diff --git a/module/spl/spl-kmem.c b/module/spl/spl-kmem.c index 16eb4f884..e07e08c91 100644 --- a/module/spl/spl-kmem.c +++ b/module/spl/spl-kmem.c @@ -840,31 +840,11 @@ kv_alloc(spl_kmem_cache_t *skc, int size, int flags) ASSERT(ISP2(size)); - if (skc->skc_flags & KMC_KMEM) { + if (skc->skc_flags & KMC_KMEM) ptr = (void *)__get_free_pages(flags, get_order(size)); - } else { - /* - * As part of vmalloc() an __pte_alloc_kernel() allocation - * may occur. This internal allocation does not honor the - * gfp flags passed to vmalloc(). This means even when - * vmalloc(GFP_NOFS) is called it is possible synchronous - * reclaim will occur. This reclaim can trigger file IO - * which can result in a deadlock. This issue can be avoided - * by explicitly setting PF_MEMALLOC on the process to - * subvert synchronous reclaim. The following bug has - * been filed at kernel.org to track the issue. - * - * https://bugzilla.kernel.org/show_bug.cgi?id=30702 - */ - if (!(flags & __GFP_FS)) - current->flags |= PF_MEMALLOC; - + else ptr = __vmalloc(size, flags | __GFP_HIGHMEM, PAGE_KERNEL); - if (!(flags & __GFP_FS)) - current->flags &= ~PF_MEMALLOC; - } - /* Resulting allocated memory will be page aligned */ ASSERT(IS_P2ALIGNED(ptr, PAGE_SIZE));