Re: [PATCH v12 01/10] arm64: Add HAVE_REGS_AND_STACK_ACCESS_API feature

From: David Long
Date: Fri May 13 2016 - 15:08:03 EST


On 04/28/2016 12:08 PM, Marc Zyngier wrote:
On 27/04/16 19:52, David Long wrote:
From: "David A. Long" <dave.long@xxxxxxxxxx>

Add HAVE_REGS_AND_STACK_ACCESS_API feature for arm64.

And clearly a lot more.


Signed-off-by: David A. Long <dave.long@xxxxxxxxxx>
---
arch/arm64/Kconfig | 1 +
arch/arm64/include/asm/ptrace.h | 33 ++++++++++-
arch/arm64/kernel/ptrace.c | 118 ++++++++++++++++++++++++++++++++++++++++
3 files changed, 151 insertions(+), 1 deletion(-)

diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
index 4f43622..8f662fd 100644
--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig
@@ -80,6 +80,7 @@ config ARM64
select HAVE_PERF_EVENTS
select HAVE_PERF_REGS
select HAVE_PERF_USER_STACK_DUMP
+ select HAVE_REGS_AND_STACK_ACCESS_API
select HAVE_RCU_TABLE_FREE
select HAVE_SYSCALL_TRACEPOINTS
select IOMMU_DMA if IOMMU_SUPPORT
diff --git a/arch/arm64/include/asm/ptrace.h b/arch/arm64/include/asm/ptrace.h
index a307eb6..ee02637 100644
--- a/arch/arm64/include/asm/ptrace.h
+++ b/arch/arm64/include/asm/ptrace.h
@@ -119,6 +119,8 @@ struct pt_regs {
u64 syscallno;
};

+#define MAX_REG_OFFSET offsetof(struct pt_regs, pstate)
+
#define arch_has_single_step() (1)

#ifdef CONFIG_COMPAT
@@ -147,6 +149,35 @@ struct pt_regs {
#define user_stack_pointer(regs) \
(!compat_user_mode(regs) ? (regs)->sp : (regs)->compat_sp)

+extern int regs_query_register_offset(const char *name);
+extern const char *regs_query_register_name(unsigned int offset);
+extern bool regs_within_kernel_stack(struct pt_regs *regs, unsigned long addr);
+extern unsigned long regs_get_kernel_stack_nth(struct pt_regs *regs,
+ unsigned int n);
+
+/**
+ * regs_get_register() - get register value from its offset
+ * @regs: pt_regs from which register value is gotten
+ * @offset: offset number of the register.

Is it the offset? or the number?


I've removed "number" from the description.

+ *
+ * regs_get_register returns the value of a register whose offset from @regs.
+ * The @offset is the offset of the register in struct pt_regs.
+ * If @offset is bigger than MAX_REG_OFFSET, this returns 0.
+ */
+static inline u64 regs_get_register(struct pt_regs *regs,
+ unsigned int offset)
+{
+ if (unlikely(offset > MAX_REG_OFFSET))
+ return 0;
+ return *(u64 *)((u64)regs + offset);

So clearly it is the offset. But is 3 a valid value? I don't think so.
How about something slightly more type safe:

u64 val = 0;

WARN_ON(offset & 7);

offset >>= 3;
switch (offset) {
case 0 ... 30:
val = regs->reg[offset];
break;
case 31:
val = regs->sp;
break;
case 32:
val = regs->pc;
break;
case 33:
val = regs->pstate;
break;
}

return val;

I'm pretty sure you could replace 31/32/33 with macros using offsetof().
The compiler may even optimize this to something similar to what you
already have.

Thanks,

M.


I'm not sure this makes sense to me since my best understanding of this suite of functions makes the "offset" argument effectively an opaque type with a value returned from looking up the register by name in a compiled in table that reliably determines the offset within the pt_regs structure (i.e.: regs_query_register_offset() and regs_query_register_name()). Now this change introduces the assumption that the gp registers start at the beginning of the structure, presumably for the future use of someone hardcoding their own number for "offset".

Nonetheless in the interest of moving forward I currently have this change sitting in my next patch set.

Thanks,
-dl