[PATCH 1/2] powerpc: Use seq_buf to avoid pr_cont() in __die()

Using pr_cont() risks having our output interleaved with other output
from other CPUs. Instead use a seq_buf to construct the line and then
print it as a whole.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
---
 arch/powerpc/kernel/traps.c | 23 ++++++++++++++++-------
 1 file changed, 16 insertions(+), 7 deletions(-)

diff --git a/arch/powerpc/kernel/traps.c b/arch/powerpc/kernel/traps.c
index 64936b60d521..431a86d3f772 100644
--- a/arch/powerpc/kernel/traps.c
+++ b/arch/powerpc/kernel/traps.c

@@ -38,6 +38,7 @@
 #include <linux/kdebug.h>
 #include <linux/ratelimit.h>
 #include <linux/context_tracking.h>
+#include <linux/seq_buf.h>
 #include <linux/smp.h>
 #include <linux/console.h>
 #include <linux/kmsg_dump.h>

@@ -255,26 +256,34 @@ NOKPROBE_SYMBOL(oops_end);
 
 static int __die(const char *str, struct pt_regs *regs, long err)
 {
+	char buf[128]; /* enough for all flags and a long platform name */
+	struct seq_buf s;
+
+	seq_buf_init(&s, buf, sizeof(buf));
+
 	printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
 
 	if (IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN))
-		printk("LE ");
+		seq_buf_puts(&s, "LE ");
 	else
-		printk("BE ");
+		seq_buf_puts(&s, "BE ");
 
 	if (IS_ENABLED(CONFIG_PREEMPT))
-		pr_cont("PREEMPT ");
+		seq_buf_puts(&s, "PREEMPT ");
 
 	if (IS_ENABLED(CONFIG_SMP))
-		pr_cont("SMP NR_CPUS=%d ", NR_CPUS);
+		seq_buf_printf(&s, "SMP NR_CPUS=%d ", NR_CPUS);
 
 	if (debug_pagealloc_enabled())
-		pr_cont("DEBUG_PAGEALLOC ");
+		seq_buf_puts(&s, "DEBUG_PAGEALLOC ");
 
 	if (IS_ENABLED(CONFIG_NUMA))
-		pr_cont("NUMA ");
+		seq_buf_puts(&s, "NUMA ");
+
+	if (ppc_md.name)
+		seq_buf_puts(&s, ppc_md.name);
 
-	pr_cont("%s\n", ppc_md.name ? ppc_md.name : "");
+	printk("%s\n", buf);
 
 	if (notify_die(DIE_OOPS, str, regs, err, 255, SIGSEGV) == NOTIFY_STOP)
 		return 1;

-- 
2.20.1

The page size the kernel is built with is useful info when debugging a
crash, so add it to the output in __die().

Result looks like eg:

  kernel BUG at drivers/misc/lkdtm/bugs.c:63!
  Oops: Exception in kernel mode, sig: 5 [#1]
  LE PAGE_SIZE=64K SMP NR_CPUS=2048 NUMA pSeries
  Modules linked in: vmx_crypto kvm binfmt_misc ip_tables

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
---
 arch/powerpc/kernel/traps.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/arch/powerpc/kernel/traps.c b/arch/powerpc/kernel/traps.c
index 431a86d3f772..fc972e4eee5f 100644
--- a/arch/powerpc/kernel/traps.c
+++ b/arch/powerpc/kernel/traps.c

@@ -268,6 +268,18 @@ static int __die(const char *str, struct pt_regs *regs, long err)
 	else
 		seq_buf_puts(&s, "BE ");
 
+	seq_buf_puts(&s, "PAGE_SIZE=");
+	if (IS_ENABLED(CONFIG_PPC_4K_PAGES))
+		seq_buf_puts(&s, "4K ");
+	else if (IS_ENABLED(CONFIG_PPC_16K_PAGES))
+		seq_buf_puts(&s, "16K ");
+	else if (IS_ENABLED(CONFIG_PPC_64K_PAGES))
+		seq_buf_puts(&s, "64K ");
+	else if (IS_ENABLED(CONFIG_PPC_256K_PAGES))
+		seq_buf_puts(&s, "256K ");
+	else
+		BUILD_BUG_ON(1);
+
 	if (IS_ENABLED(CONFIG_PREEMPT))
 		seq_buf_puts(&s, "PREEMPT ");
 

-- 
2.20.1

Le 08/01/2019 à 13:04, Michael Ellerman a écrit :

Using pr_cont() risks having our output interleaved with other output
from other CPUs. Instead use a seq_buf to construct the line and then
print it as a whole.

Why not simply doing a single printk() or similar on the same model as X86 for instance ? (https://elixir.bootlin.com/linux/v5.0-rc1/source/arch/x86/kernel/dumpstack.c#L368)

Christophe

Le 08/01/2019 à 13:05, Michael Ellerman a écrit :

Can't we build all the above at once using PAGE_SHIFT ?

Something like (untested):

"%dK ", 1 << (PAGE_SHIFT - 10)

Christophe

Le 08/01/2019 à 13:21, Christophe Leroy a écrit :

Le 08/01/2019 à 13:05, Michael Ellerman a écrit :

quoted

The page size the kernel is built with is useful info when debugging a
crash, so add it to the output in __die().

Result looks like eg:

   kernel BUG at drivers/misc/lkdtm/bugs.c:63!
   Oops: Exception in kernel mode, sig: 5 [#1]
   LE PAGE_SIZE=64K SMP NR_CPUS=2048 NUMA pSeries
   Modules linked in: vmx_crypto kvm binfmt_misc ip_tables

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
---
  arch/powerpc/kernel/traps.c | 12 ++++++++++++
  1 file changed, 12 insertions(+)

diff --git a/arch/powerpc/kernel/traps.c b/arch/powerpc/kernel/traps.c
index 431a86d3f772..fc972e4eee5f 100644
--- a/arch/powerpc/kernel/traps.c
+++ b/arch/powerpc/kernel/traps.c

@@ -268,6 +268,18 @@ static int __die(const char *str, struct pt_regs *regs, long err)

      else
          seq_buf_puts(&s, "BE ");
+    seq_buf_puts(&s, "PAGE_SIZE=");
+    if (IS_ENABLED(CONFIG_PPC_4K_PAGES))
+        seq_buf_puts(&s, "4K ");
+    else if (IS_ENABLED(CONFIG_PPC_16K_PAGES))
+        seq_buf_puts(&s, "16K ");
+    else if (IS_ENABLED(CONFIG_PPC_64K_PAGES))
+        seq_buf_puts(&s, "64K ");
+    else if (IS_ENABLED(CONFIG_PPC_256K_PAGES))
+        seq_buf_puts(&s, "256K ");

Can't we build all the above at once using PAGE_SHIFT ?

Something like (untested):

"%dK ", 1 << (PAGE_SHIFT - 10)

Or even simplier:

"%dK ", PAGE_SIZE / 1024

Christophe

Christophe

quoted

+    else
+        BUILD_BUG_ON(1);
+
      if (IS_ENABLED(CONFIG_PREEMPT))
          seq_buf_puts(&s, "PREEMPT ");

Christophe Leroy [off-list ref] writes:

Le 08/01/2019 à 13:21, Christophe Leroy a écrit :

quoted

Le 08/01/2019 à 13:05, Michael Ellerman a écrit :

quoted

The page size the kernel is built with is useful info when debugging a
crash, so add it to the output in __die().

Result looks like eg:

   kernel BUG at drivers/misc/lkdtm/bugs.c:63!
   Oops: Exception in kernel mode, sig: 5 [#1]
   LE PAGE_SIZE=64K SMP NR_CPUS=2048 NUMA pSeries
   Modules linked in: vmx_crypto kvm binfmt_misc ip_tables

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
---
  arch/powerpc/kernel/traps.c | 12 ++++++++++++
  1 file changed, 12 insertions(+)

diff --git a/arch/powerpc/kernel/traps.c b/arch/powerpc/kernel/traps.c
index 431a86d3f772..fc972e4eee5f 100644
--- a/arch/powerpc/kernel/traps.c
+++ b/arch/powerpc/kernel/traps.c

@@ -268,6 +268,18 @@ static int __die(const char *str, struct pt_regs 

*regs, long err)
      else
          seq_buf_puts(&s, "BE ");
+    seq_buf_puts(&s, "PAGE_SIZE=");
+    if (IS_ENABLED(CONFIG_PPC_4K_PAGES))
+        seq_buf_puts(&s, "4K ");
+    else if (IS_ENABLED(CONFIG_PPC_16K_PAGES))
+        seq_buf_puts(&s, "16K ");
+    else if (IS_ENABLED(CONFIG_PPC_64K_PAGES))
+        seq_buf_puts(&s, "64K ");
+    else if (IS_ENABLED(CONFIG_PPC_256K_PAGES))
+        seq_buf_puts(&s, "256K ");

Can't we build all the above at once using PAGE_SHIFT ?

Something like (untested):

"%dK ", 1 << (PAGE_SHIFT - 10)

Or even simplier:

"%dK ", PAGE_SIZE / 1024

Yep, good point.

Clearly I have forgotten how to program over the break (if I ever knew).

cheers

Christophe Leroy [off-list ref] writes:

Le 08/01/2019 à 13:04, Michael Ellerman a écrit :

quoted

Using pr_cont() risks having our output interleaved with other output
from other CPUs. Instead use a seq_buf to construct the line and then
print it as a whole.

Why not simply doing a single printk() or similar on the same model as 
X86 for instance ? 
(https://elixir.bootlin.com/linux/v5.0-rc1/source/arch/x86/kernel/dumpstack.c#L368)

Yeah we could do it that way, though it can become a bit of a mess.

In this case I guess it's not *too* bad:

	printk("%s PAGE_SIZE=%luK%s%s%s%s%s %s\n",
	       IS_ENABLED(CONFIG_CPU_LITTLE_ENDIAN) ? "LE" : "BE",
	       PAGE_SIZE / 1024,
	       IS_ENABLED(CONFIG_PREEMPT) ? " PREEMPT" : "",
	       IS_ENABLED(CONFIG_SMP) ? " SMP" : "",
	       IS_ENABLED(CONFIG_SMP) ? (" NR_CPUS=" __stringify(NR_CPUS)) : "",
	       debug_pagealloc_enabled() ? " DEBUG_PAGEALLOC" : "",
	       IS_ENABLED(CONFIG_NUMA) ? " NUMA" : "",
	       ppc_md.name ? ppc_md.name : "");

And the generated code is obviously a lot smaller.

So yeah I'll go with that. Thanks for the review.

cheers