s390/vmalloc: have separate modules area

Add a special module area on top of the vmalloc area, which may be only
used for modules and bpf jit generated code.
This makes sure that inter module branches will always happen without a
trampoline and in addition having all the code within a 2GB frame is
branch prediction unit friendly.

Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/pgtable.h

@@ -119,13 +119,12 @@ static inline int is_zero_pfn(unsigned long pfn)
 
 #ifndef __ASSEMBLY__
 /*
- * The vmalloc area will always be on the topmost area of the kernel
- * mapping. We reserve 96MB (31bit) / 128GB (64bit) for vmalloc,
- * which should be enough for any sane case.
- * By putting vmalloc at the top, we maximise the gap between physical
- * memory and vmalloc to catch misplaced memory accesses. As a side
- * effect, this also makes sure that 64 bit module code cannot be used
- * as system call address.
+ * The vmalloc and module area will always be on the topmost area of the kernel
+ * mapping. We reserve 96MB (31bit) / 128GB (64bit) for vmalloc and modules.
+ * On 64 bit kernels we have a 2GB area at the top of the vmalloc area where
+ * modules will reside. That makes sure that inter module branches always
+ * happen without trampolines and in addition the placement within a 2GB frame
+ * is branch prediction unit friendly.
  */
 extern unsigned long VMALLOC_START;
 extern unsigned long VMALLOC_END;
@@ -133,6 +132,14 @@ extern struct page *vmemmap;
 
 #define VMEM_MAX_PHYS ((unsigned long) vmemmap)
 
+#ifdef CONFIG_64BIT
+extern unsigned long MODULES_VADDR;
+extern unsigned long MODULES_END;
+#define MODULES_VADDR	MODULES_VADDR
+#define MODULES_END	MODULES_END
+#define MODULES_LEN	(1UL << 31)
+#endif
+
 /*
  * A 31 bit pagetable entry of S390 has following format:
  *  |   PFRA          |    |  OS  |

arch/s390/kernel/module.c

@@ -44,6 +44,17 @@
 #define PLT_ENTRY_SIZE 20
 #endif /* CONFIG_64BIT */
 
+#ifdef CONFIG_64BIT
+void *module_alloc(unsigned long size)
+{
+	if (PAGE_ALIGN(size) > MODULES_LEN)
+		return NULL;
+	return __vmalloc_node_range(size, 1, MODULES_VADDR, MODULES_END,
+				    GFP_KERNEL, PAGE_KERNEL, -1,
+				    __builtin_return_address(0));
+}
+#endif
+
 /* Free memory returned from module_alloc */
 void module_free(struct module *mod, void *module_region)
 {

arch/s390/kernel/setup.c

@@ -105,6 +105,11 @@ EXPORT_SYMBOL(VMALLOC_END);
 struct page *vmemmap;
 EXPORT_SYMBOL(vmemmap);
 
+#ifdef CONFIG_64BIT
+unsigned long MODULES_VADDR;
+unsigned long MODULES_END;
+#endif
+
 /* An array with a pointer to the lowcore of every CPU. */
 struct _lowcore *lowcore_ptr[NR_CPUS];
 EXPORT_SYMBOL(lowcore_ptr);
@@ -544,19 +549,23 @@ static void __init setup_memory_end(void)
 
 	/* Choose kernel address space layout: 2, 3, or 4 levels. */
 #ifdef CONFIG_64BIT
-	vmalloc_size = VMALLOC_END ?: 128UL << 30;
+	vmalloc_size = VMALLOC_END ?: (128UL << 30) - MODULES_LEN;
 	tmp = (memory_end ?: real_memory_size) / PAGE_SIZE;
 	tmp = tmp * (sizeof(struct page) + PAGE_SIZE) + vmalloc_size;
 	if (tmp <= (1UL << 42))
 		vmax = 1UL << 42;	/* 3-level kernel page table */
 	else
 		vmax = 1UL << 53;	/* 4-level kernel page table */
+	/* module area is at the end of the kernel address space. */
+	MODULES_END = vmax;
+	MODULES_VADDR = MODULES_END - MODULES_LEN;
+	VMALLOC_END = MODULES_VADDR;
 #else
 	vmalloc_size = VMALLOC_END ?: 96UL << 20;
 	vmax = 1UL << 31;		/* 2-level kernel page table */
-#endif
 	/* vmalloc area is at the end of the kernel address space. */
 	VMALLOC_END = vmax;
+#endif
 	VMALLOC_START = vmax - vmalloc_size;
 
 	/* Split remaining virtual space between 1:1 mapping & vmemmap array */

arch/s390/mm/dump_pagetables.c

@@ -18,6 +18,9 @@ enum address_markers_idx {
 	KERNEL_END_NR,
 	VMEMMAP_NR,
 	VMALLOC_NR,
+#ifdef CONFIG_64BIT
+	MODULES_NR,
+#endif
 };
 
 static struct addr_marker address_markers[] = {
@@ -26,6 +29,9 @@ static struct addr_marker address_markers[] = {
 	[KERNEL_END_NR]	  = {(unsigned long)&_end, "Kernel Image End"},
 	[VMEMMAP_NR]	  = {0, "vmemmap Area"},
 	[VMALLOC_NR]	  = {0, "vmalloc Area"},
+#ifdef CONFIG_64BIT
+	[MODULES_NR]	  = {0, "Modules Area"},
+#endif
 	{ -1, NULL }
 };
 
@@ -205,11 +211,12 @@ static int pt_dump_init(void)
 	 * kernel ASCE. We need this to keep the page table walker functions
 	 * from accessing non-existent entries.
 	 */
-#ifdef CONFIG_64BIT
+#ifdef CONFIG_32BIT
+	max_addr = 1UL << 31;
+#else
 	max_addr = (S390_lowcore.kernel_asce & _REGION_ENTRY_TYPE_MASK) >> 2;
 	max_addr = 1UL << (max_addr * 11 + 31);
-#else
-	max_addr = 1UL << 31;
+	address_markers[MODULES_NR].start_address = MODULES_VADDR;
 #endif
 	address_markers[VMEMMAP_NR].start_address = (unsigned long) vmemmap;
 	address_markers[VMALLOC_NR].start_address = VMALLOC_START;