git commit 0c8c0f03e3
"x86/fpu, sched: Dynamically allocate 'struct fpu'"
moved the thread_struct to the end of the task_struct.
This causes some of the offsets used in entry.S to overflow their
instruction operand field. To fix this use aghi to create a
dedicated pointer for the thread_struct.
Signed-off-by: Martin Schwidefsky <[email protected]>
If the kernel is compiled with gcc 5.1 and the XZ compression option
the decompress_kernel function calls _sclp_print_early in 64-bit mode
while the content of the upper register half of %r6 is non-zero.
This causes a specification exception on the servc instruction in
_sclp_servc.
The _sclp_print_early function saves and restores the upper registers
halves but it fails to clear them for the 31-bit code of the mini sclp
driver.
Cc: <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
The SCLP interface to query, configure and deconfigure CPUs actually
operates on cores. For a machine without the multi-threading faciltiy
a CPU and a core are equivalent but starting with System z13 a core
can have multiple hardware threads, also referred to as logical CPUs.
To avoid confusion replace the word 'cpu' with 'core' in the SCLP
interface. Also replace MAX_CPU_ADDRESS with SCLP_MAX_CORES.
The core-id is an 8-bit field, the maximum thread id is in the range
0-31. The theoretical limit for the CPU address is therefore 8191.
Signed-off-by: Martin Schwidefsky <[email protected]>
On a (theoretical) system where the read-cpu-info SCLP command does
not work but SMT is enabled, the sigp detection loop may not find
all configured cores. The maximum CPU address needs to be shifted
with smp_cpu_mt_shift.
Signed-off-by: Martin Schwidefsky <[email protected]>
Git commit 152125b7a8
"s390/mm: implement dirty bits for large segment table entries"
broke the pmd_pfn function, it changed the return value from
'unsigned long' to 'int'. This breaks all machine configurations
with memory above the 8TB line.
Cc: [email protected] # 3.17+
Signed-off-by: Martin Schwidefsky <[email protected]>
After the file ptes have been removed the bit combination used to
encode non-linear mappings can be reused for the swap ptes. This
frees up a precious pte software bit. Reflect the change in the
swap encoding in the comments and do some cleanup while we are
at it.
Signed-off-by: Martin Schwidefsky <[email protected]>
The dirty & young bit from the pmd is not copied correctly to the
pseudo pte in __pmd_to_pte. In fact it is not copied at all, the
bits get lost. As the old style huge page currently does not need
the dirty & young information this has no effect, but may be needed
in the future.
Signed-off-by: Martin Schwidefsky <[email protected]>
Replacing a 2K page table with a 4K page table while a VMA is active
for the affected memory region is fundamentally broken. Rip out the
page table reallocation code and replace it with a simple system
control 'vm.allocate_pgste'. If the system control is set the page
tables for all processes are allocated as full 4K pages, even for
processes that do not need it.
Signed-off-by: Martin Schwidefsky <[email protected]>
Introduce KEXEC_CONTROL_MEMORY_GFP to allow the architecture code
to override the gfp flags of the allocation for the kexec control
page. The loop in kimage_alloc_normal_control_pages allocates pages
with GFP_KERNEL until a page is found that happens to have an
address smaller than the KEXEC_CONTROL_MEMORY_LIMIT. On systems
with a large memory size but a small KEXEC_CONTROL_MEMORY_LIMIT
the loop will keep allocating memory until the oom killer steps in.
Signed-off-by: Martin Schwidefsky <[email protected]>
For compat tasks the mmap randomization does not use the maximum
randomization value from mmap_rnd_mask but the fixed value of 0x7ff.
This needs to be respected in the definition of STACK_RND_MASK as
well.
Signed-off-by: Martin Schwidefsky <[email protected]>
The switch_mm function does nothing in case the prev and next mm
are the same. It can happen that a crst_table_downgrade has changed
the top-level pgd in the meantime on a different CPU. Always store
the new ASCE to be picked up in entry.S.
[[email protected]]: Bug was introduced with git commit
53e857f308 ("s390/mm,tlb: race of lazy TLB flush vs. recreation
of TLB entries") and causes random crashes due to broken page tables
being used.
Reported-by: Dominik Vogt <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
Signed-off-by: Heiko Carstens <[email protected]>
Until we have hard performance data about the effects of CAD in the
spinlock loop disable the instruction by default.
Signed-off-by: Martin Schwidefsky <[email protected]>
The base address (STACK_TOP / 3 * 2) for a 64-bit program is two thirds
into the 4GB segment at 0x2aa00000000. The randomization added on z13
can eat another 1GB of the remaining 1.33GB to the next 4GB boundary.
In the worst case 300MB are left for the executable + bss which may
cross into the next 4GB segment. This is bad for branch prediction,
therefore align the base address to 4GB to give the program more room
before it crosses the 4GB boundary.
Signed-off-by: Martin Schwidefsky <[email protected]>
Git commit 8d8f2e18a6dbd3d09dd918788422e6ac8c878e96
"s390/vdso: ectg gettime support for CLOCK_THREAD_CPUTIME_ID"
broke clock_gettime for CLOCK_THREAD_CPUTIME_ID.
Git commit c742b31c03
"fast vdso implementation for CLOCK_THREAD_CPUTIME_ID"
introduced the ECTG for clock id -2. Correct would have been
clock id -3.
Fix the whole mess, CLOCK_THREAD_CPUTIME_ID is based on
CPUCLOCK_SCHED and can not be speed up by the vdso. A speedup
is only available for clock id -3 which is CPUCLOCK_VIRT for
the task currently running on the CPU.
Signed-off-by: Martin Schwidefsky <[email protected]>
Add the compare-and-delay instruction to the spin-lock and rw-lock
retry loops. A CPU executing the compare-and-delay instruction stops
until the lock value has changed. This is done to make the locking
code for contended locks to behave better in regard to the multi-
hreading facility. A thread of a core executing a compare-and-delay
will allow the other threads of a core to get a larger share of the
core resources.
Signed-off-by: Martin Schwidefsky <[email protected]>
The unit check handler for 3480 / 3490 tapes used to print a different
warning message for erpa code 0x57 dependent on the device type.
The warning messages have been remove in a cleanup, the if statement
is a left over. Remove it.
Reported-by: Fraser Brown <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
The multi-threading facility is introduced with the z13 processor family.
This patch adds code to detect the multi-threading facility. With the
facility enabled each core will surface multiple hardware threads to the
system. Each hardware threads looks like a normal CPU to the operating
system with all its registers and properties.
The SCLP interface reports the SMT topology indirectly via the maximum
thread id. Each reported CPU in the result of a read-scp-information
is a core representing a number of hardware threads.
To reflect the reduced CPU capacity if two hardware threads run on a
single core the MT utilization counter set is used to normalize the
raw cputime obtained by the CPU timer deltas. This scaled cputime is
reported via the taskstats interface. The normal /proc/stat numbers
are based on the raw cputime and are not affected by the normalization.
Signed-off-by: Martin Schwidefsky <[email protected]>
Avoid cache aliasing on z13 by aligning shared objects to multiples
of 512K. The virtual addresses of a page from a shared file needs
to have identical bits in the range 2^12 to 2^18.
Signed-off-by: Martin Schwidefsky <[email protected]>
pmd_to_page() is only available if USE_SPLIT_PMD_PTLOCKS is defined.
The use of pmd_to_page in the gmap code can cause compile errors if
NR_CPUS is smaller than SPLIT_PTLOCK_CPUS. Do not use pmd_to_page
outside of USE_SPLIT_PMD_PTLOCKS sections.
Reported-by: Mike Frysinger <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
git commit 8461b63ca0
"s390: translate cputime magic constants to macros"
introduce a built error for 31-bit:
kernel/built-in.o: In function `posix_cpu_timer_set':
posix-cpu-timers.c:(.text+0x2a8cc): undefined reference to `__udivdi3'
The original code is actually broken for 31-bit and has been
corrected by the above commit by forcing the compiler to use
64-bit arithmetic through the CPUTIME_PER_USEC define.
To fix the compile error replace the 64-bit division with
a call to __div().
Signed-off-by: Martin Schwidefsky <[email protected]>
The pmd_free_tlb function fails to call pgtable_pmd_page_dtor.
Without the call the ptlock for the pmd tables will not be freed.
Add the missing call.
Signed-off-by: Martin Schwidefsky <[email protected]>
To improve the output of the perf tool hide most of the symbols
from entry[64].S by using the '.L' prefix.
Signed-off-by: Martin Schwidefsky <[email protected]>
On machines with support for vector registers the signal frame includes
an area for the vector registers and the ptrace regset interface allow
read and write. This is true even if the task never used any vector
instruction. Only elf core dumps do not include the vector registers,
to make things consistent always include the vector register note in
core dumps create on a machine with vector register support.
Signed-off-by: Martin Schwidefsky <[email protected]>
The copy_thread function fails to reset the p->thread.vxrs pointer.
This causes the child to use the same vector register save area,
causing both data corruptions and multiple frees of the memory for
the save area after the tasks sharing the save area terminate.
Signed-off-by: Martin Schwidefsky <[email protected]>
Allow to specify the Compoment ID for Call Home via the kernel
configuration. This removes the need for distribution specific
patch against the sclp_async.c source file.
Signed-off-by: Martin Schwidefsky <[email protected]>
The floating point registers of a process that uses vector instruction are
not store into task->thread.fp_regs anymore but in the upper halves of the
first 16 vector registers.
The ptrace interface for the peeks and pokes to the user area fails to take
this into account. Fix __peek_user[_compat] and __poke_user[_compat]
to use the vector array for the floating pointer register if the process
has one.
Signed-off-by: Martin Schwidefsky <[email protected]>
Fix the following warnings from the sparse code checker:
arch/s390/kernel/signal.c:374:38: warning: cast removes address space of expression
arch/s390/kernel/signal.c:374:65: warning: incorrect type in initializer (different address spaces)
arch/s390/kernel/signal.c:374:65: expected unsigned short [noderef] [usertype] <asn:1>*svc
arch/s390/kernel/signal.c:374:65: got void *
arch/s390/kernel/compat_signal.c:437:38: warning: cast removes address space of expression
arch/s390/kernel/compat_signal.c:437:65: warning: incorrect type in initializer (different address spaces)
arch/s390/kernel/compat_signal.c:437:65: expected unsigned short [noderef] [usertype] <asn:1>*svc
arch/s390/kernel/compat_signal.c:437:65: got void *
Signed-off-by: Martin Schwidefsky <[email protected]>
The page table lock is acquired with a call to get_locked_pte,
replace the plain spin_unlock with the correct unlock function
pte_unmap_unlock.
Signed-off-by: Martin Schwidefsky <[email protected]>
Fix the following warnings from the sparse code checker:
arch/s390/include/asm/pci_io.h:165:49: warning: cast removes address space of expression
arch/s390/pci/pci.c:476:44: warning: cast removes address space of expression
arch/s390/pci/pci.c:491:36: warning: incorrect type in argument 2 (different address spaces)
arch/s390/pci/pci.c:491:36: expected void [noderef] <asn:2>*addr
arch/s390/pci/pci.c:491:36: got void *<noident>
Signed-off-by: Martin Schwidefsky <[email protected]>
The kernel build for s390 fails for gcc compilers with version 3.x,
set the minimum required version of gcc to version 4.3.
As the atomic builtins are available with all gcc 4.x compilers,
use the __sync_val_compare_and_swap and __sync_bool_compare_and_swap
functions to replace the complex macro and inline assembler magic
in include/asm/cmpxchg.h. The compiler can just-do-it and generates
better code with the builtins.
While we are at it use __sync_bool_compare_and_swap for the
_raw_compare_and_swap function in the spinlock code as well.
Signed-off-by: Martin Schwidefsky <[email protected]>
Found this in the message log on a s390 system:
BUG kmalloc-192 (Not tainted): Poison overwritten
Disabling lock debugging due to kernel taint
INFO: 0x00000000684761f4-0x00000000684761f7. First byte 0xff instead of 0x6b
INFO: Allocated in call_usermodehelper_setup+0x70/0x128 age=71 cpu=2 pid=648
__slab_alloc.isra.47.constprop.56+0x5f6/0x658
kmem_cache_alloc_trace+0x106/0x408
call_usermodehelper_setup+0x70/0x128
call_usermodehelper+0x62/0x90
cgroup_release_agent+0x178/0x1c0
process_one_work+0x36e/0x680
worker_thread+0x2f0/0x4f8
kthread+0x10a/0x120
kernel_thread_starter+0x6/0xc
kernel_thread_starter+0x0/0xc
INFO: Freed in call_usermodehelper_exec+0x110/0x1b8 age=71 cpu=2 pid=648
__slab_free+0x94/0x560
kfree+0x364/0x3e0
call_usermodehelper_exec+0x110/0x1b8
cgroup_release_agent+0x178/0x1c0
process_one_work+0x36e/0x680
worker_thread+0x2f0/0x4f8
kthread+0x10a/0x120
kernel_thread_starter+0x6/0xc
kernel_thread_starter+0x0/0xc
There is a use-after-free bug on the subprocess_info structure allocated
by the user mode helper. In case do_execve() returns with an error
____call_usermodehelper() stores the error code to sub_info->retval, but
sub_info can already have been freed.
Regarding UMH_NO_WAIT, the sub_info structure can be freed by
__call_usermodehelper() before the worker thread returns from
do_execve(), allowing memory corruption when do_execve() failed after
exec_mmap() is called.
Regarding UMH_WAIT_EXEC, the call to umh_complete() allows
call_usermodehelper_exec() to continue which then frees sub_info.
To fix this race the code needs to make sure that the call to
call_usermodehelper_freeinfo() is always done after the last store to
sub_info->retval.
Signed-off-by: Martin Schwidefsky <[email protected]>
Reviewed-by: Oleg Nesterov <[email protected]>
Cc: Tetsuo Handa <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>
Analog to ptep_get_and_clear_full define a variant of the
pmpd_get_and_clear primitive which gets the full hint from the
mmu_gather struct. This allows s390 to avoid a costly instruction
when destroying an address space.
Signed-off-by: Martin Schwidefsky <[email protected]>
The last high frequency call site of the STCK instruction is
do_account_vtime. Replace it with the faster STCKF instruction.
Signed-off-by: Martin Schwidefsky <[email protected]>
The vector extension introduces 32 128-bit vector registers and a set of
instruction to operate on the vector registers.
The kernel can control the use of vector registers for the problem state
program with a bit in control register 0. Once enabled for a process the
kernel needs to retain the content of the vector registers on context
switch. The signal frame is extended to include the vector registers.
Two new register sets NT_S390_VXRS_LOW and NT_S390_VXRS_HIGH are added
to the regset interface for the debugger and core dumps.
Signed-off-by: Martin Schwidefsky <[email protected]>
Move the C functions and definitions related to the idle state handling
to arch/s390/include/asm/idle.h and arch/s390/kernel/idle.c. The function
s390_get_idle_time is renamed to arch_cpu_idle_time and vtime_stop_cpu to
enabled_wait.
Signed-off-by: Martin Schwidefsky <[email protected]>
Move the nohz_delay bit from the s390_idle data structure to the
per-cpu flags. Clear the nohz delay flag in __cpu_disable and
remove the cpu hotplug notifier that used to do this.
Signed-off-by: Martin Schwidefsky <[email protected]>
The sysfs attributes /sys/devices/system/cpu/cpu0/idle_count and
/sys/devices/system/cpu/cpu0/idle_time_us are reset to zero every
time a CPU is set online. The idle and iowait fields in /proc/stat
corresponding to idle_time_us are not reset. To make things
consistent do not reset the data for the sys attributes.
Signed-off-by: Martin Schwidefsky <[email protected]>
Fix calculation to decide if a 4-level kernel page table is required.
Git commit c972cc60c2 "s390/vmalloc: have separate modules area"
added the separate module area which reduces the size of the vmalloc
area but fails to take it into account for the 3 vs 4 level page table
decision.
Signed-off-by: Martin Schwidefsky <[email protected]>
The call to topology_init is too late for the set_sched_topology call.
The initial scheduling domain structure has already been established
with default topology array. Use the smp_cpus_done() call to get the
s390 specific topology array registered early enough.
Cc: [email protected] # v3.16+
Signed-off-by: Martin Schwidefsky <[email protected]>
Make use of the load-and-add, load-and-or and load-and-and instructions
to atomically update the read-write lock without a compare-and-swap loop.
Signed-off-by: Martin Schwidefsky <[email protected]>
Set the write-lock bit in the out-of-line rwlock code to indicate that
a writer is waiting. Additional readers will no be able to get the lock
until at least one writer got the lock. Additional writers have to wait
for the first writer to release the lock again.
Signed-off-by: Martin Schwidefsky <[email protected]>
Add an owner field to the arch_rwlock_t to be able to pass the timeslice
of a virtual CPU with diagnose 0x9c to the lock owner in case the rwlock
is write-locked. The undirected yield in case the rwlock is acquired
writable but the lock is read-locked is removed.
Signed-off-by: Martin Schwidefsky <[email protected]>
The event type masks can change asynchronously. These changes are reported
by SCLP to the OS by state-change events which are retrieved with the read
event data command. The SCLP driver has a request queue, there is a window
where the read event data request has not completed yet but the SCLP console
drivers are trying to queue output requests. As the masks are not updated
yet the requests are discarded.
The simplest fix is to queue the console requests independent of the
event type masks and rely on SCLP to return with an error code if a
specific event type is not available.
Signed-off-by: Martin Schwidefsky <[email protected]>
If gettimeofday / clock_gettime are called multiple times in a row
the STCK instruction will stall until a difference in the result is
visible. This unnecessarily slows down the vdso calls, use stcke
instead of stck to get rid of the stall.
Signed-off-by: Martin Schwidefsky <[email protected]>
The explicit NULL pointer check on the timespec argument is only
required for clock_getres but not for clock_gettime.
Signed-off-by: Martin Schwidefsky <[email protected]>
The radix tree rework removed all code that uses the gmap_rmap
and gmap_pgtable data structures. Remove these outdated definitions.
Signed-off-by: Martin Schwidefsky <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>
Add an addressing limit to the gmap address spaces and only allocate
the page table levels that are needed for the given limit. The limit
is fixed and can not be changed after a gmap has been created.
Signed-off-by: Martin Schwidefsky <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>
Store the target address for the gmap segments in a radix tree
instead of using invalid segment table entries. gmap_translate
becomes a simple radix_tree_lookup, gmap_fault is split into the
address translation with gmap_translate and the part that does
the linking of the gmap shadow page table with the process page
table.
A second radix tree is used to keep the pointers to the segment
table entries for segments that are mapped in the guest address
space. On unmap of a segment the pointer is retrieved from the
radix tree and is used to carry out the segment invalidation in
the gmap shadow page table. As the radix tree can only store one
pointer, each host segment may only be mapped to exactly one
guest location.
Signed-off-by: Martin Schwidefsky <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>
Make the order of arguments for the gmap calls more consistent,
if the gmap pointer is passed it is always the first argument.
In addition distinguish between guest address and user address
by naming the variables gaddr for a guest address and vmaddr for
a user address.
Signed-off-by: Martin Schwidefsky <[email protected]>
Reviewed-by: Cornelia Huck <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>
The sclp line mode terminal driver scans the tty output for '\t',
there is no need to set the XTABS flag in c_oflag.
Signed-off-by: Martin Schwidefsky <[email protected]>
git commit 37f81fa1f6
"n_tty: do O_ONLCR translation as a single write"
surfaced a bug in the 3215 device driver. In combination this
broke tab expansion for tty ouput.
The cause is an asymmetry in the behaviour of tty3215_ops->write
vs tty3215_ops->put_char. The put_char function scans for '\t'
but the write function does not.
As the driver has logic for the '\t' expansion remove XTABS
from c_oflag of the initial termios as well.
Reported-by: Stephen Powell <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
The virtual-machine cpu information data block and the cpu-id of
the boot cpu can be used as source of device randomness.
Signed-off-by: Martin Schwidefsky <[email protected]>
The large segment table entry format has block of bits for the
ACC/F values for the large page. These bits are valid only if
another bit (AV bit 0x10000) of the segment table entry is set.
The ACC/F bits do not have a meaning if the AV bit is off.
This allows to put the THP splitting bit, the segment young bit
and the new segment dirty bit into the ACC/F bits as long as
the AV bit stays off. The dirty and young information is only
available if the pmd is large.
Signed-off-by: Martin Schwidefsky <[email protected]>
The ccw_device_start in raw3215_start_io can fail. raw3215_try_io
does not check if the request could be started and removes any
pending timer. This can leave the system in a hanging state.
Check for pending request after raw3215_start_io and start a
timer if necessary.
Signed-off-by: Martin Schwidefsky <[email protected]>
The fixup of the inline assembly to restore the floating-point-control
register needs to check for instruction address *after* the lfcp
instruction as the specification and data exceptions are suppresssing.
Signed-off-by: Martin Schwidefsky <[email protected]>
The PSW mask check of the PTRACE_POKEUSR_AREA command is incorrect.
The PSW_MASK_USER define contains the PSW_MASK_ASC bits, the ptrace
interface accepts all combinations for the address-space-control
bits. To protect the kernel space the PSW mask check in ptrace needs
to reject the address-space-control bit combination for home space.
Fixes CVE-2014-3534
Cc: [email protected]
Signed-off-by: Martin Schwidefsky <[email protected]>
The size detection for 3270 terminals with the read-partition command is
broken. The raw3270_reset_device_cb function clears the init_data array,
but if raw3270_writesf_readpart has been called the read-partition command
is queued which needs the init_data array. In this case the size detection
will fail and the invalid command does funny things to the terminal.
Signed-off-by: Martin Schwidefsky <[email protected]>
The uc_sigmask definition in the kernel differs from the one in the
glibc, the kernel uc_sigmask has 64 bits while the glibc verison
is 1024 bits. The extension of the ucontext structure for 64-bit
register support for 31-bit compat processes added a new field
uc_gprs_high which starts 8 bytes after the uc_sigmask field.
As the glibc view of the ucontext assumes a size of 128 bytes for
uc_sigmask add a 120 byte padding to the kernel structure
ucontext_extended after the 8 byte uc_sigmask.
Signed-off-by: Martin Schwidefsky <[email protected]>
This patch fixes a problem introduced with git commit beef560b4c
"s390/uaccess: simplify control register updates".
The switch_mm function is not called if the next process is a kernel
thread without an attached mm or is a nop if the mm does not change.
But CR1 still needs to be loaded with the kernel ASCE in case the
code returns to a uaccess function that uses the secondary space mode.
In addition move the set_fs call from finish_arch_switch to
finish_arch_post_lock_switch and then remove finish_arch_switch.
Signed-off-by: Martin Schwidefsky <[email protected]>
Remove the 96-byte irb array from the lowcore and create a per-cpu
variable instead. That way we will pick up any change in the definition
of the struct irb automatically.
Acked-By: Sebastian Ott <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
Add -fno-delete-null-pointer-checks to CFLAGS for the code in
arch/s390/boot. Without the option a compressed kernel built with
gcc 4.9 won't boot.
Signed-off-by: Martin Schwidefsky <[email protected]>
Analog to git commit 28b92e09e2
first cast tk->wall_to_monotonic.tv_nsec to u64 before doing
the shift with tk->shift to avoid loosing relevant bits on a
32-bit kernel.
Cc: [email protected] # 3.13+
Signed-off-by: Martin Schwidefsky <[email protected]>
The out of line _raw_read_lock_wait_flags/_raw_write_lock_wait_flags
functions for the arch_read_lock_flags/arch_write_lock_flags calls
fail to re-enable the interrupts after another unsuccessful try to
get the lock with compare-and-swap. The following wait would be
done with interrupts disabled which is suboptimal.
Signed-off-by: Martin Schwidefsky <[email protected]>
In case a lock is contended it is better to do a load-and-test first
before trying to get the lock with compare-and-swap. This helps to avoid
unnecessary cache invalidations of the cacheline for the lock if the
CPU has to wait for the lock. For an uncontended lock doing the
compare-and-swap directly is a bit better, if the CPU does not have the
cacheline in its cache yet the compare-and-swap will get it read-write
immediately while a load-and-test would get it read-only first.
Always to the load-and-test first to avoid the cacheline invalidations
for the contended case outweight the potential read-only to read-write
cacheline upgrade for the uncontended case.
Signed-off-by: Martin Schwidefsky <[email protected]>
The oi and ni instructions used in entry[64].S to set and clear bits
in the thread-flags are not guaranteed to be atomic in regard to other
CPUs. Split the TIF bits into CPU, pt_regs and thread-info specific
bits. Updates on the TIF bits are done with atomic instructions,
updates on CPU and pt_regs bits are done with non-atomic instructions.
Signed-off-by: Martin Schwidefsky <[email protected]>
Always switch to the kernel ASCE in switch_mm. Load the secondary
space ASCE in finish_arch_post_lock_switch after checking that
any pending page table operations have completed. The primary
ASCE is loaded in entry[64].S. With this the update_primary_asce
call can be removed from the switch_to macro and from the start
of switch_mm function. Remove the load_primary argument from
update_user_asce/clear_user_asce, rename update_user_asce to
set_user_asce and rename update_primary_asce to load_kernel_asce.
Signed-off-by: Martin Schwidefsky <[email protected]>
Use the mm semaphore to serialize multiple invocations of s390_enable_skey.
The second CPU faulting on a storage key operation needs to wait for the
completion of the page table update. Taking the mm semaphore writable
has the positive side-effect that it prevents any host faults from
taking place which does have implications on keys vs PGSTE.
Signed-off-by: Martin Schwidefsky <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>
The A register needs to be initialized to zero in the prolog if the
first instruction of the BPF program is BPF_S_LDX_B_MSH to prevent
leaking the content of %r5 to user space.
Cc: <[email protected]>
Signed-off-by: Martin Schwidefsky <[email protected]>
Switch the user dirty bit detection used for migration from the hardware
provided host change-bit in the pgste to a fault based detection method.
This reduced the dependency of the host from the storage key to a point
where it becomes possible to enable the RCP bypass for KVM guests.
The fault based dirty detection will only indicate changes caused
by accesses via the guest address space. The hardware based method
can detect all changes, even those caused by I/O or accesses via the
kernel page table. The KVM/qemu code needs to take this into account.
Signed-off-by: Martin Schwidefsky <[email protected]>
Signed-off-by: Dominik Dingel <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>
The git commit c63badebfe
"s390: optimize control register update" broke the update for
control register 0. After the update do the lctlg from the correct
value.
Cc: <[email protected]> # 3.14
Signed-off-by: Martin Schwidefsky <[email protected]>
The zEC12 machines introduced the local-clearing control for the IDTE
and IPTE instruction. If the control is set only the TLB of the local
CPU is cleared of entries, either all entries of a single address space
for IDTE, or the entry for a single page-table entry for IPTE.
Without the local-clearing control the TLB flush is broadcasted to all
CPUs in the configuration, which is expensive.
The reset of the bit mask of the CPUs that need flushing after a
non-local IDTE is tricky. As TLB entries for an address space remain
in the TLB even if the address space is detached a new bit field is
required to keep track of attached CPUs vs. CPUs in the need of a
flush. After a non-local flush with IDTE the bit-field of attached CPUs
is copied to the bit-field of CPUs in need of a flush. The ordering
of operations on cpu_attach_mask, attach_count and mm_cpumask(mm) is
such that an underindication in mm_cpumask(mm) is prevented but an
overindication in mm_cpumask(mm) is possible.
Signed-off-by: Martin Schwidefsky <[email protected]>
The principles of operations states that the CPU is allowed to create
TLB entries for an address space anytime while an ASCE is loaded to
the control register. This is true even if the CPU is running in the
kernel and the user address space is not (actively) accessed.
In theory this can affect two aspects of the TLB flush logic.
For full-mm flushes the ASCE of the dying process is still attached.
The approach to flush first with IDTE and then just free all page
tables can in theory lead to stale TLB entries. Use the batched
free of page tables for the full-mm flushes as well.
For operations that can have a stale ASCE in the control register,
e.g. a delayed update_user_asce in switch_mm, load the kernel ASCE
to prevent invalid TLBs from being created.
Signed-off-by: Martin Schwidefsky <[email protected]>
If the 3270 device is detached the initial reset device request will
stays pending until the device is operational. A second reset device
call will reuse the same request structure which will cause an oops.
Add a check to see if the reset device request is already pending
and do nothing in this case.
Signed-off-by: Martin Schwidefsky <[email protected]>
This patch adds a parameter 'auto_update' to the con3270
driver, causing the 'auto_update' feature to be disabled
if unset.
The 'auto_update' feature will cause the con3270 driver
to switch to the console view whenever new system messages
are displayed, which makes working on the 3270 terminal
awkward.
Signed-off-by: Martin Schwidefsky <[email protected]>
The PTRACE_SINGLEBLOCK option is used to get control whenever
the inferior has executed a successful branch. The PER option to
implement block stepping is successful-branching event, bit 32
in the PER-event mask.
Signed-off-by: Martin Schwidefsky <[email protected]>
Add airq_iv_alloc and airq_iv_free to allocate and free consecutive
ranges of irqs from the interrupt vector.
Signed-off-by: Martin Schwidefsky <[email protected]>
Signed-off-by: Christian Borntraeger <[email protected]>
Add airq_iv_alloc and airq_iv_free to allocate and free consecutive
ranges of irqs from the interrupt vector.
Signed-off-by: Martin Schwidefsky <[email protected]>
Add the pgtable_pmd_page_ctor/pgtable_pmd_page_dtor calls to the pmd
allocation and free functions and enable ARCH_ENABLE_SPLIT_PMD_PTLOCK
for 64 bit.
Signed-off-by: Martin Schwidefsky <[email protected]>
The pmd pointer passed to pmd_lockptr/pmd_huge_pte can point to any
entry in a pmd table. With USE_SPLIT_PMD_PTLOCKS==1 the code uses
virt_to_page to get a struct page for the pmd table. The virt_to_page
function automatically masks the lower PAGE_SHIFT bits from the
address. But if the size of a pmd table is larger than PAGE_SIZE the
additional bits are not removed from the pmd address and the wrong
page struct is used.
Fix this by explicitely masking the offset in the pmd table from
the pmd pointer.
Signed-off-by: Martin Schwidefsky <[email protected]>
The guest page state needs to be reset to stable for all pages
on initial program load via diagnose 0x308.
Signed-off-by: Martin Schwidefsky <[email protected]>
Git commit 050eef364a "[S390] fix tlb flushing vs. concurrent
/proc accesses" introduced the attach counter to avoid using the
mm_users value to decide between IPTE for every PTE and lazy TLB
flushing with IDTE. That fixed the problem with mm_users but it
introduced another subtle race, fortunately one that is very hard
to hit.
The background is the requirement of the architecture that a valid
PTE may not be changed while it can be used concurrently by another
cpu. The decision between IPTE and lazy TLB flushing needs to be
done while the PTE is still valid. Now if the virtual cpu is
temporarily stopped after the decision to use lazy TLB flushing but
before the invalid bit of the PTE has been set, another cpu can attach
the mm, find that flush_mm is set, do the IDTE, return to userspace,
and recreate a TLB that uses the PTE in question. When the first,
stopped cpu continues it will change the PTE while it is attached on
another cpu. The first cpu will do another IDTE shortly after the
modification of the PTE which makes the race window quite short.
To fix this race the CPU that wants to attach the address space of a
user space thread needs to wait for the end of the PTE modification.
The number of concurrent TLB flushers for an mm is tracked in the
upper 16 bits of the attach_count and finish_arch_post_lock_switch
is used to wait for the end of the flush operation if required.
Signed-off-by: Martin Schwidefsky <[email protected]>
The finish_arch_post_lock_switch is called at the end of the task
switch after all locks have been released. In concept it is paired
with the switch_mm function, but the current code only does the
call in finish_task_switch. Add the call to idle_task_exit and
use_mm. One use case for the additional calls is s390 which will
use finish_arch_post_lock_switch to wait for the completion of
TLB flush operations.
Signed-off-by: Martin Schwidefsky <[email protected]>
The kernel currently crashes with a low-address-protection exception
if a user space process executes an instruction that tries to use the
linkage stack. Set the base-ASTE origin and the subspace-ASTE origin
of the dispatchable-unit-control-table to point to a dummy ASTE.
Set up control register 15 to point to an empty linkage stack with no
room left.
A user space process with a linkage stack instruction will still crash
but with a different exception which is correctly translated to a
segmentation fault instead of a kernel oops.
Signed-off-by: Martin Schwidefsky <[email protected]>
To provide access to the set-partition-resource-parameter interface
to user space add a new attribute to hypfs/debugfs:
* s390_hypsfs/diag_304
The data for the query-partition-resource-parameters command can
be access by a read on the attribute. All other diagnose 0x304
requests need to be submitted via ioctl with CAP_SYS_ADMIN rights.
Signed-off-by: Martin Schwidefsky <[email protected]>
Skip the call to brk_rnd() if the PF_RANDOMIZE flag is not set for
the process. This avoids the costly get_random_int() call. Modify
arch_randomize_brk() as well to make it look like randomize_et_dyn().
Signed-off-by: Martin Schwidefsky <[email protected]>
The tty3270_alloc_screen function is called from tty3270_install with
swapped arguments, the number of columns instead of rows and vice versa.
The number of rows is typically smaller than the number of columns which
makes the screen array too big but the individual cell arrays for the
lines too small. Creating lines longer than the number of rows will
clobber the memory after the end of the cell array.
The fix is simple, call tty3270_alloc_screen with the correct argument
order.
Signed-off-by: Martin Schwidefsky <[email protected]>
The return code of the __put_user call to store the rt_sigreturn
system call to the user stack if not properly checked, the err
variable is only checked before to the __put_user. Use an if
statement instead.
Signed-off-by: Martin Schwidefsky <[email protected]>
The deactivation and freeing of the tty view of the 3270 device
can race with a tty3270_update invocation via the update timer.
To fix this move the del_timer_sync call for the update timer from
tty3270_free_view to tty3270_free prior to the tty3270_free_screen
call.
Signed-off-by: Martin Schwidefsky <[email protected]>
It is less expensive to update control registers 0 and 2 with two
individual stctg/lctlg instructions as with a single one that spans
control register 0, 1 and 2.
Signed-off-by: Martin Schwidefsky <[email protected]>
The user_enable_single_step() and user_disable_sindle_step() functions
are always called on the inferior, never for the currently active
process. Remove the unnecessary check for the current process and
the update_cr_regs() call from the enable/disable functions.
Signed-off-by: Martin Schwidefsky <[email protected]>
With git commit 79c74ecbeb
"s390/time,vdso: convert to the new update_vsyscall interface"
the new update_vsyscall function already does the sum of xtime
and wall_to_monotonic. The old update_vsyscall function only
copied the wall_to_monotonic offset. The vdso code needs to be
modified to take this into consideration.
Signed-off-by: Martin Schwidefsky <[email protected]>
Martin Schwidefsky