- Dec 21, 2007
-
-
Christoph Hellwig authored
Make most places that use spu_acquire/spu_acquire_saved interruptible, this allows getting out of the spufs code when e.g. pressing ctrl+c. There are a few places where we get called e.g. from spufs teardown routines were we can't simply err out so these are left with a comment. For now I've also not touched the poll routines because it's open what libspe would expect in terms of interrupted system calls. Acked-by:
Arnd Bergmann <arnd.bergmann@de.ibm.com> Signed-off-by:
Christoph Hellwig <hch@lst.de> Signed-off-by:
Jeremy Kerr <jk@ozlabs.org> Signed-off-by:
Paul Mackerras <paulus@samba.org>
-
Christoph Hellwig authored
The simple attr macros currently used by spufs can't deal with the handlers returning errors, which is required to make the state_mutex interruptible. This adds a local copy that allows for an error return from the get/set handlers. Signed-off-by:
-
Luke Browning authored
Change spufs_spu_run so that the context is queued directly to the scheduler and the controlling thread advances directly to spufs_wait() for spe errors and exceptions. nosched contexts are treated the same as before. Fixes from Christoph Hellwig <hch@lst.de> Signed-off-by:
Luke Browning <lukebr@linux.vnet.ibm.com> Signed-off-by:
-
Masato Noguchi authored
This changes the spu context switch code to not write to reserved bits of spu interrupt status register. The architecture book says the reserved fields should be set to zero. Signed-off-by:
Masato Noguchi <Masato.Noguchi@jp.sony.com> Signed-off-by:
-
Luke Browning authored
Need to re-check priority after dropping lock. Otherwise, a more favored context may be preempted. Signed-off-by:
-
Luke Browning authored
This cleans up spu_run_init so that it does all of the spu initialization for spufs_run_spu. It initializes the spu context as much as possible before it activates the spu and writes the runcntl register. Signed-off-by:
-
Jeremy Kerr authored
Based on original patches from Arnd Bergmann <arnd.bergman@de.ibm.com>; and Luke Browning <lukebr@linux.vnet.ibm.com> Currently, spu contexts need to be loaded to the SPU in order to take class 0 and class 1 exceptions. This change makes the actual interrupt-handlers much simpler (ie, set the exception information in the context save area), and defers the handling code to the spufs_handle_class[01] functions, called from spufs_run_spu. This should improve the concurrency of the spu scheduling leading to greater SPU utilization when SPUs are overcommited. Signed-off-by:
-
Jeremy Kerr authored
Add a few #defines for the class 0, 1 and 2 interrupt status bits, and use them instead of magic numbers when we're setting or checking for these interrupts. Also, add a #define for the class 2 mailbox threshold interrupt mask. Signed-off-by:
-
Jeremy Kerr authored
When doing a poll on the mbox stat file of a swapped-out context, we clear the class 0 interrupt status, rather than the class 2 interrupt status. This change corrects the poll operation to clear the correct interrupt. Signed-off-by:
-
Luke Browning authored
This change encapsulates the spu_privcntl_RW register so that it can be written through backing ops. This is necessary so that spu contexts can be initialized and queued to the scheduler in spufs_run_spu. Signed-off-by:
-
Arnd Bergmann authored
This change disables the logic that faults-in spu contexts under the covers from the page fault handler. When a fault requires a runnable context, the handler will block until the context is scheduled by other means. Signed-off-by:
-
Jeremy Kerr authored
Currently, part of the spufs code (switch.o, lscsa_alloc.o and fault.o) is compiled directly into the kernel. This change moves these components of spufs into the kernel. The lscsa and switch objects are fairly straightforward to move in. For the fault.o module, we split the fault-handling code into two parts: a/p/p/c/spu_fault.c and a/p/p/c/spufs/fault.c. The former is for the in-kernel spu_handle_mm_fault function, and we move the rest of the fault-handling code into spufs. Signed-off-by:
-
Julio M. Merino Vidal authored
Fix a few typos in the spufs scheduler comments Signed-off-by:
Julio M. Merino Vidal <jmerino@ac.upc.edu> Signed-off-by:
-
Masato Noguchi authored
Add platform specific SPU run control routines to the spufs. The current spufs implementation uses the SPU master run control bit (MFC_SR1[S]) to control SPE execution, but the PS3 hypervisor does not support the use of this feature. This change adds the run control wrapper routies spu_enable_spu() and spu_disable_spu(). The bare metal routines use the master run control bit, and the PS3 specific routines use the priv2 run control register. An outstanding enhancement for the PS3 would be to add a guard to check for incorrect access to the spu problem state when the spu context is disabled. This check could be implemented with a flag added to the spu context that would inhibit mapping problem state pages, and a routine to unmap spu problem state pages. When the spu is enabled with ps3_enable_spu() the flag would be set allowing pages to be mapped, and when the spu is disabled with ps3_disable_spu() the flag would be cleared and mapped problem state pages would be unmapped. Signed-off-by:
Geoff Levand <geoffrey.levand@am.sony.com> Signed-off-by:
-
- Dec 19, 2007
-
-
Jeremy Kerr authored
Currently, we have a possibilty that the SLBs setup during context switch don't cover the entirety of the necessary lscsa and code regions, if these regions cross a segment boundary. This change checks the start and end of each region, and inserts a SLB entry for each, if unique. We also remove the assumption that the spu_save_code and spu_restore_code reside in the same segment, by using the specific code array for save and restore. Signed-off-by:
Arnd Bergmann <arnd@arndb.de>
-
Jeremy Kerr authored
Add a function spu_64k_pages_available(), so that we can abstract the explicity use of mmu_psize_defs() in lssca_alloc.c Signed-off-by:
-
Jeremy Kerr authored
Currently, the SPU context switch code (spufs/switch.c) sets up the SPU's SLBs directly, which requires some low-level mm stuff. This change moves the kernel SLB setup to spu_base.c, by exposing a function spu_setup_kernel_slbs() to do this setup. This allows us to remove the low-level mm code from switch.c, making it possible to later move switch.c to the spufs module. Also, add a struct spu_slb for the cases where we need to deal with SLB entries. Signed-off-by:
-
- Nov 20, 2007
-
-
Jeremy Kerr authored
We can currently cause an oops by repeatedly creating and destroying contexts, while doing getdents() calls on the "/spu" directory. This is due to the context's top-level dentry remaining hashed while the context is being destroyed. Fix this by unhashing the context's dentry with the dentry->d_inode->i_mutex held. This way, we'll hit the check for d_unhashed in dentry_readdir, and won't be included in the list of subdirs for /spu. test: spufs-testsuite:tests/01-spu_create/07-destroy-vs-readdir-race Signed-off-by:
-
- Oct 19, 2007
-
-
Jan Engelhardt authored
* Convert files to UTF-8. * Also correct some people's names (one example is Eißfeldt, which was found in a source file. Given that the author used an ß at all in a source file indicates that the real name has in fact a 'ß' and not an 'ss', which is commonly used as a substitute for 'ß' when limited to 7bit.) * Correct town names (Goettingen -> Göttingen) * Update Eberhard Mönkeberg's address (http://lkml.org/lkml/2007/1/8/313 ) Signed-off-by:Jan Engelhardt <jengelh@gmx.de> Signed-off-by:
Adrian Bunk <bunk@kernel.org>
-
- Oct 17, 2007
-
-
Christoph Lameter authored
Slab constructors currently have a flags parameter that is never used. And the order of the arguments is opposite to other slab functions. The object pointer is placed before the kmem_cache pointer. Convert ctor(void *object, struct kmem_cache *s, unsigned long flags) to ctor(struct kmem_cache *s, void *object) throughout the kernel [akpm@linux-foundation.org: coupla fixes] Signed-off-by:Christoph Lameter <clameter@sgi.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
- Oct 12, 2007
-
-
Paul Mackerras authored
This makes the kernel use 1TB segments for all kernel mappings and for user addresses of 1TB and above, on machines which support them (currently POWER5+, POWER6 and PA6T). We detect that the machine supports 1TB segments by looking at the ibm,processor-segment-sizes property in the device tree. We don't currently use 1TB segments for user addresses < 1T, since that would effectively prevent 32-bit processes from using huge pages unless we also had a way to revert to using 256MB segments. That would be possible but would involve extra complications (such as keeping track of which segment size was used when HPTEs were inserted) and is not addressed here. Parts of this patch were originally written by Ben Herrenschmidt. Signed-off-by:
-
- Sep 26, 2007
-
-
Jeremy Kerr authored
The commit 8b6f50ef seems to have been affected by a mismerge of a duplicate patch (d054b36f) - both the spufs_dir_contents and spufs_dir_nosched_contents have been given write-only signal notification files. This change reverts the spufs_dir_contents array to use the readable signal notification file implementation. Signed-off-by:
-
- Sep 19, 2007
-
-
Christoph Hellwig authored
find_victim can dereference a NULL pointer when iterating over the list of victim spus because list_mutex only guarantees spu->ct to be stable, but of course not to be non-NULL. Also fix find_victim to not call spu_unbind_context without list_mutex because that violates the above guarantee. Signed-off-by:
-
Michael Ellerman authored
This patch adds DEFINE_SPUFS_ATTRIBUTE(), a wrapper around DEFINE_SIMPLE_ATTRIBUTE which does the specified locking for the get routine for us. Unfortunately we need two get routines (a locked and unlocked version) to support the coredump code. This hides one of those (the locked version) inside the macro foo. Signed-off-by:
Michael Ellerman <michael@ellerman.id.au> Signed-off-by:
-
Michael Ellerman authored
Currently the spu coredump code doesn't respect the ulimit, it should. Signed-off-by:
-
Michael Ellerman authored
Rework spufs_coredump_extra_notes_write() to check for and return errors. If we're coredumping to a pipe we can't trust file->f_pos, we need to maintain the foffset value passed to us. The cleanest way to do this is to have the low level write routine increment foffset when we've successfully written. Signed-off-by:
-
Michael Ellerman authored
Because spufs might be built as a module, we can't have other parts of the kernel calling directly into it, we need stub routines that check first if the module is loaded. Currently we have two structures which hold callbacks for these stubs, the syscalls are in spufs_calls and the coredump calls are in spufs_coredump_calls. In both cases the logic for registering/unregistering is essentially the same, so we can simplify things by combining the two. Signed-off-by:
-
Michael Ellerman authored
Signed-off-by:
-
Michael Ellerman authored
The SPUFS attribute get routines take a void * because the generic attribute code doesn't know what sort of data it's passing around. However our internal __spufs_get_foo() routines can take a spu_context * directly, which saves plonking it in and out of a void * again. Signed-off-by:
-
Michael Ellerman authored
The spufs_coredump_read array is NULL terminated, and we also store the size. We only need one or the other, and the other arrays in file.c are NULL terminated, so do that. Signed-off-by:
-
Michael Ellerman authored
The routine to dump the local store, __spufs_mem_read(), does not take the spu_lslr_RW value into account - so we shouldn't check it when we're calculating the size either. Signed-off-by:
-
Michael Ellerman authored
Unfortunately GDB expects some of the SPU coredump values to be identical in format to what is found in spufs. This means we need to dump some of the values as ASCII strings, not the actual values. Because we don't know what the values will be, we always print the values with the format "0x%.16lx", that way we know the result will be 19 bytes. do_coredump_read() doesn't take a __user buffer, so remove the annotation, and because we know that it's safe to just snprintf() directly to it. Signed-off-by:
-
Michael Ellerman authored
The spufs_coredump_reader array contains the size of the data that will be returned by the read routine. Currently these are specified as literals, and though some are obvious, sizeof(u32) == 4, others are not, 69 * 8 == ??? Instead, use sizeof() whatever type is returned by each routine, or in the case of spufs_mem_read() the #define LS_SIZE. Signed-off-by:
-
Michael Ellerman authored
It makes sense to stop the SPU processes as soon as possible. Also if we dont acquire_saved() I think there's a possibility that the value in csa.priv2.spu_lslr_RW won't be accurate. Signed-off-by:
-
Michael Ellerman authored
Remove the ctx_info struct entirely, and also the ctx_info_list. This fixes a race where two processes can clobber each other's ctx_info structs. Instead of using the list, we just repeat the search through the file descriptor table. Signed-off-by:
-
Michael Ellerman authored
Extract the logic for searching through the file descriptors for spu contexts into a separate routine, coredump_next_context(), so we can use it elsewhere in future. In the process we flatten the for loop, and move the NOSCHED test into coredump_next_context(). Signed-off-by:
-
Jeremy Kerr authored
Based on an original patch from Masato Noguchi <Masato.Noguchi@jp.sony.com>. We're currently not restoring the SPE decrementer as specified by the CBE handbook. This change fixes our implementation to match, and makes the function read more like the docs. Signed-off-by:
-
Jeremy Kerr authored
At present, a built-in spufs will not use the spufs_calls callbacks, but directly call sys_spu_create. This saves us an indirect branch, but means we have duplicated functions - one for CONFIG_SPU_FS=y and one for =m. This change unifies the spufs syscall path, and provides access to the spufs_calls structure through a get/put pair. At present, the only user of the spufs_calls structure is spu_syscalls.c, but this will facilitate adding the coredump calls later. Everyone likes numbers, right? Here's a before/after comparison with CONFIG_SPU_FS=y, doing spu_create(); close(); 64k times. Before: [jk@cell ~]$ time ./spu_create performing 65536 spu_create calls real 0m24.075s user 0m0.146s sys 0m23.925s After: [jk@cell ~]$ time ./spu_create performing 65536 spu_create calls real 0m24.777s user 0m0.141s sys 0m24.631s So, we're adding around 11us per syscall, at the benefit of having only one syscall path. Signed-off-by:
-
Andre Detsch authored
Affinity reference point location (gang->aff_ref_spu) is reset when the whole gang is descheduled. However, the last member of a gang can be descheduled while we are trying to schedule another member of the gang. This was leading to a race condition, and the code was using gang->aff_ref_spu in an unsafe manner. By holding the gang->aff_mutex a little bit longer, and increment gang->aff_sched_count (which controls when gang->aff_ref_spu should be reset) a little bit earlier, the problem is fixed. Signed-off-by:
Andre Detsch <adetsch@br.ibm.com> Signed-off-by:
-
Sebastian Siewior authored
According to the comment in spufs_init_isolated_loader(), the isolated loader should be aligned on a 16 byte boundary. ARCH_{KMALLOC,SLAB}_MINALIGN is not defined so only 8 byte alignment is guaranteed. This enforces alignment via __get_free_pages. Signed-off-by:Sebastian Siewior <sebastian@breakpoint.cc> Signed-off-by:
-
