Date: Mon, 25 Feb 2019 19:22:13 +0000 (UTC) From: Mark Johnston <markj@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r344550 - in head/sys: kern vm Message-ID: <201902251922.x1PJMD0R051425@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: markj Date: Mon Feb 25 19:22:13 2019 New Revision: 344550 URL: https://svnweb.freebsd.org/changeset/base/344550 Log: Improve vmem tuning for platforms without a direct map. On platforms without a direct map (i.e., platforms without UMA_MD_SMALL_ALLOC defined), the boundary tag allocator reserves a number of tags for use when allocating a new slab of boundary tags, as such platforms require free boundary tags in order to allocate boundary tags. r327899 increased the number of boundary tags required for a KVA allocation in the worst case, and the aforementioned reservation was not updated accordingly. In some cases, this could lead to a system hang. Fix the problem by increasing this reservation. Also reduce KVA_QUANTUM on systems lacking superpage support. The previous import quantum (4MB with a 4KB page size) was quite large for systems with limited KVA, and fragmentation in kernel_arena could cause kernel memory allocation failures even with a substantial amount of free KVA. Reported and tested by: jhibbits Reviewed by: alc, kib No objections: jeff MFC after: 2 weeks Sponsored by: The FreeBSD Foundation Differential Revision: https://reviews.freebsd.org/D19337 Modified: head/sys/kern/subr_vmem.c head/sys/vm/vm_kern.c Modified: head/sys/kern/subr_vmem.c ============================================================================== --- head/sys/kern/subr_vmem.c Mon Feb 25 19:18:32 2019 (r344549) +++ head/sys/kern/subr_vmem.c Mon Feb 25 19:22:13 2019 (r344550) @@ -689,9 +689,11 @@ vmem_startup(void) /* * Reserve enough tags to allocate new tags. We allow multiple * CPUs to attempt to allocate new tags concurrently to limit - * false restarts in UMA. + * false restarts in UMA. vmem_bt_alloc() allocates from a per-domain + * arena, which may involve importing a range from the kernel arena, + * so we need to keep at least 2 * BT_MAXALLOC tags reserved. */ - uma_zone_reserve(vmem_bt_zone, BT_MAXALLOC * (mp_ncpus + 1) / 2); + uma_zone_reserve(vmem_bt_zone, 2 * BT_MAXALLOC * mp_ncpus); uma_zone_set_allocf(vmem_bt_zone, vmem_bt_alloc); #endif } Modified: head/sys/vm/vm_kern.c ============================================================================== --- head/sys/vm/vm_kern.c Mon Feb 25 19:18:32 2019 (r344549) +++ head/sys/vm/vm_kern.c Mon Feb 25 19:22:13 2019 (r344550) @@ -124,8 +124,8 @@ SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLA #if VM_NRESERVLEVEL > 0 #define KVA_QUANTUM_SHIFT (VM_LEVEL_0_ORDER + PAGE_SHIFT) #else -/* On non-superpage architectures want large import sizes. */ -#define KVA_QUANTUM_SHIFT (10 + PAGE_SHIFT) +/* On non-superpage architectures we want large import sizes. */ +#define KVA_QUANTUM_SHIFT (8 + PAGE_SHIFT) #endif #define KVA_QUANTUM (1 << KVA_QUANTUM_SHIFT)
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201902251922.x1PJMD0R051425>