ARM: mm: allow non-text sections to be non-executable

Adds CONFIG_ARM_KERNMEM_PERMS to separate the kernel memory regions
into section-sized areas that can have different permisions. Performs
the NX permission changes during free_initmem, so that init memory can be
reclaimed.

This uses section size instead of PMD size to reduce memory lost to
padding on non-LPAE systems.

Based on work by Brad Spengler, Larry Bassel, and Laura Abbott.

Signed-off-by: Kees Cook <keescook@chromium.org>
Tested-by: Laura Abbott <lauraa@codeaurora.org>
Acked-by: Nicolas Pitre <nico@linaro.org>
diff --git a/arch/arm/kernel/vmlinux.lds.S b/arch/arm/kernel/vmlinux.lds.S
index 6f57cb9..18fd68a 100644
--- a/arch/arm/kernel/vmlinux.lds.S
+++ b/arch/arm/kernel/vmlinux.lds.S
@@ -8,6 +8,9 @@
 #include <asm/thread_info.h>
 #include <asm/memory.h>
 #include <asm/page.h>
+#ifdef CONFIG_ARM_KERNMEM_PERMS
+#include <asm/pgtable.h>
+#endif
 	
 #define PROC_INFO							\
 	. = ALIGN(4);							\
@@ -90,6 +93,11 @@
 		_text = .;
 		HEAD_TEXT
 	}
+
+#ifdef CONFIG_ARM_KERNMEM_PERMS
+	. = ALIGN(1<<SECTION_SHIFT);
+#endif
+
 	.text : {			/* Real text segment		*/
 		_stext = .;		/* Text and read-only data	*/
 			__exception_text_start = .;
@@ -145,7 +153,11 @@
 	_etext = .;			/* End of text and rodata section */
 
 #ifndef CONFIG_XIP_KERNEL
+# ifdef CONFIG_ARM_KERNMEM_PERMS
+	. = ALIGN(1<<SECTION_SHIFT);
+# else
 	. = ALIGN(PAGE_SIZE);
+# endif
 	__init_begin = .;
 #endif
 	/*
@@ -220,7 +232,11 @@
 	. = PAGE_OFFSET + TEXT_OFFSET;
 #else
 	__init_end = .;
+#ifdef CONFIG_ARM_KERNMEM_PERMS
+	. = ALIGN(1<<SECTION_SHIFT);
+#else
 	. = ALIGN(THREAD_SIZE);
+#endif
 	__data_loc = .;
 #endif
 
diff --git a/arch/arm/mm/Kconfig b/arch/arm/mm/Kconfig
index ae69809..7a0756d 100644
--- a/arch/arm/mm/Kconfig
+++ b/arch/arm/mm/Kconfig
@@ -1008,3 +1008,12 @@
 	help
 	  This option specifies the architecture can support big endian
 	  operation.
+
+config ARM_KERNMEM_PERMS
+	bool "Restrict kernel memory permissions"
+	help
+	  If this is set, kernel memory other than kernel text (and rodata)
+	  will be made non-executable. The tradeoff is that each region is
+	  padded to section-size (1MiB) boundaries (because their permissions
+	  are different and splitting the 1M pages into 4K ones causes TLB
+	  performance problems), wasting memory.
diff --git a/arch/arm/mm/init.c b/arch/arm/mm/init.c
index ad82c05..e6bfe76 100644
--- a/arch/arm/mm/init.c
+++ b/arch/arm/mm/init.c
@@ -29,6 +29,7 @@
 #include <asm/prom.h>
 #include <asm/sections.h>
 #include <asm/setup.h>
+#include <asm/system_info.h>
 #include <asm/tlb.h>
 #include <asm/fixmap.h>
 
@@ -615,7 +616,99 @@
 	}
 }
 
-void free_initmem(void)
+#ifdef CONFIG_ARM_KERNMEM_PERMS
+struct section_perm {
+	unsigned long start;
+	unsigned long end;
+	pmdval_t mask;
+	pmdval_t prot;
+};
+
+struct section_perm nx_perms[] = {
+	/* Make pages tables, etc before _stext RW (set NX). */
+	{
+		.start	= PAGE_OFFSET,
+		.end	= (unsigned long)_stext,
+		.mask	= ~PMD_SECT_XN,
+		.prot	= PMD_SECT_XN,
+	},
+	/* Make init RW (set NX). */
+	{
+		.start	= (unsigned long)__init_begin,
+		.end	= (unsigned long)_sdata,
+		.mask	= ~PMD_SECT_XN,
+		.prot	= PMD_SECT_XN,
+	},
+};
+
+/*
+ * Updates section permissions only for the current mm (sections are
+ * copied into each mm). During startup, this is the init_mm. Is only
+ * safe to be called with preemption disabled, as under stop_machine().
+ */
+static inline void section_update(unsigned long addr, pmdval_t mask,
+				  pmdval_t prot)
+{
+	struct mm_struct *mm;
+	pmd_t *pmd;
+
+	mm = current->active_mm;
+	pmd = pmd_offset(pud_offset(pgd_offset(mm, addr), addr), addr);
+
+#ifdef CONFIG_ARM_LPAE
+	pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
+#else
+	if (addr & SECTION_SIZE)
+		pmd[1] = __pmd((pmd_val(pmd[1]) & mask) | prot);
+	else
+		pmd[0] = __pmd((pmd_val(pmd[0]) & mask) | prot);
+#endif
+	flush_pmd_entry(pmd);
+	local_flush_tlb_kernel_range(addr, addr + SECTION_SIZE);
+}
+
+/* Make sure extended page tables are in use. */
+static inline bool arch_has_strict_perms(void)
+{
+	if (cpu_architecture() < CPU_ARCH_ARMv6)
+		return false;
+
+	return !!(get_cr() & CR_XP);
+}
+
+#define set_section_perms(perms, field)	{				\
+	size_t i;							\
+	unsigned long addr;						\
+									\
+	if (!arch_has_strict_perms())					\
+		return;							\
+									\
+	for (i = 0; i < ARRAY_SIZE(perms); i++) {			\
+		if (!IS_ALIGNED(perms[i].start, SECTION_SIZE) ||	\
+		    !IS_ALIGNED(perms[i].end, SECTION_SIZE)) {		\
+			pr_err("BUG: section %lx-%lx not aligned to %lx\n", \
+				perms[i].start, perms[i].end,		\
+				SECTION_SIZE);				\
+			continue;					\
+		}							\
+									\
+		for (addr = perms[i].start;				\
+		     addr < perms[i].end;				\
+		     addr += SECTION_SIZE)				\
+			section_update(addr, perms[i].mask,		\
+				       perms[i].field);			\
+	}								\
+}
+
+static inline void fix_kernmem_perms(void)
+{
+	set_section_perms(nx_perms, prot);
+}
+#else
+static inline void fix_kernmem_perms(void) { }
+#endif /* CONFIG_ARM_KERNMEM_PERMS */
+
+void free_tcmmem(void)
 {
 #ifdef CONFIG_HAVE_TCM
 	extern char __tcm_start, __tcm_end;
@@ -623,6 +716,12 @@
 	poison_init_mem(&__tcm_start, &__tcm_end - &__tcm_start);
 	free_reserved_area(&__tcm_start, &__tcm_end, -1, "TCM link");
 #endif
+}
+
+void free_initmem(void)
+{
+	fix_kernmem_perms();
+	free_tcmmem();
 
 	poison_init_mem(__init_begin, __init_end - __init_begin);
 	if (!machine_is_integrator() && !machine_is_cintegrator())
diff --git a/arch/arm/mm/mmu.c b/arch/arm/mm/mmu.c
index bdf5c94..1c52c8e 100644
--- a/arch/arm/mm/mmu.c
+++ b/arch/arm/mm/mmu.c
@@ -1373,13 +1373,20 @@
 		if (start >= end)
 			break;
 
-		if (end < kernel_x_start || start >= kernel_x_end) {
+		if (end < kernel_x_start) {
 			map.pfn = __phys_to_pfn(start);
 			map.virtual = __phys_to_virt(start);
 			map.length = end - start;
 			map.type = MT_MEMORY_RWX;
 
 			create_mapping(&map);
+		} else if (start >= kernel_x_end) {
+			map.pfn = __phys_to_pfn(start);
+			map.virtual = __phys_to_virt(start);
+			map.length = end - start;
+			map.type = MT_MEMORY_RW;
+
+			create_mapping(&map);
 		} else {
 			/* This better cover the entire kernel */
 			if (start < kernel_x_start) {