On 2022/9/21 16:31, Leizhen (ThunderTown) wrote:

On 2022/9/21 16:00, Petr Mladek wrote:

quoted

On Tue 2022-09-20 15:13:11, Zhen Lei wrote:

quoted

For a symbol, there may be more than one place that can be merged. For
example: nfs_fs_proc_net_init, there are two "f"+"s_" combinations.
And we're only compressing the first combination at the moment.

Really?

Yes, there are about 200 such functions.

quoted

quoted

diff --git a/scripts/kallsyms.c b/scripts/kallsyms.c
index 8caccc8f4a23703..3319d9f38d7a5f2 100644
--- a/scripts/kallsyms.c
+++ b/scripts/kallsyms.c

@@ -553,7 +553,7 @@ static void compress_symbols(const unsigned char *str, int idx)
 	unsigned char *p1, *p2;
 
 	for (i = 0; i < table_cnt; i++) {
-
+retry:
 		len = table[i]->len;
 		p1 = table[i]->sym;
 

@@ -585,6 +585,9 @@ static void compress_symbols(const unsigned char *str, int idx)
 
 		/* increase the counts for this symbol's new tokens */
 		learn_symbol(table[i]->sym, len);
+
+		/* May be more than one place that can be merged, try again */
+		goto retry;
 	}
 }

My understanding is that the code already tries to find the same
token several times. Here are the important parts of the existing
code:

static void compress_symbols(const unsigned char *str, int idx)
{

		p2 = find_token(p1, len, str);

		do {
			/* replace the found token with idx */
			*p2 = idx;
			[...]

			/* find the token on the symbol */
			p2 = find_token(p1, size, str);

Oh, yes, it retries. Let me reanalyze it. However, the problem is
real, and there may be a problem somewhere in the loop.

Hi, Petr:
  Thanks. I found that it's my fault. The first round skip the type
character. But the next round will incorrectly skip one character,
so for nfs_fs_proc_net_init, the next round start from s, and using
            ^
the proposed "unsigned char type" in your next reply should solve
the problem. Thank you very much.

-	for (i = 0; i < len - 1; i++)
+	for (i = sym_start_idx; i < len - 1; i++)

quoted

		} while (p2);

Best Regards,
Petr
.

-- 
Regards,
  Zhen Lei