Michael Walker
2006-May-25 22:53 UTC
[syslinux] pxelinux/mboot confused about e820 memory maps on HP Proliant BIOS's
All,
When attempting to pxeboot the latest Xen3.0 hypervisor/kernel
it was failing with the following error:
------
\ \/ /___ _ __ |___ / / _ \ _ _ _ __ ___| |_ __ _| |__ | | ___
\ // _ \ '_ \ |_ \| | | |__| | | | '_ \/ __| __/ _` | '_ \| |/
_ \
/ \ __/ | | | ___) | |_| |__| |_| | | | \__ \ || (_| | |_) | | __/
/_/\_\___|_| |_| |____(_)___/ \__,_|_| |_|___/\__\__,_|_.__/|_|\___|
http://www.cl.cam.ac.uk/netos/xen
University of Cambridge Computer Laboratory
Xen version 3.0-unstable (root at rc.cassatt.com) (gcc version 3.4.3 20050227
(Red
Hat 3.4.3-22.1)) Thu May 18 14:14:42 PDT 2006
Latest ChangeSet: Wed May 17 23:15:36 2006 +0100 10005:86d8246c6aff
(XEN) Command line: kernels/xen-debug.gz console=vga
(XEN) WARNING: Buggy e820 map detected and fixed (truncated length fields).
(XEN) Physical RAM map:
(XEN) Not enough memory to stash the DOM0 kernel image.
...
------
After some diagnosing this turned out to be a problem with mboot
and the HP Proliant DL360 & HP Proliant DL380 systems we have.
When mboot was querying the BIOS for their E820 memory maps they
were getting gobbly gook and not the proper memory maps. We did
not see this problem with our Dell Proliant 1750's.
I found a fix which I've applied to mboot.c32 and it seems to fix
the problem. The issue is that the regs_in struct passed into
the INT15/E820 call the full register structure was not being
cleared. Although the mboot.c code does clearly set the
designated fields for the e820 interrupt apparently the HP bios's
can do strange things depending upon the value of the unspecified
register fields. The regs_in struct is local variable and hence
has random fields based off of the stack. The fix below simply
does a 'memset(®s_in, 0, sizeof (regs_in))' to clear the
fields.
Also included in the diffs below are changes to permit
the syslinux package to build on the latest Fedora Core 5
release. They should make no functional change - but reflect
that CLK_TCK is not visible by default (some Xopen/Posix stuff I
think).
Please consider these changes for inclusion into the next
release of syslinux/pxelinux/mboot.
Thanx Much,
_Mike_
---
Michael Walker
Software Developer
Cassatt Corporation, San Jose, CA
----------------------------
--- baseline-syslinux/./syslinux-3.11/com32/modules/mboot.c 2006-05-25
14:03:32.000000000 -0700
+++ fixed-syslinux/./syslinux-3.11/com32/modules/mboot.c 2006-05-25
14:03:04.000000000 -0700
@@ -362,6 +362,7 @@
e820->size = sizeof(*e820) - sizeof(e820->size);
/* Ask the BIOS to fill in this descriptor */
+ memset(®s_in, 0, sizeof (regs_in));
regs_in.eax.l = 0xe820; /* "Get system memory map" */
regs_in.ebx.l = regs_out.ebx.l; /* Continuation value from last call */
regs_in.ecx.l = 20; /* Size of buffer to write into */
--- baseline-syslinux/./syslinux-3.11/com32/samples/keytest.c 2006-05-25
14:03:32.000000000 -0700
+++ fixed-syslinux/./syslinux-3.11/com32/samples/keytest.c 2006-05-25
14:03:03.000000000 -0700
@@ -72,7 +72,7 @@
{
console_ansi_raw();
- printf("CLK_TCK = %d\n", (int)CLK_TCK);
+ printf("CLOCKS_PER_SEC = %d\n", (int)CLOCKS_PER_SEC);
printf("Press keys, end with Ctrl-C...\n");
for (;;) {
--- baseline-syslinux/./syslinux-3.11/com32/libutil/get_key.c 2006-05-25
14:03:31.000000000 -0700
+++ fixed-syslinux/./syslinux-3.11/com32/libutil/get_key.c 2006-05-25
14:03:01.000000000 -0700
@@ -116,7 +116,7 @@
};
#define NCODES ((int)(sizeof keycodes/sizeof(struct keycode)))
-#define KEY_TIMEOUT ((CLK_TCK+9)/10)
+#define KEY_TIMEOUT ((CLOCKS_PER_SEC+9)/10)
int get_key(FILE *f, clock_t timeout)
{
--- baseline-syslinux/./syslinux-3.11/com32/include/sys/times.h 2006-05-25
14:03:31.000000000 -0700
+++ fixed-syslinux/./syslinux-3.11/com32/include/sys/times.h 2006-05-25
14:03:02.000000000 -0700
@@ -13,6 +13,7 @@
#define HZ 18 /* Piddly resolution... */
#define CLK_TCK HZ
+#define CLOCKS_PER_SEC HZ
typedef uint16_t clock_t;