diff --git a/util/cbfstool/cbfs-payload-linux.c b/util/cbfstool/cbfs-payload-linux.c
new file mode 100644
index 0000000..149265c
--- /dev/null
+++ b/util/cbfstool/cbfs-payload-linux.c
@@ -0,0 +1,238 @@
+/*
+ * cbfs-payload-linux
+ *
+ * Copyright (C) 2013 Patrick Georgi <patrick@georgi-clan.de>
+ *
+ * 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 <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include "common.h"
+#include "cbfs.h"
+#include "linux.h"
+
+/* TODO:
+ *   handle special arguments
+ *     mem= argument - only affects loading decisions (kernel + initrd), not e820 -> build time
+ *     vga= argument (FILO ignores this)
+ *   add support for more parameters to trampoline:
+ *     alt_mem_k, ext_mem_k (not strictly necessary since e820 takes precedence)
+ *     framebuffer/console values
+ *
+ *  larger work:
+ *     is compress() safe to use in a size constrained buffer? ie. do(es) the
+ *     compression algorithm(s) stop once the compression result reaches input
+ *     size (ie. incompressible data)?
+ */
+int parse_bzImage_to_payload(const struct buffer *input,
+			     struct buffer *output, const char *initrd_name,
+			     char *cmdline, comp_algo algo)
+{
+	int cur_len = 0;
+	int num_segments = 3; /* parameter block, real kernel, and trampoline */
+
+	comp_func_ptr compress = compression_function(algo);
+	if (!compress)
+		return -1;
+
+	unsigned int initrd_base = 64*1024*1024;
+	unsigned int initrd_size = 0;
+	void *initrd_data = NULL;
+	if (initrd_name != NULL) {
+		/* TODO: load initrd, set initrd_size */
+		num_segments++;
+		FILE *initrd_file = fopen(initrd_name, "rb");
+		if (!initrd_file) {
+			ERROR("could not open initrd.\n");
+			return -1;
+		}
+		fseek(initrd_file, 0, SEEK_END);
+		initrd_size = ftell(initrd_file);
+		fseek(initrd_file, 0, SEEK_SET);
+		initrd_data = malloc(initrd_size);
+		if (!initrd_data) {
+			ERROR("could not allocate memory for initrd.\n");
+			return -1;
+		}
+		if (fread(initrd_data, initrd_size, 1, initrd_file) != 1) {
+			ERROR("could not load initrd.\n");
+			return -1;
+		}
+		fclose(initrd_file);
+	}
+
+	unsigned int cmdline_size = 0;
+	if (cmdline != NULL) {
+		num_segments++;
+		cmdline_size = strlen(cmdline) + 1;
+	}
+
+	struct linux_header *hdr = (struct linux_header *)input->data;
+	unsigned int setup_size = 4 * 512;
+	if (hdr->setup_sects != 0) {
+		setup_size = (hdr->setup_sects + 1) * 512;
+	}
+
+	/* Setup parameter block. Imitate FILO. */
+	struct linux_params params;
+	params.mount_root_rdonly = hdr->root_flags;
+	params.orig_root_dev = hdr->root_dev;
+	/* Sensible video defaults. Might be overridden on runtime by coreboot tables. */
+	params.orig_video_mode = 3;
+	params.orig_video_cols = 80;
+	params.orig_video_lines = 25;
+	params.orig_video_isVGA = 1;
+	params.orig_video_points = 16;
+
+	params.loader_type = 0xff; /* Unregistered Linux loader */
+
+	if (cmdline != NULL) {
+		if (hdr->protocol_version < 0x202) {
+			params.cl_magic = CL_MAGIC_VALUE;
+			params.cl_offset = COMMAND_LINE_LOC - LINUX_PARAM_LOC;
+		} else {
+			params.cmd_line_ptr = COMMAND_LINE_LOC;
+		}
+	}
+
+	unsigned long kernel_base = 0x100000;
+	if ((hdr->protocol_version >= 0x200) && (!hdr->loadflags & 1)) {
+		kernel_base = 0x1000; /* zImage kernel */
+	}
+	/* kernel prefers an address, so listen */
+	if ((hdr->protocol_version >= 0x20a) && (!(hdr->pref_address >> 32))) {
+		kernel_base = hdr->pref_address;
+	}
+	if (hdr->protocol_version >= 0x205) {
+		params.relocatable_kernel = hdr->relocatable_kernel;
+		params.kernel_alignment = hdr->kernel_alignment;
+		if (hdr->relocatable_kernel != 0) {
+			/* 16 MB should be way outside coreboot's playground,
+			 * so if possible (relocatable kernel) use that to
+			 * avoid a trampoline copy. */
+			kernel_base = ALIGN(16*1024*1024, params.kernel_alignment);
+		}
+	}
+
+	/* We have a trampoline and use that, but it can simply use
+	 * this information for its jump to real Linux. */
+	params.kernel_start = kernel_base;
+
+	void *kernel_data = input->data + setup_size;
+	unsigned int kernel_size = input->size - setup_size;
+
+	if (initrd_data != NULL) {
+		/* TODO: this is a bit of a hack. Linux recommends to store
+		 * initrd near to end-of-mem, but that's hard to do on build
+		 * time. It definitely fails to read the image if it's too
+		 * close to the kernel, so give it some room.
+		 */
+		initrd_base = ALIGN(kernel_base + kernel_size, 16*1024*1024);
+
+		params.initrd_start = initrd_base;
+		params.initrd_size = initrd_size;
+	}
+
+	struct cbfs_payload_segment *segs;
+	unsigned long doffset = (num_segments + 1) * sizeof(*segs);
+
+	/* Allocate a block of memory to store the data in */
+	int isize = sizeof(params) + kernel_size + cmdline_size + initrd_size;
+	if (buffer_create(output, doffset + isize, input->name) != 0)
+		return -1;
+	memset(output->data, 0, output->size);
+
+	segs = (struct cbfs_payload_segment *)output->data;
+
+	/* parameter block */
+	segs[0].type = PAYLOAD_SEGMENT_DATA;
+	segs[0].load_addr = htonll(LINUX_PARAM_LOC);
+	segs[0].mem_len = htonl(sizeof(params));
+	segs[0].offset = htonl(doffset);
+
+	compress((void*)&params, sizeof(params), output->data + doffset, &cur_len);
+	segs[0].compression = htonl(algo);
+	segs[0].len = htonl(cur_len);
+
+	doffset += cur_len;
+
+	/* code block */
+	segs[1].type = PAYLOAD_SEGMENT_CODE;
+	segs[1].load_addr = htonll(kernel_base);
+	segs[1].mem_len = htonl(kernel_size);
+	segs[1].offset = htonl(doffset);
+
+	compress(kernel_data, kernel_size, output->data + doffset, &cur_len);
+	segs[1].compression = htonl(algo);
+	segs[1].len = htonl(cur_len);
+
+	doffset += cur_len;
+
+	/* trampoline */
+	extern void *trampoline_start;
+	extern long trampoline_size;
+
+	unsigned int entrypoint = 0x40000; /* TODO: any better place? */
+
+	segs[2].type = PAYLOAD_SEGMENT_CODE;
+	segs[2].load_addr = htonll(entrypoint);
+	segs[2].mem_len = htonl(trampoline_size);
+	segs[2].offset = htonl(doffset);
+
+	compress(trampoline_start, trampoline_size, output->data + doffset, &cur_len);
+	segs[2].compression = htonl(algo);
+	segs[2].len = htonl(cur_len);
+
+	doffset += cur_len;
+
+	if (cmdline_size > 0) {
+		/* command line block */
+		segs[3].type = PAYLOAD_SEGMENT_DATA;
+		segs[3].load_addr = htonll(COMMAND_LINE_LOC);
+		segs[3].mem_len = htonl(cmdline_size);
+		segs[3].offset = htonl(doffset);
+
+		compress(cmdline, cmdline_size, output->data + doffset, &cur_len);
+		segs[3].compression = htonl(algo);
+		segs[3].len = htonl(cur_len);
+
+		doffset += cur_len;
+	}
+
+	if (initrd_size > 0) {
+		/* setup block */
+		segs[num_segments-1].type = PAYLOAD_SEGMENT_DATA;
+		segs[num_segments-1].load_addr = htonll(initrd_base);
+		segs[num_segments-1].mem_len = htonl(initrd_size);
+		segs[num_segments-1].offset = htonl(doffset);
+
+		compress(initrd_data, initrd_size, output->data + doffset, &cur_len);
+		segs[num_segments-1].compression = htonl(algo);
+		segs[num_segments-1].len = htonl(cur_len);
+
+		doffset += cur_len;
+	}
+
+	/* prepare entry point segment */
+	segs[num_segments].type = PAYLOAD_SEGMENT_ENTRY;
+	segs[num_segments].load_addr = htonll(entrypoint);
+	output->size = doffset;
+
+	return 0;
+}
+
+