Skip to contentSkip to conversation

x86/idt: Remove X86_TRAP_BP initialization in idt_setup_traps()

PATCH RESEND New Yonghong Song 2 participants 2 messages Nov 8, 2017 x86

lore
Share & tools
lore

Yonghong Song

author
#0 lore raw

PATCH RESEND 1/1 x86/idt: Remove X86_TRAP_BP initialization in idt_setup_traps()

Commit b70543a0b2b6("x86/idt: Move regular trap init to tables") moves regular trap init for each trap vector into a table based initialization. It introduced the initialization for vector X86_TRAP_BP which was not in the original change and it breaks uprobe functionality for x86_32. A program with uprobe instrumented can have segfault on x86_32 instead of normal finish. Uprobe still works on x86_64 even with this commit.

This patch removes X86_TRAP_BP from the structure def_idts which is used in idt_setup_traps(). It also removes a redundant entry for X86_TRAP_NMI in def_idts. Tested on both x86_64 and x86_32, uprobe works fine.

Fixes b70543a0b2b6("x86/idt: Move regular trap init to tables")
Reported-and-tested-by Yonghong Song <[email protected]>
Signed-off-by Yonghong Song <[email protected]>
---
 arch/x86/kernel/idt.c | 2 --
 1 file changed, 2 deletions(-)

[RESEND with adding [email protected] as the recipient.]

arch/x86/kernel/idt.c +0 -2
index 723fa97..d985cef 100644
@@ -92,8 +92,6 @@ static const __initdata struct idt_data def_idts[] = {
 	INTG(X86_TRAP_DF,		double_fault),
 #endif
 	INTG(X86_TRAP_DB,		debug),
-	INTG(X86_TRAP_NMI,		nmi),
-	INTG(X86_TRAP_BP,		int3),
 
 #ifdef CONFIG_X86_MCE
 	INTG(X86_TRAP_MC,		&machine_check),
Signature · 1 line
2.9.5
merge bot
#1 lore raw
x86/idt: Remove X86_TRAP_BP initialization in idt_setup_traps()
Commit-ID:  d0cd64b02aa854d68ce517cb7da1fe4e4fff2653
Gitweb:     https://git.kernel.org/tip/d0cd64b02aa854d68ce517cb7da1fe4e4fff2653
Author:     Yonghong Song <[email protected]>
AuthorDate: Wed, 8 Nov 2017 11:28:45 -0800
Committer:  Thomas Gleixner <[email protected]>
CommitDate: Wed, 8 Nov 2017 21:05:23 +0100

x86/idt: Remove X86_TRAP_BP initialization in idt_setup_traps()

Commit b70543a0b2b6("x86/idt: Move regular trap init to tables") moves regular trap init for each trap vector into a table based initialization. It introduced the initialization for vector X86_TRAP_BP which was not in the code which it replaced. This breaks uprobe functionality for x86_32; the probed program segfaults instead of handling the probe proper.

The reason for this is that TRAP_BP is set up as system interrupt gate (DPL3) in the early IDT and then replaced by a regular interrupt gate (DPL0) in idt_setup_traps(). The DPL0 restriction causes the int3 trap to fail with a #GP resulting in a SIGSEGV of the probed program.

On 64bit this does not cause a problem because the IDT entry is replaced with a system interrupt gate (DPL3) with interrupt stack afterwards.

Remove X86_TRAP_BP from the def_idts table which is used in idt_setup_traps(). Remove a redundant entry for X86_TRAP_NMI in def_idts while at it. Tested on both x86_64 and x86_32.

[ tglx: Amended changelog with a description of the root cause ]

Fixes b70543a0b2b6("x86/idt: Move regular trap init to tables")
Reported-and-tested-by Yonghong Song <[email protected]>
Signed-off-by Yonghong Song <[email protected]>
Signed-off-by Thomas Gleixner <[email protected]>
Cc [email protected]
Cc [email protected]
Cc [email protected]
Cc [email protected]
Cc [email protected]
Link https://lkml.kernel.org/r/[email protected] open here
---
 arch/x86/kernel/idt.c | 2 --
 1 file changed, 2 deletions(-)
arch/x86/kernel/idt.c +0 -2
index 6107ee1..014cb2f 100644
@@ -92,8 +92,6 @@ static const __initdata struct idt_data def_idts[] = {
 	INTG(X86_TRAP_DF,		double_fault),
 #endif
 	INTG(X86_TRAP_DB,		debug),
-	INTG(X86_TRAP_NMI,		nmi),
-	INTG(X86_TRAP_BP,		int3),
 
 #ifdef CONFIG_X86_MCE
 	INTG(X86_TRAP_MC,		&machine_check),