- 03 Sep, 2004 40 commits
-
-
Andi Kleen authored
This patch fixes the argument length checking in sched_setaffinity. Previously it would error out when the length passed was smaller than sizeof(cpumask_t). And any bits beyond cpumask_s would be silently ignored. First this assumes that the user application knows the size of cpumask_t, which should be kernel internal. When you increase cpumask_t old applications break and there is no good way for the application to find out the cpumask_t size the kernel uses. This patch changes it to do similar checking to the NUMA API calls: - Any length is ok as long as all online CPUs are covered (this could still cause application breakage with more CPUs, but there is no good way around it) - When the user passes more than cpumask_t bytes the excess bytes are checked to be zero. Signed-off-by:
Andrew Morton <akpm@osdl.org> Signed-off-by:
Linus Torvalds <torvalds@osdl.org>
-
Andi Kleen authored
This patch increases some hardcoded limits people have already run into or plan to go towards eventually. Increase max MP busses on x86-64. Needed for some machines. Suggested by Andrew Chen and James Cleverdon. MAX_IRQ_SOURCES kept at 256 for now, matching i386. Increase MAX_IO_APICS to 32 following i386. Signed-off-by:
-
Andi Kleen authored
From: Andreas Gruenbacher Add brackets to boot_cpu_data to fix compilation of centrino cpufreq driver. Signed-off-by:
-
Andi Kleen authored
Following the CPUID extensions and APIC-ID setup presented by Richard Brunner at Kernel Summit. Without this the kernel's node discovery would assign the wrong nodes on dual core systems and cause performance degradation. Only handles power of two cored CPUs, others will be ignored. /proc/cpuinfo has a new field cpu cores: <number of CPU cores> and also show the other core as HT sibling (this is needed to make old CPU based license managers to count one CPU package as only one physical CPU) FIXME this currently misses the disable code for SMT idle. That's because 2.6.9rc1-bk6 removed the arch hooks need for that. Needs to be readded later. Signed-off-by:
-
Haren Myneni authored
This patch contains - Removes __initdata from lmb definition (struct lmb lmb;) and modified the existing page_is_ram function. Also changed the current argument from physical address to pfn to make it compatible across architectures. Please review them and send me your comments/suggestions. If you are OK with any one patch, please include it in the mainline kernel. Signed-off-by:
Anton Blanchard <anton@samba.org> Signed-off-by:
-
Anton Blanchard authored
Wrap a lot more of processor.h in #ifdef __KERNEL__ / __ASSEMBLY__. Remove now unised EXC_FRAME_SIZE. Signed-off-by:
-
Anton Blanchard authored
Instead of statically allocating the NUMA node structures, do it at runtime with node local memory where possible. With NR_CPUS=128 we had over 800kB of memory allocated before this patch, now we allocate only as many nodes as the machine requires. We have to do some fancy footwork in do_init_bootmem since we use both the LMB allocator and the bootmem allocator at the same time. The problem has always been there although I only recently found a test case that hit it. Wrap that logic in careful_allocation which uses the LMB allocator then falls back to the bootmem allocator if required. Now alloc_bootmem on a node with no memory doesnt panic, we dont need the hack in paging_init so remove it. Signed-off-by:
-
Anton Blanchard authored
On some machines we would print hundreds of lines of NUMA debug output. The following patch cleans it up so we only print a summary of cpus and memory vs nodes: Node 0 CPUs: 0-1 Node 1 CPUs: 16-17 Node 2 CPUs: 32-33 Node 3 CPUs: 48-49 Node 0 Memory: 0x0-0x400000000 Node 1 Memory: 0x400000000-0x800000000 I lifted the code to do this out of xmon. Signed-off-by:
-
Anton Blanchard authored
Update iSeries_defconfig. Signed-off-by:
-
Anton Blanchard authored
Update pSeries_defconfig. This config should better cover POWER5 systems with SPLPAR spinlocks, IPR SCSI, NR_CPUS = 128, NUMA, HVCS, and irq stacks enabled, VSCSI and VETH as modules. Signed-off-by:
-
Anton Blanchard authored
Now we set up cpu_possible_map early we can dynamically allocate the emergency stacks. Previously we would allocate NR_CPUS * PAGE_SIZE, potentially wasting quite a lot of memory. Clean a comment up in irqstack_early_init and use unsigned int for cpu ids while in the area. Signed-off-by:
-
Paul Mackerras authored
This patch uses the firmware-defined error log buffer length for calls to the firmware routine 'check-exception'. It also simplifies code in rtasd.c that is attempting to obtain the error log length. Signed-off-by:
Linas Vepstas <linas@linas.org> Signed-off-by:
Paul Mackerras <paulus@samba.org> Signed-off-by:
-
Olof Johansson authored
Implement the HCALLs to do more than one TCE setup or invalidation at a time on pSeries LPAR. Previous implementation did one hypervisor call per setup or teardown, resulting in significant overhead. A simple test of "time dd if=/dev/sda of=/dev/null bs=128k" shows the amount of system time go down by about 5% by using the multi-tce calls. Signed-off-by:
Olof Johansson <olof@austin.ibm.com> Signed-off-by:
-
Olof Johansson authored
To be able to use the cur_cpu_spec->firmware_features values in early setup, the function initializing them has to be moved up to before the early init calls. Signed-off-by:
-
Andy Whitcroft authored
ppc64 has constraints on the mixing of small and large pages, such that they cannot be mixed in the same 256Mb segment. This patch provides an architecture specific arch_get_unmapped_area_topdown() which implements these contraints. Signed-off-by:
Andy Whitcroft <apw@shadowen.org> Signed-off-by:
-
Andy Whitcroft authored
Recent patches introduced a top down user process address space allocation policy; further patches enable this for ppc64. Although these work correctly for normal maps, the topdown algorithm does not take into account stringent mixing constraints for small and large pages on this architecture. These patches introduce a ppc64 specific arch_get_unused_area_topdown() variant. The first introduces infrastructure to allow replacement of the generic arch_get_unused_area_topdown() and the second utilises this infrastructure. In this patch I have followed the pattern set by the arch_get_unused_area() using HAVE_ARCH_UNMAPPED_AREA_TOPDOWN to be consistent. However, it would also be possible to simply have a ppc64_get_unused_area_topdown() in the arch/ppc64/mm/mmap.c or to use weak bindings. This patch: Allow an architecture to override the default definition of arch_get_unmapped_area_topdown(). Signed-off-by:
-
Anton Blanchard authored
We should print a stack backtrace when we get EEH errors, it makes debugging the cause of the fail easier. Signed-off-by:
-
Paul Mackerras authored
This patch adds explicit checking for EEH slot isolation events into the PCI config space read path. The change itself would have been minor, except that pci config reads don't have a pointer to a struct pci_dev. Thus, I had to restructure the eeh code to accomodate this, which seems to be a good thing anyway, making it a tad cleaner. Signed-off-by:
-
Anton Blanchard authored
DEBUG_SPINLOCK_SLEEP is useful on ppc64, so allow it to be selected. Signed-off-by:
-
David Gibson authored
This has been given basic testing on Power4 pSeries and iSeries machines. At present, the SLB miss handler has to check the SLB slot it is about to use to ensure that it does not contain the SLBE for the current kernel stack - throwing out the SLBE for the kernel stack can trigger the "megabug": we have no SLBE for the stack, but we don't fault immediately because we have an ERAT entry for it. The ERAT entry is then lost due to a tlbie on another CPU during the unrecoverable section of the exception return path. This patch implements a different approach - with this patch SLB slot 2 always (well, nearly always) contains an SLBE for the stack. This slot is never cast out by the normal SLB miss path. On context switch, an SLBE for the new stack is pinned into this slot, unless the new stack is the the bolted segment. For iSeries we need a special workaround because there is no way of ensuring the stack SLBE is preserved an a shared processor switch. So, we still need to handle taking an SLB miss on the stack, in which case we must make sure it is loaded into slot 2, rather than using the normal round-robin. This approach shaves a few ns off the slb miss time (on pSeries), but more importantly makes it easier to experiment with different SLB castout aporoaches without worrying about reinstating the megabug. Signed-off-by:
David Gibson <david@gibson.dropbear.id.au> Signed-off-by:
-
Benjamin Herrenschmidt authored
... but a better solution should probably be found, that's saved_auxv in struct mm_struct is ... hrm... disgusting at least ;) So the problem is that we currently do unsigned long saved_auxv[40]; That is we assume that all archs will have an AUXV entry of no more than 2 unsigned longs and we'll have no more than 20 entries in the table. Unfortunately, this is wrong. PPC/PPC64 adds 5 platform entries. then we add 14 in the common code, and then we have if (k_platform) { NEW_AUX_ENT(AT_PLATFORM, (elf_addr_t)(unsigned long)u_platform); } if (bprm->interp_flags & BINPRM_FLAGS_EXECFD) { NEW_AUX_ENT(AT_EXECFD, (elf_addr_t) bprm->interp_data); } which may add a 20th entry... which means we are already out of room for the terminating AT_NULL entry... bad bad bad. Now I'm adding AT_SYSINFO_EHDR to ppc/ppc64 and it breaks all the time. The quick fix is to bump the number of entries in struct mm_struct to 42, though that's worth a big BOLD comment in binfmt_elf in case anybody adds a new entry in there. (Patch below). I wonder if we shouldn't add a ARCH_MAX_AUXV_ENTRIES macro to the asm-arch/elf.h though to get the max amount of entry the arch may add to the normal count, and have the mm struct array use that plus the number of "normal" entries... This patch increase the max amount of auxv entries in struct mm_struct to 42 so that ppc & ppc64 don't get a chance to blow the array up. Signed-off-by:Benjamin Herrenschmidt <benh@kernel.crashing.org> Signed-off-by:
-
Andrew Morton authored
drivers/scsi/ipr.c: In function `ipr_worker_thread': drivers/scsi/ipr.c:1825: wrong type argument to unary exclamation mark drivers/scsi/ipr.c: In function `ipr_read_dump': drivers/scsi/ipr.c:2426: wrong type argument to unary exclamation mark Cc: James Bottomley <James.Bottomley@steeleye.com> Cc: Brian King <brking@us.ibm.com> Signed-off-by:
-
John Stultz authored
I've been hunting down a bug affecting IBM x440/x445 systems where the floppy driver would get spurious interrupts and would not initialize properly. After digging James Cleverdon and Bill Irwin pointed out that target_cpus() is routing the interrupts to the clustered apic broadcast mask using destFixed deliver mode. This was causing multiple interrupts to show up, breaking the floppy init code. This fix simply changes the delivery mode to dest_LowestPrio and initially routes interrupts to the first cpu to resolve this issue. Signed-off-by:
-
Jason Davis authored
Signed-off-by:
-
Olaf Hering authored
An attempt to fix the random poking in legacy io space... ask for permissions before poking at io ports. this affects detect_and_report_winbond() and detect_and_report_smsc(). Its unclear what port number is returned in show_parconfig_winbond(), so just request the region as well and free the probing region. Signed-off-by:
Olaf Hering <olh@suse.de> Signed-off-by:
-
Andrew Morton authored
Page allocation failures are expected when add_to_swap() tries to allocate radix-tree nodes without PF_MEMALLOC. Kill the __alloc_pages() warnings. Signed-off-by:
-
Takashi Iwai authored
sound/core/pcm_native.c was assuming that all architectures implement io_remap_page_range() in the same way. They don't, so the sparc64 build broke. mmap callback is added to PCM ops. The driver can implement its own mmap method. Also, mmapping of the status and control records is available only on known coherent architectures now. Signed-off-by:
-
Andrew Morton authored
Fix the build with CONFIG_SHMEM=n Signed-off-by:
-
Andrew Morton authored
I mucked this up. Signed-off-by:
-
bk://kernel.bkbits.net/davem/net-2.6Linus Torvalds authored
into ppc970.osdl.org:/home/torvalds/v2.6/linux
-
bk://kernel.bkbits.net/davem/sparc-2.6Linus Torvalds authored
into ppc970.osdl.org:/home/torvalds/v2.6/linux
-
Kirill Korotaev authored
Dmitry Torokhov triggered a problem in the new pidhash macros: These do_each_task_pid()/while_each_task_pid() do loop forever if you use a 'continue' inside them. The end of the loop has to be inside the "while()" statement, so as to not make "continue" jump over it.
-
http://lia64.bkbits.net/linux-ia64-release-2.6.9Linus Torvalds authored
into ppc970.osdl.org:/home/torvalds/v2.6/linux
-
Tony Luck authored
into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9
-
Tony Luck authored
into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9
-
Tony Luck authored
into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9
-
Tony Luck authored
into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9
-
Tony Luck authored
into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9
-
Tony Luck authored
into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9
-
Tony Luck authored
into agluck-lia64.sc.intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.9
-
