Check the TIF_32BIT_FPREGS task setting of the tracee rather than the
tracer in determining the layout of floating-point general registers in
the floating-point context, correcting access to odd-numbered registers
for o32 tracees where the setting disagrees between the two processes.
Cc: [email protected] # 3.14+
Fixes: 597ce1723e0f ("MIPS: Support for 64-bit FP with O32 binaries")
Signed-off-by: Maciej W. Rozycki <[email protected]>
---
Hi,
These are not the usual requests used by GDB to access the floating-point
context, which is likely why it went unnoticed so long. They are only
used as a fallback in the case where PTRACE_GETFPREGS and PTRACE_SETFPREGS
requests are not supported, i.e. with ancient kernels.
However to verify an unrelated GDB bug fix I have tweaked GDB to always
use PTRACE_PEEKUSR and PTRACE_POKEUSR, and then discovered this issue in
native GDB regression testing, as it showed regressions from corrupt FGR
contents across numerous tests compared to the usual results. This fix
removed those regressions then.
Not being typically used does not mean we ought to keep the interface
broken. Therefore please apply.
Maciej
---
arch/mips/kernel/ptrace.c | 4 ++--
arch/mips/kernel/ptrace32.c | 4 ++--
2 files changed, 4 insertions(+), 4 deletions(-)
linux-mips-ptrace-test-thread-flag.diff
Index: linux/arch/mips/kernel/ptrace.c
===================================================================
--- linux.orig/arch/mips/kernel/ptrace.c 2018-05-12 22:52:19.000000000 +0100
+++ linux/arch/mips/kernel/ptrace.c 2018-05-12 22:56:07.893993000 +0100
@@ -1059,7 +1059,7 @@ long arch_ptrace(struct task_struct *chi
fregs = get_fpu_regs(child);
#ifdef CONFIG_32BIT
- if (test_thread_flag(TIF_32BIT_FPREGS)) {
+ if (test_tsk_thread_flag(child, TIF_32BIT_FPREGS)) {
/*
* The odd registers are actually the high
* order bits of the values stored in the even
@@ -1154,7 +1154,7 @@ long arch_ptrace(struct task_struct *chi
init_fp_ctx(child);
#ifdef CONFIG_32BIT
- if (test_thread_flag(TIF_32BIT_FPREGS)) {
+ if (test_tsk_thread_flag(child, TIF_32BIT_FPREGS)) {
/*
* The odd registers are actually the high
* order bits of the values stored in the even
Index: linux-mipsswbrd038/arch/mips/kernel/ptrace32.c
===================================================================
--- linux-mipsswbrd038.orig/arch/mips/kernel/ptrace32.c 2018-05-12 22:52:19.000000000 +0100
+++ linux-mipsswbrd038/arch/mips/kernel/ptrace32.c 2018-05-12 22:55:20.906637000 +0100
@@ -99,7 +99,7 @@ long compat_arch_ptrace(struct task_stru
break;
}
fregs = get_fpu_regs(child);
- if (test_thread_flag(TIF_32BIT_FPREGS)) {
+ if (test_tsk_thread_flag(child, TIF_32BIT_FPREGS)) {
/*
* The odd registers are actually the high
* order bits of the values stored in the even
@@ -212,7 +212,7 @@ long compat_arch_ptrace(struct task_stru
sizeof(child->thread.fpu));
child->thread.fpu.fcr31 = 0;
}
- if (test_thread_flag(TIF_32BIT_FPREGS)) {
+ if (test_tsk_thread_flag(child, TIF_32BIT_FPREGS)) {
/*
* The odd registers are actually the high
* order bits of the values stored in the even
On Mon, May 14, 2018 at 04:49:43PM +0100, Maciej W. Rozycki wrote:
> Check the TIF_32BIT_FPREGS task setting of the tracee rather than the
> tracer in determining the layout of floating-point general registers in
> the floating-point context, correcting access to odd-numbered registers
> for o32 tracees where the setting disagrees between the two processes.
>
> Cc: [email protected] # 3.14+
> Fixes: 597ce1723e0f ("MIPS: Support for 64-bit FP with O32 binaries")
> Signed-off-by: Maciej W. Rozycki <[email protected]>
> ---
> Hi,
>
> These are not the usual requests used by GDB to access the floating-point
> context, which is likely why it went unnoticed so long. They are only
> used as a fallback in the case where PTRACE_GETFPREGS and PTRACE_SETFPREGS
> requests are not supported, i.e. with ancient kernels.
>
> However to verify an unrelated GDB bug fix I have tweaked GDB to always
> use PTRACE_PEEKUSR and PTRACE_POKEUSR, and then discovered this issue in
> native GDB regression testing, as it showed regressions from corrupt FGR
> contents across numerous tests compared to the usual results. This fix
> removed those regressions then.
>
> Not being typically used does not mean we ought to keep the interface
> broken. Therefore please apply.
Looks good. Applied to fixes branch for 4.17.
Thanks
James