- 07 Nov, 2004 40 commits
-
-
Paolo \'Blaisorblade\' Giarrusso authored
I'm reverting the compile-time fixes to replace them with a bigger change, which also happens to fix the compilation problem. The change is in next patches. Signed-off-by:
Paolo 'Blaisorblade' Giarrusso <blaisorblade_spam@yahoo.it> Signed-off-by:
Andrew Morton <akpm@osdl.org> Signed-off-by:
Linus Torvalds <torvalds@osdl.org>
-
Paolo \'Blaisorblade\' Giarrusso authored
This patch makes UML use print_symbol() when dumping the stack. It is not yet complete (the i386 version uses the frame pointers to walk the stack, for instance), but should be at least enough. Also, it uses bust_spinlock() when dumping errors to help them reach the console. This appears not to be enough - however I analyzed the UML console code, and it seems that UML console don't get flushed *at all* when exiting. Signed-off-by:
-
Paolo \'Blaisorblade\' Giarrusso authored
From: Paolo 'Blaisorblade' Giarrusso <blaisorblade_spam@yahoo.it> Jeff Dike <jdike@addtoit.com> Since local_irq_save() and local_irq_disable() should match (apart from saving the flags), get/set_signals must match [un]block_signals, i.e. change_signals; since get/set_signals don't enable/disable SIGPROF (and this behaviour is safe as explained in the comment the patch adds, since the profiling code does not interact with the kernel code), not even change_signals must toggle it. Signed-off-by: -
Paolo \'Blaisorblade\' Giarrusso authored
Make generic_console_write() catch the EINTR error code and retry doing the calls accordingly. Signed-off-by:
-
Paolo \'Blaisorblade\' Giarrusso authored
From: Gerd Knorr <kraxel@bytesex.org>, and me, Paolo Giarrusso This is a rewrite of the mconsole_proc function. The old code had the problem that the kernel crashed after calling "uml_mconsole proc <somefile>" a few times. I havn't tracked what exactly causes the problem, I guess trying to access the procfs without actually mounting it somewhere causes some corruption of kernel data structures. The new code simply openes /proc/<file> via sys_open(). That simplifies the function alot. It also doesn't crash any more ;) Also, from Paolo Giarrusso: When printing the content of a file through the "proc" command, make it begin on his own line. PROBLEMS: Drawback is that it only works when procfs is actually mounted below /proc. And within UML, often this is false, because when building honeypots we mount HPPFS under /proc to avoid the hacker recognizing he's attacking a UML instance. One suggestion I've received to fix the later issue was to mount the procfs within a kernel thread with a private namespace, but I havn't tried that so far. Instead, we'd like to fix the actual code of the mconsole_proc function. From "Anthony Brock" Anthony_Brock (at) ous (dot) edu comes a comment, suggesting where the bug lays (as noted in the source): removing part of the code (it seems the symlink lookup) there are no more crashes. Signed-off-by:
-
Paolo \'Blaisorblade\' Giarrusso authored
From: Paolo 'Blaisorblade' Giarrusso <blaisorblade_spam@yahoo.it> Jeff Dike <jdike@addtoit.com> Currently there are two distinct definitions of current_thread_info (which has its usual meaning) and current_thread (which is the same thing); so define the second to be the first (for UML compatibility). If needed, a search & replace \<current_thread\> to current_thread_info and removing current_thread can be done (that would be a separate, low-priority patch, however). Also, since the involved headers include each other very deeply, we must fix somehow this recursive inclusion or things won't compile. Jeff Dike noted that this can be done easily, since the nasty chain of includes can be broken by eliminating an include of current.h, and implemented this. Signed-off-by: -
Paolo \'Blaisorblade\' Giarrusso authored
Avoid including some unused headers. Signed-off-by:
-
Paolo \'Blaisorblade\' Giarrusso authored
Rename headers with the same name to avoid specifying the complete path for including them. Signed-off-by:
-
Paolo \'Blaisorblade\' Giarrusso authored
Since arch/um/kernel/tt/unmap_fin.o is linked statically, it could contain some symbols also contained in the rest of vmlinux; on some systems, this actually happens (due to some glibc differences). Since that file *must* be linked statically and then relinked (its code must remap the whole .text area except the section it is contained in during UML startup), we cannot change this. So, we use objcopy to turn all symbols into local ones except for switcheroo(), which is the only function actually defined in the source (the other ones are brought in during linking). This shouldn't hurt other systems, while improving the behaviour on affected ones (there is still another issue I'm trying to fix). And it is logically always needed (i.e., without this UML has just happened to work until now). Signed-off-by:
-
Paolo \'Blaisorblade\' Giarrusso authored
Correct the entry_exit parameter for syscall_trace when tracing syscalls. It was the reverse of the needed value. (i.e. it was 1 on entry and 0 on exit). Signed-off-by:
-
Paolo \'Blaisorblade\' Giarrusso authored
This adds a check that /tmp is not mounted noexec. UML needs to be able to do PROT_EXEC mmaps of temp files. Previously, a noexec /tmp would cause an early mysterious UML crash. (Actually, not /tmp but $TMP, which often is the same). Signed-off-by:
-
Paolo \'Blaisorblade\' Giarrusso authored
Changes a comment about forcing /bin/bash as shell for the UML Makefile. Signed-off-by:
-
Paolo \'Blaisorblade\' Giarrusso authored
From: Gerd Knorr <kraxel@bytesex.org>, Paolo 'Blaisorblade' Giarrusso <blaisorblade_spam@yahoo.it> Uml was using kill(pid, SIGKILL) instead of ptrace(PTRACE_KILL, pid), which used to work, while being probably undocumented. Due to the changes in 2.6.9 to the ptrace(2) semantics, with the introduction of TASK_TRACED by Roland McGrath, this does not more work, so this patch should fix it. With help of Gerd Knorr report and analysis, and its early fix of this problem. He fixed the problem by sending SIGCONT to the child processes after SIGKILL. Which IMHO should not be needed (I think this is a regression, which he already reported). I improved his one, by replacing the SIGCONT hack with using PTRACE_KILL instead. Signed-off-by: -
Hirokazu Takata authored
Here is a patch to fix compile errors for an Mappi evaluation board. - Just change from CONFIG_M32RPCC to CONFIG_M32R_PCC. Thanks to Naoto Sugai for reporting. Signed-off-by:
Hirokazu Takata <takata@linux-m32r.org> Signed-off-by:
-
Hirokazu Takata authored
This patch fixes arch/m32r/lib/memset.S. Signed-off-by:
-
Christoph Lameter authored
- mmtimer: add support for the use of hardware timers on SGI SHubs (sn subarch) via the posix timer interface. - Allows the scheduling of up to 3 hardware interrupts per node to deliver signals to processes running on those nodes. - Periodic timers in user space capable of supporting intervals down to 10us (<10us supported but will usually result in excessive overruns). - Scheduling accuracy of 40ns. (However, typically it takes at least a few usecs for an interrupt to be forwarded to a signal in user space.) - Add missing SHub information to include/asm-ia64/sn/shub_mmr.h - Reserve timer interrupt used in include/asm-ia64/sn/intr.h Signed-off-by:
Christoph Lameter <clameter@sgi.com> Signed-off-by:
-
Andi Kleen authored
Fix an obviously broken change someone did to the x86-64 NMI handler. reason would get used uninitialized, causing random misbehaviour in the NMI handler and gcc rightly complained about it. Signed-off-by:
Andi Kleen <ak@suse.de> Signed-off-by:
-
Andi Kleen authored
Add links for p4-clockmod driver in x86-64 cpufreq. AK: Made it dependent on EMBEDDED because the driver is only useful in some special situations and should be normally not used. Signed-off-by:
"Venkatesh Pallipadi" <venkatesh.pallipadi@intel.com> Signed-off-by:
-
Paolo \'Blaisorblade\' Giarrusso authored
The DR7 register on i386/x86_64 has a special meaning, so there is a special check to do. Since the code is rather difficult, I added an explaination about it. Also, while studying the i386 Intel Manual, I saw that x86_64, even on 32bit emulation, allows using values which are disallowed on i386. It's almost obvious that what it allows is setting a 8-byte wide data watchpoint (in fact I double checked the AMD manuals, just in case, and this is true; I couldn't find a mention of this in my Intel manuals). But since the original ia32 emulation code has this comment: /* You are not expected to understand this ... I don't neither. */ I am dubious that the code in ptrace32.c is wrong: does x86_64 supports 8byte-wide watchpoints in 32-bit emulation? I've checked the AMD manual Vol. 2 (no. 24593), which says that to set the length to 8 bits "long mode must be enabled". This means, actually, that the current code is safe, notwithstanding the comment (which I removed). Signed-off-by:
-
Maciej W. Rozycki authored
While discussing races in the NMI handler's trailer fiddling with RTC registers, I've discovered we incorrectly attempt to handle NMIs coming from the system (memory errors, IOCHK# assertions, etc.) with all processors even though the interrupts are only routed to the bootstrap processor. If one of these events coincides with a NMI watchdog tick it may even be handled multiple times in parallel. Here is a fix that makes application processors ignore these events. They no longer access the NMI status bits at I/O port 0x61 which has also the advantage of removing the contention on the port when the I/O APIC NMI watchdog makes all processors arrive at the handler at the same time. Signed-off-by:
Maciej W. Rozycki <macro@linux-mips.org> Signed-off-by:
-
Maciej W. Rozycki authored
Here is a patch to do some cleanup to code affected by the recent change to the bootstrap code responsible for enabling the local APIC. It adds a message about how to get the APIC enabled if previously disabled by the BIOS to aid people expecting the former behavior. It removes some P4-specific code that's become dead as a result of the change. Finally it improves a couple of inaccurate comments. This has been agreed to be reasonable. Signed-off-by:
-
Paul Mackerras authored
This patch is from Hanna Linder. As pci_find_device is going away I've replaced it with pci_get_device and pci_dev_put. Signed-off-by:
Hanna Linder <hannal@us.ibm.com> Signed-off-by:
Maximilian Attems <janitor@sternwelten.at> Signed-off-by:
Paul Mackerras <paulus@samba.org> Signed-off-by:
-
Paul Mackerras authored
This patch is from Hanna Linder. As pci_find_device is going away using the new for_each_pci_dev macro. Signed-off-by:
-
Paul Mackerras authored
This patch is from Hanna Linder. As pci_find_device is going away using the new for_each_pci_dev macro. Signed-off-by:
-
Paul Mackerras authored
This patch is from Hanna Linder. As pci_find_device is going away using the new for_each_pci_dev macro. Signed-off-by:
-
Paul Mackerras authored
This patch is from Hanna Linder. As pci_find_device is going away using the new for_each_pci_dev macro. If someone with a PPC64 system could test it I would appreciate it. Signed-off-by:
-
Paul Mackerras authored
This patch is from John Rose <johnrose@austin.ibm.com>. This patch changes the following iommu-related things: - Renames the [i,p]series versions of iommu_devnode_init(), to keep things logically separate where possible. - Moves iommu_free_table() to generic iommu.c - Creates of_cleanup_node(), which will directly precede the dynamic removal of any device node Signed-off-by:
John Rose <johnrose@austin.ibm.com> Signed-off-by:
-
Anton Blanchard authored
Now that spinlocks are always out of line, oprofile needs to backtrace through them. The following patch adds this but also adds the ability to turn it off (via the backtrace_spinlocks option in oprofilefs). The backout option is included because the backtracing here is best effort. On ppc64 the performance monitor exception is not an NMI, we get them only when interrupts are enabled. This means we can receive a profile hit that is inside a spinlock when our PC is somewhere completely In this patch we check to make sure the PC of the performance monitor exception as well as the current PC is inside the spinlock region. If so then we find the callers PC. If this is not true we play it safe and leave the tick inside the lock region. Also, now that we execute the SLB handler in real mode we have to adjust the address range that we consider as valid real mode addresses. Otherwise the SLB miss handler will end up as unknown kernel profile hits. Signed-off-by:
Anton Blanchard <anton@samba.org> Signed-off-by:
-
Olof Johansson authored
With current firmware, the ibm,my-dma-window property now contains two panes for VSCSI server nodes. This breaks the current tests in the setup code. There's a bunch of references to pre-GA firmware bugs. That's a while ago, so we can remove the workarounds without breaking anyone. Signed-off-by:
Olof Johansson <olof@austin.ibm.com> Signed-off-by:
-
Stephen Rothwell authored
This patch just does some cleanups of iSeries_iommu.c remove lots of unneeded includes use list_for_each_entry white space formatting No semantic changes. Signed-off-by:
Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by:
-
Matt Porter authored
This patch adds an L2-Cache error handler. In addition, it reenables setup of the L2 cache based on erratum L2C0_1 now that we have been reassured that there are no more faulty parts being shipped. If there is a problem, the error handler will report it. Signed-off-by:
Eugene Surovegin <ebs@ebshome.net> Signed-off-by:
Matt Porter <mporter@kernel.crashing.org> Signed-off-by:
-
Tom Rini authored
include/asm-ppc/time.h has a slightly odd include list which means that some files include <asm/time.h> and also hope to get others that they really shouldn't be. This makes <asm/time.h> use <linux/types.h> (time_t) and <linux/rtc.h> (struct rtc_time) instead of <linux/mc16818rtc.h>, and fixes up the fallout from the change. Compile-tested on lite5200, walnut and defconfig, along with by-hand'ing everything else that included <asm/time.h>. Signed-off-by:
Tom Rini <trini@kernel.crashing.org> Signed-off-by:
-
Benjamin Herrenschmidt authored
pmac_zilog wasn't properly updating port->timeout, thus broke with serial console. I'm not sure if that timeout thing was added recently or not, since the driver used to work fine. This patch fixes that, and adds some proper SYSRQ support. By default, BREAK is used for sysrq, though some Apple zilog's seem to have non-working BREAK detection logic, thus the driver has a #define option you can manually enable when debugging to do SYSRQ with ctrl-O instead (like pSeries virtual consoles). Signed-off-by:
Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by:
-
Tom Rini authored
Adrian Cox has modified the Sandpoint code to once again work on the older X2 version of the board. The older version has a number of are on the older version. Signed-off-by:
Adrian Cox <adrian@humboldt.co.uk> Signed-off-by:
-
Tom Rini authored
Rune has determined that: res6 should be 94 bytes long. However... It does work @ 92 bytes because the compiler alligns the next struct (pci_cpm2_t) on a int boundary anyways (no packing specified fo the structs) I tested it with a module that pronted out the address of the PCI struct that follows the sit_cpm2_t struct. Imap = 0xF0000000 With res6[92] im-pci is @ 0xf0010430 With res6[94] im-pci is @ 0xf0010430 So we're saved by the compiler. Here's the patch Signed-off-by:
Rune Torgersen <runet@innovsys.com> Signed-off-by:
-
Christoph Hellwig authored
PREEMPT_BITS, SOFTIRQ_BITS, PREEMPT_SHIFT, SOFTIRQ_SHIFT and HARDIRQ_SHIFT are the same for all architectures (and chaning them in future ports doesn't make a lot of sense), so move them to <linux/hardirq.h>. Make the HARDIRQ_BITS definition guarded by #ifndef HARDIRQ_BITS instead of CONFIG_GENERIC_HARDIRQS so we have a saner way to override it once ports with lots of irq sources are changed to use the generic hardirq framework. Ingo, maybe we should change the default for it back to 8 as that's what most ports use? Signed-off-by:
Christoph Hellwig <hch@lst.de> Signed-off-by:
-
Hugh Dickins authored
My tmpfs superblock changes in 2.6.9 messed up mount -t tmpfs when CONFIG_TMPFS is not enabled: it wrongly claimed to succeed, and left the directory unusable, giving "Not a directory" errors thereafter. Signed-off-by:
Hugh Dickins <hugh@veritas.com> Signed-off-by:
-
Jesse Barnes authored
If NUMA is enabled, find_next_best_node is responsible for helping build the zonelist for each pgdat in the system. However, if one sets PENALTY_FOR_NODE_WITH_CPUS to a large value in an attempt to prefer nodes w/o CPUs, the local node is erroneously placed after all nodes w/o CPUs in the pgdat's zonelist. This small patch fixes that by just checking if the local node is part of the zonelist yet, and if not, returns it first. Signed-off-by:
Jesse Barnes <jbarnes@sgi.com> Signed-off-by:
-
Rajesh Venkatasubramanian authored
Add prio_tree.c documentation. Signed-off-by:
Rajesh Venkatasubramanian <vrajesh@umich.edu> Signed-off-by:
-
Rajesh Venkatasubramanian authored
Currently we use prio_tree_root->index_bits to optimize the height of both the initial heap-and-radix indexed levels of a prio_tree as well as the heap-and-size indexed overflow-sub-trees. Please see the accompanying prio_tree documentation patch for further details. When index_bits is increased in prio_tree_expand we shuffle the initial heap-and-radix indexed levels to make sure that vmas are placed in the tree at appropriate places. Similarly, since the overflow-sub-trees' heights also depend on prio_tree_root->index_bits we should shuffle all the overflow-sub-trees when index_bits changes. However, I missed to take care of this in my implementation. Recently Stefan Hornburg (Racke) reported the problem and patiently tested the trace patches. Hugh Dickins produced the trace patches that helped to detect the bug. Moreover, Hugh reduced the crash test case to few lines of code. Thanks to both of them. The easy fix is to disable prio_tree_expand code completely. That may lead to skewed trees in some common cases. Hence, this patch takes a different approach. This patch fixes the problem by not optimizing the height of the overflow-sub-trees using prio_tree_root->index_bits. Now all overflow-sub-trees use full BITS_PER_LONG bits of size_index to place the vmas (that have the same start_vm_pgoff) in an overflow-sub-tree. This may result in skewed overflow-sub-trees because all bits in vm_pgoff above prio_tree_root->index_bits will be 0 (zero). However, processes rarely map many vmas with the same start_vm_pgoff and different end_vm_pgoff. Therefore, such skewed sub-trees should be very rare. Signed-off-by:
-
