| Message ID | 1545159239-30628-2-git-send-email-vgupta@synopsys.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | ARC show_regs fixes | expand |
Hi Vineet, Just curious: isn't that enough to use GFP_NOWAIT instead of GFP_KERNEL when we allocate page in show_regs()? As I can see x86 use print_vma_addr() in their show_signal_msg() function which allocate page with __get_free_page(GFP_NOWAIT); On Tue, 2018-12-18 at 10:53 -0800, Vineet Gupta wrote: > Use on-stack smaller buffers instead of dynamic pages. > > The motivation for this change was to address lockdep splat when > signal handling code calls show_regs (with preemption disabled) and > ARC show_regs calls into sleepable page allocator. > > > potentially unexpected fatal signal 11. > > BUG: sleeping function called from invalid context at ../mm/page_alloc.c:4317 > > in_atomic(): 1, irqs_disabled(): 0, pid: 57, name: segv > > no locks held by segv/57. > > Preemption disabled at: > > [<8182f17e>] get_signal+0x4a6/0x7c4 > > CPU: 0 PID: 57 Comm: segv Not tainted 4.17.0+ #23 > > > > Stack Trace: > > arc_unwind_core.constprop.1+0xd0/0xf4 > > __might_sleep+0x1f6/0x234 > > __get_free_pages+0x174/0xca0 > > show_regs+0x22/0x330 > > get_signal+0x4ac/0x7c4 # print_fatal_signals() -> preempt_disable() > > do_signal+0x30/0x224 > > resume_user_mode_begin+0x90/0xd8 > > Despite this, lockdep still barfs (see next change), but this patch > still has merit as in we use smaller/localized buffers now and there's > less instructoh trace to sift thru when debugging pesky issues. > > Signed-off-by: Vineet Gupta <vgupta@synopsys.com> > --- > arch/arc/kernel/troubleshoot.c | 22 +++++++++------------- > 1 file changed, 9 insertions(+), 13 deletions(-) > > diff --git a/arch/arc/kernel/troubleshoot.c b/arch/arc/kernel/troubleshoot.c > index e8d9fb452346..2885bec71fb8 100644 > --- a/arch/arc/kernel/troubleshoot.c > +++ b/arch/arc/kernel/troubleshoot.c > @@ -58,11 +58,12 @@ static void show_callee_regs(struct callee_regs *cregs) > print_reg_file(&(cregs->r13), 13); > } > > -static void print_task_path_n_nm(struct task_struct *tsk, char *buf) > +static void print_task_path_n_nm(struct task_struct *tsk) > { > char *path_nm = NULL; > struct mm_struct *mm; > struct file *exe_file; > + char buf[256]; > > mm = get_task_mm(tsk); > if (!mm) > @@ -80,10 +81,9 @@ static void print_task_path_n_nm(struct task_struct *tsk, char *buf) > pr_info("Path: %s\n", !IS_ERR(path_nm) ? path_nm : "?"); > } > > -static void show_faulting_vma(unsigned long address, char *buf) > +static void show_faulting_vma(unsigned long address) > { > struct vm_area_struct *vma; > - char *nm = buf; > struct mm_struct *active_mm = current->active_mm; > > /* can't use print_vma_addr() yet as it doesn't check for > @@ -96,8 +96,11 @@ static void show_faulting_vma(unsigned long address, char *buf) > * if the container VMA is not found > */ > if (vma && (vma->vm_start <= address)) { > + char buf[256]; > + char *nm = "?"; > + > if (vma->vm_file) { > - nm = file_path(vma->vm_file, buf, PAGE_SIZE - 1); > + nm = file_path(vma->vm_file, buf, 256-1); > if (IS_ERR(nm)) > nm = "?"; > } > @@ -173,13 +176,8 @@ void show_regs(struct pt_regs *regs) > { > struct task_struct *tsk = current; > struct callee_regs *cregs; > - char *buf; > - > - buf = (char *)__get_free_page(GFP_KERNEL); > - if (!buf) > - return; > > - print_task_path_n_nm(tsk, buf); > + print_task_path_n_nm(tsk); > show_regs_print_info(KERN_INFO); > > show_ecr_verbose(regs); > @@ -189,7 +187,7 @@ void show_regs(struct pt_regs *regs) > (void *)regs->blink, (void *)regs->ret); > > if (user_mode(regs)) > - show_faulting_vma(regs->ret, buf); /* faulting code, not data */ > + show_faulting_vma(regs->ret); /* faulting code, not data */ > > pr_info("[STAT32]: 0x%08lx", regs->status32); > > @@ -221,8 +219,6 @@ void show_regs(struct pt_regs *regs) > cregs = (struct callee_regs *)current->thread.callee_reg; > if (cregs) > show_callee_regs(cregs); > - > - free_page((unsigned long)buf); > } > > void show_kernel_fault_diag(const char *str, struct pt_regs *regs,
On 12/19/18 9:04 AM, Eugeniy Paltsev wrote: > Just curious: isn't that enough to use GFP_NOWAIT instead > of GFP_KERNEL when we allocate page in show_regs()? > > As I can see x86 use print_vma_addr() in their show_signal_msg() > function which allocate page with __get_free_page(GFP_NOWAIT); I'm not sure if lockdep will be happy with it still. At any rate, as explained in changelog, this still has merit, since the buffer is only needed for nested d_path calls, which are better served with a smaller on-stack buffer. For cases such as kernel crash, we want lesser code/traces in fault path to sift thru ! -Vineet
> On Dec 18, 2018, at 11:53 AM, Vineet Gupta <Vineet.Gupta1@synopsys.com> wrote: > > Use on-stack smaller buffers instead of dynamic pages. > > The motivation for this change was to address lockdep splat when > signal handling code calls show_regs (with preemption disabled) and > ARC show_regs calls into sleepable page allocator. > > | potentially unexpected fatal signal 11. > | BUG: sleeping function called from invalid context at ../mm/page_alloc.c:4317 > | in_atomic(): 1, irqs_disabled(): 0, pid: 57, name: segv > | no locks held by segv/57. > | Preemption disabled at: > | [<8182f17e>] get_signal+0x4a6/0x7c4 > | CPU: 0 PID: 57 Comm: segv Not tainted 4.17.0+ #23 > | > | Stack Trace: > | arc_unwind_core.constprop.1+0xd0/0xf4 > | __might_sleep+0x1f6/0x234 > | __get_free_pages+0x174/0xca0 > | show_regs+0x22/0x330 > | get_signal+0x4ac/0x7c4 # print_fatal_signals() -> preempt_disable() > | do_signal+0x30/0x224 > | resume_user_mode_begin+0x90/0xd8 > > Despite this, lockdep still barfs (see next change), but this patch > still has merit as in we use smaller/localized buffers now and there's > less instructoh trace to sift thru when debugging pesky issues. > > Signed-off-by: Vineet Gupta <vgupta@synopsys.com> I would rather see 256 as a #define somewhere rather than a magic number sprinkled around arch/arc/kernel/troubleshoot.c. Still, that's what the existing code does, so I suppose it's OK. Otherwise the change looks good. Reviewed-by: William Kucharski <william.kucharski@oracle.com>
On 12/19/18 12:46 PM, William Kucharski wrote: > I would rather see 256 as a #define somewhere rather than a magic number sprinkled > around arch/arc/kernel/troubleshoot.c. That bothered me as well, but I was too lazy to define one and the existing ones don't apply. PATH_MAX is 4K which will blow up the stack usage. > > Still, that's what the existing code does, so I suppose it's OK. I'll define one locally. > Otherwise the change looks good. Thx for taking a look. > Reviewed-by: William Kucharski <william.kucharski@oracle.com> I'll add this to the patch. Thx, -Vineet
On 12/19/18 9:04 AM, Eugeniy Paltsev wrote: > As I can see x86 use print_vma_addr() in their show_signal_msg() > function which allocate page with __get_free_page(GFP_NOWAIT); Indeed with that the __get_free_page() lockdep splat is gone. There's a different one now hence my other patch. | [ARCLinux]# ./segv-null-ptr | potentially unexpected fatal signal 11. | BUG: sleeping function called from invalid context at kernel/fork.c:1011 | in_atomic(): 1, irqs_disabled(): 0, pid: 70, name: segv-null-ptr | no locks held by segv-null-ptr/70. | CPU: 0 PID: 70 Comm: segv-null-ptr Not tainted 4.18.0+ #69 | | Stack Trace: | arc_unwind_core+0xcc/0x100 | ___might_sleep+0x17a/0x190 | mmput+0x16/0xb8 | show_regs+0x52/0x310 | get_signal+0x5ee/0x610 | do_signal+0x2c/0x218 | resume_user_mode_begin+0x90/0xd8
On Tue 18-12-18 10:53:58, Vineet Gupta wrote: > Use on-stack smaller buffers instead of dynamic pages. > > The motivation for this change was to address lockdep splat when > signal handling code calls show_regs (with preemption disabled) and > ARC show_regs calls into sleepable page allocator. > > | potentially unexpected fatal signal 11. > | BUG: sleeping function called from invalid context at ../mm/page_alloc.c:4317 > | in_atomic(): 1, irqs_disabled(): 0, pid: 57, name: segv > | no locks held by segv/57. > | Preemption disabled at: > | [<8182f17e>] get_signal+0x4a6/0x7c4 > | CPU: 0 PID: 57 Comm: segv Not tainted 4.17.0+ #23 > | > | Stack Trace: > | arc_unwind_core.constprop.1+0xd0/0xf4 > | __might_sleep+0x1f6/0x234 > | __get_free_pages+0x174/0xca0 > | show_regs+0x22/0x330 > | get_signal+0x4ac/0x7c4 # print_fatal_signals() -> preempt_disable() > | do_signal+0x30/0x224 > | resume_user_mode_begin+0x90/0xd8 > > Despite this, lockdep still barfs (see next change), but this patch > still has merit as in we use smaller/localized buffers now and there's > less instructoh trace to sift thru when debugging pesky issues. But show_regs is called from contexts which might be called from deep call chains (e.g WARN). Is it safe to allocate such a large stack there?
On 2018/12/20 10:16, Vineet Gupta wrote: > On 12/19/18 9:04 AM, Eugeniy Paltsev wrote: >> As I can see x86 use print_vma_addr() in their show_signal_msg() >> function which allocate page with __get_free_page(GFP_NOWAIT); > > Indeed with that the __get_free_page() lockdep splat is gone. > > There's a different one now hence my other patch. > > | [ARCLinux]# ./segv-null-ptr > | potentially unexpected fatal signal 11. > | BUG: sleeping function called from invalid context at kernel/fork.c:1011 > | in_atomic(): 1, irqs_disabled(): 0, pid: 70, name: segv-null-ptr > | no locks held by segv-null-ptr/70. > | CPU: 0 PID: 70 Comm: segv-null-ptr Not tainted 4.18.0+ #69 > | > | Stack Trace: > | arc_unwind_core+0xcc/0x100 > | ___might_sleep+0x17a/0x190 > | mmput+0x16/0xb8 Then, does mmput_async() help? > | show_regs+0x52/0x310 > | get_signal+0x5ee/0x610 > | do_signal+0x2c/0x218 > | resume_user_mode_begin+0x90/0xd8
On 12/20/18 5:30 AM, Tetsuo Handa wrote: >> | mmput+0x16/0xb8 > Then, does mmput_async() help? Probably, I can try.
On 12/20/18 4:57 AM, Michal Hocko wrote: >> Despite this, lockdep still barfs (see next change), but this patch >> still has merit as in we use smaller/localized buffers now and there's >> less instructoh trace to sift thru when debugging pesky issues. > But show_regs is called from contexts which might be called from deep > call chains (e.g WARN). Is it safe to allocate such a large stack there? ARC has 8K pages and 256 additional bytes of stack usage doesn't seem absurdly high to me ! -Vineet
On 12/20/18 5:30 AM, Tetsuo Handa wrote: >> | Stack Trace: >> | arc_unwind_core+0xcc/0x100 >> | ___might_sleep+0x17a/0x190 >> | mmput+0x16/0xb8 > Then, does mmput_async() help? > It helps, but then we get the next one (w/o my patch 2/2) BUG: sleeping function called from invalid context at kernel/locking/rwsem.c:23 in_atomic(): 1, irqs_disabled(): 0, pid: 69, name: segv-null-ptr no locks held by segv-null-ptr/69. CPU: 0 PID: 69 Comm: segv-null-ptr Not tainted 4.18.0+ #72 Stack Trace: arc_unwind_core+0xcc/0x100 ___might_sleep+0x17a/0x190 down_read+0x18/0x38 show_regs+0x102/0x310 get_signal+0x5ee/0x610 do_signal+0x2c/0x218 resume_user_mode_begin+0x90/0xd8 @off 0x103d4 in [/segv-null-pt
diff --git a/arch/arc/kernel/troubleshoot.c b/arch/arc/kernel/troubleshoot.c index e8d9fb452346..2885bec71fb8 100644 --- a/arch/arc/kernel/troubleshoot.c +++ b/arch/arc/kernel/troubleshoot.c @@ -58,11 +58,12 @@ static void show_callee_regs(struct callee_regs *cregs) print_reg_file(&(cregs->r13), 13); } -static void print_task_path_n_nm(struct task_struct *tsk, char *buf) +static void print_task_path_n_nm(struct task_struct *tsk) { char *path_nm = NULL; struct mm_struct *mm; struct file *exe_file; + char buf[256]; mm = get_task_mm(tsk); if (!mm) @@ -80,10 +81,9 @@ static void print_task_path_n_nm(struct task_struct *tsk, char *buf) pr_info("Path: %s\n", !IS_ERR(path_nm) ? path_nm : "?"); } -static void show_faulting_vma(unsigned long address, char *buf) +static void show_faulting_vma(unsigned long address) { struct vm_area_struct *vma; - char *nm = buf; struct mm_struct *active_mm = current->active_mm; /* can't use print_vma_addr() yet as it doesn't check for @@ -96,8 +96,11 @@ static void show_faulting_vma(unsigned long address, char *buf) * if the container VMA is not found */ if (vma && (vma->vm_start <= address)) { + char buf[256]; + char *nm = "?"; + if (vma->vm_file) { - nm = file_path(vma->vm_file, buf, PAGE_SIZE - 1); + nm = file_path(vma->vm_file, buf, 256-1); if (IS_ERR(nm)) nm = "?"; } @@ -173,13 +176,8 @@ void show_regs(struct pt_regs *regs) { struct task_struct *tsk = current; struct callee_regs *cregs; - char *buf; - - buf = (char *)__get_free_page(GFP_KERNEL); - if (!buf) - return; - print_task_path_n_nm(tsk, buf); + print_task_path_n_nm(tsk); show_regs_print_info(KERN_INFO); show_ecr_verbose(regs); @@ -189,7 +187,7 @@ void show_regs(struct pt_regs *regs) (void *)regs->blink, (void *)regs->ret); if (user_mode(regs)) - show_faulting_vma(regs->ret, buf); /* faulting code, not data */ + show_faulting_vma(regs->ret); /* faulting code, not data */ pr_info("[STAT32]: 0x%08lx", regs->status32); @@ -221,8 +219,6 @@ void show_regs(struct pt_regs *regs) cregs = (struct callee_regs *)current->thread.callee_reg; if (cregs) show_callee_regs(cregs); - - free_page((unsigned long)buf); } void show_kernel_fault_diag(const char *str, struct pt_regs *regs,
Use on-stack smaller buffers instead of dynamic pages. The motivation for this change was to address lockdep splat when signal handling code calls show_regs (with preemption disabled) and ARC show_regs calls into sleepable page allocator. | potentially unexpected fatal signal 11. | BUG: sleeping function called from invalid context at ../mm/page_alloc.c:4317 | in_atomic(): 1, irqs_disabled(): 0, pid: 57, name: segv | no locks held by segv/57. | Preemption disabled at: | [<8182f17e>] get_signal+0x4a6/0x7c4 | CPU: 0 PID: 57 Comm: segv Not tainted 4.17.0+ #23 | | Stack Trace: | arc_unwind_core.constprop.1+0xd0/0xf4 | __might_sleep+0x1f6/0x234 | __get_free_pages+0x174/0xca0 | show_regs+0x22/0x330 | get_signal+0x4ac/0x7c4 # print_fatal_signals() -> preempt_disable() | do_signal+0x30/0x224 | resume_user_mode_begin+0x90/0xd8 Despite this, lockdep still barfs (see next change), but this patch still has merit as in we use smaller/localized buffers now and there's less instructoh trace to sift thru when debugging pesky issues. Signed-off-by: Vineet Gupta <vgupta@synopsys.com> --- arch/arc/kernel/troubleshoot.c | 22 +++++++++------------- 1 file changed, 9 insertions(+), 13 deletions(-)