Change Details

This is another page allocation routine. It returns runs of pages contiguous in the pindex space, with the intention of reducing the frequency of lock acquisitions. When VM_ALLOC_{NOWAIT,WAITFAIL} are specified, the returned run may be shorter than the one requested. The new vm_phys_alloc_npages() is used to allocate pages from the physical memory allocator when allocation from per-CPU caches or the reservation system fail. Some existing subroutines are changed. vm_reserv_reclaim_inactive() now returns the number of pages freed by the break. vm_domain_available() now returns the number of pages that may be allocated according to the request type. vm_reserv_extend() and vm_reserv_alloc_page() now optionally return a run of pages contiguous both in the pindex and the physical address space. vm_page_grab_pages() now uses vm_page_alloc_pages_after(), as do a couple of routines which allocate readahead and readbehind pages before performing a read from disk. Some possible optimizations: - Batched insertion of the run into the object radix tree.

(I created this review mainly to facilitate discussion, I'm not formally proposing it for inclusion yet. It's based on Jeff's numa branch, not head/.) This is another page allocation routine. It returns runs of pages contiguous in the pindex space, with the intention of reducing the frequency of lock acquisitions. When VM_ALLOC_{NOWAIT,WAITFAIL} are specified, the returned run may be shorter than the one requested. The new vm_phys_alloc_npages() is used to allocate pages from the physical memory allocator when allocation from per-CPU caches or the reservation system fail. Some existing subroutines are changed. vm_reserv_reclaim_inactive() now returns the number of pages freed by the break. vm_domain_available() now returns the number of pages that may be allocated according to the request type. vm_reserv_extend() and vm_reserv_alloc_page() now optionally return a run of pages contiguous both in the pindex and the physical address space. vm_page_grab_pages() now uses vm_page_alloc_pages_after(), as do a couple of routines which allocate readahead and readbehind pages before performing a read from disk. Some possible optimizations: - Batched insertion of the run into the object radix tree.

(I created this review mainly to facilitate discussion, I'm not formally proposing it for inclusion yet. It's based on Jeff's numa branch, not head/.) This is another page allocation routine. It returns runs of pages contiguous in the pindex space, with the intention of reducing the frequency of lock acquisitions. When VM_ALLOC_{NOWAIT,WAITFAIL} are specified, the returned run may be shorter than the one requested. The new vm_phys_alloc_npages() is used to allocate pages from the physical memory allocator when allocation from per-CPU caches or the reservation system fail. Some existing subroutines are changed. vm_reserv_reclaim_inactive() now returns the number of pages freed by the break. vm_domain_available() now returns the number of pages that may be allocated according to the request type. vm_reserv_extend() and vm_reserv_alloc_page() now optionally return a run of pages contiguous both in the pindex and the physical address space. vm_page_grab_pages() now uses vm_page_alloc_pages_after(), as do a couple of routines which allocate readahead and readbehind pages before performing a read from disk. Some possible optimizations: - Batched insertion of the run into the object radix tree.