From patchwork Tue Jun 2 23:03:29 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Dmitry Kozlyuk X-Patchwork-Id: 70761 X-Patchwork-Delegate: thomas@monjalon.net Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 4DD62A04EF; Wed, 3 Jun 2020 01:05:47 +0200 (CEST) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 93AF51C13A; Wed, 3 Jun 2020 01:04:21 +0200 (CEST) Received: from mail-lj1-f195.google.com (mail-lj1-f195.google.com [209.85.208.195]) by dpdk.org (Postfix) with ESMTP id A0EEF1BFB2 for ; Wed, 3 Jun 2020 01:04:20 +0200 (CEST) Received: by mail-lj1-f195.google.com with SMTP id m18so321303ljo.5 for ; Tue, 02 Jun 2020 16:04:20 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:to:cc:subject:date:message-id:in-reply-to:references :mime-version:content-transfer-encoding; bh=jKiCUl89IJVO1F8e8eIgPiVB0fncRVSC6JNRpfLUzrU=; b=JiSccJ3jjBH5oTOvYFzmO45jRmG5AcYZlT6y14ZfiZdi0jxUxqjuVhBsDspQuQNIJw zZ3LWEZ1eUvmC23O5+EOmH/XNPnjjW11sFNLu3WRhF7XHOGUqiyw5sfjSKCdeWKreEJ8 r3czE6spYTpCHZ62kMyRw3rekYXOpQ2hTVhffDmcUaPtTfFQyFrXCw7SLZ9i4wWJBNX4 /3CkyHKMDv7M4CXM2x9M51ro2eWm2Svzm10x3064Q9PAna1A4w8rPI2X4kSK9kM1Fiu/ wDY8Y2hTJMAu1Dsj/MejTXn+o3DmTAO5ajK8W1n+sMEbW7bgmRi4S0Z4tF6tF4ohP9sl X5Kw== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=jKiCUl89IJVO1F8e8eIgPiVB0fncRVSC6JNRpfLUzrU=; b=RYavhWyvu+2LfUUHJIiJQtQA7Ho71EZ9V9pyz3p30JDuXaoj8AkhG9hzX2oGS3tXhM Rbm+6ZFj55ZmJYYo5m4SmuncJy8XtV8eIb77fQT2dchIoVLSwq25vhlOSf6OhPMLshyK AVU1zyylZblcvO/nCjyyl1OkkJ6tD7YqMXZLdB6UxS+ZnXNBj7kflMvQqS8FsjtenfgS N2D8P4QWMY25CsqmJ33oQqxvwJFXIaSeZ7n7Y0ObDfjupnPIRahskXOUArz2IZECKwGq 84JUWTpvhspQaeU1elU8d81uIRNhfC1AWSQDKidP02hIDItYF59GU60hLpWbpdjS6i7s Rbhg== X-Gm-Message-State: AOAM530s+xiDpSr5C1U7SrAYD1qOsneGWQzBNBAe+SjMviJJYKBnmchl cdjwyd9gI2DRVxSg1q2yfEfdcZEjSHaW8Q== X-Google-Smtp-Source: ABdhPJwrULx6cHW3akFhROvRraw9j2RWG0/syHZasTb9nAeHp4cuEH7yjXCdBYfCqWzGMRkYnleHSQ== X-Received: by 2002:a2e:960b:: with SMTP id v11mr639188ljh.77.1591139053299; Tue, 02 Jun 2020 16:04:13 -0700 (PDT) Received: from localhost.localdomain (broadband-37-110-65-23.ip.moscow.rt.ru. [37.110.65.23]) by smtp.gmail.com with ESMTPSA id s18sm67471ljj.63.2020.06.02.16.04.09 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 02 Jun 2020 16:04:12 -0700 (PDT) From: Dmitry Kozlyuk To: dev@dpdk.org Cc: Dmitry Malloy , Narcisa Ana Maria Vasile , Fady Bader , Tal Shnaiderman , Dmitry Kozlyuk , Thomas Monjalon , Harini Ramakrishnan , Omar Cardona , Pallavi Kadam , Ranjit Menon , John McNamara , Marko Kovacevic , Anatoly Burakov Date: Wed, 3 Jun 2020 02:03:29 +0300 Message-Id: <20200602230329.17838-12-dmitry.kozliuk@gmail.com> X-Mailer: git-send-email 2.25.4 In-Reply-To: <20200602230329.17838-1-dmitry.kozliuk@gmail.com> References: <20200525003720.6410-1-dmitry.kozliuk@gmail.com> <20200602230329.17838-1-dmitry.kozliuk@gmail.com> MIME-Version: 1.0 Subject: [dpdk-dev] [PATCH v6 11/11] eal/windows: implement basic memory management X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Basic memory management supports core libraries and PMDs operating in IOVA as PA mode. It uses a kernel-mode driver, virt2phys, to obtain IOVAs of hugepages allocated from user-mode. Multi-process mode is not implemented and is forcefully disabled at startup. Assign myself as a maintainer for Windows file and memory management implementation. Signed-off-by: Dmitry Kozlyuk --- MAINTAINERS | 1 + config/meson.build | 12 +- doc/guides/windows_gsg/run_apps.rst | 54 +- lib/librte_eal/common/meson.build | 11 + lib/librte_eal/common/rte_malloc.c | 1 + lib/librte_eal/rte_eal_exports.def | 119 +++ lib/librte_eal/windows/eal.c | 63 +- lib/librte_eal/windows/eal_file.c | 123 +++ lib/librte_eal/windows/eal_memalloc.c | 441 +++++++++++ lib/librte_eal/windows/eal_memory.c | 710 ++++++++++++++++++ lib/librte_eal/windows/eal_mp.c | 103 +++ lib/librte_eal/windows/eal_windows.h | 75 ++ lib/librte_eal/windows/include/meson.build | 1 + lib/librte_eal/windows/include/rte_os.h | 17 + .../windows/include/rte_virt2phys.h | 34 + lib/librte_eal/windows/include/rte_windows.h | 2 + lib/librte_eal/windows/include/unistd.h | 3 + lib/librte_eal/windows/meson.build | 6 + 18 files changed, 1769 insertions(+), 7 deletions(-) create mode 100644 lib/librte_eal/windows/eal_file.c create mode 100644 lib/librte_eal/windows/eal_memalloc.c create mode 100644 lib/librte_eal/windows/eal_memory.c create mode 100644 lib/librte_eal/windows/eal_mp.c create mode 100644 lib/librte_eal/windows/include/rte_virt2phys.h diff --git a/MAINTAINERS b/MAINTAINERS index 19b818f69..5140756b3 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -339,6 +339,7 @@ F: doc/guides/windows_gsg/ Windows memory allocation M: Dmitry Kozlyuk F: lib/librte_eal/eal_hugepages.c +F: lib/librte_eal/eal_mem* Core Libraries diff --git a/config/meson.build b/config/meson.build index c1e80de4b..d3f05f878 100644 --- a/config/meson.build +++ b/config/meson.build @@ -261,15 +261,21 @@ if is_freebsd endif if is_windows - # Minimum supported API is Windows 7. - add_project_arguments('-D_WIN32_WINNT=0x0601', language: 'c') + # VirtualAlloc2() is available since Windows 10 / Server 2016. + add_project_arguments('-D_WIN32_WINNT=0x0A00', language: 'c') # Use MinGW-w64 stdio, because DPDK assumes ANSI-compliant formatting. if cc.get_id() == 'gcc' add_project_arguments('-D__USE_MINGW_ANSI_STDIO', language: 'c') endif - add_project_link_arguments('-ladvapi32', language: 'c') + # Contrary to docs, VirtualAlloc2() is exported by mincore.lib + # in Windows SDK, while MinGW exports it by advapi32.a. + if is_ms_linker + add_project_link_arguments('-lmincore', language: 'c') + endif + + add_project_link_arguments('-ladvapi32', '-lsetupapi', language: 'c') endif if get_option('b_lto') diff --git a/doc/guides/windows_gsg/run_apps.rst b/doc/guides/windows_gsg/run_apps.rst index 21ac7f6c1..78e5a614f 100644 --- a/doc/guides/windows_gsg/run_apps.rst +++ b/doc/guides/windows_gsg/run_apps.rst @@ -7,10 +7,10 @@ Running DPDK Applications Grant *Lock pages in memory* Privilege -------------------------------------- -Use of hugepages ("large pages" in Windows terminolocy) requires +Use of hugepages ("large pages" in Windows terminology) requires ``SeLockMemoryPrivilege`` for the user running an application. -1. Open *Local Security Policy* snap in, either: +1. Open *Local Security Policy* snap-in, either: * Control Panel / Computer Management / Local Security Policy; * or Win+R, type ``secpol``, press Enter. @@ -24,7 +24,55 @@ Use of hugepages ("large pages" in Windows terminolocy) requires See `Large-Page Support`_ in MSDN for details. -.. _Large-page Support: https://docs.microsoft.com/en-us/windows/win32/memory/large-page-support +.. _Large-Page Support: https://docs.microsoft.com/en-us/windows/win32/memory/large-page-support + + +Load virt2phys Driver +--------------------- + +Access to physical addresses is provided by a kernel-mode driver, virt2phys. +It is mandatory at least for using hardware PMDs, but may also be required +for mempools. + +Refer to documentation in ``dpdk-kmods`` repository for details on system +setup, driver build and installation. This driver is not signed, so signature +checking must be disabled to load it. + +.. warning:: + + Disabling driver signature enforcement weakens OS security. + It is discouraged in production environments. + +Compiled package consists of ``virt2phys.inf``, ``virt2phys.cat``, +and ``virt2phys.sys``. It can be installed as follows +from Elevated Command Prompt: + +.. code-block:: console + + pnputil /add-driver Z:\path\to\virt2phys.inf /install + +On Windows Server additional steps are required: + +1. From Device Manager, Action menu, select "Add legacy hardware". +2. It will launch the "Add Hardware Wizard". Click "Next". +3. Select second option "Install the hardware that I manually select + from a list (Advanced)". +4. On the next screen, "Kernel bypass" will be shown as a device class. +5. Select it, and click "Next". +6. The previously installed drivers will now be installed for the + "Virtual to physical address translator" device. + +When loaded successfully, the driver is shown in *Device Manager* as *Virtual +to physical address translator* device under *Kernel bypass* category. +Installed driver persists across reboots. + +If DPDK is unable to communicate with the driver, a warning is printed +on initialization (debug-level logs provide more details): + +.. code-block:: text + + EAL: Cannot open virt2phys driver interface + Run the ``helloworld`` Example diff --git a/lib/librte_eal/common/meson.build b/lib/librte_eal/common/meson.build index 4e9208129..310844269 100644 --- a/lib/librte_eal/common/meson.build +++ b/lib/librte_eal/common/meson.build @@ -8,13 +8,24 @@ if is_windows 'eal_common_bus.c', 'eal_common_class.c', 'eal_common_devargs.c', + 'eal_common_dynmem.c', 'eal_common_errno.c', + 'eal_common_fbarray.c', 'eal_common_launch.c', 'eal_common_lcore.c', 'eal_common_log.c', + 'eal_common_mcfg.c', + 'eal_common_memalloc.c', + 'eal_common_memory.c', + 'eal_common_memzone.c', 'eal_common_options.c', + 'eal_common_string_fns.c', + 'eal_common_tailqs.c', 'eal_common_thread.c', 'eal_common_trace_points.c', + 'malloc_elem.c', + 'malloc_heap.c', + 'rte_malloc.c', ) subdir_done() endif diff --git a/lib/librte_eal/common/rte_malloc.c b/lib/librte_eal/common/rte_malloc.c index f1b73168b..9d39e58c0 100644 --- a/lib/librte_eal/common/rte_malloc.c +++ b/lib/librte_eal/common/rte_malloc.c @@ -20,6 +20,7 @@ #include #include #include + #include #include diff --git a/lib/librte_eal/rte_eal_exports.def b/lib/librte_eal/rte_eal_exports.def index 12a6c79d6..e2eb24f01 100644 --- a/lib/librte_eal/rte_eal_exports.def +++ b/lib/librte_eal/rte_eal_exports.def @@ -1,9 +1,128 @@ EXPORTS __rte_panic + rte_calloc + rte_calloc_socket rte_eal_get_configuration + rte_eal_has_hugepages rte_eal_init + rte_eal_iova_mode rte_eal_mp_remote_launch rte_eal_mp_wait_lcore + rte_eal_process_type rte_eal_remote_launch rte_log + rte_eal_tailq_lookup + rte_eal_tailq_register + rte_eal_using_phys_addrs + rte_free + rte_malloc + rte_malloc_dump_stats + rte_malloc_get_socket_stats + rte_malloc_set_limit + rte_malloc_socket + rte_malloc_validate + rte_malloc_virt2iova + rte_mcfg_mem_read_lock + rte_mcfg_mem_read_unlock + rte_mcfg_mem_write_lock + rte_mcfg_mem_write_unlock + rte_mcfg_mempool_read_lock + rte_mcfg_mempool_read_unlock + rte_mcfg_mempool_write_lock + rte_mcfg_mempool_write_unlock + rte_mcfg_tailq_read_lock + rte_mcfg_tailq_read_unlock + rte_mcfg_tailq_write_lock + rte_mcfg_tailq_write_unlock + rte_mem_lock_page + rte_mem_virt2iova + rte_mem_virt2phy + rte_memory_get_nchannel + rte_memory_get_nrank + rte_memzone_dump + rte_memzone_free + rte_memzone_lookup + rte_memzone_reserve + rte_memzone_reserve_aligned + rte_memzone_reserve_bounded + rte_memzone_walk rte_vlog + rte_realloc + rte_zmalloc + rte_zmalloc_socket + + rte_mp_action_register + rte_mp_action_unregister + rte_mp_reply + rte_mp_sendmsg + + rte_fbarray_attach + rte_fbarray_destroy + rte_fbarray_detach + rte_fbarray_dump_metadata + rte_fbarray_find_contig_free + rte_fbarray_find_contig_used + rte_fbarray_find_idx + rte_fbarray_find_next_free + rte_fbarray_find_next_n_free + rte_fbarray_find_next_n_used + rte_fbarray_find_next_used + rte_fbarray_get + rte_fbarray_init + rte_fbarray_is_used + rte_fbarray_set_free + rte_fbarray_set_used + rte_malloc_dump_heaps + rte_mem_alloc_validator_register + rte_mem_alloc_validator_unregister + rte_mem_check_dma_mask + rte_mem_event_callback_register + rte_mem_event_callback_unregister + rte_mem_iova2virt + rte_mem_virt2memseg + rte_mem_virt2memseg_list + rte_memseg_contig_walk + rte_memseg_list_walk + rte_memseg_walk + rte_mp_request_async + rte_mp_request_sync + + rte_fbarray_find_prev_free + rte_fbarray_find_prev_n_free + rte_fbarray_find_prev_n_used + rte_fbarray_find_prev_used + rte_fbarray_find_rev_contig_free + rte_fbarray_find_rev_contig_used + rte_memseg_contig_walk_thread_unsafe + rte_memseg_list_walk_thread_unsafe + rte_memseg_walk_thread_unsafe + + rte_malloc_heap_create + rte_malloc_heap_destroy + rte_malloc_heap_get_socket + rte_malloc_heap_memory_add + rte_malloc_heap_memory_attach + rte_malloc_heap_memory_detach + rte_malloc_heap_memory_remove + rte_malloc_heap_socket_is_external + rte_mem_check_dma_mask_thread_unsafe + rte_mem_set_dma_mask + rte_memseg_get_fd + rte_memseg_get_fd_offset + rte_memseg_get_fd_offset_thread_unsafe + rte_memseg_get_fd_thread_unsafe + + rte_extmem_attach + rte_extmem_detach + rte_extmem_register + rte_extmem_unregister + + rte_fbarray_find_biggest_free + rte_fbarray_find_biggest_used + rte_fbarray_find_rev_biggest_free + rte_fbarray_find_rev_biggest_used + + rte_mem_lock + rte_mem_map + rte_mem_page_size + rte_mem_unmap diff --git a/lib/librte_eal/windows/eal.c b/lib/librte_eal/windows/eal.c index 7c2fcc860..a43649abc 100644 --- a/lib/librte_eal/windows/eal.c +++ b/lib/librte_eal/windows/eal.c @@ -94,6 +94,24 @@ eal_proc_type_detect(void) return ptype; } +enum rte_proc_type_t +rte_eal_process_type(void) +{ + return rte_config.process_type; +} + +int +rte_eal_has_hugepages(void) +{ + return !internal_config.no_hugetlbfs; +} + +enum rte_iova_mode +rte_eal_iova_mode(void) +{ + return rte_config.iova_mode; +} + /* display usage */ static void eal_usage(const char *prgname) @@ -256,7 +274,7 @@ __rte_trace_point_register(rte_trace_point_t *trace, const char *name, return -ENOTSUP; } -/* Launch threads, called at application init(). */ + /* Launch threads, called at application init(). */ int rte_eal_init(int argc, char **argv) { @@ -279,6 +297,13 @@ rte_eal_init(int argc, char **argv) if (fctret < 0) exit(1); + /* Prevent creation of shared memory files. */ + if (internal_config.in_memory == 0) { + RTE_LOG(WARNING, EAL, "Multi-process support is requested, " + "but not available.\n"); + internal_config.in_memory = 1; + } + if (!internal_config.no_hugetlbfs && (eal_hugepage_info_init() < 0)) { rte_eal_init_alert("Cannot get hugepage information"); rte_errno = EACCES; @@ -290,6 +315,42 @@ rte_eal_init(int argc, char **argv) internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE; } + if (eal_mem_win32api_init() < 0) { + rte_eal_init_alert("Cannot access Win32 memory management"); + rte_errno = ENOTSUP; + return -1; + } + + if (eal_mem_virt2iova_init() < 0) { + /* Non-fatal error if physical addresses are not required. */ + RTE_LOG(WARNING, EAL, "Cannot access virt2phys driver, " + "PA will not be available\n"); + } + + if (rte_eal_memzone_init() < 0) { + rte_eal_init_alert("Cannot init memzone"); + rte_errno = ENODEV; + return -1; + } + + if (rte_eal_memory_init() < 0) { + rte_eal_init_alert("Cannot init memory"); + rte_errno = ENOMEM; + return -1; + } + + if (rte_eal_malloc_heap_init() < 0) { + rte_eal_init_alert("Cannot init malloc heap"); + rte_errno = ENODEV; + return -1; + } + + if (rte_eal_tailqs_init() < 0) { + rte_eal_init_alert("Cannot init tail queues for objects"); + rte_errno = EFAULT; + return -1; + } + eal_thread_init_master(rte_config.master_lcore); RTE_LCORE_FOREACH_SLAVE(i) { diff --git a/lib/librte_eal/windows/eal_file.c b/lib/librte_eal/windows/eal_file.c new file mode 100644 index 000000000..dbb08456c --- /dev/null +++ b/lib/librte_eal/windows/eal_file.c @@ -0,0 +1,123 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2020 Dmitry Kozlyuk + */ + +#include +#include +#include +#include + +#include "eal_private.h" +#include "eal_windows.h" + +int +eal_file_create(const char *path) +{ + int fd, ret; + + ret = _sopen_s(&fd, path, _O_CREAT | _O_RDWR, _SH_DENYNO, _S_IWRITE); + if (ret) { + rte_errno = ret; + return -1; + } + + return fd; +} + +int +eal_file_open(const char *path, bool writable) +{ + int fd, ret, flags; + + flags = writable ? _O_RDWR : _O_RDONLY; + ret = _sopen_s(&fd, path, flags, _SH_DENYNO, 0); + if (ret < 0) { + rte_errno = errno; + return -1; + } + + return fd; +} + +int +eal_file_truncate(int fd, ssize_t size) +{ + HANDLE handle; + DWORD ret; + LONG low = (LONG)((size_t)size); + LONG high = (LONG)((size_t)size >> 32); + + handle = (HANDLE)_get_osfhandle(fd); + if (handle == INVALID_HANDLE_VALUE) { + rte_errno = EBADF; + return -1; + } + + ret = SetFilePointer(handle, low, &high, FILE_BEGIN); + if (ret == INVALID_SET_FILE_POINTER) { + RTE_LOG_WIN32_ERR("SetFilePointer()"); + rte_errno = EINVAL; + return -1; + } + + return 0; +} + +static int +lock_file(HANDLE handle, enum eal_flock_op op, enum eal_flock_mode mode) +{ + DWORD sys_flags = 0; + OVERLAPPED overlapped; + + if (op == EAL_FLOCK_EXCLUSIVE) + sys_flags |= LOCKFILE_EXCLUSIVE_LOCK; + if (mode == EAL_FLOCK_RETURN) + sys_flags |= LOCKFILE_FAIL_IMMEDIATELY; + + memset(&overlapped, 0, sizeof(overlapped)); + if (!LockFileEx(handle, sys_flags, 0, 0, 0, &overlapped)) { + if ((sys_flags & LOCKFILE_FAIL_IMMEDIATELY) && + (GetLastError() == ERROR_IO_PENDING)) { + rte_errno = EWOULDBLOCK; + } else { + RTE_LOG_WIN32_ERR("LockFileEx()"); + rte_errno = EINVAL; + } + return -1; + } + + return 0; +} + +static int +unlock_file(HANDLE handle) +{ + if (!UnlockFileEx(handle, 0, 0, 0, NULL)) { + RTE_LOG_WIN32_ERR("UnlockFileEx()"); + rte_errno = EINVAL; + return -1; + } + return 0; +} + +int +eal_file_lock(int fd, enum eal_flock_op op, enum eal_flock_mode mode) +{ + HANDLE handle = (HANDLE)_get_osfhandle(fd); + + if (handle == INVALID_HANDLE_VALUE) { + rte_errno = EBADF; + return -1; + } + + switch (op) { + case EAL_FLOCK_EXCLUSIVE: + case EAL_FLOCK_SHARED: + return lock_file(handle, op, mode); + case EAL_FLOCK_UNLOCK: + return unlock_file(handle); + default: + rte_errno = EINVAL; + return -1; + } +} diff --git a/lib/librte_eal/windows/eal_memalloc.c b/lib/librte_eal/windows/eal_memalloc.c new file mode 100644 index 000000000..a7452b6e1 --- /dev/null +++ b/lib/librte_eal/windows/eal_memalloc.c @@ -0,0 +1,441 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright (c) 2020 Dmitry Kozlyuk + */ + +#include +#include + +#include "eal_internal_cfg.h" +#include "eal_memalloc.h" +#include "eal_memcfg.h" +#include "eal_private.h" +#include "eal_windows.h" + +int +eal_memalloc_get_seg_fd(int list_idx, int seg_idx) +{ + /* Hugepages have no associated files in Windows. */ + RTE_SET_USED(list_idx); + RTE_SET_USED(seg_idx); + EAL_LOG_NOT_IMPLEMENTED(); + return -1; +} + +int +eal_memalloc_get_seg_fd_offset(int list_idx, int seg_idx, size_t *offset) +{ + /* Hugepages have no associated files in Windows. */ + RTE_SET_USED(list_idx); + RTE_SET_USED(seg_idx); + RTE_SET_USED(offset); + EAL_LOG_NOT_IMPLEMENTED(); + return -1; +} + +static int +alloc_seg(struct rte_memseg *ms, void *requested_addr, int socket_id, + struct hugepage_info *hi) +{ + HANDLE current_process; + unsigned int numa_node; + size_t alloc_sz; + void *addr; + rte_iova_t iova = RTE_BAD_IOVA; + PSAPI_WORKING_SET_EX_INFORMATION info; + PSAPI_WORKING_SET_EX_BLOCK *page; + + if (ms->len > 0) { + /* If a segment is already allocated as needed, return it. */ + if ((ms->addr == requested_addr) && + (ms->socket_id == socket_id) && + (ms->hugepage_sz == hi->hugepage_sz)) { + return 0; + } + + /* Bugcheck, should not happen. */ + RTE_LOG(DEBUG, EAL, "Attempted to reallocate segment %p " + "(size %zu) on socket %d", ms->addr, + ms->len, ms->socket_id); + return -1; + } + + current_process = GetCurrentProcess(); + numa_node = eal_socket_numa_node(socket_id); + alloc_sz = hi->hugepage_sz; + + if (requested_addr == NULL) { + /* Request a new chunk of memory from OS. */ + addr = eal_mem_alloc_socket(alloc_sz, socket_id); + if (addr == NULL) { + RTE_LOG(DEBUG, EAL, "Cannot allocate %zu bytes " + "on socket %d\n", alloc_sz, socket_id); + return -1; + } + } else { + /* Requested address is already reserved, commit memory. */ + addr = eal_mem_commit(requested_addr, alloc_sz, socket_id); + + /* During commitment, memory is temporary freed and might + * be allocated by different non-EAL thread. This is a fatal + * error, because it breaks MSL assumptions. + */ + if ((addr != NULL) && (addr != requested_addr)) { + RTE_LOG(CRIT, EAL, "Address %p occupied by an alien " + " allocation - MSL is not VA-contiguous!\n", + requested_addr); + return -1; + } + + if (addr == NULL) { + RTE_LOG(DEBUG, EAL, "Cannot commit reserved memory %p " + "(size %zu) on socket %d\n", + requested_addr, alloc_sz, socket_id); + return -1; + } + } + + /* Force OS to allocate a physical page and select a NUMA node. + * Hugepages are not pageable in Windows, so there's no race + * for physical address. + */ + *(volatile int *)addr = *(volatile int *)addr; + + /* Only try to obtain IOVA if it's available, so that applications + * that do not need IOVA can use this allocator. + */ + if (rte_eal_using_phys_addrs()) { + iova = rte_mem_virt2iova(addr); + if (iova == RTE_BAD_IOVA) { + RTE_LOG(DEBUG, EAL, + "Cannot get IOVA of allocated segment\n"); + goto error; + } + } + + /* Only "Ex" function can handle hugepages. */ + info.VirtualAddress = addr; + if (!QueryWorkingSetEx(current_process, &info, sizeof(info))) { + RTE_LOG_WIN32_ERR("QueryWorkingSetEx(%p)", addr); + goto error; + } + + page = &info.VirtualAttributes; + if (!page->Valid || !page->LargePage) { + RTE_LOG(DEBUG, EAL, "Got regular page instead of a hugepage\n"); + goto error; + } + if (page->Node != numa_node) { + RTE_LOG(DEBUG, EAL, + "NUMA node hint %u (socket %d) not respected, got %u\n", + numa_node, socket_id, page->Node); + goto error; + } + + ms->addr = addr; + ms->hugepage_sz = hi->hugepage_sz; + ms->len = alloc_sz; + ms->nchannel = rte_memory_get_nchannel(); + ms->nrank = rte_memory_get_nrank(); + ms->iova = iova; + ms->socket_id = socket_id; + + return 0; + +error: + /* Only jump here when `addr` and `alloc_sz` are valid. */ + if (eal_mem_decommit(addr, alloc_sz) && (rte_errno == EADDRNOTAVAIL)) { + /* During decommitment, memory is temporarily returned + * to the system and the address may become unavailable. + */ + RTE_LOG(CRIT, EAL, "Address %p occupied by an alien " + " allocation - MSL is not VA-contiguous!\n", addr); + } + return -1; +} + +static int +free_seg(struct rte_memseg *ms) +{ + if (eal_mem_decommit(ms->addr, ms->len)) { + if (rte_errno == EADDRNOTAVAIL) { + /* See alloc_seg() for explanation. */ + RTE_LOG(CRIT, EAL, "Address %p occupied by an alien " + " allocation - MSL is not VA-contiguous!\n", + ms->addr); + } + return -1; + } + + /* Must clear the segment, because alloc_seg() inspects it. */ + memset(ms, 0, sizeof(*ms)); + return 0; +} + +struct alloc_walk_param { + struct hugepage_info *hi; + struct rte_memseg **ms; + size_t page_sz; + unsigned int segs_allocated; + unsigned int n_segs; + int socket; + bool exact; +}; + +static int +alloc_seg_walk(const struct rte_memseg_list *msl, void *arg) +{ + struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; + struct alloc_walk_param *wa = arg; + struct rte_memseg_list *cur_msl; + size_t page_sz; + int cur_idx, start_idx, j; + unsigned int msl_idx, need, i; + + if (msl->page_sz != wa->page_sz) + return 0; + if (msl->socket_id != wa->socket) + return 0; + + page_sz = (size_t)msl->page_sz; + + msl_idx = msl - mcfg->memsegs; + cur_msl = &mcfg->memsegs[msl_idx]; + + need = wa->n_segs; + + /* try finding space in memseg list */ + if (wa->exact) { + /* if we require exact number of pages in a list, find them */ + cur_idx = rte_fbarray_find_next_n_free( + &cur_msl->memseg_arr, 0, need); + if (cur_idx < 0) + return 0; + start_idx = cur_idx; + } else { + int cur_len; + + /* we don't require exact number of pages, so we're going to go + * for best-effort allocation. that means finding the biggest + * unused block, and going with that. + */ + cur_idx = rte_fbarray_find_biggest_free( + &cur_msl->memseg_arr, 0); + if (cur_idx < 0) + return 0; + start_idx = cur_idx; + /* adjust the size to possibly be smaller than original + * request, but do not allow it to be bigger. + */ + cur_len = rte_fbarray_find_contig_free( + &cur_msl->memseg_arr, cur_idx); + need = RTE_MIN(need, (unsigned int)cur_len); + } + + for (i = 0; i < need; i++, cur_idx++) { + struct rte_memseg *cur; + void *map_addr; + + cur = rte_fbarray_get(&cur_msl->memseg_arr, cur_idx); + map_addr = RTE_PTR_ADD(cur_msl->base_va, cur_idx * page_sz); + + if (alloc_seg(cur, map_addr, wa->socket, wa->hi)) { + RTE_LOG(DEBUG, EAL, "attempted to allocate %i segments, " + "but only %i were allocated\n", need, i); + + /* if exact number wasn't requested, stop */ + if (!wa->exact) + goto out; + + /* clean up */ + for (j = start_idx; j < cur_idx; j++) { + struct rte_memseg *tmp; + struct rte_fbarray *arr = &cur_msl->memseg_arr; + + tmp = rte_fbarray_get(arr, j); + rte_fbarray_set_free(arr, j); + + if (free_seg(tmp)) + RTE_LOG(DEBUG, EAL, "Cannot free page\n"); + } + /* clear the list */ + if (wa->ms) + memset(wa->ms, 0, sizeof(*wa->ms) * wa->n_segs); + + return -1; + } + if (wa->ms) + wa->ms[i] = cur; + + rte_fbarray_set_used(&cur_msl->memseg_arr, cur_idx); + } + +out: + wa->segs_allocated = i; + if (i > 0) + cur_msl->version++; + + /* if we didn't allocate any segments, move on to the next list */ + return i > 0; +} + +struct free_walk_param { + struct hugepage_info *hi; + struct rte_memseg *ms; +}; +static int +free_seg_walk(const struct rte_memseg_list *msl, void *arg) +{ + struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config; + struct rte_memseg_list *found_msl; + struct free_walk_param *wa = arg; + uintptr_t start_addr, end_addr; + int msl_idx, seg_idx, ret; + + start_addr = (uintptr_t) msl->base_va; + end_addr = start_addr + msl->len; + + if ((uintptr_t)wa->ms->addr < start_addr || + (uintptr_t)wa->ms->addr >= end_addr) + return 0; + + msl_idx = msl - mcfg->memsegs; + seg_idx = RTE_PTR_DIFF(wa->ms->addr, start_addr) / msl->page_sz; + + /* msl is const */ + found_msl = &mcfg->memsegs[msl_idx]; + found_msl->version++; + + rte_fbarray_set_free(&found_msl->memseg_arr, seg_idx); + + ret = free_seg(wa->ms); + + return (ret < 0) ? (-1) : 1; +} + +int +eal_memalloc_alloc_seg_bulk(struct rte_memseg **ms, int n_segs, + size_t page_sz, int socket, bool exact) +{ + unsigned int i; + int ret = -1; + struct alloc_walk_param wa; + struct hugepage_info *hi = NULL; + + if (internal_config.legacy_mem) { + RTE_LOG(ERR, EAL, "dynamic allocation not supported in legacy mode\n"); + return -ENOTSUP; + } + + for (i = 0; i < internal_config.num_hugepage_sizes; i++) { + struct hugepage_info *hpi = &internal_config.hugepage_info[i]; + if (page_sz == hpi->hugepage_sz) { + hi = hpi; + break; + } + } + if (!hi) { + RTE_LOG(ERR, EAL, "cannot find relevant hugepage_info entry\n"); + return -1; + } + + memset(&wa, 0, sizeof(wa)); + wa.exact = exact; + wa.hi = hi; + wa.ms = ms; + wa.n_segs = n_segs; + wa.page_sz = page_sz; + wa.socket = socket; + wa.segs_allocated = 0; + + /* memalloc is locked, so it's safe to use thread-unsafe version */ + ret = rte_memseg_list_walk_thread_unsafe(alloc_seg_walk, &wa); + if (ret == 0) { + RTE_LOG(ERR, EAL, "cannot find suitable memseg_list\n"); + ret = -1; + } else if (ret > 0) { + ret = (int)wa.segs_allocated; + } + + return ret; +} + +struct rte_memseg * +eal_memalloc_alloc_seg(size_t page_sz, int socket) +{ + struct rte_memseg *ms = NULL; + eal_memalloc_alloc_seg_bulk(&ms, 1, page_sz, socket, true); + return ms; +} + +int +eal_memalloc_free_seg_bulk(struct rte_memseg **ms, int n_segs) +{ + int seg, ret = 0; + + /* dynamic free not supported in legacy mode */ + if (internal_config.legacy_mem) + return -1; + + for (seg = 0; seg < n_segs; seg++) { + struct rte_memseg *cur = ms[seg]; + struct hugepage_info *hi = NULL; + struct free_walk_param wa; + size_t i; + int walk_res; + + /* if this page is marked as unfreeable, fail */ + if (cur->flags & RTE_MEMSEG_FLAG_DO_NOT_FREE) { + RTE_LOG(DEBUG, EAL, "Page is not allowed to be freed\n"); + ret = -1; + continue; + } + + memset(&wa, 0, sizeof(wa)); + + for (i = 0; i < RTE_DIM(internal_config.hugepage_info); i++) { + hi = &internal_config.hugepage_info[i]; + if (cur->hugepage_sz == hi->hugepage_sz) + break; + } + if (i == RTE_DIM(internal_config.hugepage_info)) { + RTE_LOG(ERR, EAL, "Can't find relevant hugepage_info entry\n"); + ret = -1; + continue; + } + + wa.ms = cur; + wa.hi = hi; + + /* memalloc is locked, so it's safe to use thread-unsafe version + */ + walk_res = rte_memseg_list_walk_thread_unsafe(free_seg_walk, + &wa); + if (walk_res == 1) + continue; + if (walk_res == 0) + RTE_LOG(ERR, EAL, "Couldn't find memseg list\n"); + ret = -1; + } + return ret; +} + +int +eal_memalloc_free_seg(struct rte_memseg *ms) +{ + return eal_memalloc_free_seg_bulk(&ms, 1); +} + +int +eal_memalloc_sync_with_primary(void) +{ + /* No multi-process support. */ + EAL_LOG_NOT_IMPLEMENTED(); + return -1; +} + +int +eal_memalloc_init(void) +{ + /* No action required. */ + return 0; +} diff --git a/lib/librte_eal/windows/eal_memory.c b/lib/librte_eal/windows/eal_memory.c new file mode 100644 index 000000000..2739da346 --- /dev/null +++ b/lib/librte_eal/windows/eal_memory.c @@ -0,0 +1,710 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright (c) 2020 Dmitry Kozlyuk + */ + +#include +#include + +#include +#include + +#include "eal_internal_cfg.h" +#include "eal_memalloc.h" +#include "eal_memcfg.h" +#include "eal_options.h" +#include "eal_private.h" +#include "eal_windows.h" + +#include + +/* MinGW-w64 headers lack VirtualAlloc2() in some distributions. + * Provide a copy of definitions and code to load it dynamically. + * Note: definitions are copied verbatim from Microsoft documentation + * and don't follow DPDK code style. + * + * MEM_RESERVE_PLACEHOLDER being defined means VirtualAlloc2() is present too. + */ +#ifndef MEM_PRESERVE_PLACEHOLDER + +/* https://docs.microsoft.com/en-us/windows/win32/api/winnt/ne-winnt-mem_extended_parameter_type */ +typedef enum MEM_EXTENDED_PARAMETER_TYPE { + MemExtendedParameterInvalidType, + MemExtendedParameterAddressRequirements, + MemExtendedParameterNumaNode, + MemExtendedParameterPartitionHandle, + MemExtendedParameterUserPhysicalHandle, + MemExtendedParameterAttributeFlags, + MemExtendedParameterMax +} *PMEM_EXTENDED_PARAMETER_TYPE; + +#define MEM_EXTENDED_PARAMETER_TYPE_BITS 4 + +/* https://docs.microsoft.com/en-us/windows/win32/api/winnt/ns-winnt-mem_extended_parameter */ +typedef struct MEM_EXTENDED_PARAMETER { + struct { + DWORD64 Type : MEM_EXTENDED_PARAMETER_TYPE_BITS; + DWORD64 Reserved : 64 - MEM_EXTENDED_PARAMETER_TYPE_BITS; + } DUMMYSTRUCTNAME; + union { + DWORD64 ULong64; + PVOID Pointer; + SIZE_T Size; + HANDLE Handle; + DWORD ULong; + } DUMMYUNIONNAME; +} MEM_EXTENDED_PARAMETER, *PMEM_EXTENDED_PARAMETER; + +/* https://docs.microsoft.com/en-us/windows/win32/api/memoryapi/nf-memoryapi-virtualalloc2 */ +typedef PVOID (*VirtualAlloc2_type)( + HANDLE Process, + PVOID BaseAddress, + SIZE_T Size, + ULONG AllocationType, + ULONG PageProtection, + MEM_EXTENDED_PARAMETER *ExtendedParameters, + ULONG ParameterCount +); + +/* VirtualAlloc2() flags. */ +#define MEM_COALESCE_PLACEHOLDERS 0x00000001 +#define MEM_PRESERVE_PLACEHOLDER 0x00000002 +#define MEM_REPLACE_PLACEHOLDER 0x00004000 +#define MEM_RESERVE_PLACEHOLDER 0x00040000 + +/* Named exactly as the function, so that user code does not depend + * on it being found at compile time or dynamically. + */ +static VirtualAlloc2_type VirtualAlloc2; + +int +eal_mem_win32api_init(void) +{ + /* Contrary to the docs, VirtualAlloc2() is not in kernel32.dll, + * see https://github.com/MicrosoftDocs/feedback/issues/1129. + */ + static const char library_name[] = "kernelbase.dll"; + static const char function[] = "VirtualAlloc2"; + + HMODULE library = NULL; + int ret = 0; + + /* Already done. */ + if (VirtualAlloc2 != NULL) + return 0; + + library = LoadLibraryA(library_name); + if (library == NULL) { + RTE_LOG_WIN32_ERR("LoadLibraryA(\"%s\")", library_name); + return -1; + } + + VirtualAlloc2 = (VirtualAlloc2_type)( + (void *)GetProcAddress(library, function)); + if (VirtualAlloc2 == NULL) { + RTE_LOG_WIN32_ERR("GetProcAddress(\"%s\", \"%s\")\n", + library_name, function); + + /* Contrary to the docs, Server 2016 is not supported. */ + RTE_LOG(ERR, EAL, "Windows 10 or Windows Server 2019 " + " is required for memory management\n"); + ret = -1; + } + + FreeLibrary(library); + + return ret; +} + +#else + +/* Stub in case VirtualAlloc2() is provided by the compiler. */ +int +eal_mem_win32api_init(void) +{ + return 0; +} + +#endif /* defined(MEM_RESERVE_PLACEHOLDER) */ + +static HANDLE virt2phys_device = INVALID_HANDLE_VALUE; + +int +eal_mem_virt2iova_init(void) +{ + HDEVINFO list = INVALID_HANDLE_VALUE; + SP_DEVICE_INTERFACE_DATA ifdata; + SP_DEVICE_INTERFACE_DETAIL_DATA *detail = NULL; + DWORD detail_size; + int ret = -1; + + list = SetupDiGetClassDevs( + &GUID_DEVINTERFACE_VIRT2PHYS, NULL, NULL, + DIGCF_DEVICEINTERFACE | DIGCF_PRESENT); + if (list == INVALID_HANDLE_VALUE) { + RTE_LOG_WIN32_ERR("SetupDiGetClassDevs()"); + goto exit; + } + + ifdata.cbSize = sizeof(ifdata); + if (!SetupDiEnumDeviceInterfaces( + list, NULL, &GUID_DEVINTERFACE_VIRT2PHYS, 0, &ifdata)) { + RTE_LOG_WIN32_ERR("SetupDiEnumDeviceInterfaces()"); + goto exit; + } + + if (!SetupDiGetDeviceInterfaceDetail( + list, &ifdata, NULL, 0, &detail_size, NULL)) { + if (GetLastError() != ERROR_INSUFFICIENT_BUFFER) { + RTE_LOG_WIN32_ERR( + "SetupDiGetDeviceInterfaceDetail(probe)"); + goto exit; + } + } + + detail = malloc(detail_size); + if (detail == NULL) { + RTE_LOG(ERR, EAL, "Cannot allocate virt2phys " + "device interface detail data\n"); + goto exit; + } + + detail->cbSize = sizeof(*detail); + if (!SetupDiGetDeviceInterfaceDetail( + list, &ifdata, detail, detail_size, NULL, NULL)) { + RTE_LOG_WIN32_ERR("SetupDiGetDeviceInterfaceDetail(read)"); + goto exit; + } + + RTE_LOG(DEBUG, EAL, "Found virt2phys device: %s\n", detail->DevicePath); + + virt2phys_device = CreateFile( + detail->DevicePath, 0, 0, NULL, OPEN_EXISTING, 0, NULL); + if (virt2phys_device == INVALID_HANDLE_VALUE) { + RTE_LOG_WIN32_ERR("CreateFile()"); + goto exit; + } + + /* Indicate success. */ + ret = 0; + +exit: + if (detail != NULL) + free(detail); + if (list != INVALID_HANDLE_VALUE) + SetupDiDestroyDeviceInfoList(list); + + return ret; +} + +phys_addr_t +rte_mem_virt2phy(const void *virt) +{ + LARGE_INTEGER phys; + DWORD bytes_returned; + + if (virt2phys_device == INVALID_HANDLE_VALUE) + return RTE_BAD_PHYS_ADDR; + + if (!DeviceIoControl( + virt2phys_device, IOCTL_VIRT2PHYS_TRANSLATE, + &virt, sizeof(virt), &phys, sizeof(phys), + &bytes_returned, NULL)) { + RTE_LOG_WIN32_ERR("DeviceIoControl(IOCTL_VIRT2PHYS_TRANSLATE)"); + return RTE_BAD_PHYS_ADDR; + } + + return phys.QuadPart; +} + +/* Windows currently only supports IOVA as PA. */ +rte_iova_t +rte_mem_virt2iova(const void *virt) +{ + phys_addr_t phys; + + if (virt2phys_device == INVALID_HANDLE_VALUE) + return RTE_BAD_IOVA; + + phys = rte_mem_virt2phy(virt); + if (phys == RTE_BAD_PHYS_ADDR) + return RTE_BAD_IOVA; + + return (rte_iova_t)phys; +} + +/* Always using physical addresses under Windows if they can be obtained. */ +int +rte_eal_using_phys_addrs(void) +{ + return virt2phys_device != INVALID_HANDLE_VALUE; +} + +/* Approximate error mapping from VirtualAlloc2() to POSIX mmap(3). */ +static void +set_errno_from_win32_alloc_error(DWORD code) +{ + switch (code) { + case ERROR_SUCCESS: + rte_errno = 0; + break; + + case ERROR_INVALID_ADDRESS: + /* A valid requested address is not available. */ + case ERROR_COMMITMENT_LIMIT: + /* May occcur when committing regular memory. */ + case ERROR_NO_SYSTEM_RESOURCES: + /* Occurs when the system runs out of hugepages. */ + rte_errno = ENOMEM; + break; + + case ERROR_INVALID_PARAMETER: + default: + rte_errno = EINVAL; + break; + } +} + +void * +eal_mem_reserve(void *requested_addr, size_t size, int flags) +{ + HANDLE process; + void *virt; + + /* Windows requires hugepages to be committed. */ + if (flags & EAL_RESERVE_HUGEPAGES) { + rte_errno = ENOTSUP; + return NULL; + } + + process = GetCurrentProcess(); + + virt = VirtualAlloc2(process, requested_addr, size, + MEM_RESERVE | MEM_RESERVE_PLACEHOLDER, PAGE_NOACCESS, + NULL, 0); + if (virt == NULL) { + DWORD err = GetLastError(); + RTE_LOG_WIN32_ERR("VirtualAlloc2()"); + set_errno_from_win32_alloc_error(err); + return NULL; + } + + if ((flags & EAL_RESERVE_FORCE_ADDRESS) && (virt != requested_addr)) { + if (!VirtualFreeEx(process, virt, 0, MEM_RELEASE)) + RTE_LOG_WIN32_ERR("VirtualFreeEx()"); + rte_errno = ENOMEM; + return NULL; + } + + return virt; +} + +void * +eal_mem_alloc_socket(size_t size, int socket_id) +{ + DWORD flags = MEM_RESERVE | MEM_COMMIT; + void *addr; + + flags = MEM_RESERVE | MEM_COMMIT | MEM_LARGE_PAGES; + addr = VirtualAllocExNuma(GetCurrentProcess(), NULL, size, flags, + PAGE_READWRITE, eal_socket_numa_node(socket_id)); + if (addr == NULL) + rte_errno = ENOMEM; + return addr; +} + +void * +eal_mem_commit(void *requested_addr, size_t size, int socket_id) +{ + HANDLE process; + MEM_EXTENDED_PARAMETER param; + DWORD param_count = 0; + DWORD flags; + void *addr; + + process = GetCurrentProcess(); + + if (requested_addr != NULL) { + MEMORY_BASIC_INFORMATION info; + + if (VirtualQueryEx(process, requested_addr, &info, + sizeof(info)) != sizeof(info)) { + RTE_LOG_WIN32_ERR("VirtualQuery(%p)", requested_addr); + return NULL; + } + + /* Split reserved region if only a part is committed. */ + flags = MEM_RELEASE | MEM_PRESERVE_PLACEHOLDER; + if ((info.RegionSize > size) && !VirtualFreeEx( + process, requested_addr, size, flags)) { + RTE_LOG_WIN32_ERR( + "VirtualFreeEx(%p, %zu, preserve placeholder)", + requested_addr, size); + return NULL; + } + + /* Temporarily release the region to be committed. + * + * There is an inherent race for this memory range + * if another thread allocates memory via OS API. + * However, VirtualAlloc2(MEM_REPLACE_PLACEHOLDER) + * doesn't work with MEM_LARGE_PAGES on Windows Server. + */ + if (!VirtualFreeEx(process, requested_addr, 0, MEM_RELEASE)) { + RTE_LOG_WIN32_ERR("VirtualFreeEx(%p, 0, release)", + requested_addr); + return NULL; + } + } + + if (socket_id != SOCKET_ID_ANY) { + param_count = 1; + memset(¶m, 0, sizeof(param)); + param.Type = MemExtendedParameterNumaNode; + param.ULong = eal_socket_numa_node(socket_id); + } + + flags = MEM_RESERVE | MEM_COMMIT | MEM_LARGE_PAGES; + addr = VirtualAlloc2(process, requested_addr, size, + flags, PAGE_READWRITE, ¶m, param_count); + if (addr == NULL) { + /* Logging may overwrite GetLastError() result. */ + DWORD err = GetLastError(); + RTE_LOG_WIN32_ERR("VirtualAlloc2(%p, %zu, commit large pages)", + requested_addr, size); + set_errno_from_win32_alloc_error(err); + return NULL; + } + + if ((requested_addr != NULL) && (addr != requested_addr)) { + /* We lost the race for the requested_addr. */ + if (!VirtualFreeEx(process, addr, 0, MEM_RELEASE)) + RTE_LOG_WIN32_ERR("VirtualFreeEx(%p, release)", addr); + + rte_errno = EADDRNOTAVAIL; + return NULL; + } + + return addr; +} + +int +eal_mem_decommit(void *addr, size_t size) +{ + HANDLE process; + void *stub; + DWORD flags; + + process = GetCurrentProcess(); + + /* Hugepages cannot be decommited on Windows, + * so free them and replace the block with a placeholder. + * There is a race for VA in this block until VirtualAlloc2 call. + */ + if (!VirtualFreeEx(process, addr, 0, MEM_RELEASE)) { + RTE_LOG_WIN32_ERR("VirtualFreeEx(%p, 0, release)", addr); + return -1; + } + + flags = MEM_RESERVE | MEM_RESERVE_PLACEHOLDER; + stub = VirtualAlloc2( + process, addr, size, flags, PAGE_NOACCESS, NULL, 0); + if (stub == NULL) { + /* We lost the race for the VA. */ + if (!VirtualFreeEx(process, stub, 0, MEM_RELEASE)) + RTE_LOG_WIN32_ERR("VirtualFreeEx(%p, release)", stub); + rte_errno = EADDRNOTAVAIL; + return -1; + } + + /* No need to join reserved regions adjascent to the freed one: + * eal_mem_commit() will just pick up the page-size placeholder + * created here. + */ + return 0; +} + +/** + * Free a reserved memory region in full or in part. + * + * @param addr + * Starting address of the area to free. + * @param size + * Number of bytes to free. Must be a multiple of page size. + * @param reserved + * Fail if the region is not in reserved state. + * @return + * * 0 on successful deallocation; + * * 1 if region mut be in reserved state but it is not; + * * (-1) on system API failures. + */ +static int +mem_free(void *addr, size_t size, bool reserved) +{ + MEMORY_BASIC_INFORMATION info; + HANDLE process; + + process = GetCurrentProcess(); + + if (VirtualQueryEx( + process, addr, &info, sizeof(info)) != sizeof(info)) { + RTE_LOG_WIN32_ERR("VirtualQueryEx(%p)", addr); + return -1; + } + + if (reserved && (info.State != MEM_RESERVE)) + return 1; + + /* Free complete region. */ + if ((addr == info.AllocationBase) && (size == info.RegionSize)) { + if (!VirtualFreeEx(process, addr, 0, MEM_RELEASE)) { + RTE_LOG_WIN32_ERR("VirtualFreeEx(%p, 0, release)", + addr); + } + return 0; + } + + /* Split the part to be freed and the remaining reservation. */ + if (!VirtualFreeEx(process, addr, size, + MEM_RELEASE | MEM_PRESERVE_PLACEHOLDER)) { + RTE_LOG_WIN32_ERR( + "VirtualFreeEx(%p, %zu, preserve placeholder)", + addr, size); + return -1; + } + + /* Actually free reservation part. */ + if (!VirtualFreeEx(process, addr, 0, MEM_RELEASE)) { + RTE_LOG_WIN32_ERR("VirtualFreeEx(%p, 0, release)", addr); + return -1; + } + + return 0; +} + +void +eal_mem_free(void *virt, size_t size) +{ + mem_free(virt, size, false); +} + +int +eal_mem_set_dump(void *virt, size_t size, bool dump) +{ + RTE_SET_USED(virt); + RTE_SET_USED(size); + RTE_SET_USED(dump); + + /* Windows does not dump reserved memory by default. + * + * There is to include or exclude regions from the dump, + * but this is not currently required by EAL. + */ + + rte_errno = ENOTSUP; + return -1; +} + +void * +rte_mem_map(void *requested_addr, size_t size, int prot, int flags, + int fd, size_t offset) +{ + HANDLE file_handle = INVALID_HANDLE_VALUE; + HANDLE mapping_handle = INVALID_HANDLE_VALUE; + DWORD sys_prot = 0; + DWORD sys_access = 0; + DWORD size_high = (DWORD)(size >> 32); + DWORD size_low = (DWORD)size; + DWORD offset_high = (DWORD)(offset >> 32); + DWORD offset_low = (DWORD)offset; + LPVOID virt = NULL; + + if (prot & RTE_PROT_EXECUTE) { + if (prot & RTE_PROT_READ) { + sys_prot = PAGE_EXECUTE_READ; + sys_access = FILE_MAP_READ | FILE_MAP_EXECUTE; + } + if (prot & RTE_PROT_WRITE) { + sys_prot = PAGE_EXECUTE_READWRITE; + sys_access = FILE_MAP_WRITE | FILE_MAP_EXECUTE; + } + } else { + if (prot & RTE_PROT_READ) { + sys_prot = PAGE_READONLY; + sys_access = FILE_MAP_READ; + } + if (prot & RTE_PROT_WRITE) { + sys_prot = PAGE_READWRITE; + sys_access = FILE_MAP_WRITE; + } + } + + if (flags & RTE_MAP_PRIVATE) + sys_access |= FILE_MAP_COPY; + + if ((flags & RTE_MAP_ANONYMOUS) == 0) + file_handle = (HANDLE)_get_osfhandle(fd); + + mapping_handle = CreateFileMapping( + file_handle, NULL, sys_prot, size_high, size_low, NULL); + if (mapping_handle == INVALID_HANDLE_VALUE) { + RTE_LOG_WIN32_ERR("CreateFileMapping()"); + return NULL; + } + + /* There is a race for the requested_addr between mem_free() + * and MapViewOfFileEx(). MapViewOfFile3() that can replace a reserved + * region with a mapping in a single operation, but it does not support + * private mappings. + */ + if (requested_addr != NULL) { + int ret = mem_free(requested_addr, size, true); + if (ret) { + if (ret > 0) { + RTE_LOG(ERR, EAL, "Cannot map memory " + "to a region not reserved\n"); + rte_errno = EADDRNOTAVAIL; + } + return NULL; + } + } + + virt = MapViewOfFileEx(mapping_handle, sys_access, + offset_high, offset_low, size, requested_addr); + if (!virt) { + RTE_LOG_WIN32_ERR("MapViewOfFileEx()"); + return NULL; + } + + if ((flags & RTE_MAP_FORCE_ADDRESS) && (virt != requested_addr)) { + if (!UnmapViewOfFile(virt)) + RTE_LOG_WIN32_ERR("UnmapViewOfFile()"); + virt = NULL; + } + + if (!CloseHandle(mapping_handle)) + RTE_LOG_WIN32_ERR("CloseHandle()"); + + return virt; +} + +int +rte_mem_unmap(void *virt, size_t size) +{ + RTE_SET_USED(size); + + if (!UnmapViewOfFile(virt)) { + RTE_LOG_WIN32_ERR("UnmapViewOfFile()"); + rte_errno = EINVAL; + return -1; + } + return 0; +} + +uint64_t +eal_get_baseaddr(void) +{ + /* Windows strategy for memory allocation is undocumented. + * Returning 0 here effectively disables address guessing + * unless user provides an address hint. + */ + return 0; +} + +size_t +rte_mem_page_size(void) +{ + static SYSTEM_INFO info; + + if (info.dwPageSize == 0) + GetSystemInfo(&info); + + return info.dwPageSize; +} + +int +rte_mem_lock(const void *virt, size_t size) +{ + /* VirtualLock() takes `void*`, work around compiler warning. */ + void *addr = (void *)((uintptr_t)virt); + + if (!VirtualLock(addr, size)) { + RTE_LOG_WIN32_ERR("VirtualLock(%p %#zx)", virt, size); + return -1; + } + + return 0; +} + +int +rte_eal_memseg_init(void) +{ + if (rte_eal_process_type() != RTE_PROC_PRIMARY) { + EAL_LOG_NOT_IMPLEMENTED(); + return -1; + } + + return eal_dynmem_memseg_lists_init(); +} + +static int +eal_nohuge_init(void) +{ + struct rte_mem_config *mcfg; + struct rte_memseg_list *msl; + int n_segs; + uint64_t mem_sz, page_sz; + void *addr; + + mcfg = rte_eal_get_configuration()->mem_config; + + /* nohuge mode is legacy mode */ + internal_config.legacy_mem = 1; + + msl = &mcfg->memsegs[0]; + + mem_sz = internal_config.memory; + page_sz = RTE_PGSIZE_4K; + n_segs = mem_sz / page_sz; + + if (eal_memseg_list_init_named( + msl, "nohugemem", page_sz, n_segs, 0, true)) { + return -1; + } + + addr = VirtualAlloc( + NULL, mem_sz, MEM_RESERVE | MEM_COMMIT, PAGE_READWRITE); + if (addr == NULL) { + RTE_LOG_WIN32_ERR("VirtualAlloc(size=%#zx)", mem_sz); + RTE_LOG(ERR, EAL, "Cannot allocate memory\n"); + return -1; + } + + msl->base_va = addr; + msl->len = mem_sz; + + eal_memseg_list_populate(msl, addr, n_segs); + + if (mcfg->dma_maskbits && + rte_mem_check_dma_mask_thread_unsafe(mcfg->dma_maskbits)) { + RTE_LOG(ERR, EAL, + "%s(): couldn't allocate memory due to IOVA " + "exceeding limits of current DMA mask.\n", __func__); + return -1; + } + + return 0; +} + +int +rte_eal_hugepage_init(void) +{ + return internal_config.no_hugetlbfs ? + eal_nohuge_init() : eal_dynmem_hugepage_init(); +} + +int +rte_eal_hugepage_attach(void) +{ + EAL_LOG_NOT_IMPLEMENTED(); + return -1; +} diff --git a/lib/librte_eal/windows/eal_mp.c b/lib/librte_eal/windows/eal_mp.c new file mode 100644 index 000000000..16a5e8ba0 --- /dev/null +++ b/lib/librte_eal/windows/eal_mp.c @@ -0,0 +1,103 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright (c) 2020 Dmitry Kozlyuk + */ + +/** + * @file Multiprocess support stubs + * + * Stubs must log an error until implemented. If success is required + * for non-multiprocess operation, stub must log a warning and a comment + * must document what requires success emulation. + */ + +#include +#include + +#include "eal_private.h" +#include "eal_windows.h" +#include "malloc_mp.h" + +void +rte_mp_channel_cleanup(void) +{ + EAL_LOG_NOT_IMPLEMENTED(); +} + +int +rte_mp_action_register(const char *name, rte_mp_t action) +{ + RTE_SET_USED(name); + RTE_SET_USED(action); + EAL_LOG_NOT_IMPLEMENTED(); + return -1; +} + +void +rte_mp_action_unregister(const char *name) +{ + RTE_SET_USED(name); + EAL_LOG_NOT_IMPLEMENTED(); +} + +int +rte_mp_sendmsg(struct rte_mp_msg *msg) +{ + RTE_SET_USED(msg); + EAL_LOG_NOT_IMPLEMENTED(); + return -1; +} + +int +rte_mp_request_sync(struct rte_mp_msg *req, struct rte_mp_reply *reply, + const struct timespec *ts) +{ + RTE_SET_USED(req); + RTE_SET_USED(reply); + RTE_SET_USED(ts); + EAL_LOG_NOT_IMPLEMENTED(); + return -1; +} + +int +rte_mp_request_async(struct rte_mp_msg *req, const struct timespec *ts, + rte_mp_async_reply_t clb) +{ + RTE_SET_USED(req); + RTE_SET_USED(ts); + RTE_SET_USED(clb); + EAL_LOG_NOT_IMPLEMENTED(); + return -1; +} + +int +rte_mp_reply(struct rte_mp_msg *msg, const char *peer) +{ + RTE_SET_USED(msg); + RTE_SET_USED(peer); + EAL_LOG_NOT_IMPLEMENTED(); + return -1; +} + +int +register_mp_requests(void) +{ + /* Non-stub function succeeds if multi-process is not supported. */ + EAL_LOG_STUB(); + return 0; +} + +int +request_to_primary(struct malloc_mp_req *req) +{ + RTE_SET_USED(req); + EAL_LOG_NOT_IMPLEMENTED(); + return -1; +} + +int +request_sync(void) +{ + /* Common memory allocator depends on this function success. */ + EAL_LOG_STUB(); + return 0; +} diff --git a/lib/librte_eal/windows/eal_windows.h b/lib/librte_eal/windows/eal_windows.h index 390d2fd66..caabffedf 100644 --- a/lib/librte_eal/windows/eal_windows.h +++ b/lib/librte_eal/windows/eal_windows.h @@ -9,8 +9,24 @@ * @file Facilities private to Windows EAL */ +#include #include +/** + * Log current function as not implemented and set rte_errno. + */ +#define EAL_LOG_NOT_IMPLEMENTED() \ + do { \ + RTE_LOG(DEBUG, EAL, "%s() is not implemented\n", __func__); \ + rte_errno = ENOTSUP; \ + } while (0) + +/** + * Log current function as a stub. + */ +#define EAL_LOG_STUB() \ + RTE_LOG(DEBUG, EAL, "Windows: %s() is a stub\n", __func__) + /** * Create a map of processors and cores on the system. */ @@ -36,4 +52,63 @@ int eal_thread_create(pthread_t *thread); */ unsigned int eal_socket_numa_node(unsigned int socket_id); +/** + * Open virt2phys driver interface device. + * + * @return 0 on success, (-1) on failure. + */ +int eal_mem_virt2iova_init(void); + +/** + * Locate Win32 memory management routines in system libraries. + * + * @return 0 on success, (-1) on failure. + */ +int eal_mem_win32api_init(void); + +/** + * Allocate new memory in hugepages on the specified NUMA node. + * + * @param size + * Number of bytes to allocate. Must be a multiple of huge page size. + * @param socket_id + * Socket ID. + * @return + * Address of the memory allocated on success or NULL on failure. + */ +void *eal_mem_alloc_socket(size_t size, int socket_id); + +/** + * Commit memory previously reserved with eal_mem_reserve() + * or decommitted from hugepages by eal_mem_decommit(). + * + * @param requested_addr + * Address within a reserved region. Must not be NULL. + * @param size + * Number of bytes to commit. Must be a multiple of page size. + * @param socket_id + * Socket ID to allocate on. Can be SOCKET_ID_ANY. + * @return + * On success, address of the committed memory, that is, requested_addr. + * On failure, NULL and rte_errno is set. + */ +void *eal_mem_commit(void *requested_addr, size_t size, int socket_id); + +/** + * Put allocated or committed memory back into reserved state. + * + * @param addr + * Address of the region to decommit. + * @param size + * Number of bytes to decommit, must be the size of a page + * (hugepage or regular one). + * + * The *addr* and *size* must match location and size + * of a previously allocated or committed region. + * + * @return + * 0 on success, (-1) on failure. + */ +int eal_mem_decommit(void *addr, size_t size); + #endif /* _EAL_WINDOWS_H_ */ diff --git a/lib/librte_eal/windows/include/meson.build b/lib/librte_eal/windows/include/meson.build index 5fb1962ac..b3534b025 100644 --- a/lib/librte_eal/windows/include/meson.build +++ b/lib/librte_eal/windows/include/meson.build @@ -5,5 +5,6 @@ includes += include_directories('.') headers += files( 'rte_os.h', + 'rte_virt2phys.h', 'rte_windows.h', ) diff --git a/lib/librte_eal/windows/include/rte_os.h b/lib/librte_eal/windows/include/rte_os.h index 510e39e03..cb10d6494 100644 --- a/lib/librte_eal/windows/include/rte_os.h +++ b/lib/librte_eal/windows/include/rte_os.h @@ -14,6 +14,7 @@ #include #include #include +#include #ifdef __cplusplus extern "C" { @@ -36,6 +37,9 @@ extern "C" { #define strncasecmp(s1, s2, count) _strnicmp(s1, s2, count) +#define close _close +#define unlink _unlink + /* cpu_set macros implementation */ #define RTE_CPU_AND(dst, src1, src2) CPU_AND(dst, src1, src2) #define RTE_CPU_OR(dst, src1, src2) CPU_OR(dst, src1, src2) @@ -46,6 +50,7 @@ extern "C" { typedef long long ssize_t; #ifndef RTE_TOOLCHAIN_GCC + static inline int asprintf(char **buffer, const char *format, ...) { @@ -72,6 +77,18 @@ asprintf(char **buffer, const char *format, ...) } return ret; } + +static inline const char * +eal_strerror(int code) +{ + static char buffer[128]; + + strerror_s(buffer, sizeof(buffer), code); + return buffer; +} + +#define strerror eal_strerror + #endif /* RTE_TOOLCHAIN_GCC */ #ifdef __cplusplus diff --git a/lib/librte_eal/windows/include/rte_virt2phys.h b/lib/librte_eal/windows/include/rte_virt2phys.h new file mode 100644 index 000000000..4bb2b4aaf --- /dev/null +++ b/lib/librte_eal/windows/include/rte_virt2phys.h @@ -0,0 +1,34 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright (c) 2020 Dmitry Kozlyuk + */ + +/** + * @file virt2phys driver interface + */ + +/** + * Driver device interface GUID {539c2135-793a-4926-afec-d3a1b61bbc8a}. + */ +DEFINE_GUID(GUID_DEVINTERFACE_VIRT2PHYS, + 0x539c2135, 0x793a, 0x4926, + 0xaf, 0xec, 0xd3, 0xa1, 0xb6, 0x1b, 0xbc, 0x8a); + +/** + * Driver device type for IO control codes. + */ +#define VIRT2PHYS_DEVTYPE 0x8000 + +/** + * Translate a valid non-paged virtual address to a physical address. + * + * Note: A physical address zero (0) is reported if input address + * is paged out or not mapped. However, if input is a valid mapping + * of I/O port 0x0000, output is also zero. There is no way + * to distinguish between these cases by return value only. + * + * Input: a non-paged virtual address (PVOID). + * + * Output: the corresponding physical address (LARGE_INTEGER). + */ +#define IOCTL_VIRT2PHYS_TRANSLATE CTL_CODE( \ + VIRT2PHYS_DEVTYPE, 0x800, METHOD_BUFFERED, FILE_ANY_ACCESS) diff --git a/lib/librte_eal/windows/include/rte_windows.h b/lib/librte_eal/windows/include/rte_windows.h index ed6e4c148..899ed7d87 100644 --- a/lib/librte_eal/windows/include/rte_windows.h +++ b/lib/librte_eal/windows/include/rte_windows.h @@ -23,6 +23,8 @@ #include #include +#include +#include /* Have GUIDs defined. */ #ifndef INITGUID diff --git a/lib/librte_eal/windows/include/unistd.h b/lib/librte_eal/windows/include/unistd.h index 757b7f3c5..6b33005b2 100644 --- a/lib/librte_eal/windows/include/unistd.h +++ b/lib/librte_eal/windows/include/unistd.h @@ -9,4 +9,7 @@ * as Microsoft libc does not contain unistd.h. This may be removed * in future releases. */ + +#include + #endif /* _UNISTD_H_ */ diff --git a/lib/librte_eal/windows/meson.build b/lib/librte_eal/windows/meson.build index 52978e9d7..ded5a2b80 100644 --- a/lib/librte_eal/windows/meson.build +++ b/lib/librte_eal/windows/meson.build @@ -6,10 +6,16 @@ subdir('include') sources += files( 'eal.c', 'eal_debug.c', + 'eal_file.c', 'eal_hugepages.c', 'eal_lcore.c', 'eal_log.c', + 'eal_memalloc.c', + 'eal_memory.c', + 'eal_mp.c', 'eal_thread.c', 'fnmatch.c', 'getopt.c', ) + +dpdk_conf.set10('RTE_EAL_NUMA_AWARE_HUGEPAGES', true)