linux-stable.git/arch/parisc/kernel/entry.S, branch linux-4.3.y Linux kernel stable tree parisc: Fix some PTE/TLB race conditions and optimize __flush_tlb_range based on timing results 2015-07-10T19:47:47+00:00 John David Anglin dave.anglin@bell.net 2015-07-01T21:18:37+00:00 01ab60570427caa24b9debc369e452e86cd9beb4 The increased use of pdtlb/pitlb instructions seemed to increase the frequency of random segmentation faults building packages. Further, we had a number of cases where TLB inserts would repeatedly fail and all forward progress would stop. The Haskell ghc package caused a lot of trouble in this area. The final indication of a race in pte handling was this syslog entry on sibaris (C8000): swap_free: Unused swap offset entry 00000004 BUG: Bad page map in process mysqld pte:00000100 pmd:019bbec5 addr:00000000ec464000 vm_flags:00100073 anon_vma:0000000221023828 mapping: (null) index:ec464 CPU: 1 PID: 9176 Comm: mysqld Not tainted 4.0.0-2-parisc64-smp #1 Debian 4.0.5-1 Backtrace: [<0000000040173eb0>] show_stack+0x20/0x38 [<0000000040444424>] dump_stack+0x9c/0x110 [<00000000402a0d38>] print_bad_pte+0x1a8/0x278 [<00000000402a28b8>] unmap_single_vma+0x3d8/0x770 [<00000000402a4090>] zap_page_range+0xf0/0x198 [<00000000402ba2a4>] SyS_madvise+0x404/0x8c0 Note that the pte value is 0 except for the accessed bit 0x100. This bit shouldn't be set without the present bit. It should be noted that the madvise system call is probably a trigger for many of the random segmentation faults. In looking at the kernel code, I found the following problems: 1) The pte_clear define didn't take TLB lock when clearing a pte. 2) We didn't test pte present bit inside lock in exception support. 3) The pte and tlb locks needed to merged in order to ensure consistency between page table and TLB. This also has the effect of serializing TLB broadcasts on SMP systems. The attached change implements the above and a few other tweaks to try to improve performance. Based on the timing code, TLB purges are very slow (e.g., ~ 209 cycles per page on rp3440). Thus, I think it beneficial to test the split_tlb variable to avoid duplicate purges. Probably, all PA 2.0 machines have combined TLBs. I dropped using __flush_tlb_range in flush_tlb_mm as I realized all applications and most threads have a stack size that is too large to make this useful. I added some comments to this effect. Since implementing 1 through 3, I haven't had any random segmentation faults on mx3210 (rp3440) in about one week of building code and running as a Debian buildd. Signed-off-by: John David Anglin <dave.anglin@bell.net> Cc: stable@vger.kernel.org # v3.18+ Signed-off-by: Helge Deller <deller@gmx.de> 
The increased use of pdtlb/pitlb instructions seemed to increase the
frequency of random segmentation faults building packages. Further, we
had a number of cases where TLB inserts would repeatedly fail and all
forward progress would stop. The Haskell ghc package caused a lot of
trouble in this area. The final indication of a race in pte handling was
this syslog entry on sibaris (C8000):

 swap_free: Unused swap offset entry 00000004
 BUG: Bad page map in process mysqld  pte:00000100 pmd:019bbec5
 addr:00000000ec464000 vm_flags:00100073 anon_vma:0000000221023828 mapping: (null) index:ec464
 CPU: 1 PID: 9176 Comm: mysqld Not tainted 4.0.0-2-parisc64-smp #1 Debian 4.0.5-1
 Backtrace:
  [<0000000040173eb0>] show_stack+0x20/0x38
  [<0000000040444424>] dump_stack+0x9c/0x110
  [<00000000402a0d38>] print_bad_pte+0x1a8/0x278
  [<00000000402a28b8>] unmap_single_vma+0x3d8/0x770
  [<00000000402a4090>] zap_page_range+0xf0/0x198
  [<00000000402ba2a4>] SyS_madvise+0x404/0x8c0

Note that the pte value is 0 except for the accessed bit 0x100. This bit
shouldn't be set without the present bit.

It should be noted that the madvise system call is probably a trigger for many
of the random segmentation faults.

In looking at the kernel code, I found the following problems:

1) The pte_clear define didn't take TLB lock when clearing a pte.
2) We didn't test pte present bit inside lock in exception support.
3) The pte and tlb locks needed to merged in order to ensure consistency
between page table and TLB. This also has the effect of serializing TLB
broadcasts on SMP systems.

The attached change implements the above and a few other tweaks to try
to improve performance. Based on the timing code, TLB purges are very
slow (e.g., ~ 209 cycles per page on rp3440). Thus, I think it
beneficial to test the split_tlb variable to avoid duplicate purges.
Probably, all PA 2.0 machines have combined TLBs.

I dropped using __flush_tlb_range in flush_tlb_mm as I realized all
applications and most threads have a stack size that is too large to
make this useful. I added some comments to this effect.

Since implementing 1 through 3, I haven't had any random segmentation
faults on mx3210 (rp3440) in about one week of building code and running
as a Debian buildd.

Signed-off-by: John David Anglin <dave.anglin@bell.net>
Cc: stable@vger.kernel.org # v3.18+
Signed-off-by: Helge Deller <deller@gmx.de>
parisc: expose number of page table levels on Kconfig level 2015-04-14T23:49:02+00:00 Kirill A. Shutemov kirill.shutemov@linux.intel.com 2015-04-14T22:45:54+00:00 f24ffde43237755b290c46306a3dd2deb1428700 We would want to use number of page table level to define mm_struct. Let's expose it as CONFIG_PGTABLE_LEVELS. Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: "James E.J. Bottomley" <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Tested-by: Guenter Roeck <linux@roeck-us.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> 
We would want to use number of page table level to define mm_struct.
Let's expose it as CONFIG_PGTABLE_LEVELS.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: "James E.J. Bottomley" <jejb@parisc-linux.org>
Cc: Helge Deller <deller@gmx.de>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
parisc: hpux - Remove hpux gateway page 2015-02-16T21:35:06+00:00 Helge Deller deller@gmx.de 2015-02-16T21:21:00+00:00 c78c2b7e0409feed41ba1b5e84bff5d901c9b65f Drop code to create HP-UX gateway page and syscall entry code. Signed-off-by: Helge Deller <deller@gmx.de> 
Drop code to create HP-UX gateway page and syscall entry code.

Signed-off-by: Helge Deller <deller@gmx.de>
parisc: make interrupt and interruption stack allocation reentrant 2013-05-24T20:35:36+00:00 John David Anglin dave.anglin@bell.net 2013-05-20T16:42:53+00:00 b63a2bbc0b9b106a93e11952ab057e2408f2eb02 The get_stack_use_cr30 and get_stack_use_r30 macros allocate a stack frame for external interrupts and interruptions requiring a stack frame. They are currently not reentrant in that they save register context before the stack is set or adjusted. I have observed a number of system crashes where there was clear evidence of stack corruption during interrupt processing, and as a result register corruption. Some interruptions can still occur during interruption processing, however external interrupts are disabled and data TLB misses don't occur for absolute accesses. So, it's not entirely clear what triggers this issue. Also, if an interruption occurs when Q=0, it is generally not possible to recover as the shadowed registers are not copied. The attached patch reworks the get_stack_use_cr30 and get_stack_use_r30 macros to allocate stack before doing register saves. The new code is a couple of instructions shorter than the old implementation. Thus, it's an improvement even if it doesn't fully resolve the stack corruption issue. Based on limited testing, it improves SMP system stability. Signed-off-by: John David Anglin <dave.anglin@bell.net> Signed-off-by: Helge Deller <deller@gmx.de> 
The get_stack_use_cr30 and get_stack_use_r30 macros allocate a stack
frame for external interrupts and interruptions requiring a stack frame.
They are currently not reentrant in that they save register context
before the stack is set or adjusted.

I have observed a number of system crashes where there was clear
evidence of stack corruption during interrupt processing, and as a
result register corruption. Some interruptions can still occur during
interruption processing, however external interrupts are disabled and
data TLB misses don't occur for absolute accesses. So, it's not entirely
clear what triggers this issue. Also, if an interruption occurs when
Q=0, it is generally not possible to recover as the shadowed registers
are not copied.

The attached patch reworks the get_stack_use_cr30 and get_stack_use_r30
macros to allocate stack before doing register saves. The new code is a
couple of instructions shorter than the old implementation. Thus, it's
an improvement even if it doesn't fully resolve the stack corruption
issue. Based on limited testing, it improves SMP system stability.

Signed-off-by: John David Anglin <dave.anglin@bell.net>
Signed-off-by: Helge Deller <deller@gmx.de>
parisc: fix SMP races when updating PTE and TLB entries in entry.S 2013-05-11T19:13:04+00:00 John David Anglin dave.anglin@bell.net 2013-05-10T23:21:38+00:00 f0a18819e261afc5fdbd8c5c6f9943123c5461ba Currently, race conditions exist in the handling of TLB interruptions in entry.S. In particular, dirty bit updates can be lost if an accessed interruption occurs just after the dirty bit interruption on a different cpu. Lost dirty bit updates result in user pages not being flushed and general system instability. This change adds lock and unlock macros to synchronize all PTE and TLB updates done in entry.S. As a result, userspace stability is significantly improved. Signed-off-by: John David Anglin <dave.anglin@bell.net> Signed-off-by: Helge Deller <deller@gmx.de> 
Currently, race conditions exist in the handling of TLB interruptions in
entry.S.  In particular, dirty bit updates can be lost if an accessed
interruption occurs just after the dirty bit interruption on a different
cpu.  Lost dirty bit updates result in user pages not being flushed and
general system instability.  This change adds lock and unlock macros to
synchronize all PTE and TLB updates done in entry.S.  As a result,
userspace stability is significantly improved.

Signed-off-by: John David Anglin  <dave.anglin@bell.net>
Signed-off-by: Helge Deller <deller@gmx.de>
parisc: implement irq stacks 2013-05-07T20:30:35+00:00 Helge Deller deller@gmx.de 2013-05-07T20:25:42+00:00 200c880420a2c02a0899120ce52d801fad705b90 Default kernel stack size on parisc is 16k. During tests we found that the kernel stack can easily grow beyond 13k, which leaves 3k left for irq processing. This patch adds the possibility to activate an additional stack of 16k per CPU which is being used during irq processing. This implementation does not yet uses this irq stack for the irq bh handler. The assembler code for call_on_stack was heavily cleaned up by John David Anglin. CC: John David Anglin <dave.anglin@bell.net> Signed-off-by: Helge Deller <deller@gmx.de> 
Default kernel stack size on parisc is 16k.  During tests we found that the
kernel stack can easily grow beyond 13k, which leaves 3k left for irq
processing.

This patch adds the possibility to activate an additional stack of 16k per CPU
which is being used during irq processing.  This implementation does not yet
uses this irq stack for the irq bh handler.

The assembler code for call_on_stack was heavily cleaned up by John
David Anglin.

CC: John David Anglin <dave.anglin@bell.net>
Signed-off-by: Helge Deller <deller@gmx.de>
parisc: only re-enable interrupts if we need to schedule or deliver signals when returning to userspace 2013-05-07T18:33:03+00:00 John David Anglin dave.anglin@bell.net 2013-05-07T00:07:25+00:00 c207a76bf155cb5cf24cf849c08f6555e9180594 Helge and I have found that we have a kernel stack overflow problem which causes a variety of random failures. Currently, we re-enable interrupts when returning from an external interrupt incase we need to schedule or delivery signals. As a result, a potentially unlimited number of interrupts can occur while we are running on the kernel stack. It is very limited in space (currently, 16k). This change defers enabling interrupts until we have actually decided to schedule or delivery signals. This only occurs when we about to return to userspace. This limits the number of interrupts on the kernel stack to one. In other cases, interrupts remain disabled until the final return from interrupt (rfi). Signed-off-by: John David Anglin <dave.anglin@bell.net> Signed-off-by: Helge Deller <deller@gmx.de> 
Helge and I have found that we have a kernel stack overflow problem
which causes a variety of random failures.
Currently, we re-enable interrupts when returning from an external
interrupt incase we need to schedule or delivery
signals.  As a result, a potentially unlimited number of interrupts
can occur while we are running on the kernel
stack.  It is very limited in space (currently, 16k).  This change
defers enabling interrupts until we have
actually decided to schedule or delivery signals.  This only occurs
when we about to return to userspace.  This
limits the number of interrupts on the kernel stack to one.  In other
cases, interrupts remain disabled until the
final return from interrupt (rfi).

Signed-off-by: John David Anglin  <dave.anglin@bell.net>
Signed-off-by: Helge Deller <deller@gmx.de>
parisc: use long branch in fork_like macro 2013-05-06T21:09:48+00:00 John David Anglin dave.anglin@bell.net 2013-05-04T19:16:41+00:00 bbbfde782084b4f0d85ddffb88f1cf4650ff40e4 The "b" branch instruction used in the fork_like macro only can handle 17-bit pc-relative offsets. This fails with an out of range offset with some .config files. Rewrite to use the "be" instruction which can branch to any address in a space. Signed-off-by: John David Anglin <dave.anglin@bell.net> Signed-off-by: Helge Deller <deller@gmx.de> 
The "b" branch instruction used in the fork_like macro only can handle
17-bit pc-relative offsets.
This fails with an out of range offset with some .config files.
Rewrite to use the "be" instruction which
can branch to any address in a space.

Signed-off-by: John David Anglin  <dave.anglin@bell.net>
Signed-off-by: Helge Deller <deller@gmx.de>
parisc: fix partly 16/64k PAGE_SIZE boot 2013-05-06T21:08:32+00:00 Helge Deller deller@gmx.de 2013-05-02T20:41:45+00:00 6a45716abbf9dc0f397946306db1f78b2eba3086 This patch fixes partly PAGE_SIZEs of 16K or 64K by adjusting the assembler PTE lookup code and the assembler TEMPALIAS code. Furthermore some data alignments for PAGE_SIZE have been limited to 4K (or less) to not waste too much memory with greater page sizes. As a side note, the palo loader can (currently) only handle up to 10 ELF segments which is fixed with tighter aligning as well. My testings indicated that the ldci command in the sba iommu coding needed adjustment by the PAGE_SHIFT value and that the I/O PDIR Page size was only set to 4K for my machine (C3000). All this fixes partly the boot, but there are still quite some caching problems left. Examples are e.g. the symbios logic driver which is failing: sym0: <896> rev 0x7 at pci 0000:00:0f.0 irq 69 sym0: PA-RISC Firmware, ID 7, Fast-40, SE, parity checking CACHE TEST FAILED: DMA error (dstat=0x81).sym0: CACHE INCORRECTLY CONFIGURED. and the tulip network driver which doesn't seem to work correctly either: Sending BOOTP requests .net eth0: Setting full-duplex based on MII#1 link partner capability of 05e1 ..... timed out! Beside those kernel fixes glibc will need fixes too to be able to handle >4K page sizes. Signed-off-by: Helge Deller <deller@gmx.de> 
This patch fixes partly PAGE_SIZEs of 16K or 64K by adjusting the
assembler PTE lookup code and the assembler TEMPALIAS code.  Furthermore
some data alignments for PAGE_SIZE have been limited to 4K (or less) to
not waste too much memory with greater page sizes. As a side note, the
palo loader can (currently) only handle up to 10 ELF segments which is
fixed with tighter aligning as well.

My testings indicated that the ldci command in the sba iommu coding
needed adjustment by the PAGE_SHIFT value and that the I/O PDIR Page
size was only set to 4K for my machine (C3000).

All this fixes partly the boot, but there are still quite some caching
problems left.  Examples are e.g. the symbios logic driver which is
failing:

sym0: <896> rev 0x7 at pci 0000:00:0f.0 irq 69
sym0: PA-RISC Firmware, ID 7, Fast-40, SE, parity checking
CACHE TEST FAILED: DMA error (dstat=0x81).sym0: CACHE INCORRECTLY CONFIGURED.

and the tulip network driver which doesn't seem to work correctly
either:

Sending BOOTP requests .net eth0: Setting full-duplex based on MII#1
link partner capability of 05e1
..... timed out!

Beside those kernel fixes glibc will need fixes too to be able to handle
>4K page sizes.

Signed-off-by: Helge Deller <deller@gmx.de>
Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/signal 2013-02-24T02:50:11+00:00 Linus Torvalds torvalds@linux-foundation.org 2013-02-24T02:50:11+00:00 9e2d59ad580d590134285f361a0e80f0e98c0207 Pull signal handling cleanups from Al Viro: "This is the first pile; another one will come a bit later and will contain SYSCALL_DEFINE-related patches. - a bunch of signal-related syscalls (both native and compat) unified. - a bunch of compat syscalls switched to COMPAT_SYSCALL_DEFINE (fixing several potential problems with missing argument validation, while we are at it) - a lot of now-pointless wrappers killed - a couple of architectures (cris and hexagon) forgot to save altstack settings into sigframe, even though they used the (uninitialized) values in sigreturn; fixed. - microblaze fixes for delivery of multiple signals arriving at once - saner set of helpers for signal delivery introduced, several architectures switched to using those." * 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/signal: (143 commits) x86: convert to ksignal sparc: convert to ksignal arm: switch to struct ksignal * passing alpha: pass k_sigaction and siginfo_t using ksignal pointer burying unused conditionals make do_sigaltstack() static arm64: switch to generic old sigaction() (compat-only) arm64: switch to generic compat rt_sigaction() arm64: switch compat to generic old sigsuspend arm64: switch to generic compat rt_sigqueueinfo() arm64: switch to generic compat rt_sigpending() arm64: switch to generic compat rt_sigprocmask() arm64: switch to generic sigaltstack sparc: switch to generic old sigsuspend sparc: COMPAT_SYSCALL_DEFINE does all sign-extension as well as SYSCALL_DEFINE sparc: kill sign-extending wrappers for native syscalls kill sparc32_open() sparc: switch to use of generic old sigaction sparc: switch sys_compat_rt_sigaction() to COMPAT_SYSCALL_DEFINE mips: switch to generic sys_fork() and sys_clone() ... 
Pull signal handling cleanups from Al Viro:
 "This is the first pile; another one will come a bit later and will
  contain SYSCALL_DEFINE-related patches.

   - a bunch of signal-related syscalls (both native and compat)
     unified.

   - a bunch of compat syscalls switched to COMPAT_SYSCALL_DEFINE
     (fixing several potential problems with missing argument
     validation, while we are at it)

   - a lot of now-pointless wrappers killed

   - a couple of architectures (cris and hexagon) forgot to save
     altstack settings into sigframe, even though they used the
     (uninitialized) values in sigreturn; fixed.

   - microblaze fixes for delivery of multiple signals arriving at once

   - saner set of helpers for signal delivery introduced, several
     architectures switched to using those."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/signal: (143 commits)
  x86: convert to ksignal
  sparc: convert to ksignal
  arm: switch to struct ksignal * passing
  alpha: pass k_sigaction and siginfo_t using ksignal pointer
  burying unused conditionals
  make do_sigaltstack() static
  arm64: switch to generic old sigaction() (compat-only)
  arm64: switch to generic compat rt_sigaction()
  arm64: switch compat to generic old sigsuspend
  arm64: switch to generic compat rt_sigqueueinfo()
  arm64: switch to generic compat rt_sigpending()
  arm64: switch to generic compat rt_sigprocmask()
  arm64: switch to generic sigaltstack
  sparc: switch to generic old sigsuspend
  sparc: COMPAT_SYSCALL_DEFINE does all sign-extension as well as SYSCALL_DEFINE
  sparc: kill sign-extending wrappers for native syscalls
  kill sparc32_open()
  sparc: switch to use of generic old sigaction
  sparc: switch sys_compat_rt_sigaction() to COMPAT_SYSCALL_DEFINE
  mips: switch to generic sys_fork() and sys_clone()
  ...