public inbox for u-boot@lists.denx.de
 help / color / mirror / Atom feed
From: Sean Anderson <seanga2@gmail.com>
To: u-boot@lists.denx.de, Tom Rini <trini@konsulko.com>
Cc: Harald Seiler <hws@denx.de>, Simon Glass <sjg@chromium.org>,
	Heinrich Schuchardt <xypron.glpk@gmx.de>,
	Sean Anderson <seanga2@gmail.com>
Subject: [PATCH v2 24/29] test: spl: Add functions to create filesystems
Date: Sat, 14 Oct 2023 16:48:00 -0400	[thread overview]
Message-ID: <20231014204805.439009-25-seanga2@gmail.com> (raw)
In-Reply-To: <20231014204805.439009-1-seanga2@gmail.com>

Add some functions for creating fat/ext2 filesystems with a single file and
a test for them. Filesystems require block devices, and it is easiest to
just use MMC for this. To get an MMC, we must also pull in the test device
tree. SPL_TIMER is necessary for SPL_MMC, perhaps because it uses a timeout.

Signed-off-by: Sean Anderson <seanga2@gmail.com>
---

Changes in v2:
- Explicitly invalidate FAT block device cache before testing
- Remove unused dm.h include in spl_load.c
- Add note about why we enable SPL_TIMER
- Split off ext/-T changes into their own commits

 configs/sandbox_noinst_defconfig |   7 +
 include/test/spl.h               |   3 +
 test/image/Kconfig               |  11 ++
 test/image/Makefile              |   1 +
 test/image/spl_load_fs.c         | 306 +++++++++++++++++++++++++++++++
 5 files changed, 328 insertions(+)
 create mode 100644 test/image/spl_load_fs.c

diff --git a/configs/sandbox_noinst_defconfig b/configs/sandbox_noinst_defconfig
index 908155be8a3..0a542cfb6aa 100644
--- a/configs/sandbox_noinst_defconfig
+++ b/configs/sandbox_noinst_defconfig
@@ -6,10 +6,12 @@ CONFIG_NR_DRAM_BANKS=1
 CONFIG_ENV_SIZE=0x2000
 CONFIG_DEFAULT_DEVICE_TREE="sandbox"
 CONFIG_DM_RESET=y
+CONFIG_SPL_MMC=y
 CONFIG_SPL_SERIAL=y
 CONFIG_SPL_DRIVERS_MISC=y
 CONFIG_SPL_SYS_MALLOC_F_LEN=0x8000
 CONFIG_SPL=y
+CONFIG_SPL_FS_FAT=y
 CONFIG_SYS_LOAD_ADDR=0x0
 CONFIG_PCI=y
 CONFIG_SANDBOX_SPL=y
@@ -39,7 +41,9 @@ CONFIG_SPL_HAS_CUSTOM_MALLOC_START=y
 CONFIG_SPL_CUSTOM_SYS_MALLOC_ADDR=0xa000000
 CONFIG_SPL_SYS_MALLOC_SIZE=0x4000000
 CONFIG_SPL_ENV_SUPPORT=y
+CONFIG_SPL_FS_EXT4=y
 CONFIG_SPL_I2C=y
+CONFIG_SPL_MMC_WRITE=y
 CONFIG_SPL_RTC=y
 CONFIG_CMD_CPU=y
 CONFIG_CMD_LICENSE=y
@@ -97,6 +101,7 @@ CONFIG_AMIGA_PARTITION=y
 CONFIG_OF_CONTROL=y
 CONFIG_SPL_OF_CONTROL=y
 CONFIG_SPL_OF_PLATDATA=y
+CONFIG_SPL_OF_REAL=y
 CONFIG_ENV_IS_NOWHERE=y
 CONFIG_ENV_IS_IN_EXT4=y
 CONFIG_ENV_EXT4_INTERFACE="host"
@@ -159,6 +164,7 @@ CONFIG_CROS_EC_SPI=y
 CONFIG_P2SB=y
 CONFIG_PWRSEQ=y
 CONFIG_SPL_PWRSEQ=y
+CONFIG_FS_LOADER=y
 CONFIG_MMC_SANDBOX=y
 CONFIG_SPI_FLASH_SANDBOX=y
 CONFIG_SPI_FLASH_ATMEL=y
@@ -220,6 +226,7 @@ CONFIG_SYSRESET=y
 CONFIG_SPL_SYSRESET=y
 CONFIG_DM_THERMAL=y
 CONFIG_TIMER=y
+CONFIG_SPL_TIMER=y
 CONFIG_TIMER_EARLY=y
 CONFIG_SANDBOX_TIMER=y
 CONFIG_USB=y
diff --git a/include/test/spl.h b/include/test/spl.h
index 4c6b789e478..7dba3f03d07 100644
--- a/include/test/spl.h
+++ b/include/test/spl.h
@@ -114,4 +114,7 @@ SPL_TEST(func##_##type, flags)
 /* More than a couple blocks, and will not be aligned to anything */
 #define SPL_TEST_DATA_SIZE	4099
 
+/* Flags necessary for accessing DM devices */
+#define DM_FLAGS (UT_TESTF_DM | UT_TESTF_SCAN_FDT)
+
 #endif /* TEST_SPL_H */
diff --git a/test/image/Kconfig b/test/image/Kconfig
index 70ffe0ff276..963c86cc290 100644
--- a/test/image/Kconfig
+++ b/test/image/Kconfig
@@ -10,6 +10,17 @@ config SPL_UT_LOAD
 
 if SPL_UT_LOAD
 
+config SPL_UT_LOAD_FS
+	bool "Unit tests for filesystems"
+	depends on SANDBOX && SPL_OF_REAL
+	depends on FS_LOADER
+	depends on SPL_FS_FAT
+	depends on SPL_FS_EXT4
+	depends on SPL_MMC_WRITE
+	default y
+	help
+	  Test filesystems in SPL.
+
 config SPL_UT_LOAD_OS
 	bool "Test loading from the host OS"
 	depends on SANDBOX && SPL_LOAD_FIT
diff --git a/test/image/Makefile b/test/image/Makefile
index f7ae996bc86..6bd77bf0aab 100644
--- a/test/image/Makefile
+++ b/test/image/Makefile
@@ -3,4 +3,5 @@
 # Copyright 2021 Google LLC
 
 obj-y += spl_load.o
+obj-$(CONFIG_SPL_UT_LOAD_FS) += spl_load_fs.o
 obj-$(CONFIG_SPL_UT_LOAD_OS) += spl_load_os.o
diff --git a/test/image/spl_load_fs.c b/test/image/spl_load_fs.c
new file mode 100644
index 00000000000..0ee1ede134a
--- /dev/null
+++ b/test/image/spl_load_fs.c
@@ -0,0 +1,306 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * Copyright (C) 2023 Sean Anderson <seanga2@gmail.com>
+ */
+
+#include <common.h>
+#include <blk.h>
+#include <ext_common.h>
+#include <ext4fs.h>
+#include <fat.h>
+#include <fs.h>
+#include <memalign.h>
+#include <asm/io.h>
+#include <linux/stat.h>
+#include <test/spl.h>
+#include <test/ut.h>
+
+/**
+ * create_ext2() - Create an "ext2" filesystem with a single file
+ * @dst: The location of the new filesystem; MUST be zeroed
+ * @size: The size of the file
+ * @filename: The name of the file
+ * @data_offset: Filled with the offset of the file data from @dst
+ *
+ * Budget mke2fs. We use 1k blocks (to reduce overhead) with a single block
+ * group, which limits us to 8M of data. Almost every feature which increases
+ * complexity (checksums, hash tree directories, etc.) is disabled. We do cheat
+ * a little and use extents from ext4 to save having to deal with indirects, but
+ * U-Boot doesn't care.
+ *
+ * If @dst is %NULL, nothing is copied.
+ *
+ * Return: The size of the filesystem in bytes
+ */
+static size_t create_ext2(void *dst, size_t size, const char *filename,
+			  size_t *data_offset)
+{
+	u32 super_block = 1;
+	u32 group_block = 2;
+	u32 block_bitmap_block = 3;
+	u32 inode_bitmap_block = 4;
+	u32 inode_table_block = 5;
+	u32 root_block = 6;
+	u32 file_block = 7;
+
+	u32 root_ino = EXT2_ROOT_INO;
+	u32 file_ino = EXT2_BOOT_LOADER_INO;
+
+	u32 block_size = EXT2_MIN_BLOCK_SIZE;
+	u32 inode_size = sizeof(struct ext2_inode);
+
+	u32 file_blocks = (size + block_size - 1) / block_size;
+	u32 blocks = file_block + file_blocks;
+	u32 inodes = block_size / inode_size;
+	u32 filename_len = strlen(filename);
+	u32 dirent_len = ALIGN(filename_len, sizeof(struct ext2_dirent)) +
+			    sizeof(struct ext2_dirent);
+
+	struct ext2_sblock *sblock = dst + super_block * block_size;
+	struct ext2_block_group *bg = dst + group_block * block_size;
+	struct ext2_inode *inode_table = dst + inode_table_block * block_size;
+	struct ext2_inode *root_inode = &inode_table[root_ino - 1];
+	struct ext2_inode *file_inode = &inode_table[file_ino - 1];
+	struct ext4_extent_header *ext_block = (void *)&file_inode->b;
+	struct ext4_extent *extent = (void *)(ext_block + 1);
+	struct ext2_dirent *dot = dst + root_block * block_size;
+	struct ext2_dirent *dotdot = dot + 2;
+	struct ext2_dirent *dirent = dotdot + 2;
+	struct ext2_dirent *last = ((void *)dirent) + dirent_len;
+
+	/* Make sure we fit in one block group */
+	if (blocks > block_size * 8)
+		return 0;
+
+	if (filename_len > EXT2_NAME_LEN)
+		return 0;
+
+	if (data_offset)
+		*data_offset = file_block * block_size;
+
+	if (!dst)
+		goto out;
+
+	sblock->total_inodes = cpu_to_le32(inodes);
+	sblock->total_blocks = cpu_to_le32(blocks);
+	sblock->first_data_block = cpu_to_le32(super_block);
+	sblock->blocks_per_group = cpu_to_le32(blocks);
+	sblock->fragments_per_group = cpu_to_le32(blocks);
+	sblock->inodes_per_group = cpu_to_le32(inodes);
+	sblock->magic = cpu_to_le16(EXT2_MAGIC);
+	/* Done mostly so we can pretend to be (in)compatible */
+	sblock->revision_level = cpu_to_le32(EXT2_DYNAMIC_REV);
+	/* Not really accurate but it doesn't matter */
+	sblock->first_inode = cpu_to_le32(EXT2_GOOD_OLD_FIRST_INO);
+	sblock->inode_size = cpu_to_le32(inode_size);
+	sblock->feature_incompat = cpu_to_le32(EXT4_FEATURE_INCOMPAT_EXTENTS);
+
+	bg->block_id = cpu_to_le32(block_bitmap_block);
+	bg->inode_id = cpu_to_le32(inode_bitmap_block);
+	bg->inode_table_id = cpu_to_le32(inode_table_block);
+
+	/*
+	 * All blocks/inodes are in-use. I don't want to have to deal with
+	 * endianness, so just fill everything in.
+	 */
+	memset(dst + block_bitmap_block * block_size, 0xff, block_size * 2);
+
+	root_inode->mode = cpu_to_le16(S_IFDIR | 0755);
+	root_inode->size = cpu_to_le32(block_size);
+	root_inode->nlinks = cpu_to_le16(3);
+	root_inode->blockcnt = cpu_to_le32(1);
+	root_inode->flags = cpu_to_le32(EXT4_TOPDIR_FL);
+	root_inode->b.blocks.dir_blocks[0] = root_block;
+
+	file_inode->mode = cpu_to_le16(S_IFREG | 0644);
+	file_inode->size = cpu_to_le32(size);
+	file_inode->nlinks = cpu_to_le16(1);
+	file_inode->blockcnt = cpu_to_le32(file_blocks);
+	file_inode->flags = cpu_to_le32(EXT4_EXTENTS_FL);
+	ext_block->eh_magic = cpu_to_le16(EXT4_EXT_MAGIC);
+	ext_block->eh_entries = cpu_to_le16(1);
+	ext_block->eh_max = cpu_to_le16(sizeof(file_inode->b) /
+					sizeof(*ext_block) - 1);
+	extent->ee_len = cpu_to_le16(file_blocks);
+	extent->ee_start_lo = cpu_to_le16(file_block);
+
+	/* I'm not sure we need these, but it can't hurt */
+	dot->inode = cpu_to_le32(root_ino);
+	dot->direntlen = cpu_to_le16(2 * sizeof(*dot));
+	dot->namelen = 1;
+	dot->filetype = FILETYPE_DIRECTORY;
+	memcpy(dot + 1, ".", dot->namelen);
+
+	dotdot->inode = cpu_to_le32(root_ino);
+	dotdot->direntlen = cpu_to_le16(2 * sizeof(*dotdot));
+	dotdot->namelen = 2;
+	dotdot->filetype = FILETYPE_DIRECTORY;
+	memcpy(dotdot + 1, "..", dotdot->namelen);
+
+	dirent->inode = cpu_to_le32(file_ino);
+	dirent->direntlen = cpu_to_le16(dirent_len);
+	dirent->namelen = filename_len;
+	dirent->filetype = FILETYPE_REG;
+	memcpy(dirent + 1, filename, filename_len);
+
+	last->direntlen = block_size - dirent_len;
+
+out:
+	return (size_t)blocks * block_size;
+}
+
+/**
+ * create_fat() - Create a FAT32 filesystem with a single file
+ * @dst: The location of the new filesystem; MUST be zeroed
+ * @size: The size of the file
+ * @filename: The name of the file
+ * @data_offset: Filled with the offset of the file data from @dst
+ *
+ * Budget mkfs.fat. We use FAT32 (so I don't have to deal with FAT12) with no
+ * info sector, and a single one-sector FAT. This limits us to 64k of data
+ * (enough for anyone). The filename must fit in 8.3.
+ *
+ * If @dst is %NULL, nothing is copied.
+ *
+ * Return: The size of the filesystem in bytes
+ */
+static size_t create_fat(void *dst, size_t size, const char *filename,
+			 size_t *data_offset)
+{
+	u16 boot_sector = 0;
+	u16 fat_sector = 1;
+	u32 root_sector = 2;
+	u32 file_sector = 3;
+
+	u16 sector_size = 512;
+	u32 file_sectors = (size + sector_size - 1) / sector_size;
+	u32 sectors = file_sector + file_sectors;
+
+	char *ext;
+	size_t filename_len, ext_len;
+	int i;
+
+	struct boot_sector *bs = dst + boot_sector * sector_size;
+	struct volume_info *vi = (void *)(bs + 1);
+	__le32 *fat = dst + fat_sector * sector_size;
+	struct dir_entry *dirent = dst + root_sector * sector_size;
+
+	/* Make sure we fit in the FAT */
+	if (sectors > sector_size / sizeof(u32))
+		return 0;
+
+	ext = strchr(filename, '.');
+	if (ext) {
+		filename_len = ext - filename;
+		ext++;
+		ext_len = strlen(ext);
+	} else {
+		filename_len = strlen(filename);
+		ext_len = 0;
+	}
+
+	if (filename_len > 8 || ext_len > 3)
+		return 0;
+
+	if (data_offset)
+		*data_offset = file_sector * sector_size;
+
+	if (!dst)
+		goto out;
+
+	bs->sector_size[0] = sector_size & 0xff;
+	bs->sector_size[1] = sector_size >> 8;
+	bs->cluster_size = 1;
+	bs->reserved = cpu_to_le16(fat_sector);
+	bs->fats = 1;
+	bs->media = 0xf8;
+	bs->total_sect = cpu_to_le32(sectors);
+	bs->fat32_length = cpu_to_le32(1);
+	bs->root_cluster = cpu_to_le32(root_sector);
+
+	vi->ext_boot_sign = 0x29;
+	memcpy(vi->fs_type, FAT32_SIGN, sizeof(vi->fs_type));
+
+	memcpy(dst + 0x1fe, "\x55\xAA", 2);
+
+	fat[0] = cpu_to_le32(0x0ffffff8);
+	fat[1] = cpu_to_le32(0x0fffffff);
+	fat[2] = cpu_to_le32(0x0ffffff8);
+	for (i = file_sector; file_sectors > 1; file_sectors--, i++)
+		fat[i] = cpu_to_le32(i + 1);
+	fat[i] = cpu_to_le32(0x0ffffff8);
+
+	for (i = 0; i < sizeof(dirent->nameext.name); i++) {
+		if (i < filename_len)
+			dirent->nameext.name[i] = toupper(filename[i]);
+		else
+			dirent->nameext.name[i] = ' ';
+	}
+
+	for (i = 0; i < sizeof(dirent->nameext.ext); i++) {
+		if (i < ext_len)
+			dirent->nameext.ext[i] = toupper(ext[i]);
+		else
+			dirent->nameext.ext[i] = ' ';
+	}
+
+	dirent->start = cpu_to_le16(file_sector);
+	dirent->size = cpu_to_le32(size);
+
+out:
+	return sectors * sector_size;
+}
+
+typedef size_t (*create_fs_t)(void *, size_t, const char *, size_t *);
+
+static int spl_test_fs(struct unit_test_state *uts, const char *test_name,
+		       create_fs_t create)
+{
+	const char *filename = CONFIG_SPL_FS_LOAD_PAYLOAD_NAME;
+	struct blk_desc *dev_desc;
+	char *data_write, *data_read;
+	void *fs;
+	size_t fs_size, fs_data, fs_blocks, data_size = SPL_TEST_DATA_SIZE;
+	loff_t actread;
+
+	fs_size = create(NULL, data_size, filename, &fs_data);
+	ut_assert(fs_size);
+	fs = calloc(fs_size, 1);
+	ut_assertnonnull(fs);
+
+	data_write = fs + fs_data;
+	generate_data(data_write, data_size, test_name);
+	ut_asserteq(fs_size, create(fs, data_size, filename, NULL));
+
+	dev_desc = blk_get_devnum_by_uclass_id(UCLASS_MMC, 0);
+	ut_assertnonnull(dev_desc);
+	ut_asserteq(512, dev_desc->blksz);
+	fs_blocks = fs_size / dev_desc->blksz;
+	ut_asserteq(fs_blocks, blk_dwrite(dev_desc, 0, fs_blocks, fs));
+
+	/* We have to use malloc so we can call virt_to_phys */
+	data_read = malloc_cache_aligned(data_size);
+	ut_assertnonnull(data_read);
+	ut_assertok(fs_set_blk_dev_with_part(dev_desc, 0));
+	ut_assertok(fs_read("/" CONFIG_SPL_FS_LOAD_PAYLOAD_NAME,
+			    virt_to_phys(data_read), 0, data_size, &actread));
+	ut_asserteq(data_size, actread);
+	ut_asserteq_mem(data_write, data_read, data_size);
+
+	free(data_read);
+	free(fs);
+	return 0;
+}
+
+static int spl_test_ext(struct unit_test_state *uts)
+{
+	return spl_test_fs(uts, __func__, create_ext2);
+}
+SPL_TEST(spl_test_ext, DM_FLAGS);
+
+static int spl_test_fat(struct unit_test_state *uts)
+{
+	spl_fat_force_reregister();
+	return spl_test_fs(uts, __func__, create_fat);
+}
+SPL_TEST(spl_test_fat, DM_FLAGS);
-- 
2.37.1


  parent reply	other threads:[~2023-10-15  7:07 UTC|newest]

Thread overview: 43+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-10-14 20:47 [PATCH v2 00/29] test: spl: Test some load methods Sean Anderson
2023-10-14 20:47 ` [PATCH v2 01/29] spl: legacy: Fix referencing _image_binary_end Sean Anderson
2023-10-14 20:47 ` [PATCH v2 02/29] spl: nor: Don't allocate header on stack Sean Anderson
2023-10-14 20:47 ` [PATCH v2 03/29] spl: fit: Fix entry point for SPL_LOAD_FIT_FULL Sean Anderson
2023-10-14 20:47 ` [PATCH v2 04/29] arm: imx: Fix i.MX8 container load address Sean Anderson
2023-10-14 20:47 ` [PATCH v2 05/29] arm: imx: Add newlines after error messages Sean Anderson
2023-10-14 20:47 ` [PATCH v2 06/29] arm: imx: Use log_err for errors in read_auth_container Sean Anderson
2023-10-14 20:47 ` [PATCH v2 07/29] arm: imx: Add function to validate i.MX8 containers Sean Anderson
2023-10-14 20:47 ` [PATCH v2 08/29] arm: imx: Check header before calling spl_load_imx_container Sean Anderson
2023-10-14 20:47 ` [PATCH v2 09/29] Move i.MX8 container image loading support to common/spl Sean Anderson
2023-10-14 20:47 ` [PATCH v2 10/29] spl: Allow enabling SPL_OF_REAL and SPL_OF_PLATDATA at the same time Sean Anderson
2023-10-14 20:47 ` [PATCH v2 11/29] lib: acpi: Fix linking SPL when ACPIGEN is enabled Sean Anderson
2023-10-14 20:47 ` [PATCH v2 12/29] fs: Disable sandbox filesystem in SPL Sean Anderson
2023-10-14 21:14   ` Heinrich Schuchardt
2023-10-14 21:24     ` Sean Anderson
2023-10-14 21:45   ` Simon Glass
2023-10-14 20:47 ` [PATCH v2 13/29] fs: ext4: Fix building ext4 in SPL if write is enabled Sean Anderson
2023-10-14 20:47 ` [PATCH v2 14/29] fs: ext4: Add some defines for testing Sean Anderson
2023-10-14 23:35   ` Simon Glass
2023-10-14 20:47 ` [PATCH v2 15/29] net: Fix compiling SPL when fastboot is enabled Sean Anderson
2023-10-14 20:47 ` [PATCH v2 16/29] net: bootp: Move port numbers to header Sean Anderson
2023-10-14 21:20   ` Heinrich Schuchardt
2023-10-14 21:22     ` Tom Rini
2023-10-14 20:47 ` [PATCH v2 17/29] net: bootp: Fall back to BOOTP from DHCP when unit testing Sean Anderson
2023-10-14 21:22   ` Heinrich Schuchardt
2023-10-14 21:27     ` Sean Anderson
2023-10-15 14:39       ` Simon Glass
2023-10-15 19:20         ` Tom Rini
2023-10-14 20:47 ` [PATCH v2 18/29] spl: Add callbacks to invalidate cached devices Sean Anderson
2023-10-14 20:47 ` [PATCH v2 19/29] spl: Use map_sysmem where appropriate Sean Anderson
2023-10-14 20:47 ` [PATCH v2 20/29] sandbox: Support -T in spl Sean Anderson
2023-10-14 23:35   ` Simon Glass
2023-10-14 20:47 ` [PATCH v2 21/29] test: spl: Split tests up and use some configs Sean Anderson
2023-10-14 20:47 ` [PATCH v2 22/29] test: spl: Fix spl_test_load not failing if fname doesn't exist Sean Anderson
2023-10-14 20:47 ` [PATCH v2 23/29] test: spl: Add functions to create images Sean Anderson
2023-10-14 20:48 ` Sean Anderson [this message]
2023-10-14 23:35   ` [PATCH v2 24/29] test: spl: Add functions to create filesystems Simon Glass
2023-10-14 20:48 ` [PATCH v2 25/29] test: spl: Add a test for spl_blk_load_image Sean Anderson
2023-10-14 20:48 ` [PATCH v2 26/29] test: spl: Add a test for the MMC load method Sean Anderson
2023-10-14 20:48 ` [PATCH v2 27/29] test: spl: Add a test for the NET " Sean Anderson
2023-10-14 20:48 ` [PATCH v2 28/29] test: spl: Add a test for the NOR " Sean Anderson
2023-10-14 20:48 ` [PATCH v2 29/29] test: spl: Add a test for the SPI " Sean Anderson
2023-10-18 12:30 ` [PATCH v2 00/29] test: spl: Test some load methods Tom Rini

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20231014204805.439009-25-seanga2@gmail.com \
    --to=seanga2@gmail.com \
    --cc=hws@denx.de \
    --cc=sjg@chromium.org \
    --cc=trini@konsulko.com \
    --cc=u-boot@lists.denx.de \
    --cc=xypron.glpk@gmx.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox