[Devel] 2.6.18-128.1.1.el5.028stab062.3 build fixes
Solar Designer
solar at openwall.com
Sun Jul 12 13:25:06 PDT 2009
Hi,
Attached is a patch against 2.6.18-128.1.1.el5.028stab062.3 that makes
it actually compile and link for me, as tested on Owl
( http://www.openwall.com/Owl/ ) using gcc 3.4.5, for 32-bit x86 only so
far. We're using the 92.1.18.el5.028stab060.8 revision (from April 1st)
a lot (x86 and x86-64), and we thought we'd skip over this newer broken
revision, letting you fix stuff for us... yet it remains the latest for
a while, so I gave it another spin today. I have not tried actually
booting into the resulting kernels yet.
Here are the specific build errors:
kernel/sched.c:2431: error: redefinition of 'set_cpus_allowed_ptr'
include/linux/sched.h:1277: error: previous definition of 'set_cpus_allowed_ptr' was here
Fixed by adding #ifdef CONFIG_SMP ... #endif around the function in
kernel/sched.c (the above was a non-SMP build).
In file included from include/linux/fsnotify.h:17,
from fs/open.c:13:
include/linux/inotify.h: In function `inotify_unmount_mnt':
include/linux/inotify.h:234: error: parameter name omitted
Fixed by adding the missing parameter name, apparently required by our
older gcc. The breakage would be worse with CONFIG_INOTIFY set (it was
not for this build, and my patch does not include fixes for the case
when CONFIG_INOTIFY is set, even though this is needed).
drivers/scsi/scsi_dh_export.c:26: error: redefinition of 'store_scsi_dh_data'
include/scsi/scsi_dh.h:78: error: previous definition of 'store_scsi_dh_data' was here
drivers/scsi/scsi_dh_export.c:34: error: redefinition of 'retrieve_scsi_dh_data'
include/scsi/scsi_dh.h:81: error: previous definition of 'retrieve_scsi_dh_data' was here
Added to drivers/scsi/scsi_dh_export.c:
#if defined(CONFIG_SCSI_DH) || defined(CONFIG_SCSI_DH_MODULE)
...
#endif
All of the above was for a "default" config - I simply ran
"make menuconfig", followed by Exit (saving the "changes"). So this was
a fairly weird non-SMP build, yet that's what you have as the defaults -
which got to work, right?
Well, after the above fixes, it failed on linkage badly. It turned out
that the default config was inconsistent (and this was not detected
anywhere sooner than linkage). It had:
# CONFIG_X86_XEN is not set
CONFIG_XEN_BLKDEV_FRONTEND=y
CONFIG_XEN_NETDEV_FRONTEND=y
CONFIG_XEN_PV_ON_HVM=y
So I went to "make menuconfig" again and changed:
Xen PV-ON-HVM Configuration --->
[ ] XEN_PV_ON_HVM Support
to get:
# CONFIG_XEN_PV_ON_HVM is not set
(the *_FRONTEND=y's disappeared).
Then it almost linked, giving only this error:
kernel/built-in.o(.init.text+0x1b4): In function `sched_init':
: undefined reference to `arch_update_cpu_topology'
Apparently, this is caused by our older gcc (yet our gcc is capable of
building 92.1.18.el5.028stab060.8 just fine, with no changes needed).
Fixed by adding some #ifdef CONFIG_S390's, because the function is only
defined for real for that arch.
Then it linked:
System is 1723 kB
Now let's try SMP (note: CONFIG_PM disabled):
kernel/sched.c: In function `wake_idle':
kernel/sched.c:2358: sorry, unimplemented: inlining failed in call to 'task_hot': function body not available
kernel/sched.c:2397: sorry, unimplemented: called from here
Moved the actual task_hot() definition up in that file. This helped.
Then things failed on linkage:
drivers/built-in.o(.text+0x2955): In function `pci_save_state':
: undefined reference to `pci_save_msi_state'
drivers/built-in.o(.text+0x2968): In function `pci_save_state':
: undefined reference to `pci_save_msix_state'
drivers/built-in.o(.text+0x29fa): In function `pci_restore_state':
: undefined reference to `pci_restore_msi_state'
drivers/built-in.o(.text+0x2a01): In function `pci_restore_state':
: undefined reference to `pci_restore_msix_state'
To fix this, I reverted a portion of drivers/pci/pci.h to whatever was
found in the 92.1.18.el5.028stab060.8 patch. I also removed:
static inline void pci_restore_msi_state(struct pci_dev *dev) {}
...
extern void pci_restore_msi_state(struct pci_dev *dev);
from include/linux/pci.h (would be redundant with my change to
drivers/pci/pci.h, producing warnings).
Got it to build:
System is 1770 kB
Finally, same SMP config, but with PM & ACPI enabled, primarily to test
that my changes have not broken that.
System is 1898 kB
There are still plenty of warnings, some of which look nasty. I have
only verified that my patch does not add any warnings (on the three
builds above). Someone will need to review the warnings and fix the
code (at least for things more serious than unused variables).
Finally, we're trying to get the Owl userland to build with OpenVZ
kernel headers. Currently, three packages do not build, and two of the
build failures look like they will need to be patched in OpenVZ kernel
headers rather than in the packages. I will likely post another message
on that.
I hope this helps.
Alexander
-------------- next part --------------
diff -urpX nopatch linux-2.6.18-128.1.1.el5.028stab062.3/drivers/pci/pci.h linux/drivers/pci/pci.h
--- linux-2.6.18-128.1.1.el5.028stab062.3/drivers/pci/pci.h 2009-05-10 17:51:03 +0000
+++ linux/drivers/pci/pci.h 2009-07-12 19:38:01 +0000
@@ -46,14 +46,19 @@ extern unsigned int pci_pm_d3_delay;
#ifdef CONFIG_PCI_MSI
void disable_msi_mode(struct pci_dev *dev, int pos, int type);
void pci_no_msi(void);
+#else
+static inline void disable_msi_mode(struct pci_dev *dev, int pos, int type) { }
+static inline void pci_no_msi(void) { }
+#endif
+#if defined(CONFIG_PCI_MSI) && defined(CONFIG_PM)
int pci_save_msi_state(struct pci_dev *dev);
int pci_save_msix_state(struct pci_dev *dev);
+void pci_restore_msi_state(struct pci_dev *dev);
void pci_restore_msix_state(struct pci_dev *dev);
#else
-static inline void disable_msi_mode(struct pci_dev *dev, int pos, int type) { }
-static inline void pci_no_msi(void) { }
static inline int pci_save_msi_state(struct pci_dev *dev) { return 0; }
static inline int pci_save_msix_state(struct pci_dev *dev) { return 0; }
+static inline void pci_restore_msi_state(struct pci_dev *dev) {}
static inline void pci_restore_msix_state(struct pci_dev *dev) {}
#endif
diff -urpX nopatch linux-2.6.18-128.1.1.el5.028stab062.3/drivers/scsi/scsi_dh_export.c linux/drivers/scsi/scsi_dh_export.c
--- linux-2.6.18-128.1.1.el5.028stab062.3/drivers/scsi/scsi_dh_export.c 2009-05-10 17:51:01 +0000
+++ linux/drivers/scsi/scsi_dh_export.c 2009-07-12 17:30:55 +0000
@@ -22,6 +22,7 @@
#include <scsi/scsi_dh.h>
+#if defined(CONFIG_SCSI_DH) || defined(CONFIG_SCSI_DH_MODULE)
void store_scsi_dh_data(struct scsi_device *sdev, struct scsi_dh_data *data)
{
struct scsi_device_dh_data *s = container_of(sdev,
@@ -37,4 +38,4 @@ struct scsi_dh_data *retrieve_scsi_dh_da
return s->scsi_dh_data;
}
EXPORT_SYMBOL_GPL(retrieve_scsi_dh_data);
-
+#endif
diff -urpX nopatch linux-2.6.18-128.1.1.el5.028stab062.3/include/linux/inotify.h linux/include/linux/inotify.h
--- linux-2.6.18-128.1.1.el5.028stab062.3/include/linux/inotify.h 2009-05-10 17:51:04 +0000
+++ linux/include/linux/inotify.h 2009-07-12 17:23:28 +0000
@@ -231,7 +231,7 @@ static inline void inotify_unmount_inode
{
}
-static inline void inotify_unmount_mnt(struct vfsmount *)
+static inline void inotify_unmount_mnt(struct vfsmount *mnt)
{
}
diff -urpX nopatch linux-2.6.18-128.1.1.el5.028stab062.3/include/linux/pci.h linux/include/linux/pci.h
--- linux-2.6.18-128.1.1.el5.028stab062.3/include/linux/pci.h 2009-05-10 17:51:03 +0000
+++ linux/include/linux/pci.h 2009-07-12 19:41:22 +0000
@@ -657,7 +657,6 @@ static inline int pci_enable_msix(struct
struct msix_entry *entries, int nvec) {return -1;}
static inline void pci_disable_msix(struct pci_dev *dev) {}
static inline void msi_remove_pci_irq_vectors(struct pci_dev *dev) {}
-static inline void pci_restore_msi_state(struct pci_dev *dev) {}
#else
extern void pci_scan_msi_device(struct pci_dev *dev);
extern int pci_enable_msi(struct pci_dev *dev);
@@ -666,7 +665,6 @@ extern int pci_enable_msix(struct pci_de
struct msix_entry *entries, int nvec);
extern void pci_disable_msix(struct pci_dev *dev);
extern void msi_remove_pci_irq_vectors(struct pci_dev *dev);
-extern void pci_restore_msi_state(struct pci_dev *dev);
#endif
#ifndef CONFIG_EEH
diff -urpX nopatch linux-2.6.18-128.1.1.el5.028stab062.3/include/linux/topology.h linux/include/linux/topology.h
--- linux-2.6.18-128.1.1.el5.028stab062.3/include/linux/topology.h 2009-05-10 17:50:57 +0000
+++ linux/include/linux/topology.h 2009-07-12 17:42:25 +0000
@@ -50,7 +50,9 @@
for_each_online_node(node) \
if (nr_cpus_node(node))
-void arch_update_cpu_topology(void);
+#ifdef CONFIG_S390
+extern void arch_update_cpu_topology(void);
+#endif
/* Conform to ACPI 2.0 SLIT distance definitions */
#define LOCAL_DISTANCE 10
diff -urpX nopatch linux-2.6.18-128.1.1.el5.028stab062.3/kernel/sched.c linux/kernel/sched.c
--- linux-2.6.18-128.1.1.el5.028stab062.3/kernel/sched.c 2009-05-10 17:51:05 +0000
+++ linux/kernel/sched.c 2009-07-12 19:25:59 +0000
@@ -2354,8 +2354,14 @@ nextlevel:
return cpu;
}
-static inline int task_hot(struct task_struct *p, unsigned long long now,
- struct sched_domain *sd);
+/*
+ * Is this task likely cache-hot:
+ */
+static inline int
+task_hot(struct task_struct *p, unsigned long long now, struct sched_domain *sd)
+{
+ return (long long)(now - p->last_ran) < (long long)sd->cache_hot_time;
+}
#endif /* CONFIG_SMP */
@@ -2418,6 +2424,7 @@ static inline vcpu_t wake_idle(vcpu_t cp
}
#endif
+#ifdef CONFIG_SMP
/*
* Change a given task's CPU affinity. Migrate the thread to a
* proper CPU and schedule it away if the CPU it's executing on
@@ -2462,6 +2469,7 @@ out:
return ret;
}
+#endif
/***
* try_to_wake_up - wake up a thread
@@ -3429,15 +3437,6 @@ out:
#ifdef CONFIG_SMP
/*
- * Is this task likely cache-hot:
- */
-static inline int
-task_hot(struct task_struct *p, unsigned long long now, struct sched_domain *sd)
-{
- return (long long)(now - p->last_ran) < (long long)sd->cache_hot_time;
-}
-
-/*
* double_lock_balance - lock the busiest runqueue, this_rq is locked already.
*/
static void double_lock_balance(struct rq *this_rq, struct rq *busiest)
@@ -8426,10 +8425,6 @@ static void detach_destroy_domains(const
arch_destroy_sched_domains(cpu_map);
}
-void __attribute__((weak)) arch_update_cpu_topology(void)
-{
-}
-
/*
* Partition sched domains as specified by the cpumasks below.
* This attaches all cpus from the cpumasks to the NULL domain,
@@ -9126,7 +9121,9 @@ void __init sched_init(void)
init_preempt_notifiers();
+#ifdef CONFIG_S390
arch_update_cpu_topology();
+#endif
/*
* The boot idle thread does lazy MMU switching as well:
*/
More information about the Devel
mailing list