- 21 Jan, 2005 30 commits
-
-
Andi Kleen authored
Compiling an allyesconfig kernel straight with a gcc 4.0 snapshot gives nearly 10k new warnings like: warning: pointer targets in passing argument 5 of `cpuid' differ in signedness Since the sheer number of these warnings was too much even for the most determined kernel janitors (I actually asked ;-) and I don't think it's a very serious issue to have these mismatches I submitted an new option to gcc to disable it. It was incorporated in gcc mainline now. This patch makes the kernel compilation use it. There are still quite a lot of new warnings with 4.0 (mostly about uninitialized variables), but the compile log looks much nicer nnow. Signed-off-by:
Andi Kleen <ak@suse.de> Signed-off-by:
Andrew Morton <akpm@osdl.org> Signed-off-by:
Linus Torvalds <torvalds@osdl.org>
-
Chris Wright authored
Most of the include/asm-*/resource.h headers are the same as one another. This patch makes one generic version, include/asm-generic/resource.h, and uses that when appropriate. The only vaguely interesting things here are that the generic version introduces a new _STK_LIM_MAX macro, which can be populated by an arch (ia64 and parisc needed that). Also, some arches hid RLIM_INFINITY under __KERNEL__, while others did not. The generic version does not, so the following arches will see that change: arm, arm26, mips, ppc, ppc64, sh (and hence sh64) And, finally, some arches maintain their own order for the resource numbers. This is now marked by __ARCH_RLIMIT_ORDER, and is used by the following arches: alpha, mips, sparc, and sparc64. This actually uncovered a mips bug (fix already sent, this patch is relative to that fix), where the default RLIMIT_MEMLOCK was set to RLIM_INFINITY and RLIMIT_NPROC set to MLOCK_LIMIT (the latter is no big deal because RLIMIT_NPROC default is overwritten dynamically during bootup in fork_init()). Also, this change makes alpha's default for RLIMIT_NPROC change from RLIM_INFINITY to 0, but again...no problem as it's dynamically overwritten during bootup. The following arches are left untouched: m68knommu: untouched (uses m68k/resource.h) sh64: untouched (uses asm-sh/resource.h) um: untouched (uses arch code already) Signed-off-by:Chris Wright <chrisw@osdl.org> Signed-off-by:
-
Eric Lammerts authored
I have a problem with errors=panic on ext3. When a panic occurs, the error event is not recorded anywhere. So after the reboot, e2fsck doesn't kick in, the file system gets mounted again and the box panics again... Patch below moves the ERRORS_PANIC test down a bit so the journal is aborted before panic() is called. Signed-off-by:
Eric Lammerts <eric@lammerts.org> Signed-off-by:
-
H. Peter Anvin authored
The use of U+F800 to U+F804 has been deprecated since 2003; this makes the deprecation effective by replacing these characters with the officially assigned U+23BA to U+23BD. It also updates unicode.txt to match the latest version from the LANANA webpage. Signed-off-by:
-
Egbert Eich authored
So far the X.Org/XFree86 Xservers use internal functions to perform saving/restoring of console fonts. This doesn't work under all circumstances and has disadvantages: 1. On some platforms (IA64) X needs to POST the BIOS before it even has a chance to access the hardware itself. This POSTing will usually undo any changes to the graphics hardware that the kernel may have done. 2. More and more drivers fully rely on BIOS support. However the BIOS functions which could be used to save/restore register settings may be broken so the only way of mode save/restore is getting/setting the BIOS mode ID. This will restore the BIOSes default fonts, not the custom font that the user may have loaded. I would like to utilize the kernel's save/restore console fonts ioctls to save/restore the text mode console fonts from inside X. The patch does two things: 1. Eliminates the optimization that returns from vgacon_adjust_height() early (without reprogramming the HW) when the hight of the font hasn't changed. 2. Resets the stored 'from' and 'to' lines for the text cursor in vgacon_adjust_height() to cause vgacon_set_cursor_size() reprogram the textcursor start and end line. These are necessary to sanitize the HW in case something other than the kernel has changed the register values without restoring them properly. I'm fully aware that in the long run we will need to look into a new driver model for graphics where no two instances fight over who gets register access. However such a model won't be created nor will we get the majority of the drivers ported over night. Therefore we need to find an interim solution for the most pressing problems. Signed-off-by:
Egbert Eich <eich@freedesktop.org> Approved-by:
Alan Cox <alan@redhat.com> Signed-off-by:
-
Keith Owens authored
scripts/reference*.pl - treat built-in.o as conglomerate. Ignore references from altinstructions to init text/data. Signed-off-by:
Keith Owens <kaos@ocs.com.au> Signed-off-by:
-
Christoph Hellwig authored
The newly merged frv do_IRQ code calls softirq_pending(), but always with the current cpu as argument - switch to local_softirq_pending(). Btw, this usage look bogus to me, any reason you need to call do_softirq again after you did four lines above in irq_exit(), David? Signed-off-by:
-
Christoph Hellwig authored
Acked by Mikael Starvik. Signed-off-by:
-
Christoph Hellwig authored
arch/mips/kernel/binfmt_elfn32.c needs to override e_flags for elf coredumps. It already defines ELF_CORE_EFLAGS before including binfmt_elf.c, but the latter doesn't pick up the define yet. Signed-off-by:
-
Ingo Molnar authored
We think it'll work OK. Signed-off-by:
Ingo Molnar <mingo@elte.hu> Signed-off-by:
-
David Mosberger authored
The "addr" member in the time-interpolator is sometimes used as a function-pointer and sometimes as an I/O-memory pointer. The attached patch tells sparse that this is OK. Signed-off-by:
David Mosberger-Tang <davidm@hpl.hp.com> Signed-off-by:
-
Randy Dunlap authored
Use correct data type for module_param: drivers/media/radio/radio-typhoon.c:317: warning: return from incompatible pointer type Signed-off-by:
Randy Dunlap <rddunlap@osdl.org> Signed-off-by:
-
Christoph Hellwig authored
Signed-off-by:
-
Christoph Hellwig authored
Currently it's up to the filesystem ->permission method to check whether the filesystem is readonly or the file marked immutable. But this is really a VFS decision, and the distintion becomes more important when moving to per-mountpoint read only flags. For most filesystems that have been using generic_permission this is not change in behaviour. For those that we're missing the check (cifs without CIFS_MOUNT_NO_PERM and coda [1]) this is a bugfix. Both reiserfs and xfs have this check still in their ->permission routine because they call it from other places aswell. I'll try switching them over to generic_permission and will take care of this one. [1] coda_ioctl_permission always returns 0, aka always grants access, which looks more than fishy to me. Signed-off-by: -
Jan Kara authored
Remove unnecessary division and modulo from ext3 code in often used paths. Without the patch an oprofile of dbench load shows ext3_get_group_desc() uses 0.84% and ext3_group_sparse() 1.51%, with the patch the numbers are 0.33% and 0.27% respectively. Signed-off-by:
Jan Kara <jack@suse.cz> Signed-off-by:
Andreas Dilger <adilger@clusterfs.com> Signed-off-by:
-
Prasanna Meda authored
expand_files() cleanup: Make expand_files() common code for fork.c:fork/copy_files(), open.c:open/get_unused_fd() and fcntl.c:dup/locate_fd(). expand_files() does both expand_fd_array and expand_fd_set based on the need. This is used in dup(). open() and fork() duplicates the work of expand files. At all places we check for expanding fd array, we also check for expanding fdset. There is no need of checking and calling them seperately. This change also moves the expand_files to file.c from fcntl.c, and makes the expand_fd_array and expand_fd_set local to that file. Signed-off-by:
Prasanna Meda <pmeda@akamai.com> Signed-off-by:
-
Jeff Dike authored
Signed-off-by:
Jeff Dike <jdike@addtoit.com> Signed-off-by:
-
Jeff Dike authored
Fix a race where signals could be handled to the parent of a new process on the kernel stack of the child, corrupting that stack, and crashing UML when the next first runs. Signed-off-by:
-
Jeff Dike authored
Signed-off-by:
-
Jeff Dike authored
Fixes the "profiling timer expired" message on shutting down with profiling enabled. Signed-off-by:
-
Jeff Dike authored
Signed-off-by:
-
Jeff Dike authored
If the file backing a ubd device is not an even blocklength, then the last partial block is now readable, and it is padded with zeros. Signed-off-by:
-
Jeff Dike authored
Provide definitions of __initcall, __exitcall, and __init_call for userspace code. Signed-off-by:
-
Jeff Dike authored
Replace FRAME_SIZE_OFFSET with MAX_REG_OFFSET because different arches have Signed-off-by:
-
Pavel Machek authored
This passes pm_message_t down to PCI drivers. Ugly translation code can be removed and this will allow PCI devices to do right thing during swsusp snapshot -- like not unneccessarily blanking display. Only obscure /sysfs code passes anything but 3 to pci_device_suspend, anyway, so this is pretty close to a nop. Signed-off-by:
Pavel Machek <pavel@suse.cz> Signed-off-by:
-
Pavel Machek authored
This removes another O(n^2) algorithm from page relocation in swsusp. Relocation took as long as reading pages from disk on my machine, and it took minutes for some poor testers. Signed-off-by:
-
Yoshinori Sato authored
o include/asm-h8300/mmu.h: Changeset 1.2053 support o mm/nommu.c: add __vm_enough_memory Signed-off-by:
Yoshinori Sato <ysato@users.sourceforge.jp> Signed-off-by:
-
Yoshinori Sato authored
Signed-off-by:
-
Prasanna S. Panchamukhi authored
Minor changes to the kprobes code to provide memory allocation for x86_64 architecture outside kprobes spin lock. Signed-off-by:
Prasanna S Panchamukhi <prasanna@in.ibm.com> Signed-off-by:
-
Andi Kleen authored
Small cleanup: save the CPUID level once instead of regetting it all the time. Needed for some other patches. Signed-off-by:
-
- 20 Jan, 2005 10 commits
-
-
Venkatesh Pallipadi authored
do_suspend_lowlevel doesn't have an argument anymore, but the assembly code would still test it and fail randomly. Remove the bogus test. Signed-off-by:
-
Andi Kleen authored
With Martin Schwidefsky <schwidefsky@de.ibm.com> Fix typo in ioremap pud conversion Signed-off-by:
-
Andi Kleen authored
Fix SMP race in NMI watchdog on i386/x86-64 Fix a long standing SMP Setup race in the NMI watchdog. The watchdog would tick from very early and check if all CPUs increase their timer interrupts. For that it would check the cpu_online_map. Now if a CPU took too long to boot the watchdog would trigger prematurely because the CPU didn't increase its timer count yet. Fix is to check cpu_callin_map instead of cpu_online_map because the first is only set when a CPU started its timer interrupt. I fixed it on i386 and x86-64. Description of the problem from Manpreet Singh. Thanks. Cc: <manpreet@fabric7.com> Signed-off-by:
-
Andi Kleen authored
Fix a long standing race in x86-64 SMP TLB handling. When a mm is freed and another thread exits to a lazy TLB thread (like idle) the freed user page tables would be still kept loaded in the idle thread. When an interrupt does a prefetch on NULL the CPU would try to follow it and read random data. This could lead to machine checks on Opterons in some cases. Credit goes to some unnamed debugging wizards at AMD who described the problem. All blame to me. I did the fix based on their description. Signed-off-by:
-
Andi Kleen authored
Check if node exists during CMP setup and fall back to other node if needed. This avoids a crash when the NUMA node discovery fails (e.g. due to node interleaving in the BIOS or numa=off) Signed-off-by:
-
Matthew Dobson authored
More missing nodemask.h inclusions. Signed-off-by:
-
Vincent Hanquez authored
Hi, the following patch: - correct the err variable tested twice when _NSIG_WORDS == 1 (unlikely to happen, but ..) - remove some |= in favor of = because we don't need to 'pack' err Signed-off-by:
Vincent Hanquez <tab@snarc.org> Acked-by:
-
Randy Dunlap authored
Correct .text references to .init.data; init_intel_cacheinfo() can be __init. Signed-off-by:
-
Suresh B. Siddha authored
Current code can lead to corruption. Use cpumask_t instead of unsigned long. Signed-off-by:
Suresh Siddha <suresh.b.siddha@intel.com> Signed-off-by:
-
Rik van Riel authored
In unprivileged Xen domains, all that __ioremap() does is a "return NULL", which causes dmi_iterate() to crash the kernel at boot time. This trivial check bails dmi_iterate() out of the loop when it finds that the ioremap() returned a NULL pointer. (akpm: this fix goes beyond xen - ioremap can return NULL) Signed-off-by:
Rik van Riel <riel@redhat.com> Signed-off-by:
-
