Re: [sparc64] ftrace: kernel startup-tests unaligned access

From: Steven Rostedt
Date: Mon Dec 14 2020 - 11:27:43 EST


On Mon, 14 Dec 2020 11:15:12 -0500
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> Does sparc64 require 8 byte alignment for 8 byte words?
>

In other words, does this patch fix anything?

-- Steve

diff --git a/arch/Kconfig b/arch/Kconfig
index 56b6ccc0e32d..fa716994f77e 100644
--- a/arch/Kconfig
+++ b/arch/Kconfig
@@ -143,6 +143,22 @@ config UPROBES
managed by the kernel and kept transparent to the probed
application. )

+config HAVE_64BIT_ALIGNED_ACCESS
+ def_bool 64BIT && !HAVE_EFFICIENT_UNALIGNED_ACCESS
+ help
+ Some architectures require 64 bit accesses to be 64 bit
+ aligned, which also requires structs containing 64 bit values
+ to be 64 bit aligned too. This includes some 32 bit
+ architectures which can do 64 bit accesses, as well as 64 bit
+ architectures without unaligned access.
+
+ This symbol should be selected by an architecture if 64 bit
+ accesses are required to be 64 bit aligned in this way even
+ though it is not a 64 bit architecture.
+
+ See Documentation/unaligned-memory-access.txt for more
+ information on the topic of unaligned memory accesses.
+
config HAVE_EFFICIENT_UNALIGNED_ACCESS
bool
help
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index f09d3f5911cb..623657f84b72 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -130,7 +130,16 @@ int ring_buffer_print_entry_header(struct trace_seq *s)
#define RB_ALIGNMENT 4U
#define RB_MAX_SMALL_DATA (RB_ALIGNMENT * RINGBUF_TYPE_DATA_TYPE_LEN_MAX)
#define RB_EVNT_MIN_SIZE 8U /* two 32bit words */
-#define RB_ALIGN_DATA __aligned(RB_ALIGNMENT)
+
+#ifndef CONFIG_HAVE_64BIT_ALIGNED_ACCESS
+# define RB_FORCE_8BYTE_ALIGNMENT 0
+# define RB_ARCH_ALIGNMENT RB_ALIGNMENT
+#else
+# define RB_FORCE_8BYTE_ALIGNMENT 1
+# define RB_ARCH_ALIGNMENT 8U
+#endif
+
+#define RB_ALIGN_DATA __aligned(RB_ARCH_ALIGNMENT)

/* define RINGBUF_TYPE_DATA for 'case RINGBUF_TYPE_DATA:' */
#define RINGBUF_TYPE_DATA 0 ... RINGBUF_TYPE_DATA_TYPE_LEN_MAX
@@ -2717,7 +2726,7 @@ rb_update_event(struct ring_buffer_per_cpu *cpu_buffer,

event->time_delta = delta;
length -= RB_EVNT_HDR_SIZE;
- if (length > RB_MAX_SMALL_DATA) {
+ if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT) {
event->type_len = 0;
event->array[0] = length;
} else
@@ -2732,11 +2741,11 @@ static unsigned rb_calculate_event_length(unsigned length)
if (!length)
length++;

- if (length > RB_MAX_SMALL_DATA)
+ if (length > RB_MAX_SMALL_DATA || RB_FORCE_8BYTE_ALIGNMENT)
length += sizeof(event.array[0]);

length += RB_EVNT_HDR_SIZE;
- length = ALIGN(length, RB_ALIGNMENT);
+ length = ALIGN(length, RB_ARCH_ALIGNMENT);

/*
* In case the time delta is larger than the 27 bits for it