Syslinux - Feb 2015 - [PATCH 0/1] dprintf: add debug console support

This patch adds support for printing messages through a debug console.  QEMU,
for example, supports this through the debugcon facility.  The benefit is that
it's *much* faster than printing over a serial port.

To print to I/O port 0x402 (the default used by SeaBIOS and OVMF),
add "-DDEBUG_IO_PORT=0x402 -DCORE_DEBUG=1" to the build CFLAGS.

To enable a debug console in QEMU that listens to I/O port 0x402 and prints to
stdio, add "-debugcon stdio -global isa-debugcon.iobase=0x402" to the
QEMU
command line.

Jonathan Boeing (1):
  dprintf: add debug console support

 com32/include/dprintf.h      |  2 +-
 com32/lib/dprintf.c          |  4 ++--
 com32/lib/vdprintf.c         | 21 ++++++++++++++++++++-
 core/lwip/src/netif/undiif.c |  2 +-
 mk/devel.mk                  |  1 +
 5 files changed, 25 insertions(+), 5 deletions(-)

-- 
2.3.0

Add a vdprintf implementation that prints to a QEMU/Bochs style debug
console.

A debug console is an I/O port; writes to it are trapped by the host
and printed to a target.

Signed-off-by: Jonathan Boeing <jonathan.n.boeing at gmail.com>
---
 com32/include/dprintf.h      |  2 +-
 com32/lib/dprintf.c          |  4 ++--
 com32/lib/vdprintf.c         | 21 ++++++++++++++++++++-
 core/lwip/src/netif/undiif.c |  2 +-
 mk/devel.mk                  |  1 +
 5 files changed, 25 insertions(+), 5 deletions(-)

diff --git a/com32/include/dprintf.h b/com32/include/dprintf.h
index b3f1b46..24cab7b 100644
--- a/com32/include/dprintf.h
+++ b/com32/include/dprintf.h
@@ -7,7 +7,7 @@
 
 #include <syslinux/debug.h>
 
-#if !defined(DEBUG_PORT) && !defined(DEBUG_STDIO)
+#if !defined(DEBUG_PORT) && !defined(DEBUG_STDIO) &&
!defined(DEBUG_IO_PORT)
 # undef CORE_DEBUG
 #endif
 
diff --git a/com32/lib/dprintf.c b/com32/lib/dprintf.c
index dea77b3..e9dacb3 100644
--- a/com32/lib/dprintf.c
+++ b/com32/lib/dprintf.c
@@ -5,7 +5,7 @@
 #include <stdio.h>
 #include <stdarg.h>
 
-#ifdef DEBUG_PORT
+#if defined(DEBUG_PORT) || defined(DEBUG_IO_PORT)
 
 void vdprintf(const char *, va_list);
 
@@ -18,4 +18,4 @@ void dprintf(const char *format, ...)
     va_end(ap);
 }
 
-#endif /* DEBUG_PORT */
+#endif /* DEBUG_PORT || DEBUG_IO_PORT */
diff --git a/com32/lib/vdprintf.c b/com32/lib/vdprintf.c
index bcf55bb..95c10e9 100644
--- a/com32/lib/vdprintf.c
+++ b/com32/lib/vdprintf.c
@@ -10,7 +10,26 @@
 #include <sys/io.h>
 #include <sys/cpu.h>
 
-#ifdef DEBUG_PORT
+#ifdef DEBUG_IO_PORT
+
+#define BUFFER_SIZE 0x100
+
+void vdprintf(const char *format, va_list ap)
+{
+    int rv;
+    char buffer[BUFFER_SIZE];
+    char *p;
+
+    rv = vsnprintf(buffer, BUFFER_SIZE, format, ap);
+    if (rv < 0)
+        return;
+
+    p = buffer;
+    while(*p)
+        outb(*p++, DEBUG_IO_PORT);
+}
+
+#elif defined DEBUG_PORT
 
 #define BUFFER_SIZE	4096
 
diff --git a/core/lwip/src/netif/undiif.c b/core/lwip/src/netif/undiif.c
index e62a984..05606e6 100644
--- a/core/lwip/src/netif/undiif.c
+++ b/core/lwip/src/netif/undiif.c
@@ -52,7 +52,7 @@
 # ifndef DEBUG
 #  define DEBUG 1
 # endif
-# ifndef DEBUG_PORT
+# if !defined(DEBUG_PORT) && !defined(DEBUG_IO_PORT)
 #  define DEBUG_PORT 0x3f8
 # endif
 #endif /* UNDIIF_ID_FULL_DEBUG */
diff --git a/mk/devel.mk b/mk/devel.mk
index b1fca87..1e57dcf 100644
--- a/mk/devel.mk
+++ b/mk/devel.mk
@@ -2,6 +2,7 @@
 GCCWARN += -Wno-clobbered 
 #GCCWARN += -DDEBUG_MALLOC
 # GCCWARN += -DDEBUG_PORT=0x3f8 -DCORE_DEBUG=1
+# GCCWARN += -DDEBUG_IO_PORT=0x402 -DCORE_DEBUG=1
 GCCWARN += -DDYNAMIC_DEBUG
 
 ## The following will enable printing ethernet/arp/ip/icmp/tcp/udp headers
-- 
2.3.0

On Sun, Feb 08, 2015 at 09:13:02AM -0700, Jonathan Boeing via Syslinux
wrote:
> This patch adds support for printing messages through a debug console. 
QEMU,
> for example, supports this through the debugcon facility.  The benefit is
that
> it's *much* faster than printing over a serial port.
> 
> To print to I/O port 0x402 (the default used by SeaBIOS and OVMF),
> add "-DDEBUG_IO_PORT=0x402 -DCORE_DEBUG=1" to the build CFLAGS.
> 
> To enable a debug console in QEMU that listens to I/O port 0x402 and prints
to
> stdio, add "-debugcon stdio -global isa-debugcon.iobase=0x402" to
the QEMU
> command line.

When enabled, what will the effect on real hardware without the diagnostic
hardware ?

When enabled, what will the effect on real hardware ?

What (real) diagnostic hardware is adviced?



Oops, I should have start with:

Thank you for providing these development tools.


Cheers
Geert Stappers

On Sun, 8 Feb 2015 17:41:58 +0100
Geert Stappers via Syslinux <syslinux at zytor.com> wrote:
> On Sun, Feb 08, 2015 at 09:13:02AM -0700, Jonathan Boeing via
> Syslinux wrote:
> > This patch adds support for printing messages through a debug
> > console.  QEMU, for example, supports this through the debugcon
> > facility.  The benefit is that it's *much* faster than printing
> > over a serial port.
> > 
> > To print to I/O port 0x402 (the default used by SeaBIOS and OVMF),
> > add "-DDEBUG_IO_PORT=0x402 -DCORE_DEBUG=1" to the build
CFLAGS.
> > 
> > To enable a debug console in QEMU that listens to I/O port 0x402
> > and prints to stdio, add "-debugcon stdio -global
> > isa-debugcon.iobase=0x402" to the QEMU command line.
> 
> 
> When enabled, what will the effect on real hardware without the
> diagnostic hardware ?
> 
> When enabled, what will the effect on real hardware ?
> 
> What (real) diagnostic hardware is adviced?
> 
This doesn't support real hardware[1]; it's a VM-only thing.

If run on real hardware, it shouldn't have any effect[2] as long as it's
writing to an unused I/O port.  Port 0x402 seems to be unused (based on
some google searching) outside of debug consoles.  Bochs has a list at
http://bochs.sourceforge.net/techspec/PORTS.LST that shows it's in the
EISA range, but not specifically used.

So, it should be safe on real hardware, but I'd recommend it only for
testing and development on a VM.

Also, syslinux works correctly if it's enabled and run on a VM without a
debug console.  The writes are just silently discarded.



[1] Specialized (read expensive) CPU debugging hardware could capture
these writes.  I'm not sure if Intel System Studio can capture them,
but it's still $2399 for the JTAG-equipped version.

[2] It's CPU implementation specific, but it shouldn't have any visible
effect.

Regards,
Jonathan Boeing