Skip to contentSkip to conversation

FAILED: patch "[PATCH] parisc: fix race condition in patching code" failed to apply to 5.2-stable tree

PATCH [email protected] 3 participants 5 messages Aug 5, 2019

Subsystems (3)
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

The patch below does not apply to the 5.2-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From 740f05f30a8c49ec63668055d28feedd906d3c50 Mon Sep 17 00:00:00 2001
From: Sven Schnelle <[email protected]>
Date: Tue, 23 Jul 2019 22:37:54 +0200
Subject: [PATCH] parisc: fix race condition in patching code

Assume the following ftrace code sequence that was patched in earlier by ftrace_make_call():

PAGE A:
ffc: addr of ftrace_caller()
PAGE B:
000: 0x6fc10080 /* stw,ma r1,40(sp) */
004: 0x48213fd1 /* ldw -18(r1),r1 */
008: 0xe820c002 /* bv,n r0(r1) */
00c: 0xe83f1fdf /* b,l,n .-c,r1 */

When a Code sequences that is to be patched spans a page break, we might have already cleared the part on the PAGE A. If an interrupt is coming in during the remap of the fixed mapping to PAGE B, it might execute the patched function with only parts of the FTRACE code cleared. To prevent this, clear the jump to our mini trampoline first, and clear the remaining parts after this. This might also happen when patch_text() patches a function that it calls during remap.

arch/parisc/kernel/ftrace.c +2 -1
index d784ccdd8fef..b6fb30f2e4bf 100644
@@ -181,8 +181,9 @@ int ftrace_make_nop(struct module *mod, struct dyn_ftrace *rec,
 	for (i = 0; i < ARRAY_SIZE(insn); i++)
 		insn[i] = INSN_NOP;
 
+	__patch_text((void *)rec->ip, INSN_NOP);
 	__patch_text_multiple((void *)rec->ip + 4 - sizeof(insn),
-			      insn, sizeof(insn));
+			      insn, sizeof(insn)-4);
 	return 0;
 }
 #endif
#1 lore raw
Quoted 5 lines · On Mon, Aug 05, 2019 at 07:30:55AM +0200, [email protected] wrote:

The patch below does not apply to the 5.2-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

The reason is that 5.2 doesn't have DYNAMIC_FTRACE suport, so this can be ignored.

Regards
Sven

author maintainer
#2 lore raw
Quoted 9 lines · Sven Schnelle
Quoted 5 lines · On Mon, Aug 05, 2019 at 07:30:55AM +0200, [email protected] wrote:

The patch below does not apply to the 5.2-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

The reason is that 5.2 doesn't have DYNAMIC_FTRACE suport, so this can be ignored.

Then why did the patch have:

	    Cc: <[email protected]> # 5.2+

in it?

confused,

greg k-h

#3 lore raw

Hi Greg,

Quoted 8 lines · Greg KH
Quoted 7 lines · Sven Schnelle
Quoted 5 lines · On Mon, Aug 05, 2019 at 07:30:55AM +0200, [email protected] wrote:

The patch below does not apply to the 5.2-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

Quoted 6 lines
Quoted 2 lines

The reason is that 5.2 doesn't have DYNAMIC_FTRACE suport, so this can be ignored.

Then why did the patch have:

	    Cc: <[email protected]> # 5.2+

in it?

My Fault. Helge asked me whether this patch should be added to 5.2 stable, and i said "yes", assuming that DYNAMIC_FTRACE for PA-RISC is already in Linux-5.2, which isn't the case. It got merged in 5.3.

Regards
Sven

maintainer
#4 lore raw
Quoted 12 lines · Greg KH
Quoted 11 lines · Sven Schnelle
Quoted 9 lines · On Mon, Aug 05, 2019 at 07:30:55AM +0200, [email protected] wrote:
Quoted 5 lines

The patch below does not apply to the 5.2-stable tree.
If someone wants it applied there, or to any other stable or longterm tree, then please email the backport, including the original git commit id to <[email protected]>.

The reason is that 5.2 doesn't have DYNAMIC_FTRACE suport, so this can be ignored.

Quoted 7 lines
Quoted 3 lines

Then why did the patch have:

	    Cc: <[email protected]> # 5.2+

in it?

My Fault. Helge asked me whether this patch should be added to 5.2 stable, and i said "yes", assuming that DYNAMIC_FTRACE for PA-RISC is already in Linux-5.2, which isn't the case. It got merged in 5.3.

Greg, It's actually my fault.
I did asked Sven, but usually I do test if the patches apply cleanly too. So either I did missed that testing for this patch, or something else must have been wrong on my side.
Sorry for that.
Helge