| Message ID | 20180301193803.129195-1-dianders@chromium.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Thu, Mar 01, 2018 at 11:38:03AM -0800, Douglas Anderson wrote: > This is the equivalent of commit 001bf455d206 ("ARM: 8428/1: kgdb: Fix > registers on sleeping tasks") but for arm64. Nuff said. It's a pity that 001bf455d206 doesn't explain *why* past_pt_regs doesn't work. Anyway, does this mean you're actually using kgdb on arm64? Does the rest of it appear to work? Will > diff --git a/arch/arm64/kernel/kgdb.c b/arch/arm64/kernel/kgdb.c > index 2122cd187f19..01285d4dcdc3 100644 > --- a/arch/arm64/kernel/kgdb.c > +++ b/arch/arm64/kernel/kgdb.c > @@ -138,14 +138,26 @@ int dbg_set_reg(int regno, void *mem, struct pt_regs *regs) > void > sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *task) > { > - struct pt_regs *thread_regs; > + struct thread_struct *thread = &task->thread; > + struct cpu_context *cpu_context = &thread->cpu_context; > > /* Initialize to zero */ > memset((char *)gdb_regs, 0, NUMREGBYTES); > - thread_regs = task_pt_regs(task); > - memcpy((void *)gdb_regs, (void *)thread_regs->regs, GP_REG_BYTES); > - /* Special case for PSTATE (check comments in asm/kgdb.h for details) */ > - dbg_get_reg(33, gdb_regs + GP_REG_BYTES, thread_regs); > + > + gdb_regs[19] = cpu_context->x19; > + gdb_regs[20] = cpu_context->x20; > + gdb_regs[21] = cpu_context->x21; > + gdb_regs[22] = cpu_context->x22; > + gdb_regs[23] = cpu_context->x23; > + gdb_regs[24] = cpu_context->x24; > + gdb_regs[25] = cpu_context->x25; > + gdb_regs[26] = cpu_context->x26; > + gdb_regs[27] = cpu_context->x27; > + gdb_regs[28] = cpu_context->x28; > + gdb_regs[29] = cpu_context->fp; > + > + gdb_regs[31] = cpu_context->sp; > + gdb_regs[32] = cpu_context->pc; > } > > void kgdb_arch_set_pc(struct pt_regs *regs, unsigned long pc) > -- > 2.16.2.395.g2e18187dfd-goog >
On Fri, Mar 02, 2018 at 06:01:31PM +0000, Will Deacon wrote: > On Thu, Mar 01, 2018 at 11:38:03AM -0800, Douglas Anderson wrote: > > This is the equivalent of commit 001bf455d206 ("ARM: 8428/1: kgdb: Fix > > registers on sleeping tasks") but for arm64. Nuff said. > > It's a pity that 001bf455d206 doesn't explain *why* past_pt_regs doesn't > work. The task_pt_regs are the userspace regs at the highest address on the kernel stack: #define task_pt_regs(p) \ ((struct pt_regs *)(THREAD_SIZE + task_stack_page(p)) - 1) ... for kernel tasks, that's meaningless, and for user tasks, that won't correspond to kernel state. > > diff --git a/arch/arm64/kernel/kgdb.c b/arch/arm64/kernel/kgdb.c > > index 2122cd187f19..01285d4dcdc3 100644 > > --- a/arch/arm64/kernel/kgdb.c > > +++ b/arch/arm64/kernel/kgdb.c > > @@ -138,14 +138,26 @@ int dbg_set_reg(int regno, void *mem, struct pt_regs *regs) > > void > > sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *task) > > { > > - struct pt_regs *thread_regs; > > + struct thread_struct *thread = &task->thread; > > + struct cpu_context *cpu_context = &thread->cpu_context; We can do this in one go: struct cpu_context *cpu_context = &task->thread.cpu_context; ... since we don't need the thread variable otherwise. > > > > /* Initialize to zero */ > > memset((char *)gdb_regs, 0, NUMREGBYTES); > > - thread_regs = task_pt_regs(task); > > - memcpy((void *)gdb_regs, (void *)thread_regs->regs, GP_REG_BYTES); > > - /* Special case for PSTATE (check comments in asm/kgdb.h for details) */ > > - dbg_get_reg(33, gdb_regs + GP_REG_BYTES, thread_regs); > > + > > + gdb_regs[19] = cpu_context->x19; > > + gdb_regs[20] = cpu_context->x20; > > + gdb_regs[21] = cpu_context->x21; > > + gdb_regs[22] = cpu_context->x22; > > + gdb_regs[23] = cpu_context->x23; > > + gdb_regs[24] = cpu_context->x24; > > + gdb_regs[25] = cpu_context->x25; > > + gdb_regs[26] = cpu_context->x26; > > + gdb_regs[27] = cpu_context->x27; > > + gdb_regs[28] = cpu_context->x28; > > + gdb_regs[29] = cpu_context->fp; > > + > > + gdb_regs[31] = cpu_context->sp; > > + gdb_regs[32] = cpu_context->pc; Are the other reg fields initialised elsewhere? We might want to zero them here. Thanks, Mark.
Hi, On Fri, Mar 2, 2018 at 10:01 AM, Will Deacon <will.deacon@arm.com> wrote: > On Thu, Mar 01, 2018 at 11:38:03AM -0800, Douglas Anderson wrote: >> This is the equivalent of commit 001bf455d206 ("ARM: 8428/1: kgdb: Fix >> registers on sleeping tasks") but for arm64. Nuff said. > > It's a pity that 001bf455d206 doesn't explain *why* past_pt_regs doesn't > work. Ah, sorry. I can add that to the description but Mark Rutland also explained it. task_pt_regs are the userspace registers. That's not what kgdb needs. > Anyway, does this mean you're actually using kgdb on arm64? Does the rest of > it appear to work? Yeah, I've been using it for years. I had this patch on my list to upstream but it conflicted with commit 0d15ef677839 ("arm64: kgdb: Match pstate size with gdbserver protocol") and I never had time to figure out if that mattered, so it's been sitting at http://crosreview.com/371942 forever. NOTE: in general I haven't been using single step and breakpoints since those have always been a bit iffy on serial kgdb and especially since we have read-only text on our system, but maybe I'll try that now that I'm running with a newer kernel. -Doug
Hi, On Fri, Mar 2, 2018 at 10:16 AM, Mark Rutland <mark.rutland@arm.com> wrote: > On Fri, Mar 02, 2018 at 06:01:31PM +0000, Will Deacon wrote: >> On Thu, Mar 01, 2018 at 11:38:03AM -0800, Douglas Anderson wrote: >> > This is the equivalent of commit 001bf455d206 ("ARM: 8428/1: kgdb: Fix >> > registers on sleeping tasks") but for arm64. Nuff said. >> >> It's a pity that 001bf455d206 doesn't explain *why* past_pt_regs doesn't >> work. > > The task_pt_regs are the userspace regs at the highest address on the > kernel stack: > > #define task_pt_regs(p) \ > ((struct pt_regs *)(THREAD_SIZE + task_stack_page(p)) - 1) > > ... for kernel tasks, that's meaningless, and for user tasks, that won't > correspond to kernel state. > >> > diff --git a/arch/arm64/kernel/kgdb.c b/arch/arm64/kernel/kgdb.c >> > index 2122cd187f19..01285d4dcdc3 100644 >> > --- a/arch/arm64/kernel/kgdb.c >> > +++ b/arch/arm64/kernel/kgdb.c >> > @@ -138,14 +138,26 @@ int dbg_set_reg(int regno, void *mem, struct pt_regs *regs) >> > void >> > sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *task) >> > { >> > - struct pt_regs *thread_regs; >> > + struct thread_struct *thread = &task->thread; >> > + struct cpu_context *cpu_context = &thread->cpu_context; > > We can do this in one go: > > struct cpu_context *cpu_context = &task->thread.cpu_context; > > ... since we don't need the thread variable otherwise. Sure, I can spin it that way if it makes you happy. >> > >> > /* Initialize to zero */ >> > memset((char *)gdb_regs, 0, NUMREGBYTES); >> > - thread_regs = task_pt_regs(task); >> > - memcpy((void *)gdb_regs, (void *)thread_regs->regs, GP_REG_BYTES); >> > - /* Special case for PSTATE (check comments in asm/kgdb.h for details) */ >> > - dbg_get_reg(33, gdb_regs + GP_REG_BYTES, thread_regs); >> > + >> > + gdb_regs[19] = cpu_context->x19; >> > + gdb_regs[20] = cpu_context->x20; >> > + gdb_regs[21] = cpu_context->x21; >> > + gdb_regs[22] = cpu_context->x22; >> > + gdb_regs[23] = cpu_context->x23; >> > + gdb_regs[24] = cpu_context->x24; >> > + gdb_regs[25] = cpu_context->x25; >> > + gdb_regs[26] = cpu_context->x26; >> > + gdb_regs[27] = cpu_context->x27; >> > + gdb_regs[28] = cpu_context->x28; >> > + gdb_regs[29] = cpu_context->fp; >> > + >> > + gdb_regs[31] = cpu_context->sp; >> > + gdb_regs[32] = cpu_context->pc; > > Are the other reg fields initialised elsewhere? > > We might want to zero them here. Isn't that covered by the the "/* Initialize to zero */" comment and and "memset((char *)gdb_regs, 0, NUMREGBYTES);" ...or are you thinking of initting something else to zero? -Doug
On Fri, Mar 02, 2018 at 10:45:25AM -0800, Doug Anderson wrote: > Hi, > > On Fri, Mar 2, 2018 at 10:16 AM, Mark Rutland <mark.rutland@arm.com> wrote: > > On Fri, Mar 02, 2018 at 06:01:31PM +0000, Will Deacon wrote: > >> On Thu, Mar 01, 2018 at 11:38:03AM -0800, Douglas Anderson wrote: > >> > /* Initialize to zero */ > >> > memset((char *)gdb_regs, 0, NUMREGBYTES); > >> > - thread_regs = task_pt_regs(task); > >> > - memcpy((void *)gdb_regs, (void *)thread_regs->regs, GP_REG_BYTES); > >> > - /* Special case for PSTATE (check comments in asm/kgdb.h for details) */ > >> > - dbg_get_reg(33, gdb_regs + GP_REG_BYTES, thread_regs); > >> > + > >> > + gdb_regs[19] = cpu_context->x19; > >> > + gdb_regs[20] = cpu_context->x20; > >> > + gdb_regs[21] = cpu_context->x21; > >> > + gdb_regs[22] = cpu_context->x22; > >> > + gdb_regs[23] = cpu_context->x23; > >> > + gdb_regs[24] = cpu_context->x24; > >> > + gdb_regs[25] = cpu_context->x25; > >> > + gdb_regs[26] = cpu_context->x26; > >> > + gdb_regs[27] = cpu_context->x27; > >> > + gdb_regs[28] = cpu_context->x28; > >> > + gdb_regs[29] = cpu_context->fp; > >> > + > >> > + gdb_regs[31] = cpu_context->sp; > >> > + gdb_regs[32] = cpu_context->pc; > > > > Are the other reg fields initialised elsewhere? > > > > We might want to zero them here. > > Isn't that covered by the the "/* Initialize to zero */" comment and > and "memset((char *)gdb_regs, 0, NUMREGBYTES);" I'd misread the patch and thought that was part of the deleted lines. That looks fine, then. Mark.
diff --git a/arch/arm64/kernel/kgdb.c b/arch/arm64/kernel/kgdb.c index 2122cd187f19..01285d4dcdc3 100644 --- a/arch/arm64/kernel/kgdb.c +++ b/arch/arm64/kernel/kgdb.c @@ -138,14 +138,26 @@ int dbg_set_reg(int regno, void *mem, struct pt_regs *regs) void sleeping_thread_to_gdb_regs(unsigned long *gdb_regs, struct task_struct *task) { - struct pt_regs *thread_regs; + struct thread_struct *thread = &task->thread; + struct cpu_context *cpu_context = &thread->cpu_context; /* Initialize to zero */ memset((char *)gdb_regs, 0, NUMREGBYTES); - thread_regs = task_pt_regs(task); - memcpy((void *)gdb_regs, (void *)thread_regs->regs, GP_REG_BYTES); - /* Special case for PSTATE (check comments in asm/kgdb.h for details) */ - dbg_get_reg(33, gdb_regs + GP_REG_BYTES, thread_regs); + + gdb_regs[19] = cpu_context->x19; + gdb_regs[20] = cpu_context->x20; + gdb_regs[21] = cpu_context->x21; + gdb_regs[22] = cpu_context->x22; + gdb_regs[23] = cpu_context->x23; + gdb_regs[24] = cpu_context->x24; + gdb_regs[25] = cpu_context->x25; + gdb_regs[26] = cpu_context->x26; + gdb_regs[27] = cpu_context->x27; + gdb_regs[28] = cpu_context->x28; + gdb_regs[29] = cpu_context->fp; + + gdb_regs[31] = cpu_context->sp; + gdb_regs[32] = cpu_context->pc; } void kgdb_arch_set_pc(struct pt_regs *regs, unsigned long pc)
This is the equivalent of commit 001bf455d206 ("ARM: 8428/1: kgdb: Fix registers on sleeping tasks") but for arm64. Nuff said. Signed-off-by: Douglas Anderson <dianders@chromium.org> --- arch/arm64/kernel/kgdb.c | 22 +++++++++++++++++----- 1 file changed, 17 insertions(+), 5 deletions(-)