x86: add MIRROR_PAYLOAD_TO_RAM_BEFORE_LOADING option

Boot speeds can be sped up by mirroring the payload into
main memory before doing the actual loading. Systems that
would benefit from this are typically Intel ones whose SPI
are memory mapped. Without the SPI being cached all accesses
to the payload in SPI while being loaded result in uncacheable
accesses. Instead take advantage of the on-board SPI controller
which has an internal cache and prefetcher by copying 64-byte
cachelines using 32-bit word copies.

Change-Id: I4aac856b1b5130fa2d68a6c45a96cfeead472a52
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/5305
Tested-by: build bot (Jenkins)
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Vladimir Serbinenko <phcoder@gmail.com>
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>

src/cpu/x86/Kconfig

@@ -121,3 +121,12 @@ config BACKUP_DEFAULT_SMM_REGION
 	help
 	 The cpu support will select this option if the default SMM region
 	 needs to be backed up for suspend/resume purposes.
+
+config MIRROR_PAYLOAD_TO_RAM_BEFORE_LOADING
+	def_bool n
+	help
+	  On certain platforms a boot speed gain can be realized if mirroring
+	  the payload data stored in non-volatile storage. On x86 systems the
+	  payload would typically live in a memory-mapped SPI part. Copying
+	  the SPI contents to ram before performing the load can speed up
+	  the boot process.

src/cpu/x86/Makefile.inc

@@ -3,6 +3,7 @@ romstage-$(CONFIG_HAVE_ACPI_RESUME) += car.c
 
 subdirs-$(CONFIG_PARALLEL_MP) += name
 ramstage-$(CONFIG_PARALLEL_MP) += mp_init.c
+ramstage-$(CONFIG_MIRROR_PAYLOAD_TO_RAM_BEFORE_LOADING) += mirror_payload.c
 
 SIPI_ELF=$(obj)/cpu/x86/sipi_vector.elf
 SIPI_BIN=$(SIPI_ELF:.elf=)

src/cpu/x86/mirror_payload.c

@@ -0,0 +1,71 @@
+/*
+ * This file is part of the coreboot project.
+ *
+ * Copyright (C) 2014 Google Inc.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <console/console.h>
+#include <bootmem.h>
+#include <payload_loader.h>
+
+void mirror_payload(struct payload *payload)
+{
+	char *buffer;
+	size_t size;
+	char *src;
+	uintptr_t alignment_diff;
+	const unsigned long cacheline_size = 64;
+	const uintptr_t intra_cacheline_mask = cacheline_size - 1;
+	const uintptr_t cacheline_mask = ~intra_cacheline_mask;
+
+	src = payload->backing_store.data;
+	size = payload->backing_store.size;
+
+	/*
+	 * Adjust size so that the start and end points are aligned to a
+	 * cacheline. The SPI hardware controllers on Intel machines should
+	 * cache full length cachelines as well as prefetch data.  Once the
+	 * data is mirrored in memory all accesses should hit the CPU's cache.
+	 */
+	alignment_diff = (intra_cacheline_mask & (uintptr_t)src);
+	size += alignment_diff;
+
+	size = ALIGN(size, cacheline_size);
+
+	printk(BIOS_DEBUG, "Payload aligned size: 0x%zx\n", size);
+
+	buffer = bootmem_allocate_buffer(size);
+
+	if (buffer == NULL) {
+		printk(BIOS_DEBUG, "No buffer for mirroring payload.\n");
+		return;
+	}
+
+	src = (void *)(cacheline_mask & (uintptr_t)src);
+
+	/*
+	 * Note that if mempcy is not using 32-bit moves the performance will
+	 * degrade because the SPI hardware prefetchers look for
+	 * cacheline-aligned 32-bit accesses to kick in.
+	 */
+	memcpy(buffer, src, size);
+
+	/* Update the payload's backing store. */
+	payload->backing_store.data = &buffer[alignment_diff];
+}

src/include/payload_loader.h

@@ -44,6 +44,9 @@ struct payload *payload_load(void);
 /* Run the loaded payload. */
 void payload_run(const struct payload *payload);
 
+/* Mirror the payload to be loaded. */
+void mirror_payload(struct payload *payload);
+
 /* architecture specific function to run payload. */
 void arch_payload_run(const struct payload *payload);
 

src/lib/loaders/load_and_run_payload.c

@@ -39,6 +39,11 @@ static struct payload global_payload = {
 	.name = CONFIG_CBFS_PREFIX "/payload",
 };
 
+void __attribute__((weak)) mirror_payload(struct payload *payload)
+{
+	return;
+}
+
 struct payload *payload_load(void)
 {
 	int i;
@@ -62,6 +67,8 @@ struct payload *payload_load(void)
 	if (i == ARRAY_SIZE(payload_ops))
 		return NULL;
 
+	mirror_payload(payload);
+
 	entry = selfload(payload);
 
 	if (entry == NULL)