Re: [PATCH 10/10] powerpc: move norestart trap flag to bit 0

[Christophe Leroy]

Le 16/03/2021 à 08:11, Nicholas Piggin a écrit :

Excerpts from Christophe Leroy's message of March 15, 2021 6:14 pm:

Le 15/03/2021 à 04:17, Nicholas Piggin a écrit :

Compact the trap flags down to use the low 4 bits of regs.trap.

A few 64e interrupt trap numbers set bit 4. Although they tended to be
trivial so it wasn't a real problem[1], it is not the right thing to do,
and confusing.

[*] E.g., 0x310 hypercall goes to unknown_exception, which prints
      regs->trap directly so 0x310 will appear fine, and only the syscall
      interrupt will test norestart, so it won't be confused by 0x310.

Signed-off-by: Nicholas Piggin <npig...@gmail.com>
---
   arch/powerpc/include/asm/ptrace.h | 14 ++++++++++----
   1 file changed, 10 insertions(+), 4 deletions(-)

diff --git a/arch/powerpc/include/asm/ptrace.h 
b/arch/powerpc/include/asm/ptrace.h
index 91194fdd5d01..6a04abfe5eb6 100644
--- a/arch/powerpc/include/asm/ptrace.h
+++ b/arch/powerpc/include/asm/ptrace.h
@@ -185,15 +185,21 @@ static inline void regs_set_return_value(struct pt_regs 
*regs, unsigned long rc)
   #define current_pt_regs() \
        ((struct pt_regs *)((unsigned long)task_stack_page(current) + 
THREAD_SIZE) - 1)

+/*

+ * The 4 low bits (0xf) are available as flags to overload the trap word,
+ * because interrupt vectors have minimum alignment of 0x10. TRAP_FLAGS_MASK
+ * must cover the bits used as flags, including bit 0 which is used as the
+ * "norestart" bit.
+ */
   #ifdef __powerpc64__
-#define TRAP_FLAGS_MASK                0x10
+#define TRAP_FLAGS_MASK                0x1
   #define TRAP(regs)           ((regs)->trap & ~TRAP_FLAGS_MASK)
   #else
   /*
    * On 4xx we use bit 1 in the trap word to indicate whether the exception
    * is a critical exception (1 means it is).
    */
-#define TRAP_FLAGS_MASK                0x1E
+#define TRAP_FLAGS_MASK                0xf

Could we set 0xf for all and remove the ifdef __powerpc64__ ?

I like that it documents the bit number allocation so I prefer to leave
it, but TRAP() does not have to be defined twice at least.

   #define TRAP(regs)           ((regs)->trap & ~TRAP_FLAGS_MASK)
   #define IS_CRITICAL_EXC(regs)        (((regs)->trap & 2) != 0)
   #define IS_MCHECK_EXC(regs)  (((regs)->trap & 4) != 0)
@@ -222,12 +228,12 @@ static inline bool trap_is_syscall(struct pt_regs *regs)

static inline bool trap_norestart(struct pt_regs *regs)

   {
-       return regs->trap & 0x10;
+       return regs->trap & 0x1;
   }

static inline void set_trap_norestart(struct pt_regs *regs)

   {
-       regs->trap |= 0x10;
+       regs->trap |= 0x1;
   }

#define arch_has_single_step() (1)

While we are playing with ->trap, in mm/book3s64/hash_utils.c there is an if 
(regs->trap == 0x400).
Should be TRAP(regs) == 0x400 ?

Yes I would say so, if you want to do a patch you can add
Acked-by: Nicholas Piggin <npig...@gmail.com>

Otherwise I can do it.

Yes please do.

Thanks
Christophe