Commit ad79d603 authored by Simon Glass's avatar Simon Glass Committed by Bin Meng

cbfs: Rename camel-case variables

Rename some camel-case variables to match U-Boot style.

Camel case is not generally allowed in U-Boot. Rename this variable to fit
in with the style.
Signed-off-by: Simon Glass's avatarSimon Glass <sjg@chromium.org>
Reviewed-by: Bin Meng's avatarBin Meng <bmeng.cn@gmail.com>
Tested-by: Bin Meng's avatarBin Meng <bmeng.cn@gmail.com>
parent 630b2f39
......@@ -73,14 +73,14 @@ static void swap_file_header(struct cbfs_fileheader *dest,
* @param start The location in memory to start from.
* @param size The size of the memory region to search.
* @param align The alignment boundaries to check on.
* @param newNode A pointer to the file structure to load.
* @param new_node A pointer to the file structure to load.
* @param used A pointer to the count of of bytes scanned through,
* including the file if one is found.
*
* @return 1 if a file is found, 0 if one isn't.
*/
static int file_cbfs_next_file(struct cbfs_priv *priv, u8 *start, u32 size,
u32 align, struct cbfs_cachenode *newNode,
u32 align, struct cbfs_cachenode *new_node,
u32 *used)
{
struct cbfs_fileheader header;
......@@ -88,34 +88,34 @@ static int file_cbfs_next_file(struct cbfs_priv *priv, u8 *start, u32 size,
*used = 0;
while (size >= align) {
const struct cbfs_fileheader *fileHeader =
const struct cbfs_fileheader *file_header =
(const struct cbfs_fileheader *)start;
u32 name_len;
u32 step;
/* Check if there's a file here. */
if (memcmp(good_file_magic, &(fileHeader->magic),
sizeof(fileHeader->magic))) {
if (memcmp(good_file_magic, &file_header->magic,
sizeof(file_header->magic))) {
*used += align;
size -= align;
start += align;
continue;
}
swap_file_header(&header, fileHeader);
swap_file_header(&header, file_header);
if (header.offset < sizeof(struct cbfs_fileheader)) {
priv->result = CBFS_BAD_FILE;
return -1;
}
newNode->next = NULL;
newNode->type = header.type;
newNode->data = start + header.offset;
newNode->data_length = header.len;
new_node->next = NULL;
new_node->type = header.type;
new_node->data = start + header.offset;
new_node->data_length = header.len;
name_len = header.offset - sizeof(struct cbfs_fileheader);
newNode->name = (char *)fileHeader +
new_node->name = (char *)file_header +
sizeof(struct cbfs_fileheader);
newNode->name_length = name_len;
newNode->attributes_offset = header.attributes_offset;
new_node->name_length = name_len;
new_node->attributes_offset = header.attributes_offset;
step = header.len;
if (step % align)
......@@ -132,15 +132,15 @@ static void file_cbfs_fill_cache(struct cbfs_priv *priv, u8 *start, u32 size,
u32 align)
{
struct cbfs_cachenode *cache_node;
struct cbfs_cachenode *newNode;
struct cbfs_cachenode *new_node;
struct cbfs_cachenode **cache_tail = &priv->file_cache;
/* Clear out old information. */
cache_node = priv->file_cache;
while (cache_node) {
struct cbfs_cachenode *oldNode = cache_node;
struct cbfs_cachenode *old_node = cache_node;
cache_node = cache_node->next;
free(oldNode);
free(old_node);
}
priv->file_cache = NULL;
......@@ -148,20 +148,20 @@ static void file_cbfs_fill_cache(struct cbfs_priv *priv, u8 *start, u32 size,
int result;
u32 used;
newNode = (struct cbfs_cachenode *)
new_node = (struct cbfs_cachenode *)
malloc(sizeof(struct cbfs_cachenode));
result = file_cbfs_next_file(priv, start, size, align, newNode,
result = file_cbfs_next_file(priv, start, size, align, new_node,
&used);
if (result < 0) {
free(newNode);
free(new_node);
return;
} else if (result == 0) {
free(newNode);
free(new_node);
break;
}
*cache_tail = newNode;
cache_tail = &newNode->next;
*cache_tail = new_node;
cache_tail = &new_node->next;
size -= used;
start += used;
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment