mm/pmem: avoid inserting hugepage PTE entry with fsdax if hugepage su…

…pport is disabled

[ Upstream commit bae8495 ]

Differentiate between hardware not supporting hugepages and user disabling
THP via 'echo never > /sys/kernel/mm/transparent_hugepage/enabled'

For the devdax namespace, the kernel handles the above via the
supported_alignment attribute and failing to initialize the namespace if
the namespace align value is not supported on the platform.

For the fsdax namespace, the kernel will continue to initialize the
namespace.  This can result in the kernel creating a huge pte entry even
though the hardware don't support the same.

We do want hugepage support with pmem even if the end-user disabled THP
via sysfs file (/sys/kernel/mm/transparent_hugepage/enabled).  Hence
differentiate between hardware/firmware lacking support vs user-controlled
disable of THP and prevent a huge fault if the hardware lacks hugepage
support.

Link: https://lkml.kernel.org/r/20210205023956.417587-1-aneesh.kumar@linux.ibm.com
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
Reviewed-by: Dan Williams <dan.j.williams@intel.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Jan Kara <jack@suse.cz>
Cc: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>

include/linux/huge_mm.h

@@ -84,6 +84,7 @@ static inline vm_fault_t vmf_insert_pfn_pud(struct vm_fault *vmf, pfn_t pfn,
 }
 
 enum transparent_hugepage_flag {
+	TRANSPARENT_HUGEPAGE_NEVER_DAX,
 	TRANSPARENT_HUGEPAGE_FLAG,
 	TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG,
 	TRANSPARENT_HUGEPAGE_DEFRAG_DIRECT_FLAG,
@@ -129,6 +130,13 @@ extern unsigned long transparent_hugepage_flags;
  */
 static inline bool __transparent_hugepage_enabled(struct vm_area_struct *vma)
 {
+
+	/*
+	 * If the hardware/firmware marked hugepage support disabled.
+	 */
+	if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_NEVER_DAX))
+		return false;
+
 	if (vma->vm_flags & VM_NOHUGEPAGE)
 		return false;
 
@@ -140,12 +148,7 @@ static inline bool __transparent_hugepage_enabled(struct vm_area_struct *vma)
 
 	if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_FLAG))
 		return true;
-	/*
-	 * For dax vmas, try to always use hugepage mappings. If the kernel does
-	 * not support hugepages, fsdax mappings will fallback to PAGE_SIZE
-	 * mappings, and device-dax namespaces, that try to guarantee a given
-	 * mapping size, will fail to enable
-	 */
+
 	if (vma_is_dax(vma))
 		return true;
 

mm/huge_memory.c

@@ -375,7 +375,11 @@ static int __init hugepage_init(void)
 	struct kobject *hugepage_kobj;
 
 	if (!has_transparent_hugepage()) {
-		transparent_hugepage_flags = 0;
+		/*
+		 * Hardware doesn't support hugepages, hence disable
+		 * DAX PMD support.
+		 */
+		transparent_hugepage_flags = 1 << TRANSPARENT_HUGEPAGE_NEVER_DAX;
 		return -EINVAL;
 	}