On Fri, Mar 24, 2017 at 02:44:10PM +0530, Aneesh Kumar K.V wrote:

On Friday 24 March 2017 02:34 PM, Kirill A. Shutemov wrote:

quoted

On Mon, Mar 20, 2017 at 10:40:20AM +0530, Aneesh Kumar K.V wrote:

quoted

"Kirill A. Shutemov" [off-list ref] writes:
 @@ -168,6 +182,10 @@ arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,

quoted

 	unsigned long addr = addr0;
 	struct vm_unmapped_area_info info;

+	addr = mpx_unmapped_area_check(addr, len, flags);
+	if (IS_ERR_VALUE(addr))
+		return addr;
+
 	/* requested length too big for entire address space */
 	if (len > TASK_SIZE)
 		return -ENOMEM;

@@ -192,6 +210,14 @@ arch_get_unmapped_area_topdown(struct file *filp, const unsigned long addr0,
 	info.length = len;
 	info.low_limit = PAGE_SIZE;
 	info.high_limit = mm->mmap_base;
+
+	/*
+	 * If hint address is above DEFAULT_MAP_WINDOW, look for unmapped area
+	 * in the full address space.
+	 */
+	if (addr > DEFAULT_MAP_WINDOW)
+		info.high_limit += TASK_SIZE - DEFAULT_MAP_WINDOW;
+

Is this ok for 32 bit application ?

DEFAULT_MAP_WINDOW is equal to TASK_SIZE on 32-bit, so it's nop and will
be compile out.

That is not about CONFIG_X86_32 but about 32 bit application on a 64 bit
kernel. I guess we will never find addr > DEFAULT_MAP_WINDOW with
a 32 bit app ?

I have local change to avoid this within 32-bit syscall, but I'll need to
recheck everthing.

-- 
 Kirill A. Shutemov