From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from merlin.infradead.org ([205.233.59.134]:42328 "EHLO merlin.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751152AbeDNMAJ (ORCPT ); Sat, 14 Apr 2018 08:00:09 -0400 Received: from [216.160.245.99] (helo=kernel.dk) by merlin.infradead.org with esmtpsa (Exim 4.90_1 #2 (Red Hat Linux)) id 1f7Jqi-00021s-0P for fio@vger.kernel.org; Sat, 14 Apr 2018 12:00:08 +0000 Subject: Recent changes (master) From: Jens Axboe Message-Id: <20180414120002.7214A2C0079@kernel.dk> Date: Sat, 14 Apr 2018 06:00:02 -0600 (MDT) Sender: fio-owner@vger.kernel.org List-Id: fio@vger.kernel.org To: fio@vger.kernel.org The following changes since commit 4fe721ac83e84df7c6be07394d1963fd1ec5d9a6: os/os-dragonfly: sync with header file changes in upstream (2018-04-10 09:17:22 -0600) are available in the git repository at: git://git.kernel.dk/fio.git master for you to fetch changes up to c479640d6208236744f0562b1e79535eec290e2b: Merge branch 'proc_group' of https://github.com/sitsofe/fio (2018-04-13 17:25:35 -0600) ---------------------------------------------------------------- Jens Axboe (1): Merge branch 'proc_group' of https://github.com/sitsofe/fio Sitsofe Wheeler (7): windows: update EULA windows: prepare for Windows build split windows: target Windows 7 and add support for more than 64 CPUs doc: add Windows processor group behaviour and Windows target option configure/Makefile: make Cygwin force less appveyor: make 32 bit build target XP + minor fixes doc: add cpus_allowed reference to log_compression_cpus HOWTO | 47 +++-- Makefile | 3 - README | 11 +- appveyor.yml | 4 +- configure | 29 ++- fio.1 | 47 +++-- os/os-windows-7.h | 367 +++++++++++++++++++++++++++++++++++ os/os-windows-xp.h | 70 +++++++ os/os-windows.h | 85 +------- os/windows/eula.rtf | Bin 1075 -> 1077 bytes os/windows/posix.c | 2 + os/windows/posix/include/arpa/inet.h | 2 + os/windows/posix/include/poll.h | 9 + server.c | 6 +- 14 files changed, 551 insertions(+), 131 deletions(-) create mode 100644 os/os-windows-7.h create mode 100644 os/os-windows-xp.h --- Diff of recent changes: diff --git a/HOWTO b/HOWTO index dbbbfaa..5c8623d 100644 --- a/HOWTO +++ b/HOWTO @@ -2377,24 +2377,27 @@ Threads, processes and job synchronization Set the I/O priority class. See man :manpage:`ionice(1)`. -.. option:: cpumask=int - - Set the CPU affinity of this job. The parameter given is a bit mask of - allowed CPUs the job may run on. So if you want the allowed CPUs to be 1 - and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man - :manpage:`sched_setaffinity(2)`. This may not work on all supported - operating systems or kernel versions. This option doesn't work well for a - higher CPU count than what you can store in an integer mask, so it can only - control cpus 1-32. For boxes with larger CPU counts, use - :option:`cpus_allowed`. - .. option:: cpus_allowed=str Controls the same options as :option:`cpumask`, but accepts a textual - specification of the permitted CPUs instead. So to use CPUs 1 and 5 you - would specify ``cpus_allowed=1,5``. This option also allows a range of CPUs - to be specified -- say you wanted a binding to CPUs 1, 5, and 8 to 15, you - would set ``cpus_allowed=1,5,8-15``. + specification of the permitted CPUs instead and CPUs are indexed from 0. So + to use CPUs 0 and 5 you would specify ``cpus_allowed=0,5``. This option also + allows a range of CPUs to be specified -- say you wanted a binding to CPUs + 0, 5, and 8 to 15, you would set ``cpus_allowed=0,5,8-15``. + + On Windows, when ``cpus_allowed`` is unset only CPUs from fio's current + processor group will be used and affinity settings are inherited from the + system. An fio build configured to target Windows 7 makes options that set + CPUs processor group aware and values will set both the processor group + and a CPU from within that group. For example, on a system where processor + group 0 has 40 CPUs and processor group 1 has 32 CPUs, ``cpus_allowed`` + values between 0 and 39 will bind CPUs from processor group 0 and + ``cpus_allowed`` values between 40 and 71 will bind CPUs from processor + group 1. When using ``cpus_allowed_policy=shared`` all CPUs specified by a + single ``cpus_allowed`` option must be from the same processor group. For + Windows fio builds not built for Windows 7, CPUs will only be selected from + (and be relative to) whatever processor group fio happens to be running in + and CPUs from other processor groups cannot be used. .. option:: cpus_allowed_policy=str @@ -2411,6 +2414,17 @@ Threads, processes and job synchronization enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs in the set. +.. option:: cpumask=int + + Set the CPU affinity of this job. The parameter given is a bit mask of + allowed CPUs the job may run on. So if you want the allowed CPUs to be 1 + and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man + :manpage:`sched_setaffinity(2)`. This may not work on all supported + operating systems or kernel versions. This option doesn't work well for a + higher CPU count than what you can store in an integer mask, so it can only + control cpus 1-32. For boxes with larger CPU counts, use + :option:`cpus_allowed`. + .. option:: numa_cpu_nodes=str Set this job running on specified NUMA nodes' CPUs. The arguments allow @@ -2921,7 +2935,8 @@ Measurements and reporting Define the set of CPUs that are allowed to handle online log compression for the I/O jobs. This can provide better isolation between performance - sensitive jobs, and background compression work. + sensitive jobs, and background compression work. See + :option:`cpus_allowed` for the format used. .. option:: log_store_compressed=bool diff --git a/Makefile b/Makefile index cc4b71f..357ae98 100644 --- a/Makefile +++ b/Makefile @@ -59,9 +59,6 @@ ifdef CONFIG_LIBHDFS SOURCE += engines/libhdfs.c endif -ifdef CONFIG_64BIT_LLP64 - CFLAGS += -DBITS_PER_LONG=32 -endif ifdef CONFIG_64BIT CFLAGS += -DBITS_PER_LONG=64 endif diff --git a/README b/README index fba5f10..38022bb 100644 --- a/README +++ b/README @@ -172,15 +172,18 @@ directory. How to compile fio on 64-bit Windows: 1. Install Cygwin (http://www.cygwin.com/). Install **make** and all - packages starting with **mingw64-i686** and **mingw64-x86_64**. Ensure - **mingw64-i686-zlib** and **mingw64-x86_64-zlib** are installed if you wish + packages starting with **mingw64-x86_64**. Ensure + **mingw64-x86_64-zlib** are installed if you wish to enable fio's log compression functionality. 2. Open the Cygwin Terminal. 3. Go to the fio directory (source files). 4. Run ``make clean && make -j``. -To build fio on 32-bit Windows, run ``./configure --build-32bit-win`` before -``make``. +To build fio for 32-bit Windows, ensure the -i686 versions of the previously +mentioned -x86_64 packages are installed and run ``./configure +--build-32bit-win`` before ``make``. To build an fio that supports versions of +Windows below Windows 7/Windows Server 2008 R2 also add ``--target-win-ver=xp`` +to the end of the configure line that you run before doing ``make``. It's recommended that once built or installed, fio be run in a Command Prompt or other 'native' console such as console2, since there are known to be display and diff --git a/appveyor.yml b/appveyor.yml index 09ebccf..ca8b2ab 100644 --- a/appveyor.yml +++ b/appveyor.yml @@ -10,10 +10,10 @@ environment: CONFIGURE_OPTIONS: - platform: x86 PACKAGE_ARCH: i686 - CONFIGURE_OPTIONS: --build-32bit-win + CONFIGURE_OPTIONS: --build-32bit-win --target-win-ver=xp install: - - '%CYG_ROOT%\setup-x86_64.exe --quiet-mode --no-shortcuts --only-site --site "%CYG_MIRROR%" --packages "mingw64-%PACKAGE_ARCH%-zlib" > NULL' + - '%CYG_ROOT%\setup-x86_64.exe --quiet-mode --no-shortcuts --only-site --site "%CYG_MIRROR%" --packages "mingw64-%PACKAGE_ARCH%-zlib" > NUL' - SET PATH=%CYG_ROOT%\bin;%PATH% #��NB: Changed env variables persist to later sections build_script: diff --git a/configure b/configure index 38706a9..32baec6 100755 --- a/configure +++ b/configure @@ -167,6 +167,8 @@ for opt do ;; --build-32bit-win) build_32bit_win="yes" ;; + --target-win-ver=*) target_win_ver="$optarg" + ;; --build-static) build_static="yes" ;; --enable-gfio) gfio_check="yes" @@ -213,6 +215,7 @@ if test "$show_help" = "yes" ; then echo "--cc= Specify compiler to use" echo "--extra-cflags= Specify extra CFLAGS to pass to compiler" echo "--build-32bit-win Enable 32-bit build on Windows" + echo "--target-win-ver= Minimum version of Windows to target (XP or 7)" echo "--build-static Build a static fio" echo "--esx Configure build options for esx" echo "--enable-gfio Enable building of gtk gfio" @@ -329,20 +332,27 @@ CYGWIN*) cc="x86_64-w64-mingw32-gcc" fi fi - if test ! -z "$build_32bit_win" && test "$build_32bit_win" = "yes"; then - output_sym "CONFIG_32BIT" + + target_win_ver=$(echo "$target_win_ver" | tr '[:lower:]' '[:upper:]') + if test -z "$target_win_ver"; then + # Default Windows API target + target_win_ver="7" + fi + if test "$target_win_ver" = "XP"; then + output_sym "CONFIG_WINDOWS_XP" + elif test "$target_win_ver" = "7"; then + output_sym "CONFIG_WINDOWS_7" + CFLAGS="$CFLAGS -D_WIN32_WINNT=0x0601" else - output_sym "CONFIG_64BIT_LLP64" + fatal "Unknown target Windows version" fi + # We need this to be output_sym'd here because this is Windows specific. # The regular configure path never sets this config. output_sym "CONFIG_WINDOWSAIO" # We now take the regular configuration path without having exit 0 here. # Flags below are still necessary mostly for MinGW. socklen_t="yes" - sfaa="yes" - sync_sync="yes" - cmp_swap="yes" rusage_thread="yes" fdatasync="yes" clock_gettime="yes" # clock_monotonic probe has dependency on this @@ -350,11 +360,7 @@ CYGWIN*) gettimeofday="yes" sched_idle="yes" tcp_nodelay="yes" - tls_thread="yes" - static_assert="yes" ipv6="yes" - mkdir_two="no" - echo "BUILD_CFLAGS=$CFLAGS -include config-host.h -D_GNU_SOURCE" >> $config_host_mak ;; esac @@ -498,6 +504,9 @@ fi print_config "Operating system" "$targetos" print_config "CPU" "$cpu" print_config "Big endian" "$bigendian" +if test ! -z "$target_win_ver"; then + print_config "Target Windows version" "$target_win_ver" +fi print_config "Compiler" "$cc" print_config "Cross compile" "$cross_compile" echo diff --git a/fio.1 b/fio.1 index 5ca57ce..dd4f9cb 100644 --- a/fio.1 +++ b/fio.1 @@ -2091,22 +2091,28 @@ systems since meaning of priority may differ. .BI prioclass \fR=\fPint Set the I/O priority class. See man \fBionice\fR\|(1). .TP -.BI cpumask \fR=\fPint -Set the CPU affinity of this job. The parameter given is a bit mask of -allowed CPUs the job may run on. So if you want the allowed CPUs to be 1 -and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man -\fBsched_setaffinity\fR\|(2). This may not work on all supported -operating systems or kernel versions. This option doesn't work well for a -higher CPU count than what you can store in an integer mask, so it can only -control cpus 1\-32. For boxes with larger CPU counts, use -\fBcpus_allowed\fR. -.TP .BI cpus_allowed \fR=\fPstr Controls the same options as \fBcpumask\fR, but accepts a textual -specification of the permitted CPUs instead. So to use CPUs 1 and 5 you -would specify `cpus_allowed=1,5'. This option also allows a range of CPUs -to be specified \-\- say you wanted a binding to CPUs 1, 5, and 8 to 15, you -would set `cpus_allowed=1,5,8\-15'. +specification of the permitted CPUs instead and CPUs are indexed from 0. So +to use CPUs 0 and 5 you would specify `cpus_allowed=0,5'. This option also +allows a range of CPUs to be specified \-\- say you wanted a binding to CPUs +0, 5, and 8 to 15, you would set `cpus_allowed=0,5,8\-15'. +.RS +.P +On Windows, when `cpus_allowed' is unset only CPUs from fio's current +processor group will be used and affinity settings are inherited from the +system. An fio build configured to target Windows 7 makes options that set +CPUs processor group aware and values will set both the processor group +and a CPU from within that group. For example, on a system where processor +group 0 has 40 CPUs and processor group 1 has 32 CPUs, `cpus_allowed' +values between 0 and 39 will bind CPUs from processor group 0 and +`cpus_allowed' values between 40 and 71 will bind CPUs from processor +group 1. When using `cpus_allowed_policy=shared' all CPUs specified by a +single `cpus_allowed' option must be from the same processor group. For +Windows fio builds not built for Windows 7, CPUs will only be selected from +(and be relative to) whatever processor group fio happens to be running in +and CPUs from other processor groups cannot be used. +.RE .TP .BI cpus_allowed_policy \fR=\fPstr Set the policy of how fio distributes the CPUs specified by @@ -2127,6 +2133,16 @@ enough CPUs are given for the jobs listed, then fio will roundrobin the CPUs in the set. .RE .TP +.BI cpumask \fR=\fPint +Set the CPU affinity of this job. The parameter given is a bit mask of +allowed CPUs the job may run on. So if you want the allowed CPUs to be 1 +and 5, you would pass the decimal value of (1 << 1 | 1 << 5), or 34. See man +\fBsched_setaffinity\fR\|(2). This may not work on all supported +operating systems or kernel versions. This option doesn't work well for a +higher CPU count than what you can store in an integer mask, so it can only +control cpus 1\-32. For boxes with larger CPU counts, use +\fBcpus_allowed\fR. +.TP .BI numa_cpu_nodes \fR=\fPstr Set this job running on specified NUMA nodes' CPUs. The arguments allow comma delimited list of cpu numbers, A\-B ranges, or `all'. Note, to enable @@ -2603,7 +2619,8 @@ zlib. .BI log_compression_cpus \fR=\fPstr Define the set of CPUs that are allowed to handle online log compression for the I/O jobs. This can provide better isolation between performance -sensitive jobs, and background compression work. +sensitive jobs, and background compression work. See \fBcpus_allowed\fR for +the format used. .TP .BI log_store_compressed \fR=\fPbool If set, fio will store the log files in a compressed format. They can be diff --git a/os/os-windows-7.h b/os/os-windows-7.h new file mode 100644 index 0000000..f5ddb8e --- /dev/null +++ b/os/os-windows-7.h @@ -0,0 +1,367 @@ +#define FIO_MAX_CPUS 512 /* From Hyper-V 2016's max logical processors */ +#define FIO_CPU_MASK_STRIDE 64 +#define FIO_CPU_MASK_ROWS (FIO_MAX_CPUS / FIO_CPU_MASK_STRIDE) + +typedef struct { + uint64_t row[FIO_CPU_MASK_ROWS]; +} os_cpu_mask_t; + +#define FIO_HAVE_CPU_ONLINE_SYSCONF +/* Return all processors regardless of processor group */ +static inline unsigned int cpus_online(void) +{ + return GetMaximumProcessorCount(ALL_PROCESSOR_GROUPS); +} + +static inline void print_mask(os_cpu_mask_t *cpumask) +{ + for (int i = 0; i < FIO_CPU_MASK_ROWS; i++) + dprint(FD_PROCESS, "cpumask[%d]=%lu\n", i, cpumask->row[i]); +} + +/* Return the index of the least significant set CPU in cpumask or -1 if no + * CPUs are set */ +static inline int first_set_cpu(os_cpu_mask_t *cpumask) +{ + int cpus_offset, mask_first_cpu, row; + + cpus_offset = 0; + row = 0; + mask_first_cpu = -1; + while (mask_first_cpu < 0 && row < FIO_CPU_MASK_ROWS) { + int row_first_cpu; + + row_first_cpu = __builtin_ffsll(cpumask->row[row]) - 1; + dprint(FD_PROCESS, "row_first_cpu=%d cpumask->row[%d]=%lu\n", + row_first_cpu, row, cpumask->row[row]); + if (row_first_cpu > -1) { + mask_first_cpu = cpus_offset + row_first_cpu; + dprint(FD_PROCESS, "first set cpu in mask is at index %d\n", + mask_first_cpu); + } else { + cpus_offset += FIO_CPU_MASK_STRIDE; + row++; + } + } + + return mask_first_cpu; +} + +/* Return the index of the most significant set CPU in cpumask or -1 if no + * CPUs are set */ +static inline int last_set_cpu(os_cpu_mask_t *cpumask) +{ + int cpus_offset, mask_last_cpu, row; + + cpus_offset = (FIO_CPU_MASK_ROWS - 1) * FIO_CPU_MASK_STRIDE; + row = FIO_CPU_MASK_ROWS - 1; + mask_last_cpu = -1; + while (mask_last_cpu < 0 && row >= 0) { + int row_last_cpu; + + if (cpumask->row[row] == 0) + row_last_cpu = -1; + else { + uint64_t tmp = cpumask->row[row]; + + row_last_cpu = 0; + while (tmp >>= 1) + row_last_cpu++; + } + + dprint(FD_PROCESS, "row_last_cpu=%d cpumask->row[%d]=%lu\n", + row_last_cpu, row, cpumask->row[row]); + if (row_last_cpu > -1) { + mask_last_cpu = cpus_offset + row_last_cpu; + dprint(FD_PROCESS, "last set cpu in mask is at index %d\n", + mask_last_cpu); + } else { + cpus_offset -= FIO_CPU_MASK_STRIDE; + row--; + } + } + + return mask_last_cpu; +} + +static inline int mask_to_group_mask(os_cpu_mask_t *cpumask, int *processor_group, uint64_t *affinity_mask) +{ + WORD online_groups, group, group_size; + bool found; + int cpus_offset, search_cpu, last_cpu, bit_offset, row, end; + uint64_t group_cpumask; + + search_cpu = first_set_cpu(cpumask); + if (search_cpu < 0) { + log_info("CPU mask doesn't set any CPUs\n"); + return 1; + } + + /* Find processor group first set CPU applies to */ + online_groups = GetActiveProcessorGroupCount(); + group = 0; + found = false; + cpus_offset = 0; + group_size = 0; + while (!found && group < online_groups) { + group_size = GetMaximumProcessorCount(group); + dprint(FD_PROCESS, "group=%d group_start=%d group_size=%u search_cpu=%d\n", + group, cpus_offset, group_size, search_cpu); + if (cpus_offset + group_size > search_cpu) + found = true; + else { + cpus_offset += group_size; + group++; + } + } + + if (!found) { + log_err("CPU mask contains processor beyond last active processor index (%d)\n", + cpus_offset - 1); + print_mask(cpumask); + return 1; + } + + /* Check all the CPUs in the mask apply to ONLY that processor group */ + last_cpu = last_set_cpu(cpumask); + if (last_cpu > (cpus_offset + group_size - 1)) { + log_info("CPU mask cannot bind CPUs (e.g. %d, %d) that are " + "in different processor groups\n", search_cpu, + last_cpu); + print_mask(cpumask); + return 1; + } + + /* Extract the current processor group mask from the cpumask */ + row = cpus_offset / FIO_CPU_MASK_STRIDE; + bit_offset = cpus_offset % FIO_CPU_MASK_STRIDE; + group_cpumask = cpumask->row[row] >> bit_offset; + end = bit_offset + group_size; + if (end > FIO_CPU_MASK_STRIDE && (row + 1 < FIO_CPU_MASK_ROWS)) { + /* Some of the next row needs to be part of the mask */ + int needed, needed_shift, needed_mask_shift; + uint64_t needed_mask; + + needed = end - FIO_CPU_MASK_STRIDE; + needed_shift = FIO_CPU_MASK_STRIDE - bit_offset; + needed_mask_shift = FIO_CPU_MASK_STRIDE - needed; + needed_mask = (uint64_t)-1 >> needed_mask_shift; + dprint(FD_PROCESS, "bit_offset=%d end=%d needed=%d needed_shift=%d needed_mask=%ld needed_mask_shift=%d\n", bit_offset, end, needed, needed_shift, needed_mask, needed_mask_shift); + group_cpumask |= (cpumask->row[row + 1] & needed_mask) << needed_shift; + } + group_cpumask &= (uint64_t)-1 >> (FIO_CPU_MASK_STRIDE - group_size); + + /* Return group and mask */ + dprint(FD_PROCESS, "Returning group=%d group_mask=%lu\n", group, group_cpumask); + *processor_group = group; + *affinity_mask = group_cpumask; + + return 0; +} + +static inline int fio_setaffinity(int pid, os_cpu_mask_t cpumask) +{ + HANDLE handle = NULL; + int group, ret; + uint64_t group_mask = 0; + GROUP_AFFINITY new_group_affinity; + + ret = -1; + + if (mask_to_group_mask(&cpumask, &group, &group_mask) != 0) + goto err; + + handle = OpenThread(THREAD_QUERY_INFORMATION | THREAD_SET_INFORMATION, + TRUE, pid); + if (handle == NULL) { + log_err("fio_setaffinity: failed to get handle for pid %d\n", pid); + goto err; + } + + /* Set group and mask. + * Note: if the GROUP_AFFINITY struct's Reserved members are not + * initialised to 0 then SetThreadGroupAffinity will fail with + * GetLastError() set to ERROR_INVALID_PARAMETER */ + new_group_affinity.Mask = (KAFFINITY) group_mask; + new_group_affinity.Group = group; + new_group_affinity.Reserved[0] = 0; + new_group_affinity.Reserved[1] = 0; + new_group_affinity.Reserved[2] = 0; + if (SetThreadGroupAffinity(handle, &new_group_affinity, NULL) != 0) + ret = 0; + else { + log_err("fio_setaffinity: failed to set thread affinity " + "(pid %d, group %d, mask %" PRIx64 ", " + "GetLastError=%d)\n", pid, group, group_mask, + GetLastError()); + goto err; + } + +err: + if (handle) + CloseHandle(handle); + return ret; +} + +static inline void cpu_to_row_offset(int cpu, int *row, int *offset) +{ + *row = cpu / FIO_CPU_MASK_STRIDE; + *offset = cpu << FIO_CPU_MASK_STRIDE * *row; +} + +static inline int fio_cpuset_init(os_cpu_mask_t *mask) +{ + for (int i = 0; i < FIO_CPU_MASK_ROWS; i++) + mask->row[i] = 0; + return 0; +} + +/* + * fio_getaffinity() should not be called once a fio_setaffinity() call has + * been made because fio_setaffinity() may put the process into multiple + * processor groups + */ +static inline int fio_getaffinity(int pid, os_cpu_mask_t *mask) +{ + int ret; + int row, offset, end, group, group_size, group_start_cpu; + DWORD_PTR process_mask, system_mask; + HANDLE handle; + PUSHORT current_groups; + USHORT group_count; + WORD online_groups; + + ret = -1; + current_groups = NULL; + handle = OpenProcess(PROCESS_QUERY_INFORMATION, TRUE, pid); + if (handle == NULL) { + log_err("fio_getaffinity: failed to get handle for pid %d\n", + pid); + goto err; + } + + group_count = 1; + /* + * GetProcessGroupAffinity() seems to expect more than the natural + * alignment for a USHORT from the area pointed to by current_groups so + * arrange for maximum alignment by allocating via malloc() + */ + current_groups = malloc(sizeof(USHORT)); + if (!current_groups) { + log_err("fio_getaffinity: malloc failed\n"); + goto err; + } + if (GetProcessGroupAffinity(handle, &group_count, current_groups) == 0) { + /* NB: we also fail here if we are a multi-group process */ + log_err("fio_getaffinity: failed to get single group affinity for pid %d\n", pid); + goto err; + } + GetProcessAffinityMask(handle, &process_mask, &system_mask); + + /* Convert group and group relative mask to full CPU mask */ + online_groups = GetActiveProcessorGroupCount(); + if (online_groups == 0) { + log_err("fio_getaffinity: error retrieving total processor groups\n"); + goto err; + } + + group = 0; + group_start_cpu = 0; + group_size = 0; + dprint(FD_PROCESS, "current_groups=%d group_count=%d\n", + current_groups[0], group_count); + while (true) { + group_size = GetMaximumProcessorCount(group); + if (group_size == 0) { + log_err("fio_getaffinity: error retrieving size of " + "processor group %d\n", group); + goto err; + } else if (group >= current_groups[0] || group >= online_groups) + break; + else { + group_start_cpu += group_size; + group++; + } + } + + if (group != current_groups[0]) { + log_err("fio_getaffinity: could not find processor group %d\n", + current_groups[0]); + goto err; + } + + dprint(FD_PROCESS, "group_start_cpu=%d, group size=%u\n", + group_start_cpu, group_size); + if ((group_start_cpu + group_size) >= FIO_MAX_CPUS) { + log_err("fio_getaffinity failed: current CPU affinity (group " + "%d, group_start_cpu %d, group_size %d) extends " + "beyond mask's highest CPU (%d)\n", group, + group_start_cpu, group_size, FIO_MAX_CPUS); + goto err; + } + + fio_cpuset_init(mask); + cpu_to_row_offset(group_start_cpu, &row, &offset); + mask->row[row] = process_mask; + mask->row[row] <<= offset; + end = offset + group_size; + if (end > FIO_CPU_MASK_STRIDE) { + int needed; + uint64_t needed_mask; + + needed = FIO_CPU_MASK_STRIDE - end; + needed_mask = (uint64_t)-1 >> (FIO_CPU_MASK_STRIDE - needed); + row++; + mask->row[row] = process_mask; + mask->row[row] >>= needed; + mask->row[row] &= needed_mask; + } + ret = 0; + +err: + if (handle) + CloseHandle(handle); + if (current_groups) + free(current_groups); + + return ret; +} + +static inline void fio_cpu_clear(os_cpu_mask_t *mask, int cpu) +{ + int row, offset; + cpu_to_row_offset(cpu, &row, &offset); + + mask->row[row] &= ~(1ULL << offset); +} + +static inline void fio_cpu_set(os_cpu_mask_t *mask, int cpu) +{ + int row, offset; + cpu_to_row_offset(cpu, &row, &offset); + + mask->row[row] |= 1ULL << offset; +} + +static inline int fio_cpu_isset(os_cpu_mask_t *mask, int cpu) +{ + int row, offset; + cpu_to_row_offset(cpu, &row, &offset); + + return (mask->row[row] & (1ULL << offset)) != 0; +} + +static inline int fio_cpu_count(os_cpu_mask_t *mask) +{ + int count = 0; + + for (int i = 0; i < FIO_CPU_MASK_ROWS; i++) + count += hweight64(mask->row[i]); + + return count; +} + +static inline int fio_cpuset_exit(os_cpu_mask_t *mask) +{ + return 0; +} diff --git a/os/os-windows-xp.h b/os/os-windows-xp.h new file mode 100644 index 0000000..1ce9ab3 --- /dev/null +++ b/os/os-windows-xp.h @@ -0,0 +1,70 @@ +#define FIO_MAX_CPUS MAXIMUM_PROCESSORS + +typedef DWORD_PTR os_cpu_mask_t; + +static inline int fio_setaffinity(int pid, os_cpu_mask_t cpumask) +{ + HANDLE h; + BOOL bSuccess = FALSE; + + h = OpenThread(THREAD_QUERY_INFORMATION | THREAD_SET_INFORMATION, TRUE, pid); + if (h != NULL) { + bSuccess = SetThreadAffinityMask(h, cpumask); + if (!bSuccess) + log_err("fio_setaffinity failed: failed to set thread affinity (pid %d, mask %.16llx)\n", pid, cpumask); + + CloseHandle(h); + } else { + log_err("fio_setaffinity failed: failed to get handle for pid %d\n", pid); + } + + return (bSuccess)? 0 : -1; +} + +static inline int fio_getaffinity(int pid, os_cpu_mask_t *mask) +{ + os_cpu_mask_t systemMask; + + HANDLE h = OpenProcess(PROCESS_QUERY_INFORMATION, TRUE, pid); + + if (h != NULL) { + GetProcessAffinityMask(h, mask, &systemMask); + CloseHandle(h); + } else { + log_err("fio_getaffinity failed: failed to get handle for pid %d\n", pid); + return -1; + } + + return 0; +} + +static inline void fio_cpu_clear(os_cpu_mask_t *mask, int cpu) +{ + *mask &= ~(1ULL << cpu); +} + +static inline void fio_cpu_set(os_cpu_mask_t *mask, int cpu) +{ + *mask |= 1ULL << cpu; +} + +static inline int fio_cpu_isset(os_cpu_mask_t *mask, int cpu) +{ + return (*mask & (1ULL << cpu)) != 0; +} + +static inline int fio_cpu_count(os_cpu_mask_t *mask) +{ + return hweight64(*mask); +} + +static inline int fio_cpuset_init(os_cpu_mask_t *mask) +{ + *mask = 0; + return 0; +} + +static inline int fio_cpuset_exit(os_cpu_mask_t *mask) +{ + return 0; +} diff --git a/os/os-windows.h b/os/os-windows.h index 9b04579..01f555e 100644 --- a/os/os-windows.h +++ b/os/os-windows.h @@ -13,6 +13,7 @@ #include #include "../smalloc.h" +#include "../debug.h" #include "../file.h" #include "../log.h" #include "../lib/hweight.h" @@ -21,7 +22,7 @@ #include "windows/posix.h" -/* Cygwin doesn't define rand_r if C99 or newer is being used */ +/* MinGW won't declare rand_r unless _POSIX is defined */ #if defined(WIN32) && !defined(rand_r) int rand_r(unsigned *); #endif @@ -40,16 +41,12 @@ int rand_r(unsigned *); #define FIO_PREFERRED_CLOCK_SOURCE CS_CGETTIME #define FIO_OS_PATH_SEPARATOR '\\' -#define FIO_MAX_CPUS MAXIMUM_PROCESSORS - #define OS_MAP_ANON MAP_ANON #define fio_swap16(x) _byteswap_ushort(x) #define fio_swap32(x) _byteswap_ulong(x) #define fio_swap64(x) _byteswap_uint64(x) -typedef DWORD_PTR os_cpu_mask_t; - #define _SC_PAGESIZE 0x1 #define _SC_NPROCESSORS_ONLN 0x2 #define _SC_PHYS_PAGES 0x4 @@ -77,11 +74,6 @@ typedef DWORD_PTR os_cpu_mask_t; /* Winsock doesn't support MSG_WAIT */ #define OS_MSG_DONTWAIT 0 -#define POLLOUT 1 -#define POLLIN 2 -#define POLLERR 0 -#define POLLHUP 1 - #define SIGCONT 0 #define SIGUSR1 1 #define SIGUSR2 2 @@ -172,73 +164,6 @@ static inline int gettid(void) return GetCurrentThreadId(); } -static inline int fio_setaffinity(int pid, os_cpu_mask_t cpumask) -{ - HANDLE h; - BOOL bSuccess = FALSE; - - h = OpenThread(THREAD_QUERY_INFORMATION | THREAD_SET_INFORMATION, TRUE, pid); - if (h != NULL) { - bSuccess = SetThreadAffinityMask(h, cpumask); - if (!bSuccess) - log_err("fio_setaffinity failed: failed to set thread affinity (pid %d, mask %.16llx)\n", pid, cpumask); - - CloseHandle(h); - } else { - log_err("fio_setaffinity failed: failed to get handle for pid %d\n", pid); - } - - return (bSuccess)? 0 : -1; -} - -static inline int fio_getaffinity(int pid, os_cpu_mask_t *mask) -{ - os_cpu_mask_t systemMask; - - HANDLE h = OpenProcess(PROCESS_QUERY_INFORMATION, TRUE, pid); - - if (h != NULL) { - GetProcessAffinityMask(h, mask, &systemMask); - CloseHandle(h); - } else { - log_err("fio_getaffinity failed: failed to get handle for pid %d\n", pid); - return -1; - } - - return 0; -} - -static inline void fio_cpu_clear(os_cpu_mask_t *mask, int cpu) -{ - *mask &= ~(1ULL << cpu); -} - -static inline void fio_cpu_set(os_cpu_mask_t *mask, int cpu) -{ - *mask |= 1ULL << cpu; -} - -static inline int fio_cpu_isset(os_cpu_mask_t *mask, int cpu) -{ - return (*mask & (1ULL << cpu)) != 0; -} - -static inline int fio_cpu_count(os_cpu_mask_t *mask) -{ - return hweight64(*mask); -} - -static inline int fio_cpuset_init(os_cpu_mask_t *mask) -{ - *mask = 0; - return 0; -} - -static inline int fio_cpuset_exit(os_cpu_mask_t *mask) -{ - return 0; -} - static inline int init_random_seeds(unsigned long *rand_seeds, int size) { HCRYPTPROV hCryptProv; @@ -261,12 +186,16 @@ static inline int init_random_seeds(unsigned long *rand_seeds, int size) return 0; } - static inline int fio_set_sched_idle(void) { /* SetThreadPriority returns nonzero for success */ return (SetThreadPriority(GetCurrentThread(), THREAD_PRIORITY_IDLE))? 0 : -1; } +#ifdef CONFIG_WINDOWS_XP +#include "os-windows-xp.h" +#else +#include "os-windows-7.h" +#endif #endif /* FIO_OS_WINDOWS_H */ diff --git a/os/windows/eula.rtf b/os/windows/eula.rtf index b2798bb..01472be 100755 Binary files a/os/windows/eula.rtf and b/os/windows/eula.rtf differ diff --git a/os/windows/posix.c b/os/windows/posix.c index ecc8c40..d33250d 100755 --- a/os/windows/posix.c +++ b/os/windows/posix.c @@ -959,6 +959,7 @@ in_addr_t inet_network(const char *cp) return hbo; } +#ifdef CONFIG_WINDOWS_XP const char* inet_ntop(int af, const void *restrict src, char *restrict dst, socklen_t size) { @@ -1039,3 +1040,4 @@ int inet_pton(int af, const char *restrict src, void *restrict dst) return ret; } +#endif /* CONFIG_WINDOWS_XP */ diff --git a/os/windows/posix/include/arpa/inet.h b/os/windows/posix/include/arpa/inet.h index 30498c6..056f1dd 100644 --- a/os/windows/posix/include/arpa/inet.h +++ b/os/windows/posix/include/arpa/inet.h @@ -12,8 +12,10 @@ typedef int in_addr_t; in_addr_t inet_network(const char *cp); +#ifdef CONFIG_WINDOWS_XP const char *inet_ntop(int af, const void *restrict src, char *restrict dst, socklen_t size); int inet_pton(int af, const char *restrict src, void *restrict dst); +#endif #endif /* ARPA_INET_H */ diff --git a/os/windows/posix/include/poll.h b/os/windows/posix/include/poll.h index f064e2b..25b8183 100644 --- a/os/windows/posix/include/poll.h +++ b/os/windows/posix/include/poll.h @@ -1,8 +1,11 @@ #ifndef POLL_H #define POLL_H +#include + typedef int nfds_t; +#ifdef CONFIG_WINDOWS_XP struct pollfd { int fd; @@ -10,6 +13,12 @@ struct pollfd short revents; }; +#define POLLOUT 1 +#define POLLIN 2 +#define POLLERR 0 +#define POLLHUP 1 +#endif /* CONFIG_WINDOWS_XP */ + int poll(struct pollfd fds[], nfds_t nfds, int timeout); #endif /* POLL_H */ diff --git a/server.c b/server.c index 2e08c66..12c8d68 100644 --- a/server.c +++ b/server.c @@ -2144,14 +2144,14 @@ static int fio_init_server_ip(void) #endif if (use_ipv6) { - const void *src = &saddr_in6.sin6_addr; + void *src = &saddr_in6.sin6_addr; addr = (struct sockaddr *) &saddr_in6; socklen = sizeof(saddr_in6); saddr_in6.sin6_family = AF_INET6; str = inet_ntop(AF_INET6, src, buf, sizeof(buf)); } else { - const void *src = &saddr_in.sin_addr; + void *src = &saddr_in.sin_addr; addr = (struct sockaddr *) &saddr_in; socklen = sizeof(saddr_in); @@ -2219,7 +2219,7 @@ static int fio_init_server_connection(void) if (!bind_sock) { char *p, port[16]; - const void *src; + void *src; int af; if (use_ipv6) {