#include <linux/uaccess.h> +#include <linux/delay.h> +#include <linux/etmem.h> + +static ssize_t swap_pages_write(struct file *file, const char __user *buf, + size_t count, loff_t *ppos) +{ + char *p, *data, *data_ptr_res; + unsigned long vaddr; + struct mm_struct *mm = file->private_data; + struct page *page; + LIST_HEAD(pagelist); + int ret = 0; + + if (!mm || !mmget_not_zero(mm)) { + ret = -ESRCH; + goto out; + } + + if (count < 0) { + ret = -EOPNOTSUPP; + goto out_mm; + } + + data = memdup_user_nul(buf, count); + if (IS_ERR(data)) { + ret = PTR_ERR(data); + goto out_mm; + } + + data_ptr_res = data; + while ((p = strsep(&data, "\n")) != NULL) { + if (!*p) + continue; + + ret = kstrtoul(p, 16, &vaddr); + if (ret != 0) + continue; + + /* If get page struct failed, ignore it, get next page */ + page = get_page_from_vaddr(mm, vaddr); + if (!page) + continue; + + add_page_for_swap(page, &pagelist); + } + + if (!list_empty(&pagelist)) + reclaim_pages(&pagelist); + + ret = count; + kfree(data_ptr_res); +out_mm: + mmput(mm); +out: + return ret; +} + +static int swap_pages_open(struct inode *inode, struct file *file) +{ + if (!try_module_get(THIS_MODULE)) + return -EBUSY; + + return 0; +} + +static int swap_pages_release(struct inode *inode, struct file *file) +{ + module_put(THIS_MODULE); + return 0; +} + +extern struct file_operations proc_swap_pages_operations; + +static int swap_pages_entry(void) +{ + proc_swap_pages_operations.flock(NULL, 1, NULL); + proc_swap_pages_operations.owner = THIS_MODULE; + proc_swap_pages_operations.write = swap_pages_write; + proc_swap_pages_operations.open = swap_pages_open; + proc_swap_pages_operations.release = swap_pages_release; + proc_swap_pages_operations.flock(NULL, 0, NULL); + + return 0; +} + +static void swap_pages_exit(void) +{ + proc_swap_pages_operations.flock(NULL, 1, NULL); + proc_swap_pages_operations.owner = NULL; + proc_swap_pages_operations.write = NULL; + proc_swap_pages_operations.open = NULL; + proc_swap_pages_operations.release = NULL; + proc_swap_pages_operations.flock(NULL, 0, NULL); +} + +MODULE_LICENSE("GPL"); +module_init(swap_pages_entry); +module_exit(swap_pages_exit); diff --git a/fs/proc/internal.h b/fs/proc/internal.h index a9615455b709..be6d5dfc330c 100644 --- a/fs/proc/internal.h +++ b/fs/proc/internal.h @@ -305,6 +305,7 @@ extern const struct file_operations proc_clear_refs_operations; extern const struct file_operations proc_pagemap_operations; #ifdef CONFIG_ETMEM extern const struct file_operations proc_mm_idle_operations; +extern const struct file_operations proc_mm_swap_operations; #endif

extern unsigned long task_vsize(struct mm_struct *); diff --git a/include/linux/etmem.h b/include/linux/etmem.h index e8a2585f3891..5ebd1c3274b7 100644 --- a/include/linux/etmem.h +++ b/include/linux/etmem.h @@ -22,6 +22,19 @@ static inline struct kvm *mm_kvm(struct mm_struct *mm) } #endif

+extern int add_page_for_swap(struct page *page, struct list_head *pagelist); +extern struct page *get_page_from_vaddr(struct mm_struct *mm, + unsigned long vaddr); +#else /* !CONFIG_ETMEM */ +static inline int add_page_for_swap(struct page *page, struct list_head *pagelist) +{ + return 0; +}

+static inline struct page *get_page_from_vaddr(struct mm_struct *mm, + unsigned long vaddr) +{ + return NULL; +} #endif /* #ifdef CONFIG_ETMEM */ #endif /* define __MM_ETMEM_H_ */ diff --git a/mm/Kconfig b/mm/Kconfig index 980eaf75d87b..57818763e3c7 100644 --- a/mm/Kconfig +++ b/mm/Kconfig @@ -1359,6 +1359,16 @@ config ETMEM_SCAN scan results back to user space. etmem scan also supports virtual machine

+config ETMEM_SWAP + tristate "module: etmem page swap for etmem support" + depends on ETMEM + help + etmem swap is a critical component of the etmem feature. + When using etmem slide engine, etmem_swap.ko will add appointed pages + (ideally all of which are all rarely used, "cold" pages) to swapcache + proactively, which will later be reclaimed and added to swap space, + making room for more frequently used, "hot" pages. + config ETMEM bool "Enable etmem feature" depends on MMU diff --git a/mm/Makefile b/mm/Makefile index 6759053ed782..3d15ba814dd7 100644 --- a/mm/Makefile +++ b/mm/Makefile @@ -142,5 +142,6 @@ obj-$(CONFIG_GENERIC_IOREMAP) += ioremap.o obj-$(CONFIG_SHRINKER_DEBUG) += shrinker_debug.o obj-$(CONFIG_SHARE_POOL) += share_pool.o obj-$(CONFIG_MEMCG_MEMFS_INFO) += memcg_memfs_info.o +obj-$(CONFIG_ETMEM) += etmem.o obj-$(CONFIG_PAGE_CACHE_LIMIT) += page_cache_limit.o obj-$(CONFIG_CLEAR_FREELIST_PAGE) += clear_freelist_page.o diff --git a/mm/etmem.c b/mm/etmem.c new file mode 100644 index 000000000000..9a89bfcc1058 --- /dev/null +++ b/mm/etmem.c @@ -0,0 +1,64 @@ +// SPDX-License-Identifier: GPL-2.0 +#include <linux/mm.h> +#include <linux/swap.h> +#include <linux/pagemap.h> +#include <linux/memcontrol.h> +#include <linux/gfp.h> +#include <linux/mm_inline.h> +#include <linux/sysctl.h> +#include <linux/etmem.h> +#include "internal.h" + +int add_page_for_swap(struct page *page, struct list_head *pagelist) +{ + int err = -EBUSY; + struct page *head; + + /* If the page is mapped by more than one process, do not swap it */ + if (page_mapcount(page) > 1) + return -EACCES; + + if (PageHuge(page)) + return -EACCES; + + head = compound_head(page); + if (!folio_isolate_lru(page_folio(head))) { + put_page(page); + return err; + } + put_page(page); + if (PageUnevictable(page)) + putback_lru_page(page); + else + list_add_tail(&head->lru, pagelist); + + err = 0; + return err; +} +EXPORT_SYMBOL_GPL(add_page_for_swap); + +struct page *get_page_from_vaddr(struct mm_struct *mm, unsigned long vaddr) +{ + struct page *page; + struct vm_area_struct *vma; + unsigned int follflags; + + mmap_read_lock(mm); + + vma = find_vma(mm, vaddr); + if (!vma || vaddr < vma->vm_start || vma->vm_flags & VM_LOCKED) { + mmap_read_unlock(mm); + return NULL; + } + + follflags = FOLL_GET | FOLL_DUMP; + page = follow_page(vma, vaddr, follflags); + if (IS_ERR(page) || !page) { + mmap_read_unlock(mm); + return NULL; + } + + mmap_read_unlock(mm); + return page; +} +EXPORT_SYMBOL_GPL(get_page_from_vaddr);

euleros inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I8T1MB?from=project-issue CVE: NA

-------------------------------------------------

etmem, the memory vertical expansion technology,

In the current etmem process, memory page swapping is implemented by invoking shrink_page_list. When this interface is invoked for the first time, pages are added to the swap cache and written to disks.The swap cache page is reclaimed only when this interface is invoked for the second time and no process accesses the page.However, in the etmem process, the user mode scans pages that have been accessed, and the migration is not delivered to pages that are not accessed by processes. Therefore, the swap cache may always be occupied. To solve the preceding problem, add the logic for actively reclaiming the swap cache.When the swap cache occupies a large amount of memory, the system proactively scans the LRU linked list and reclaims the swap cache to save memory within the specified range.

Signed-off-by: liubo liubo254@huawei.com Signed-off-by: Yuchen Tang tangyuchen5@huawei.com --- fs/proc/etmem_swap.c | 174 +++++++++++++++++++++++++ include/linux/etmem.h | 28 ++++ mm/etmem.c | 289 ++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 491 insertions(+)

diff --git a/fs/proc/etmem_swap.c b/fs/proc/etmem_swap.c index 4aad6b9db9a6..d6461e48cd3a 100644 --- a/fs/proc/etmem_swap.c +++ b/fs/proc/etmem_swap.c @@ -10,8 +10,26 @@ #include <linux/mempolicy.h> #include <linux/uaccess.h> #include <linux/delay.h> +#include <linux/numa.h> +#include <linux/freezer.h> +#include <linux/kthread.h> +#include <linux/mm_inline.h> #include <linux/etmem.h>

+#define RECLAIM_SWAPCACHE_MAGIC 0X77 +#define SET_SWAPCACHE_WMARK _IOW(RECLAIM_SWAPCACHE_MAGIC, 0x02, unsigned int) +#define RECLAIM_SWAPCACHE_ON _IOW(RECLAIM_SWAPCACHE_MAGIC, 0x01, unsigned int) +#define RECLAIM_SWAPCACHE_OFF _IOW(RECLAIM_SWAPCACHE_MAGIC, 0x00, unsigned int) + +#define WATERMARK_MAX 100 +#define SWAP_SCAN_NUM_MAX 32 + +static struct task_struct *reclaim_swapcache_tk; +static bool enable_swapcache_reclaim; +static unsigned long swapcache_watermark[ETMEM_SWAPCACHE_NR_WMARK]; + +static DECLARE_WAIT_QUEUE_HEAD(reclaim_queue); + static ssize_t swap_pages_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos) { @@ -80,8 +98,153 @@ static int swap_pages_release(struct inode *inode, struct file *file) return 0; }

+/* check if swapcache meet requirements */ +static bool swapcache_balanced(void) +{ + return total_swapcache_pages() < swapcache_watermark[ETMEM_SWAPCACHE_WMARK_HIGH]; +} + +/* the flag present if swapcache reclaim is started */ +static bool swapcache_reclaim_enabled(void) +{ + return READ_ONCE(enable_swapcache_reclaim); +} + +static void start_swapcache_reclaim(void) +{ + if (swapcache_balanced()) + return; + /* RECLAIM_SWAPCACHE_ON trigger the thread to start running. */ + if (!waitqueue_active(&reclaim_queue)) + return; + + WRITE_ONCE(enable_swapcache_reclaim, true); + wake_up_interruptible(&reclaim_queue); +} + +static void stop_swapcache_reclaim(void) +{ + WRITE_ONCE(enable_swapcache_reclaim, false); +} + +static bool should_goto_sleep(void) +{ + if (swapcache_balanced()) + stop_swapcache_reclaim(); + + if (swapcache_reclaim_enabled()) + return false; + + return true; +} + +static int get_swapcache_watermark(unsigned int ratio) +{ + unsigned int low_watermark; + unsigned int high_watermark; + + low_watermark = ratio & 0xFF; + high_watermark = (ratio >> 8) & 0xFF; + if (low_watermark > WATERMARK_MAX || + high_watermark > WATERMARK_MAX || + low_watermark > high_watermark) + return -EPERM; + + swapcache_watermark[ETMEM_SWAPCACHE_WMARK_LOW] = totalram_pages() * + low_watermark / WATERMARK_MAX; + swapcache_watermark[ETMEM_SWAPCACHE_WMARK_HIGH] = totalram_pages() * + high_watermark / WATERMARK_MAX; + + return 0; +} + extern struct file_operations proc_swap_pages_operations;

+static void reclaim_swapcache_try_to_sleep(void) +{ + DEFINE_WAIT(wait); + + if (freezing(current) || kthread_should_stop()) + return; + + prepare_to_wait(&reclaim_queue, &wait, TASK_INTERRUPTIBLE); + if (should_goto_sleep()) { + if (!kthread_should_stop()) + schedule(); + } + finish_wait(&reclaim_queue, &wait); +} + +static void etmem_reclaim_swapcache(void) +{ + do_swapcache_reclaim(swapcache_watermark, + ARRAY_SIZE(swapcache_watermark)); + stop_swapcache_reclaim(); +} + +static int reclaim_swapcache_proactive(void *para) +{ + set_freezable(); + + while (1) { + bool ret; + + reclaim_swapcache_try_to_sleep(); + ret = try_to_freeze(); + if (kthread_should_stop()) + break; + + if (ret) + continue; + + etmem_reclaim_swapcache(); + } + + return 0; +} + +static int reclaim_swapcache_run(void) +{ + int ret = 0; + + reclaim_swapcache_tk = kthread_run(reclaim_swapcache_proactive, NULL, + "etmem_recalim_swapcache"); + if (IS_ERR(reclaim_swapcache_tk)) { + ret = PTR_ERR(reclaim_swapcache_tk); + reclaim_swapcache_tk = NULL; + } + return ret; +} + +static long swap_page_ioctl(struct file *filp, unsigned int cmd, + unsigned long arg) +{ + void __user *argp = (void __user *)arg; + unsigned int ratio; + + switch (cmd) { + case RECLAIM_SWAPCACHE_ON: + if (swapcache_reclaim_enabled()) + return 0; + start_swapcache_reclaim(); + break; + case RECLAIM_SWAPCACHE_OFF: + stop_swapcache_reclaim(); + break; + case SET_SWAPCACHE_WMARK: + if (get_user(ratio, (unsigned int __user *)argp)) + return -EFAULT; + + if (get_swapcache_watermark(ratio) != 0) + return -EFAULT; + break; + default: + return -EPERM; + } + + return 0; +} + static int swap_pages_entry(void) { proc_swap_pages_operations.flock(NULL, 1, NULL); @@ -89,8 +252,12 @@ static int swap_pages_entry(void) proc_swap_pages_operations.write = swap_pages_write; proc_swap_pages_operations.open = swap_pages_open; proc_swap_pages_operations.release = swap_pages_release; + proc_swap_pages_operations.unlocked_ioctl = swap_page_ioctl; proc_swap_pages_operations.flock(NULL, 0, NULL);

+ enable_swapcache_reclaim = false; + reclaim_swapcache_run(); + return 0; }

@@ -101,7 +268,14 @@ static void swap_pages_exit(void) proc_swap_pages_operations.write = NULL; proc_swap_pages_operations.open = NULL; proc_swap_pages_operations.release = NULL; + proc_swap_pages_operations.unlocked_ioctl = NULL; proc_swap_pages_operations.flock(NULL, 0, NULL); + + if (!IS_ERR(reclaim_swapcache_tk)) { + kthread_stop(reclaim_swapcache_tk); + reclaim_swapcache_tk = NULL; + } + return; }

MODULE_LICENSE("GPL"); diff --git a/include/linux/etmem.h b/include/linux/etmem.h index 9ec9657e56ed..73483a6fc7d7 100644 --- a/include/linux/etmem.h +++ b/include/linux/etmem.h @@ -9,6 +9,20 @@ #include <linux/page-flags.h>

#ifdef CONFIG_ETMEM +/** + * list_for_each_entry_safe_reverse_from + * @pos: the type * to use as a loop cursor. + * @n: another type * to use as temporary storage + * @head: the head for your list. + * @member: the name of the list_head within the struct. + * + * Iterate backwards over list of given type from current point, safe against + * removal of list entry. + */ +#define list_for_each_entry_safe_reverse_from(pos, n, head, member) \ + for (n = list_prev_entry(pos, member); \ + &pos->member != (head); \ + pos = n, n = list_prev_entry(n, member))

#if IS_ENABLED(CONFIG_KVM) static inline struct kvm *mm_kvm(struct mm_struct *mm) @@ -22,11 +36,19 @@ static inline struct kvm *mm_kvm(struct mm_struct *mm) } #endif

+enum etmem_swapcache_watermark_en { + ETMEM_SWAPCACHE_WMARK_LOW, + ETMEM_SWAPCACHE_WMARK_HIGH, + ETMEM_SWAPCACHE_NR_WMARK +}; + extern int add_page_for_swap(struct page *page, struct list_head *pagelist); extern struct page *get_page_from_vaddr(struct mm_struct *mm, unsigned long vaddr); extern struct kobj_attribute kernel_swap_enable_attr; extern bool kernel_swap_enabled(void); +extern int do_swapcache_reclaim(unsigned long *swapcache_watermark, + unsigned int watermark_nr); #else /* !CONFIG_ETMEM */ static inline int add_page_for_swap(struct page *page, struct list_head *pagelist) { @@ -43,5 +65,11 @@ static inline bool kernel_swap_enabled(void) { return true; } + +static inline int do_swapcache_reclaim(unsigned long *swapcache_watermark, + unsigned int watermark_nr) +{ + return 0; +} #endif /* #ifdef CONFIG_ETMEM */ #endif /* define __MM_ETMEM_H_ */ diff --git a/mm/etmem.c b/mm/etmem.c index acd32e71a643..fcde2f604eea 100644 --- a/mm/etmem.c +++ b/mm/etmem.c @@ -93,3 +93,292 @@ struct page *get_page_from_vaddr(struct mm_struct *mm, unsigned long vaddr) return page; } EXPORT_SYMBOL_GPL(get_page_from_vaddr); + +static int add_page_for_reclaim_swapcache(struct page *page, + struct list_head *pagelist, struct lruvec *lruvec, enum lru_list lru) +{ + struct page *head; + + /* If the page is mapped by more than one process, do not swap it */ + if (page_mapcount(page) > 1) + return -EINVAL; + + if (PageHuge(page)) + return -EINVAL; + + head = compound_head(page); + if (!PageLRU(head) || PageUnevictable(head)) + return -EBUSY; + + if (unlikely(!get_page_unless_zero(page))) + return -EBUSY; + + if (!TestClearPageLRU(page)) { + /* + * This page may in other isolation path, + * but we still hold lru_lock. + */ + put_page(page); + return -EBUSY; + } + + list_move(&head->lru, pagelist); + update_lru_size(lruvec, lru, page_zonenum(head), -thp_nr_pages(head)); + + return 0; +} + +static unsigned long reclaim_swapcache_pages_from_list(int nid, + struct list_head *page_list, unsigned long reclaim_num, bool putback_flag) +{ + unsigned long nr_reclaimed = 0; + unsigned long nr_moved = 0; + struct page *page, *next; + LIST_HEAD(swap_pages); + struct pglist_data *pgdat = NULL; + + pgdat = NODE_DATA(nid); + + if (putback_flag) + goto putback_list; + + if (reclaim_num == 0) + return 0; + + list_for_each_entry_safe(page, next, page_list, lru) { + if (!page_is_file_lru(page) && !__PageMovable(page) + && PageSwapCache(page)) { + ClearPageActive(page); + list_move(&page->lru, &swap_pages); + nr_moved++; + } + + if (nr_moved >= reclaim_num) + break; + } + + /* swap the pages */ + if (pgdat) + nr_reclaimed = reclaim_pages(&swap_pages); + + return nr_reclaimed; + +putback_list: + while (!list_empty(page_list)) { + page = lru_to_page(page_list); + list_del(&page->lru); + putback_lru_page(page); + } + + return nr_reclaimed; +} + +#define SWAP_SCAN_NUM_MAX 32 + +static bool swapcache_below_watermark(unsigned long *swapcache_watermark) +{ + return total_swapcache_pages() < swapcache_watermark[ETMEM_SWAPCACHE_WMARK_LOW]; +} + +static unsigned long get_swapcache_reclaim_num(unsigned long *swapcache_watermark) +{ + return total_swapcache_pages() > + swapcache_watermark[ETMEM_SWAPCACHE_WMARK_LOW] ? + (total_swapcache_pages() - swapcache_watermark[ETMEM_SWAPCACHE_WMARK_LOW]) : 0; +} + +/* + * The main function to reclaim swapcache, the whole reclaim process is + * divided into 3 steps. + * 1. get the total_swapcache_pages num to reclaim. + * 2. scan the LRU linked list of each memory node to obtain the + * swapcache pages that can be reclaimd. + * 3. reclaim the swapcache page until the requirements are meet. + */ +int do_swapcache_reclaim(unsigned long *swapcache_watermark, + unsigned int watermark_nr) +{ + int err = -EINVAL; + unsigned long swapcache_to_reclaim = 0; + unsigned long nr_reclaimed = 0; + unsigned long swapcache_total_reclaimable = 0; + unsigned long reclaim_page_count = 0; + + unsigned long *nr = NULL; + unsigned long *nr_to_reclaim = NULL; + struct list_head *swapcache_list = NULL; + + int nid = 0; + struct lruvec *lruvec = NULL; + struct list_head *src = NULL; + struct page *page = NULL; + struct page *next = NULL; + struct page *pos = NULL; + + struct mem_cgroup *memcg = NULL; + struct mem_cgroup *target_memcg = NULL; + + pg_data_t *pgdat = NULL; + unsigned int scan_count = 0; + int nid_num = 0; + + if (swapcache_watermark == NULL || + watermark_nr < ETMEM_SWAPCACHE_NR_WMARK) + return err; + + /* get the total_swapcache_pages num to reclaim. */ + swapcache_to_reclaim = get_swapcache_reclaim_num(swapcache_watermark); + if (swapcache_to_reclaim <= 0) + return err; + + nr = kcalloc(MAX_NUMNODES, sizeof(unsigned long), GFP_KERNEL); + if (nr == NULL) + return -ENOMEM; + + nr_to_reclaim = kcalloc(MAX_NUMNODES, sizeof(unsigned long), GFP_KERNEL); + if (nr_to_reclaim == NULL) { + kfree(nr); + return -ENOMEM; + } + + swapcache_list = kcalloc(MAX_NUMNODES, sizeof(struct list_head), GFP_KERNEL); + if (swapcache_list == NULL) { + kfree(nr); + kfree(nr_to_reclaim); + return -ENOMEM; + } + + /* + * scan the LRU linked list of each memory node to obtain the + * swapcache pages that can be reclaimd. + */ + for_each_node_state(nid, N_MEMORY) { + INIT_LIST_HEAD(&swapcache_list[nid_num]); + cond_resched(); + + pgdat = NODE_DATA(nid); + + memcg = mem_cgroup_iter(target_memcg, NULL, NULL); + do { + cond_resched(); + pos = NULL; + lruvec = mem_cgroup_lruvec(memcg, pgdat); + src = &(lruvec->lists[LRU_INACTIVE_ANON]); + spin_lock_irq(&lruvec->lru_lock); + scan_count = 0; + + /* + * Scan the swapcache pages that are not mapped from + * the end of the LRU linked list, scan SWAP_SCAN_NUM_MAX + * pages each time, and record the scan end point page. + */ + + pos = list_last_entry(src, struct page, lru); + spin_unlock_irq(&lruvec->lru_lock); +do_scan: + cond_resched(); + scan_count = 0; + spin_lock_irq(&lruvec->lru_lock); + + /* + * check if pos page is been released or not in LRU list, if true, + * cancel the subsequent page scanning of the current node. + */ + if (!pos || list_entry_is_head(pos, src, lru)) { + spin_unlock_irq(&lruvec->lru_lock); + continue; + } + + if (!PageLRU(pos) || folio_lru_list(page_folio(pos)) != LRU_INACTIVE_ANON) { + spin_unlock_irq(&lruvec->lru_lock); + continue; + } + + page = pos; + pos = NULL; + /* Continue to scan down from the last scan breakpoint */ + list_for_each_entry_safe_reverse_from(page, next, src, lru) { + scan_count++; + pos = next; + if (scan_count >= SWAP_SCAN_NUM_MAX) + break; + + if (!PageSwapCache(page)) + continue; + + if (page_mapped(page)) + continue; + + if (add_page_for_reclaim_swapcache(page, + &swapcache_list[nid_num], + lruvec, LRU_INACTIVE_ANON) != 0) + continue; + + nr[nid_num]++; + swapcache_total_reclaimable++; + } + spin_unlock_irq(&lruvec->lru_lock); + + /* + * Check whether the scanned pages meet + * the reclaim requirements. + */ + if (swapcache_total_reclaimable <= swapcache_to_reclaim || + scan_count >= SWAP_SCAN_NUM_MAX) + goto do_scan; + + } while ((memcg = mem_cgroup_iter(target_memcg, memcg, NULL))); + + /* Start reclaiming the next memory node. */ + nid_num++; + } + + /* reclaim the swapcache page until the requirements are meet. */ + do { + nid_num = 0; + reclaim_page_count = 0; + + /* start swapcache page reclaim for each node. */ + for_each_node_state(nid, N_MEMORY) { + cond_resched(); + + nr_to_reclaim[nid_num] = (swapcache_total_reclaimable == 0) ? 0 : + ((swapcache_to_reclaim * nr[nid_num]) / + swapcache_total_reclaimable); + + reclaim_page_count += reclaim_swapcache_pages_from_list(nid, + &swapcache_list[nid_num], + nr_to_reclaim[nid_num], false); + nid_num++; + } + + nr_reclaimed += reclaim_page_count; + + /* + * Check whether the swapcache page reaches the reclaim requirement or + * the number of the swapcache page reclaimd is 0. Stop reclaim. + */ + if (nr_reclaimed >= swapcache_to_reclaim || reclaim_page_count == 0) + goto exit; + } while (!swapcache_below_watermark(swapcache_watermark) || + nr_reclaimed < swapcache_to_reclaim); +exit: + nid_num = 0; + /* + * Repopulate the swapcache pages that are not reclaimd back + * to the LRU linked list. + */ + for_each_node_state(nid, N_MEMORY) { + cond_resched(); + reclaim_swapcache_pages_from_list(nid, + &swapcache_list[nid_num], 0, true); + nid_num++; + } + + kfree(nr); + kfree(nr_to_reclaim); + kfree(swapcache_list); + + return 0; +} +EXPORT_SYMBOL_GPL(do_swapcache_reclaim);