[PATCH v2 0/1] MR5213: Fix Cyberpunk 2077 spawning only one dispatcher thread in FreeBSD
Normally Cyberpunk 2077 spawns #cpu_threads of dispatcher threads. However with Freebsd, since create_logical_proc_info was a stub, the game would instead spawn a single dispatcher thread, causing the game to just crawl. Working around the issue by adding a barebone implementation of create_logical_proc_info for FreeBSD so that the game would at least get a thread count through ntdll. -- v2: Fix Cyberpunk 2077 spawning only one dispatcher thread in FreeBSD https://gitlab.winehq.org/wine/wine/-/merge_requests/5213
From: Katharine Chui <kwchuiaa(a)connect.ust.hk> Normally Cyberpunk 2077 spawns #cpu_threads of dispatcher threads. However with Freebsd, since create_logical_proc_info was a stub, the game would instead spawn a single dispatcher thread, causing the game to just crawl. Working around the issue by adding a barebone implementation of create_logical_proc_info for FreeBSD so that the game would at least get a thread count through ntdll. --- dlls/ntdll/unix/system.c | 92 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 92 insertions(+) diff --git a/dlls/ntdll/unix/system.c b/dlls/ntdll/unix/system.c index e9ed8b6f3a4..72dc12ba5a5 100644 --- a/dlls/ntdll/unix/system.c +++ b/dlls/ntdll/unix/system.c @@ -1329,6 +1329,98 @@ static NTSTATUS create_logical_proc_info(void) return STATUS_SUCCESS; } +#elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__) + +/* + * this is currently mostly a reduced clone of the macos implementation + * it only reports thread counts so that cb2077 don't spawn a single dispatch thread only, and would spawn #threads instead + * TODO list: + * smt information + * numa information + * cache information + * package count + * it is also tempting to rely on lscpu output, however that would make freebsd wine packaging dependent on lscpu, don't know if that's a good idea + * or perhaps go for cpuid directly? however that'd would introduce quite the amount of boiler plates + * yet another alternative would be to rely on /compat/linux/sys, however FreeBSD only lists the number of online cores without smt, numa, cache nor package information, and linux compat has to be enabled + */ +static NTSTATUS create_logical_proc_info(void) +{ + unsigned int pkgs_no, cores_no, lcpu_no, lcpu_per_core, cores_per_package; + ULONG_PTR all_cpus_mask = 0; + unsigned int i, p, j, k; + CACHE_DESCRIPTOR cache[10]; + + /* HW_NCPU works in FreeBSD */ + lcpu_no = peb->NumberOfProcessors; + + /* TODO assume one package */ + pkgs_no = 1; + + /* TODO assume no SMT */ + /* physical core number can be fetch with sysctl kern.smp.cores */ + cores_no = lcpu_no; + + TRACE("%u logical CPUs from %u physical cores across %u packages\n", + lcpu_no, cores_no, pkgs_no); + + lcpu_per_core = lcpu_no / cores_no; + cores_per_package = cores_no / pkgs_no; + + for(p = 0; p < pkgs_no; ++p) + { + for(j = 0; j < cores_per_package && p * cores_per_package + j < cores_no; ++j) + { + ULONG_PTR mask = 0; + DWORD phys_core; + + for(k = 0; k < lcpu_per_core; ++k) mask |= (ULONG_PTR)1 << (j * lcpu_per_core + k); + + all_cpus_mask |= mask; + + /* add to package */ + if(!logical_proc_info_add_by_id( RelationProcessorPackage, p, mask )) + return STATUS_NO_MEMORY; + + /* add new core */ + phys_core = p * cores_per_package + j; + if(!logical_proc_info_add_by_id( RelationProcessorCore, phys_core, mask )) + return STATUS_NO_MEMORY; + } + } + + /* TODO cache information */ + memset(cache, 0, sizeof(cache)); + cache[1].Level = 1; + cache[1].Type = CacheInstruction; + cache[1].Associativity = 8; /* reasonable default */ + cache[1].LineSize = 0x40; /* reasonable default */ + cache[2].Level = 1; + cache[2].Type = CacheData; + cache[2].Associativity = 8; + cache[2].LineSize = 0x40; + cache[3].Level = 2; + cache[3].Type = CacheUnified; + cache[3].Associativity = 8; + cache[3].LineSize = 0x40; + cache[4].Level = 3; + cache[4].Type = CacheUnified; + cache[4].Associativity = 12; + cache[4].LineSize = 0x40; + for(i = 0; i < 5; i++) + { + if(!logical_proc_info_add_cache( all_cpus_mask, &cache[i] )) + return STATUS_NO_MEMORY; + } + + /* TODO numa infomration */ + if(!logical_proc_info_add_numa_node( all_cpus_mask, 0 )) + return STATUS_NO_MEMORY; + + logical_proc_info_add_group( lcpu_no, all_cpus_mask ); + + return STATUS_SUCCESS; +} + #else static NTSTATUS create_logical_proc_info(void) -- GitLab https://gitlab.winehq.org/wine/wine/-/merge_requests/5213
I think the correct fix for this would be to parse the CPU topology info out of the [`kern.sched.topology_spec`](https://man.freebsd.org/cgi/man.cgi?query=smp&sektion=4&manpath=FreeBSD+14.2...) sysctl, unfortunately it’s XML. Using the built-in [bsdxml](https://man.freebsd.org/cgi/man.cgi?query=libbsdxml&apropos=0&sektion=0&manp...) library (a renamed expat) would likely be the easiest way to do that without adding an XML library or dependency on an external one. I’ve been meaning to give this a try, but probably wont get to it until next year. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/5213#note_89855
On Sat Dec 7 04:25:25 2024 +0000, Brendan Shanks wrote:
I think the correct fix for this would be to parse the CPU topology info out of the [`kern.sched.topology_spec`](https://man.freebsd.org/cgi/man.cgi?query=smp&sektion=4&manpath=FreeBSD+14.2...) sysctl, unfortunately it’s XML. Using the built-in [bsdxml](https://man.freebsd.org/cgi/man.cgi?query=libbsdxml&apropos=0&sektion=0&manp...) library (a renamed expat) would likely be the easiest way to do that without adding an XML library or dependency on an external one. I’ve been meaning to give this a try, but probably wont get to it until next year. Wine is also vendoring the libxml2 library (so that could be used too if compiled as an Unix library)
-- https://gitlab.winehq.org/wine/wine/-/merge_requests/5213#note_90089
!7339 parses `kern.sched.topology_spec`. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/5213#note_94827
This merge request was closed by Brendan Shanks. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/5213
This is obsoleted by !7339 -- https://gitlab.winehq.org/wine/wine/-/merge_requests/5213#note_118027
participants (4)
-
Aida Jonikienė -
Brendan Shanks (＠bshanks)
-
Katharine Chui
-
Katharine Chui (＠katharinechui)