| Message ID | 1448622572-16900-6-git-send-email-pmladek@suse.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
> On Nov 27, 2015, at 19:09, Petr Mladek <pmladek@suse.com> wrote: > > Testing has shown that the backtrace sometimes does not fit > into the 4kB temporary buffer that is used in NMI context. > > The warnings are gone when I double the temporary buffer size. > > Note that this problem existed even in the x86-specific > implementation that was added by the commit a9edc8809328 > ("x86/nmi: Perform a safe NMI stack trace on all CPUs"). > Nobody noticed it because it did not print any warnings. > > Signed-off-by: Petr Mladek <pmladek@suse.com> > --- > kernel/printk/nmi.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/kernel/printk/nmi.c b/kernel/printk/nmi.c > index 8af1e4016719..6111644d5f01 100644 > --- a/kernel/printk/nmi.c > +++ b/kernel/printk/nmi.c > @@ -42,7 +42,7 @@ atomic_t nmi_message_lost; > struct nmi_seq_buf { > atomic_t len; /* length of written data */ > struct irq_work work; /* IRQ work that flushes the buffer */ > - unsigned char buffer[PAGE_SIZE - sizeof(atomic_t) - > + unsigned char buffer[2 * PAGE_SIZE - sizeof(atomic_t) - > sizeof(struct irq_work)]; > }; > why not define like this: union { struct {atomic_t len; struct irq_work work; } unsigned char buffer[PAGE_SIZE * 2] ; } we can make sure the union is 2 PAGE_SIZE . Thanks
From: yalin wang > Sent: 30 November 2015 16:42 > > On Nov 27, 2015, at 19:09, Petr Mladek <pmladek@suse.com> wrote: > > > > Testing has shown that the backtrace sometimes does not fit > > into the 4kB temporary buffer that is used in NMI context. > > > > The warnings are gone when I double the temporary buffer size. You are wasting a lot of memory for something that is infrequently used. There ought to be some way of copying partial tracebacks into the main buffer. David
On Wed 2015-12-02 16:20:41, David Laight wrote: > From: yalin wang > > Sent: 30 November 2015 16:42 > > > On Nov 27, 2015, at 19:09, Petr Mladek <pmladek@suse.com> wrote: > > > > > > Testing has shown that the backtrace sometimes does not fit > > > into the 4kB temporary buffer that is used in NMI context. > > > > > > The warnings are gone when I double the temporary buffer size. > > You are wasting a lot of memory for something that is infrequently used. > There ought to be some way of copying partial tracebacks into the > main buffer. I have already tried to use a separate ring buffer that might be shared between all CPUs. But it was rejected because it was too complex. See http://thread.gmane.org/gmane.linux.kernel/1700059/focus=1700066 If we would want to crate a lockless access to the main ring buffer, we would end up with something like kernel/trace/ring_buffer.c. It is even more complicated. And reading of the messages is pretty slow. Note that we already have this buffer allocated on x86 and arm. It is used there for printing backtrace from all CPUs. This patchset makes it usable for all NMI messages. But I'll make the size configurable in the next version. Thanks for review, Petr
On Mon 2015-11-30 08:42:04, yalin wang wrote: > > > On Nov 27, 2015, at 19:09, Petr Mladek <pmladek@suse.com> wrote: > > > > Testing has shown that the backtrace sometimes does not fit > > into the 4kB temporary buffer that is used in NMI context. > > > > The warnings are gone when I double the temporary buffer size. > > > > Note that this problem existed even in the x86-specific > > implementation that was added by the commit a9edc8809328 > > ("x86/nmi: Perform a safe NMI stack trace on all CPUs"). > > Nobody noticed it because it did not print any warnings. > > > > Signed-off-by: Petr Mladek <pmladek@suse.com> > > --- > > kernel/printk/nmi.c | 2 +- > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > diff --git a/kernel/printk/nmi.c b/kernel/printk/nmi.c > > index 8af1e4016719..6111644d5f01 100644 > > --- a/kernel/printk/nmi.c > > +++ b/kernel/printk/nmi.c > > @@ -42,7 +42,7 @@ atomic_t nmi_message_lost; > > struct nmi_seq_buf { > > atomic_t len; /* length of written data */ > > struct irq_work work; /* IRQ work that flushes the buffer */ > > - unsigned char buffer[PAGE_SIZE - sizeof(atomic_t) - > > + unsigned char buffer[2 * PAGE_SIZE - sizeof(atomic_t) - > > sizeof(struct irq_work)]; > > }; > > > > why not define like this: > > union { > struct {atomic_t len; > struct irq_work work; > } > unsigned char buffer[PAGE_SIZE * 2] ; > } > > we can make sure the union is 2 PAGE_SIZE . IMHO, this would add more confusion. It would just move the computation somewhere else. The union will have 2*PAGE_SIZE but the beginning of "buffer" will be shared with "len" and "work". Therefore, we would need to skip the beginning of the buffer when storing the data. By other words, we still will be able to use only (sizeof(buffer) - sizeof(atomic_t) - sizeof(struct irq_work)] of the "buffer". Or did I miss something, please? Best Regards, Petr
diff --git a/kernel/printk/nmi.c b/kernel/printk/nmi.c index 8af1e4016719..6111644d5f01 100644 --- a/kernel/printk/nmi.c +++ b/kernel/printk/nmi.c @@ -42,7 +42,7 @@ atomic_t nmi_message_lost; struct nmi_seq_buf { atomic_t len; /* length of written data */ struct irq_work work; /* IRQ work that flushes the buffer */ - unsigned char buffer[PAGE_SIZE - sizeof(atomic_t) - + unsigned char buffer[2 * PAGE_SIZE - sizeof(atomic_t) - sizeof(struct irq_work)]; }; static DEFINE_PER_CPU(struct nmi_seq_buf, nmi_print_seq);
Testing has shown that the backtrace sometimes does not fit into the 4kB temporary buffer that is used in NMI context. The warnings are gone when I double the temporary buffer size. Note that this problem existed even in the x86-specific implementation that was added by the commit a9edc8809328 ("x86/nmi: Perform a safe NMI stack trace on all CPUs"). Nobody noticed it because it did not print any warnings. Signed-off-by: Petr Mladek <pmladek@suse.com> --- kernel/printk/nmi.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)