Date: Mon, 20 Oct 2025 16:47:19 GMT From: Mark Johnston <markj@FreeBSD.org> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-branches@FreeBSD.org Subject: git: 511311e4f7a6 - stable/15 - vm: Fix iterator usage in vm_thread_stack_create() Message-ID: <202510201647.59KGlJEF099702@gitrepo.freebsd.org>
index | next in thread | raw e-mail
The branch stable/15 has been updated by markj: URL: https://cgit.FreeBSD.org/src/commit/?id=511311e4f7a6220234bc5bced1e8e0fae96d02a6 commit 511311e4f7a6220234bc5bced1e8e0fae96d02a6 Author: Mark Johnston <markj@FreeBSD.org> AuthorDate: 2025-10-16 23:12:54 +0000 Commit: Mark Johnston <markj@FreeBSD.org> CommitDate: 2025-10-20 16:46:02 +0000 vm: Fix iterator usage in vm_thread_stack_create() After commit 7a79d0669761 we no longer hold the VM object lock when initializing or advancing the domainset iterator in vm_thread_stack_create(). We could lift the lock out of vm_thread_stack_back() but that poses complications since we cannot allocate KVA with a VM object lock held. Instead of overridding the object's iterator, just borrow that of the current thread. Kernel stacks are basically always allocated with a DOMAINSET_POLICY_PREFER policy, so it's not very important to maintain a global iterator for round-robin allocation. As a part of this, fix up flag handling: make sure we handle M_NOWAIT/M_WAITOK from the caller. Delete a comment in vm_thread_new() which refers to a non-existent consideration (stack swapping has been removed). I suspect vm_thread_new() can use M_WAITOK but opted not to make that change here. Reported by: olce Reviewed by: olce, alc, kib Fixes: 7a79d0669761 ("vm: improve kstack_object pindex calculation to avoid pindex holes") MFC after: 1 week Differential Revision: https://reviews.freebsd.org/D52982 (cherry picked from commit 3b9b64457676561b8de2bec7c94c561bbd807b0c) --- sys/vm/vm_glue.c | 24 ++++++++---------------- 1 file changed, 8 insertions(+), 16 deletions(-) diff --git a/sys/vm/vm_glue.c b/sys/vm/vm_glue.c index e0f1807a1b32..18d789c59281 100644 --- a/sys/vm/vm_glue.c +++ b/sys/vm/vm_glue.c @@ -441,19 +441,16 @@ vm_thread_kstack_arena_release(void *arena, vmem_addr_t addr, vmem_size_t size) * Create the kernel stack for a new thread. */ static vm_offset_t -vm_thread_stack_create(struct domainset *ds, int pages) +vm_thread_stack_create(struct domainset *ds, int pages, int flags) { vm_page_t ma[KSTACK_MAX_PAGES]; struct vm_domainset_iter di; - int req = VM_ALLOC_NORMAL; - vm_object_t obj; + int req; vm_offset_t ks; int domain, i; - obj = vm_thread_kstack_size_to_obj(pages); - if (vm_ndomains > 1) - obj->domain.dr_policy = ds; - vm_domainset_iter_page_init(&di, obj, 0, &domain, &req); + vm_domainset_iter_policy_init(&di, ds, &domain, &flags); + req = malloc2vm_flags(flags); do { /* * Get a kernel virtual address for this thread's kstack. @@ -480,7 +477,7 @@ vm_thread_stack_create(struct domainset *ds, int pages) vm_page_valid(ma[i]); pmap_qenter(ks, ma, pages); return (ks); - } while (vm_domainset_iter_page(&di, obj, &domain, NULL) == 0); + } while (vm_domainset_iter_policy(&di, &domain) == 0); return (0); } @@ -532,15 +529,9 @@ vm_thread_new(struct thread *td, int pages) ks = 0; if (pages == kstack_pages && kstack_cache != NULL) ks = (vm_offset_t)uma_zalloc(kstack_cache, M_NOWAIT); - - /* - * Ensure that kstack objects can draw pages from any memory - * domain. Otherwise a local memory shortage can block a process - * swap-in. - */ if (ks == 0) ks = vm_thread_stack_create(DOMAINSET_PREF(PCPU_GET(domain)), - pages); + pages, M_NOWAIT); if (ks == 0) return (0); @@ -660,7 +651,8 @@ kstack_import(void *arg, void **store, int cnt, int domain, int flags) ds = DOMAINSET_PREF(domain); for (i = 0; i < cnt; i++) { - store[i] = (void *)vm_thread_stack_create(ds, kstack_pages); + store[i] = (void *)vm_thread_stack_create(ds, kstack_pages, + flags); if (store[i] == NULL) break; }home | help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?202510201647.59KGlJEF099702>