From patchwork Fri Dec 18 14:25:09 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: hejingxian X-Patchwork-Id: 11982417 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-10.8 required=3.0 tests=BAYES_00, HEADER_FROM_DIFFERENT_DOMAINS,HTML_MESSAGE,INCLUDES_CR_TRAILER,INCLUDES_PATCH, MAILING_LIST_MULTI,SPF_HELO_NONE,SPF_PASS,URIBL_BLOCKED autolearn=ham autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id 22057C4361B for ; Fri, 18 Dec 2020 15:06:09 +0000 (UTC) Received: from kanga.kvack.org (kanga.kvack.org [205.233.56.17]) by mail.kernel.org (Postfix) with ESMTP id 1BC0823AFD for ; Fri, 18 Dec 2020 15:06:07 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 1BC0823AFD Authentication-Results: mail.kernel.org; dmarc=none (p=none dis=none) header.from=huawei.com Authentication-Results: mail.kernel.org; spf=pass smtp.mailfrom=owner-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix) id 44D668D0001; Fri, 18 Dec 2020 10:06:07 -0500 (EST) Received: by kanga.kvack.org (Postfix, from userid 40) id 3D4C26B006C; Fri, 18 Dec 2020 10:06:07 -0500 (EST) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id 1B72C8D0001; Fri, 18 Dec 2020 10:06:07 -0500 (EST) X-Delivered-To: linux-mm@kvack.org Received: from forelay.hostedemail.com (smtprelay0207.hostedemail.com [216.40.44.207]) by kanga.kvack.org (Postfix) with ESMTP id CAFB46B005D for ; Fri, 18 Dec 2020 10:06:06 -0500 (EST) Received: from smtpin03.hostedemail.com (10.5.19.251.rfc1918.com [10.5.19.251]) by forelay04.hostedemail.com (Postfix) with ESMTP id 682ED1EF3 for ; Fri, 18 Dec 2020 15:06:06 +0000 (UTC) X-FDA: 77606728332.03.push43_4800a952743e Received: from filter.hostedemail.com (10.5.16.251.rfc1918.com [10.5.16.251]) by smtpin03.hostedemail.com (Postfix) with ESMTP id 7969214DA0 for ; Fri, 18 Dec 2020 15:05:56 +0000 (UTC) X-HE-Tag: push43_4800a952743e X-Filterd-Recvd-Size: 256710 Received: from szxga01-in.huawei.com (szxga01-in.huawei.com [45.249.212.187]) by imf46.hostedemail.com (Postfix) with ESMTP for ; Fri, 18 Dec 2020 15:05:52 +0000 (UTC) Received: from dggeme702-chm.china.huawei.com (unknown [172.30.72.55]) by szxga01-in.huawei.com (SkyGuard) with ESMTP id 4CyB2W0wXSzXpkh; Fri, 18 Dec 2020 22:24:35 +0800 (CST) Received: from dggema764-chm.china.huawei.com (10.1.198.206) by dggeme702-chm.china.huawei.com (10.1.199.98) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_CBC_SHA256_P256) id 15.1.1913.5; Fri, 18 Dec 2020 22:25:09 +0800 Received: from dggema764-chm.china.huawei.com ([10.9.49.86]) by dggema764-chm.china.huawei.com ([10.9.49.86]) with mapi id 15.01.1913.007; Fri, 18 Dec 2020 22:25:09 +0800 From: hejingxian To: Andrew Morton , "linux-kernel@vger.kernel.org" , "linux-mm@kvack.org" CC: Hushiyuan , "hewenliang (C)" Subject: [PATCH] add pin memory method for checkout add restore Thread-Topic: [PATCH] add pin memory method for checkout add restore Thread-Index: AdbVSZMYq89Z1i35SCKgJJwENVHCjQ== Date: Fri, 18 Dec 2020 14:25:09 +0000 Message-ID: Accept-Language: zh-CN, en-US Content-Language: zh-CN X-MS-Has-Attach: X-MS-TNEF-Correlator: x-originating-ip: [10.174.177.226] MIME-Version: 1.0 X-CFilter-Loop: Reflected X-Bogosity: Ham, tests=bogofilter, spamicity=0.000000, version=1.2.4 Sender: owner-linux-mm@kvack.org Precedence: bulk X-Loop: owner-majordomo@kvack.org List-ID: From: Jingxian He > Date: Thu, 10 Dec 2020 20:31:15 +0800 Subject: [PATCH] add pin memory method for checkout add restore We can use the checkpoint and restore in userspace(criu) method to dump and restore tasks when updating the kernel. Currently, criu needs dump all memory data of tasks to files. When the memory size is very large(larger than 1G), the cost time of the dumping data will be very long(more than 1 min). We can pin the memory data of tasks and collect the corresponding physical pages mapping info in checkpoint process, and remap the physical pages to restore tasks after kernel is updated. The pin memory area info is saved in the reserve memblock named nvwa_res_first, which can keep usable in the kernel update process. The pin memory driver provides the following ioctl command for criu: 1) SET_PIN_MEM_AREA: set pin memory area, which can be remap to the restore task. 2) CLEAR_PIN_MEM_AREA: clear the pin memory area info, which enable user reset the pin data. 3) REMAP_PIN_MEM_AREA: remap the pages of the pin memory to the restore task. Signed-off-by: Jingxian He > --- arch/arm64/kernel/setup.c | 7 + arch/arm64/mm/init.c | 62 +++- drivers/char/Kconfig | 7 + drivers/char/Makefile | 1 + drivers/char/pin_memory.c | 198 +++++++++++++ include/linux/crash_core.h | 5 + include/linux/pin_mem.h | 62 ++++ kernel/crash_core.c | 11 + mm/Kconfig | 6 + mm/Makefile | 1 + mm/huge_memory.c | 61 ++++ mm/memory.c | 68 +++++ mm/pin_mem.c | 691 +++++++++++++++++++++++++++++++++++++++++++++ 13 files changed, 1179 insertions(+), 1 deletion(-) create mode 100644 drivers/char/pin_memory.c create mode 100644 include/linux/pin_mem.h create mode 100644 mm/pin_mem.c -- 1.8.3.1 diff --git a/arch/arm64/kernel/setup.c b/arch/arm64/kernel/setup.c index 56f6645..40751ed 100644 --- a/arch/arm64/kernel/setup.c +++ b/arch/arm64/kernel/setup.c @@ -50,6 +50,9 @@ #include #include #include +#ifdef CONFIG_PIN_MEMORY +#include +#endif static int num_standard_resources; static struct resource *standard_resources; @@ -243,6 +246,10 @@ static void __init request_standard_resources(void) crashk_res.end <= res->end) request_resource(res, &crashk_res); #endif +#ifdef CONFIG_PIN_MEMORY + if (pin_memory_resource.end) + insert_resource(&iomem_resource, &pin_memory_resource); +#endif } } diff --git a/arch/arm64/mm/init.c b/arch/arm64/mm/init.c index b65dffd..dee3192 100644 --- a/arch/arm64/mm/init.c +++ b/arch/arm64/mm/init.c @@ -41,7 +41,9 @@ #include #include #include - +#ifdef CONFIG_PIN_MEMORY +#include +#endif #define ARM64_ZONE_DMA_BITS 30 /* @@ -68,6 +70,16 @@ phys_addr_t arm64_dma_phys_limit __ro_after_init; static phys_addr_t arm64_dma32_phys_limit __ro_after_init; +#ifdef CONFIG_PIN_MEMORY +struct resource pin_memory_resource = { + .name = "Pin memory maps", + .start = 0, + .end = 0, + .flags = IORESOURCE_MEM, + .desc = IORES_DESC_PIN_MEM_MAPS +}; +#endif + #ifdef CONFIG_KEXEC_CORE /* * reserve_crashkernel() - reserves memory for crash kernel @@ -129,6 +141,47 @@ static void __init reserve_crashkernel(void) } #endif /* CONFIG_KEXEC_CORE */ +#ifdef CONFIG_PIN_MEMORY +static void __init reserve_pin_memory_res(void) +{ + unsigned long long mem_start, mem_len; + int ret; + + ret = parse_pin_memory(boot_command_line, memblock_phys_mem_size(), + &mem_len, &mem_start); + if (ret || !mem_len) + return; + + mem_len = PAGE_ALIGN(mem_len); + + if (!memblock_is_region_memory(mem_start, mem_len)) { + pr_warn("cannot reserve for pin memory: region is not memory!\n"); + return; + } + + if (memblock_is_region_reserved(mem_start, mem_len)) { + pr_warn("cannot reserve for pin memory: region overlaps reserved memory!\n"); + return; + } + + if (!IS_ALIGNED(mem_start, SZ_2M)) { + pr_warn("cannot reserve for pin memory: base address is not 2MB aligned\n"); + return; + } + + memblock_reserve(mem_start, mem_len); + pr_debug("pin memory resource reserved: 0x%016llx - 0x%016llx (%lld MB)\n", + mem_start, mem_start + mem_len, mem_len >> 20); + + pin_memory_resource.start = mem_start; + pin_memory_resource.end = mem_start + mem_len - 1; +} +#else +static void __init reserve_pin_memory_res(void) +{ +} +#endif /* CONFIG_PIN_MEMORY */ + #ifdef CONFIG_CRASH_DUMP static int __init early_init_dt_scan_elfcorehdr(unsigned long node, const char *uname, int depth, void *data) @@ -452,6 +505,8 @@ void __init arm64_memblock_init(void) reserve_crashkernel(); + reserve_pin_memory_res(); + reserve_elfcorehdr(); high_memory = __va(memblock_end_of_DRAM() - 1) + 1; @@ -573,6 +628,11 @@ void __init mem_init(void) /* this will put all unused low memory onto the freelists */ memblock_free_all(); +#ifdef CONFIG_PIN_MEMORY + /* pre alloc the pages for pin memory */ + init_reserve_page_map((unsigned long)pin_memory_resource.start, + (unsigned long)(pin_memory_resource.end - pin_memory_resource.start)); +#endif mem_init_print_info(NULL); /* diff --git a/drivers/char/Kconfig b/drivers/char/Kconfig index 26956c0..73af2f0 100644 --- a/drivers/char/Kconfig +++ b/drivers/char/Kconfig @@ -560,3 +560,10 @@ config RANDOM_TRUST_BOOTLOADER booloader is trustworthy so it will be added to the kernel's entropy pool. Otherwise, say N here so it will be regarded as device input that only mixes the entropy pool. + +config PIN_MEMORY_DEV + bool "/dev/pinmem character device" + depends PIN_MEMORY + default n + help + pin memory driver diff --git a/drivers/char/Makefile b/drivers/char/Makefile index 7c5ea6f..1941642 100644 --- a/drivers/char/Makefile +++ b/drivers/char/Makefile @@ -52,3 +52,4 @@ js-rtc-y = rtc.o obj-$(CONFIG_XILLYBUS) += xillybus/ obj-$(CONFIG_POWERNV_OP_PANEL) += powernv-op-panel.o obj-$(CONFIG_ADI) += adi.o +obj-$(CONFIG_PIN_MEMORY_DEV) += pin_memory.o diff --git a/drivers/char/pin_memory.c b/drivers/char/pin_memory.c new file mode 100644 index 00000000..a0464e1 --- /dev/null +++ b/drivers/char/pin_memory.c @@ -0,0 +1,198 @@ +/* + * Copyright @ Huawei Technologies Co., Ltd. 2020-2020. ALL rights reserved. + * Description: Euler pin memory driver + */ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define MAX_PIN_MEM_AREA_NUM 16 +struct _pin_mem_area { + unsigned long virt_start; + unsigned long virt_end; +}; + +struct pin_mem_area_set { + unsigned int pid; + unsigned int area_num; + struct _pin_mem_area mem_area[MAX_PIN_MEM_AREA_NUM]; +}; + +#define PIN_MEM_MAGIC 0x59 +#define _SET_PIN_MEM_AREA 1 +#define _CLEAR_PIN_MEM_AREA 2 +#define _REMAP_PIN_MEM_AREA 3 +#define SET_PIN_MEM_AREA _IOW(PIN_MEM_MAGIC, _SET_PIN_MEM_AREA, struct pin_mem_area_set) +#define CLEAR_PIN_MEM_AREA _IOW(PIN_MEM_MAGIC, _CLEAR_PIN_MEM_AREA, int) +#define REMAP_PIN_MEM_AREA _IOW(PIN_MEM_MAGIC, _REMAP_PIN_MEM_AREA, int) + +static int set_pin_mem(struct pin_mem_area_set *pmas) +{ + int i; + int ret = 0; + struct _pin_mem_area *pma; + struct mm_struct *mm; + struct task_struct *task; + struct pid *pid_s; + + pid_s = find_get_pid(pmas->pid); + if (!pid_s) { + pr_warn("Get pid struct fail:%d.\n", pmas->pid); + goto fail; + } + rcu_read_lock(); + task = pid_task(pid_s, PIDTYPE_PID); + if (!task) { + pr_warn("Get task struct fail:%d.\n", pmas->pid); + goto fail; + } + mm = get_task_mm(task); + for (i = 0; i < pmas->area_num; i++) { + pma = &(pmas->mem_area[i]); + ret = pin_mem_area(task, mm, pma->virt_start, pma->virt_end); + if (ret) { + mmput(mm); + goto fail; + } + } + mmput(mm); + rcu_read_unlock(); + return ret; + +fail: + rcu_read_unlock(); + return -EFAULT; +} + +static int set_pin_mem_area(unsigned long arg) +{ + struct pin_mem_area_set pmas; + void __user *buf = (void __user *)arg; + + if (!access_ok(buf, sizeof(pmas))) + return -EFAULT; + if (copy_from_user(&pmas, buf, sizeof(pmas))) + return -EINVAL; + if (pmas.area_num > MAX_PIN_MEM_AREA_NUM) { + pr_warn("Input area_num is too large.\n"); + return -EINVAL; + } + + return set_pin_mem(&pmas); +} + +static int pin_mem_remap(unsigned long arg) +{ + int pid; + struct task_struct *task; + struct mm_struct *mm; + vm_fault_t ret; + void __user *buf = (void __user *)arg; + struct pid *pid_s; + + if (!access_ok(buf, sizeof(int))) + return -EINVAL; + if (copy_from_user(&pid, buf, sizeof(int))) + return -EINVAL; + + pid_s = find_get_pid(pid); + if (!pid_s) { + pr_warn("Get pid struct fail:%d.\n", pid); + return -EINVAL; + } + rcu_read_lock(); + task = pid_task(pid_s, PIDTYPE_PID); + if (!task) { + pr_warn("Get task struct fail:%d.\n", pid); + goto fault; + } + mm = get_task_mm(task); + ret = do_mem_remap(pid, mm); + if (ret) { + pr_warn("Handle pin memory remap fail.\n"); + mmput(mm); + goto fault; + } + mmput(mm); + rcu_read_unlock(); + return 0; + +fault: + rcu_read_unlock(); + return -EFAULT; +} + +static long pin_memory_ioctl(struct file *file, unsigned cmd, unsigned long arg) +{ + long ret = 0; + + if (_IOC_TYPE(cmd) != PIN_MEM_MAGIC) + return -EINVAL; + if (_IOC_NR(cmd) > _REMAP_PIN_MEM_AREA) + return -EINVAL; + + switch (cmd) { + case SET_PIN_MEM_AREA: + ret = set_pin_mem_area(arg); + break; + case CLEAR_PIN_MEM_AREA: + clear_pin_memory_record(); + break; + case REMAP_PIN_MEM_AREA: + ret = pin_mem_remap(arg); + break; + default: + return -EINVAL; + } + return ret; +} + +static const struct file_operations pin_memory_fops = { + .owner = THIS_MODULE, + .unlocked_ioctl = pin_memory_ioctl, + .compat_ioctl = pin_memory_ioctl, +}; + +static struct miscdevice pin_memory_miscdev = { + .minor = MISC_DYNAMIC_MINOR, + .name = "pinmem", + .fops = &pin_memory_fops, +}; + +static int pin_memory_init(void) +{ + int err = misc_register(&pin_memory_miscdev); + if (!err) { + pr_info("pin_memory init\n"); + } else { + pr_warn("pin_memory init failed!\n"); + } + return err; +} + +static void pin_memory_exit(void) +{ + misc_deregister(&pin_memory_miscdev); + pr_info("pin_memory ko exists!\n"); +} + +module_init(pin_memory_init); +module_exit(pin_memory_exit); + +MODULE_LICENSE("GPL"); +MODULE_AUTHOR("Euler"); +MODULE_DESCRIPTION("pin memory"); diff --git a/include/linux/crash_core.h b/include/linux/crash_core.h index 525510a..5baf40d 100644 --- a/include/linux/crash_core.h +++ b/include/linux/crash_core.h @@ -75,4 +75,9 @@ int parse_crashkernel_high(char *cmdline, unsigned long long system_ram, int parse_crashkernel_low(char *cmdline, unsigned long long system_ram, unsigned long long *crash_size, unsigned long long *crash_base); +#ifdef CONFIG_PIN_MEMORY +int __init parse_pin_memory(char *cmdline, unsigned long long system_ram, + unsigned long long *pin_size, unsigned long long *pin_base); +#endif + #endif /* LINUX_CRASH_CORE_H */ diff --git a/include/linux/pin_mem.h b/include/linux/pin_mem.h new file mode 100644 index 00000000..0ca44ac --- /dev/null +++ b/include/linux/pin_mem.h @@ -0,0 +1,62 @@ +/* + * Copyright (C) 2020. Huawei Technologies Co., Ltd. All rights reserved. + * Provide the pin memory method for check point and restore task. + */ +#ifndef _LINUX_PIN_MEMORY_H +#define _LINUX_PIN_MEMORY_H + +#ifdef CONFIG_PIN_MEMORY +#include +#include +#include +#include +#ifdef CONFIG_ARM64 +#include +#endif + +#define PAGE_BUDDY_MAPCOUNT_VALUE (~PG_buddy) + +#define COLLECT_PAGES_FINISH 1 +#define COLLECT_PAGES_NEED_CONTINUE -1 +#define COLLECT_PAGES_FAIL 0 + +#define COMPOUND_PAD_MASK 0xffffffff +#define COMPOUND_PAD_START 0x88 +#define COMPOUND_PAD_DELTA 0x40 +#define LIST_POISON4 0xdead000000000400 + +#define next_pme(pme) ((unsigned long *)(pme + 1) + pme->nr_pages) + +struct page_map_entry { + unsigned long virt_addr; + unsigned int nr_pages; + unsigned int is_huge_page; + unsigned long phy_addr_array[0]; +}; + +struct page_map_info { + int pid; + int pid_reserved; + unsigned int entry_num; + struct page_map_entry *pme; +}; + +extern struct page_map_info *get_page_map_info(int pid); +extern struct page_map_info *create_page_map_info(int pid); +extern vm_fault_t do_mem_remap(int pid, struct mm_struct *mm); +extern vm_fault_t do_anon_page_remap(struct vm_area_struct *vma, unsigned long address, + pmd_t *pmd, struct page *page); +extern void clear_pin_memory_record(void); +extern int pin_mem_area(struct task_struct *task, struct mm_struct *mm, + unsigned long start_addr, unsigned long end_addr); +extern vm_fault_t do_anon_huge_page_remap(struct vm_area_struct *vma, unsigned long address, + pmd_t *pmd, struct page *page); + +/* reserve space for pin memory*/ +#ifdef CONFIG_ARM64 +extern struct resource pin_memory_resource; +#endif +extern void init_reserve_page_map(unsigned long map_addr, unsigned long map_size); + +#endif /* CONFIG_PIN_MEMORY */ +#endif /* _LINUX_PIN_MEMORY_H */ diff --git a/kernel/crash_core.c b/kernel/crash_core.c index 9f1557b..7512696 100644 --- a/kernel/crash_core.c +++ b/kernel/crash_core.c @@ -292,6 +292,17 @@ int __init parse_crashkernel_low(char *cmdline, "crashkernel=", suffix_tbl[SUFFIX_LOW]); } +#ifdef CONFIG_PIN_MEMORY +int __init parse_pin_memory(char *cmdline, + unsigned long long system_ram, + unsigned long long *pin_size, + unsigned long long *pin_base) +{ + return __parse_crashkernel(cmdline, system_ram, pin_size, pin_base, + "pinmemory=", NULL); +} +#endif + Elf_Word *append_elf_note(Elf_Word *buf, char *name, unsigned int type, void *data, size_t data_len) { diff --git a/mm/Kconfig b/mm/Kconfig index ab80933..c2dd088 100644 --- a/mm/Kconfig +++ b/mm/Kconfig @@ -739,4 +739,10 @@ config ARCH_HAS_HUGEPD config MAPPING_DIRTY_HELPERS bool +config PIN_MEMORY + bool "Support for pin memory" + depends on CHECKPOINT_RESTORE + help + Say y here to enable the pin memory feature for checkpoint + and restore. endmenu diff --git a/mm/Makefile b/mm/Makefile index 1937cc2..7e1984e 100644 --- a/mm/Makefile +++ b/mm/Makefile @@ -108,3 +108,4 @@ obj-$(CONFIG_ZONE_DEVICE) += memremap.o obj-$(CONFIG_HMM_MIRROR) += hmm.o obj-$(CONFIG_MEMFD_CREATE) += memfd.o obj-$(CONFIG_MAPPING_DIRTY_HELPERS) += mapping_dirty_helpers.o +obj-$(CONFIG_PIN_MEMORY) += pin_mem.o diff --git a/mm/huge_memory.c b/mm/huge_memory.c index a880932..93dc582 100644 --- a/mm/huge_memory.c +++ b/mm/huge_memory.c @@ -3083,4 +3083,65 @@ void remove_migration_pmd(struct page_vma_mapped_walk *pvmw, struct page *new) mlock_vma_page(new); update_mmu_cache_pmd(vma, address, pvmw->pmd); } + +#ifdef CONFIG_PIN_MEMORY +vm_fault_t do_anon_huge_page_remap(struct vm_area_struct *vma, unsigned long address, + pmd_t *pmd, struct page *page) +{ + gfp_t gfp; + pgtable_t pgtable; + spinlock_t *ptl; + pmd_t entry; + vm_fault_t ret = 0; + struct mem_cgroup *memcg; + + if (unlikely(anon_vma_prepare(vma))) + return VM_FAULT_OOM; + if (unlikely(khugepaged_enter(vma, vma->vm_flags))) + return VM_FAULT_OOM; + gfp = alloc_hugepage_direct_gfpmask(vma); + prep_transhuge_page(page); + if (mem_cgroup_try_charge_delay(page, vma->vm_mm, gfp, &memcg, true)) { + put_page(page); + count_vm_event(THP_FAULT_FALLBACK); + return VM_FAULT_FALLBACK; + } + pgtable = pte_alloc_one(vma->vm_mm, address); + if (unlikely(!pgtable)) { + ret = VM_FAULT_OOM; + goto release; + } + __SetPageUptodate(page); + ptl = pmd_lock(vma->vm_mm, pmd); + if (unlikely(!pmd_none(*pmd))) { + goto unlock_release; + } else { + ret = check_stable_address_space(vma->vm_mm); + if (ret) + goto unlock_release; + entry = mk_huge_pmd(page, vma->vm_page_prot); + entry = maybe_pmd_mkwrite(pmd_mkdirty(entry), vma); + page_add_new_anon_rmap(page, vma, address, true); + mem_cgroup_commit_charge(page, memcg, false, true); + lru_cache_add_active_or_unevictable(page, vma); + pgtable_trans_huge_deposit(vma->vm_mm, pmd, pgtable); + set_pmd_at(vma->vm_mm, address, pmd, entry); + add_mm_counter(vma->vm_mm, MM_ANONPAGES, HPAGE_PMD_NR); + mm_inc_nr_ptes(vma->vm_mm); + spin_unlock(ptl); + count_vm_event(THP_FAULT_ALLOC); + } + + return 0; +unlock_release: + spin_unlock(ptl); +release: + if (pgtable) + pte_free(vma->vm_mm, pgtable); + mem_cgroup_cancel_charge(page, memcg, true); + put_page(page); + return ret; +} +#endif + #endif diff --git a/mm/memory.c b/mm/memory.c index 45442d9..dd416fd 100644 --- a/mm/memory.c +++ b/mm/memory.c @@ -4799,4 +4799,72 @@ void ptlock_free(struct page *page) { kmem_cache_free(page_ptl_cachep, page->ptl); } + +#ifdef CONFIG_PIN_MEMORY +vm_fault_t do_anon_page_remap(struct vm_area_struct *vma, unsigned long address, + pmd_t *pmd, struct page *page) +{ + struct mem_cgroup *memcg; + pte_t entry; + spinlock_t *ptl; + pte_t *pte; + vm_fault_t ret = 0; + + if (pte_alloc(vma->vm_mm, pmd, address)) + return VM_FAULT_OOM; + + /* See the comment in pte_alloc_one_map() */ + if (unlikely(pmd_trans_unstable(pmd))) + return 0; + + /* Allocate our own private page. */ + if (unlikely(anon_vma_prepare(vma))) + goto oom; + + if (mem_cgroup_try_charge_delay(page, vma->vm_mm, GFP_KERNEL, &memcg, + false)) + goto oom_free_page; + + /* + * The memory barrier inside __SetPageUptodate makes sure that + * preceeding stores to the page contents become visible before + * the set_pte_at() write. + */ + __SetPageUptodate(page); + + entry = mk_pte(page, vma->vm_page_prot); + if (vma->vm_flags & VM_WRITE) + entry = pte_mkwrite(pte_mkdirty(entry)); + pte = pte_offset_map_lock(vma->vm_mm, pmd, address, + &ptl); + if (!pte_none(*pte)) { + ret = VM_FAULT_FALLBACK; + goto release; + } + + ret = check_stable_address_space(vma->vm_mm); + if (ret) + goto release; + inc_mm_counter_fast(vma->vm_mm, MM_ANONPAGES); + page_add_new_anon_rmap(page, vma, address, false); + mem_cgroup_commit_charge(page, memcg, false, false); + lru_cache_add_active_or_unevictable(page, vma); + + set_pte_at(vma->vm_mm, address, pte, entry); + /* No need to invalidate - it was non-present before */ + update_mmu_cache(vma, address, pte); +unlock: + pte_unmap_unlock(pte, ptl); + return ret; +release: + mem_cgroup_cancel_charge(page, memcg, false); + put_page(page); + goto unlock; +oom_free_page: + put_page(page); +oom: + return VM_FAULT_OOM; +} +#endif + #endif diff --git a/mm/pin_mem.c b/mm/pin_mem.c new file mode 100644 index 00000000..ca3f23a --- /dev/null +++ b/mm/pin_mem.c @@ -0,0 +1,691 @@ +/* + * Copyright (C) 2020. Huawei Technologies Co., Ltd. All rights reserved. + * Provide the pin memory method for check point and restore task. + */ +#ifdef CONFIG_PIN_MEMORY +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#define MAX_PIN_PID_NUM 128 +static DEFINE_SPINLOCK(page_map_entry_lock); + +unsigned int pin_pid_num; +static unsigned int *pin_pid_num_addr; +static unsigned long __page_map_entry_start; +static unsigned long page_map_entry_end; +static struct page_map_info *user_space_reserve_start; +static struct page_map_entry *page_map_entry_start; +unsigned int max_pin_pid_num __read_mostly; + +static int __init setup_max_pin_pid_num(char *str) +{ + int ret = 1; + + if (!str) + goto out; + + ret = kstrtouint(str, 10, &max_pin_pid_num); +out: + if (ret) { + pr_warn("Unable to parse max pin pid num.\n"); + } else { + if (max_pin_pid_num > MAX_PIN_PID_NUM) { + max_pin_pid_num = 0; + pr_warn("Input max_pin_pid_num is too large.\n"); + } + } + return ret; +} +early_param("max_pin_pid_num", setup_max_pin_pid_num); + +struct page_map_info *create_page_map_info(int pid) +{ + struct page_map_info *new; + + if (!user_space_reserve_start) + return NULL; + + if (pin_pid_num >= max_pin_pid_num) { + pr_warn("Pin pid num too large than max_pin_pid_num, fail create: %d!", pid); + return NULL; + } + new = (struct page_map_info *)(user_space_reserve_start + pin_pid_num); + new->pid = pid; + new->pme = NULL; + new->entry_num = 0; + new->pid_reserved = false; + (*pin_pid_num_addr)++; + pin_pid_num++; + return new; +} +EXPORT_SYMBOL_GPL(create_page_map_info); + +struct page_map_info *get_page_map_info(int pid) +{ + int i; + + if (!user_space_reserve_start) + return NULL; + + for (i = 0; i < pin_pid_num; i++) { + if (user_space_reserve_start[i].pid == pid) { + return &(user_space_reserve_start[i]); + } + } + return NULL; +} +EXPORT_SYMBOL_GPL(get_page_map_info); + +static struct page *find_head_page(struct page *page) +{ + struct page *p = page; + + while (!PageBuddy(p)) { + if (PageLRU(p)) + return NULL; + p--; + } + return p; +} + +static void spilt_page_area_left(struct zone *zone, struct free_area *area, struct page *page, + unsigned long size, int order) +{ + unsigned long cur_size = 1 << order; + unsigned long total_size = 0; + struct page *tmp; + unsigned long tmp_size = size; + + while (size && cur_size > size) { + cur_size >>= 1; + order--; + area--; + if (cur_size <= size) { + list_add(&page[total_size].lru, &area->free_list[MIGRATE_MOVABLE]); + atomic_set(&(page[total_size]._mapcount), PAGE_BUDDY_MAPCOUNT_VALUE); + set_page_private(&page[total_size], order); + set_pageblock_migratetype(&page[total_size], MIGRATE_MOVABLE); + area->nr_free++; + total_size += cur_size; + size -= cur_size; + } + } +} + +static void spilt_page_area_right(struct zone *zone, struct free_area *area, struct page *page, + unsigned long size, int order) +{ + unsigned long cur_size = 1 << order; + struct page *right_page, *head_page; + unsigned long tmp_size = size; + + right_page = page + size; + while (size && cur_size > size) { + cur_size >>= 1; + order--; + area--; + if (cur_size <= size) { + head_page = right_page - cur_size; + list_add(&head_page->lru, &area->free_list[MIGRATE_MOVABLE]); + atomic_set(&(head_page->_mapcount), PAGE_BUDDY_MAPCOUNT_VALUE); + set_page_private(head_page, order); + set_pageblock_migratetype(head_page, MIGRATE_MOVABLE); + area->nr_free++; + size -= cur_size; + right_page = head_page; + } + } +} + +void reserve_page_from_buddy(unsigned long nr_pages, struct page *page) +{ + unsigned int current_order; + struct page *page_end; + struct free_area *area; + struct zone *zone; + struct page *head_page; + + head_page = find_head_page(page); + if (!head_page) { + pr_warn("Find page head fail."); + return; + } + current_order = head_page->private; + page_end = head_page + (1 << current_order); + zone = page_zone(head_page); + area = &(zone->free_area[current_order]); + list_del(&head_page->lru); + atomic_set(&head_page->_mapcount, -1); + set_page_private(head_page, 0); + area->nr_free--; + if (head_page != page) + spilt_page_area_left(zone, area, head_page, + (unsigned long)(page - head_page), current_order); + page = page + nr_pages; + if (page < page_end) { + spilt_page_area_right(zone, area, page, + (unsigned long)(page_end - page), current_order); + } else if (page > page_end) { + pr_warn("Find page end smaller than page."); + } +} + +static inline void reserve_user_normal_pages(struct page *page) +{ + if (!atomic_read(&page->_refcount)) { + atomic_inc(&page->_refcount); + reserve_page_from_buddy(1, page); + } else { + pr_warn("Page %pK refcount %d large than zero, no need reserve.\n", + page, page->_refcount.counter); + } +} + +static void init_huge_pmd_pages(struct page *head_page) +{ + int i = 0; + struct page *page = head_page; + unsigned long *temp; + unsigned long compound_pad = COMPOUND_PAD_START; + + __set_bit(PG_head, &page->flags); + __set_bit(PG_active, &page->flags); + atomic_set(&page->_refcount, 1); + page++; + i++; + page->compound_head = (unsigned long)head_page + 1; + page->_compound_pad_2 = (unsigned long)head_page & COMPOUND_PAD_MASK; + temp = (unsigned long *)(&(page->_compound_pad_2)); + temp[1] = LIST_POISON4; + page->compound_dtor = HUGETLB_PAGE_DTOR + 1; + page->compound_order = HPAGE_PMD_ORDER; + page++; + i++; + page->compound_head = (unsigned long)head_page + 1; + page->_compound_pad_2 = (unsigned long)head_page + compound_pad; + i++; + INIT_LIST_HEAD(&(page->deferred_list)); + for (; i < HPAGE_PMD_NR; i++) { + page = head_page + i; + page->compound_head = (unsigned long)head_page + 1; + compound_pad += COMPOUND_PAD_DELTA; + page->_compound_pad_2 = (unsigned long)head_page + compound_pad; + temp = (unsigned long *)(&(page->_compound_pad_2)); + temp[1] = LIST_POISON4; + } +} + +static void reserve_user_huge_pmd_pages(struct page *page) +{ + struct page *head_page; + + if (!atomic_read(&page->_refcount)) { + atomic_inc(&page->_refcount); + head_page = find_head_page(page); + reserve_page_from_buddy((1 << HPAGE_PMD_ORDER), page); + init_huge_pmd_pages(page); + } else { + pr_warn("Page %pK refcount %d large than zero, no need reserve.\n", + page, page->_refcount.counter); + } +} + +static void reserve_user_space_map_pages(void) +{ + struct page_map_info *pmi; + struct page_map_entry *pme; + unsigned int i, j, index; + struct page *page; + unsigned long flags; + unsigned long page_size; + int err = 0; + unsigned long phy_addr; + + if (!user_space_reserve_start) + return; + spin_lock_irqsave(&page_map_entry_lock, flags); + for (index = 0; index < pin_pid_num; index++) { + pmi = &(user_space_reserve_start[index]); + pme = pmi->pme; + + for (i = 0; i < pmi->entry_num; i++) { + err = 0; + for (j = 0; j < pme->nr_pages; j++) { + phy_addr = pme->phy_addr_array[j]; + if (!phy_addr) + continue; + page = phys_to_page(phy_addr); + if (atomic_read(&page->_refcount)) { + pme->phy_addr_array[j] = 0; + page_size = pme->is_huge_page ? HPAGE_PMD_SIZE : PAGE_SIZE; + continue; + } + if (!pme->is_huge_page) { + reserve_user_normal_pages(page); + } else { + reserve_user_huge_pmd_pages(page); + } + } + pme = (struct page_map_entry *)next_pme(pme); + if (err) + err_phy_num++; + } + page_size = pme->is_huge_page ? HPAGE_PMD_SIZE : PAGE_SIZE; + } + spin_unlock(&page_map_entry_lock); +} + + +/* The whole page map entry collect process must be Sequentially. + The user_space_reserve_start points to the first page map info for + the first dump task. And the page_map_entry_start points to + the first page map entry of the first dump vma. */ +static void init_page_map_info(unsigned int *map_addr) +{ + unsigned long map_len = pin_memory_resource.end - pin_memory_resource.start; + + if (user_space_reserve_start || !max_pin_pid_num) + return; + pin_pid_num = *map_addr; + pin_pid_num_addr = map_addr; + user_space_reserve_start = + (struct kup_page_map_info *)(map_addr + 1); + page_map_entry_start = + (struct page_map_entry *)(user_space_reserve_start + max_pin_pid_num); + page_map_entry_end = (unsigned long)map_addr + map_len; + if (pin_pid_num > 0) + reserve_user_space_map_pages(); +} + +int collect_pmd_huge_pages(struct task_struct *task, + unsigned long start_addr, unsigned long end_addr, struct page_map_entry *pme) +{ + long res; + int index = 0; + unsigned long start = start_addr; + struct page *temp_page; + + while (start < end_addr) { + temp_page = NULL; + res = get_user_pages_remote(task, task->mm, start, 1, + FOLL_TOUCH|FOLL_GET, &temp_page, NULL, NULL); + if (!res) { + pr_warn("Get huge page for addr(%lx) fail.", start); + return COLLECT_PAGES_FAIL; + } + if (PageHead(temp_page)) { + start += HPAGE_PMD_SIZE; + pme->phy_addr_array[index] = page_to_phys(temp_page); + index++; + } else { + pme->nr_pages = index; + atomic_dec(&((temp_page)->_refcount)); + return COLLECT_PAGES_NEED_CONTINUE; + } + } + pme->nr_pages = index; + return COLLECT_PAGES_FINISH; +} + +int collect_normal_pages(struct task_struct *task, + unsigned long start_addr, unsigned long end_addr, struct page_map_entry *pme) +{ + int res; + unsigned long next; + unsigned long i, nr_pages; + struct page *tmp_page; + unsigned long *phy_addr_array = pme->phy_addr_array; + struct page **page_array = (struct page **)pme->phy_addr_array; + + next = (start_addr & HPAGE_PMD_MASK) + HPAGE_PMD_SIZE; + next = (next > end_addr) ? end_addr : next; + pme->nr_pages = 0; + while (start_addr < next) { + nr_pages = (next - start_addr) / PAGE_SIZE; + res = get_user_pages_remote(task, task->mm, start_addr, 1, + FOLL_TOUCH|FOLL_GET, &tmp_page, NULL, NULL); + if (!res) { + pr_warn("Get user pages of %lx fail.\n", start_addr); + return COLLECT_PAGES_FAIL; + } + if (PageHead(tmp_page)) { + atomic_dec(&(tmp_page->_refcount)); + return COLLECT_PAGES_NEED_CONTINUE; + } + atomic_dec(&(tmp_page->_refcount)); + if (PageTail(tmp_page)) { + start_addr = next; + pme->virt_addr = start_addr; + next = (next + HPAGE_PMD_SIZE) > end_addr ? end_addr : (next + HPAGE_PMD_SIZE); + continue; + } + res = get_user_pages_remote(task, task->mm, start_addr, nr_pages, + FOLL_TOUCH|FOLL_GET, page_array, NULL, NULL); + if (!res) { + pr_warn("Get user pages of %lx fail.\n", start_addr); + return COLLECT_PAGES_FAIL; + } + for (i = 0; i < nr_pages; i++) { + phy_addr_array[i] = page_to_phys(page_array[i]); + } + pme->nr_pages += nr_pages; + page_array += nr_pages; + phy_addr_array += nr_pages; + start_addr = next; + next = (next + HPAGE_PMD_SIZE) > end_addr ? end_addr : (next + HPAGE_PMD_SIZE); + } + return COLLECT_PAGES_FINISH; +} + +/* Users make sure that the pin memory belongs to anonymous vma. */ +int pin_mem_area(struct task_struct *task, struct mm_struct *mm, + unsigned long start_addr, unsigned long end_addr) +{ + int pid, ret; + int is_huge_page = false; + unsigned int page_size; + unsigned long nr_pages, flags; + struct page_map_entry *pme; + struct page_map_info *pmi; + struct vm_area_struct *vma; + unsigned long i; + struct page *tmp_page; + + if (!page_map_entry_start + || !task || !mm + || start_addr >= end_addr) + return -EFAULT; + + pid = task->pid; + spin_lock_irqsave(&page_map_entry_lock, flags); + nr_pages = ((end_addr - start_addr) / PAGE_SIZE); + if ((unsigned long)page_map_entry_start + nr_pages * sizeof(struct page *) + >= page_map_entry_end) { + pr_warn("Page map entry use up!\n"); + ret = -EFAULT; + goto finish; + } + vma = find_extend_vma(mm, start_addr); + if (!vma) { + pr_warn("Find no match vma!\n"); + ret = -EFAULT; + goto finish; + } + if (start_addr == (start_addr & HPAGE_PMD_MASK) && + transparent_hugepage_enabled(vma)) { + page_size = HPAGE_PMD_SIZE; + is_huge_page = true; + } else { + page_size = PAGE_SIZE; + } + pme = page_map_entry_start; + pme->virt_addr = start_addr; + pme->is_huge_page = is_huge_page; + memset(pme->phy_addr_array, 0, nr_pages * sizeof(unsigned long)); + down_write(&mm->mmap_sem); + if (!is_huge_page) { + ret = collect_normal_pages(task, start_addr, end_addr, pme); + if (!pme->nr_pages) { + if (ret == COLLECT_PAGES_FINISH) { + ret = 0; + up_write(&mm->mmap_sem); + goto finish; + } + pme->is_huge_page = true; + page_size = HPAGE_PMD_SIZE; + ret = collect_pmd_huge_pages(task, pme->virt_addr, end_addr, pme); + } + } else { + ret = collect_pmd_huge_pages(task, start_addr, end_addr, pme); + if (!pme->nr_pages) { + if (ret == COLLECT_PAGES_FINISH) { + ret = 0; + up_write(&mm->mmap_sem); + goto finish; + } + pme->is_huge_page = false; + page_size = PAGE_SIZE; + ret = collect_normal_pages(task, pme->virt_addr, end_addr, pme); + } + } + up_write(&mm->mmap_sem); + if (ret == COLLECT_PAGES_FAIL) { + ret = -EFAULT; + goto finish; + } + + /* check for zero pages */ + for (i = 0; i < pme->nr_pages; i++) { + tmp_page = phys_to_page(pme->phy_addr_array[i]); + if (!pme->is_huge_page) { + if (page_to_pfn(tmp_page) == my_zero_pfn(pme->virt_addr + i * PAGE_SIZE)) + pme->phy_addr_array[i] = 0; + } else if (is_huge_zero_page(tmp_page)) + pme->phy_addr_array[i] = 0; + } + + page_map_entry_start = (struct page_map_entry *)(next_pme(pme)); + pmi = get_page_map_info(pid); + if (!pmi) + pmi = create_page_map_info(pid); + if (!pmi) { + pr_warn("Create page map info fail for pid: %d!\n", pid); + ret = -EFAULT; + goto finish; + } + if (!pmi->pme) + pmi->pme = pme; + pmi->entry_num++; + + if (ret == COLLECT_PAGES_NEED_CONTINUE) { + ret = pin_mem_area(task, mm, pme->virt_addr + pme->nr_pages * page_size, end_addr); + } + +finish: + spin_unlock_irqrestore(&page_map_entry_lock, flags); + return ret; +} +EXPORT_SYMBOL_GPL(pin_mem_area); + +vm_fault_t remap_normal_pages(struct mm_struct *mm, struct vm_area_struct *vma, + struct page_map_entry *pme) +{ + int ret; + unsigned int j; + pgd_t *pgd; + p4d_t *p4d; + pmd_t *pmd; + pud_t *pud; + struct page *page; + unsigned long address; + unsigned long phy_addr; + + for (j = 0; j < pme->nr_pages; j++) { + address = pme->virt_addr + j * PAGE_SIZE; + phy_addr = pme->phy_addr_array[j]; + if (!phy_addr) + continue; + page = phys_to_page(phy_addr); + if (page->flags & (1 << PG_reserved)) + page->flags -= (1 << PG_reserved); + if (page_to_pfn(page) == my_zero_pfn(address)) { + pme->phy_addr_array[j] = 0; + continue; + } + page->mapping = NULL; + pgd = pgd_offset(mm, address); + p4d = p4d_alloc(mm, pgd, address); + if (!p4d) + return VM_FAULT_OOM; + pud = pud_alloc(mm, p4d, address); + if (!pud) + return VM_FAULT_OOM; + pmd = pmd_alloc(mm, pud, address); + if (!pmd) + return VM_FAULT_OOM; + ret = do_anon_page_remap(vma, address, pmd, page); + if (ret == VM_FAULT_OOM) + return ret; + } + return 0; +} + +vm_fault_t remap_huge_pmd_pages(struct mm_struct *mm, struct vm_area_struct *vma, + struct page_map_entry *pme) +{ + int ret; + unsigned int j; + pgd_t *pgd; + p4d_t *p4d; + pmd_t *pmd; + pud_t *pud; + struct page *page; + unsigned long address; + unsigned long phy_addr; + + for (j = 0; j < pme->nr_pages; j++) { + address = pme->virt_addr + j * HPAGE_PMD_SIZE; + phy_addr = pme->phy_addr_array[j]; + if (!phy_addr) + continue; + page = phys_to_page(phy_addr); + if (page->flags & (1 << PG_reserved)) + page->flags -= (1 << PG_reserved); + if (is_huge_zero_page(page)) { + pme->phy_addr_array[j] = 0; + continue; + } + pgd = pgd_offset(mm, address); + p4d = p4d_alloc(mm, pgd, address); + if (!p4d) + return VM_FAULT_OOM; + pud = pud_alloc(mm, p4d, address); + if (!pud) + return VM_FAULT_OOM; + pmd = pmd_alloc(mm, pud, address); + if (!pmd) + return VM_FAULT_OOM; + ret = do_anon_huge_page_remap(vma, address, pmd, page); + if (ret == VM_FAULT_OOM) + return ret; + } + return 0; +} + +vm_fault_t do_mem_remap(int pid, struct mm_struct *mm) +{ + unsigned int i = 0; + vm_fault_t ret = 0; + struct vm_area_struct *vma; + struct page_map_info *pmi; + struct page_map_entry *pme; + + pmi = get_page_map_info(pid); + if (!pmi) + return -EFAULT; + down_write(&mm->mmap_sem); + pme = pmi->pme; + vma = mm->mmap; + while ((i < pmi->entry_num) && (vma != NULL)) { + if (pme->virt_addr >= vma->vm_start && pme->virt_addr < vma->vm_end) { + i++; + if (!vma_is_anonymous(vma)) { + pme = (struct page_map_entry *)(next_pme(pme)); + continue; + } + if (!pme->is_huge_page) { + ret = remap_normal_pages(mm, vma, pme); + if (ret < 0) + goto out; + } else { + ret = remap_huge_pmd_pages(mm, vma, pme); + if (ret < 0) + goto out; + } + pme = (struct page_map_entry *)(next_pme(pme)); + } else { + vma = vma->vm_next; + } + } +out: + up_write(&mm->mmap_sem); + return ret; +} +EXPORT_SYMBOL_GPL(do_mem_remap); + +#if defined(CONFIG_ARM64) +void init_reserve_page_map(unsigned long map_addr, unsigned long map_size) +{ + void *addr; + + if (!map_addr || !map_size) + return; + addr = phys_to_virt(map_addr); + init_page_map_info((unsigned int *)addr); +} +#else +void init_reserve_page_map(unsigned long map_addr, unsigned long map_size) +{ +} +#endif + +/* Clear all pin memory record. */ +void clear_pin_memory_record(void) +{ + if (pin_pid_num_addr) { + *pin_pid_num_addr = 0; + pin_pid_num = 0; + page_map_entry_start = (struct page_map_entry *)__page_map_entry_start; + } + if (kernel_space_reserve_start && kernel_pin_space_size > 0) { + *(unsigned long *)kernel_space_reserve_start = 0; + } +} +EXPORT_SYMBOL_GPL(clear_pin_memory_record); + +vm_fault_t reserve_kernel_space_mem(unsigned long start_addr, unsigned int pages) +{ + unsigned long i; + unsigned long entry_num; + struct page_map_entry *pme, *pme_start; + + + entry_num = *(unsigned long *)kernel_space_reserve_start; + pme_start = (struct page_map_entry *)(kernel_space_reserve_start + sizeof(entry_num)); + pme = pme_start; + spin_lock(&page_map_entry_lock); + for (i = 0; i < entry_num; i++) { + if (start_addr == pme->virt_addr) { + spin_unlock(&page_map_entry_lock); + return 0; + } + pme = pme + 1; + } + if ((unsigned long)(pme_start + entry_num) >= kernel_space_reserve_end) { + spin_unlock(&page_map_entry_lock); + return VM_FAULT_OOM; + } + pme = pme_start + entry_num; + pme->virt_addr = start_addr; + pme->nr_pages = pages; + pme->is_huge_page = false; + *(unsigned long *)kernel_space_reserve_start = entry_num + 1; + spin_unlock(&page_map_entry_lock); + return 0; +} +EXPORT_SYMBOL_GPL(reserve_kernel_space_mem); + +#endif /* CONFIG_PIN_MEMORY */