commit 5fb1d3ce3e74a4530042795e1e065422295f1371 upstream.

When the kernel leaves to userspace via syscall_restore_rfi(), the
W bit is not set in the new PSW. This doesn't cause any problems
because there's no 64 bit userspace for parisc. Simple static binaries
are usually loaded at addresses way below the 32 bit limit so the W bit
doesn't matter.

Fix this by setting the W bit when TIF_32BIT is not set.

Signed-off-by: Sven Schnelle <svens@stackframe.org>
Cc: stable@vger.kernel.org
Signed-off-by: Helge Deller <deller@gmx.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

commit 1aa4524c0c1b54842c4c0a370171d11b12d0709b upstream.

In wide mode, the IASQ contain the upper part of the GVA
during interruption. This needs to be reversed before
the space is used - otherwise it contains parts of IAOQ.
See Page 2-13 "Processing Resources / Interruption Instruction
Address Queues" in the Parisc 2.0 Architecture Manual page 2-13
for an explanation.

The IAOQ/IASQ space_adjust was skipped for other interruptions
than itlb misses. However, the code in handle_interruption()
checks whether iasq[0] contains a valid space. Due to the not
masked out bits this match failed and the process was killed.

Also add space_adjust for IAOQ1/IASQ1 so ptregs contains sane values.

Signed-off-by: Sven Schnelle <svens@stackframe.org>
Cc: stable@vger.kernel.org # v6.0+
Signed-off-by: Helge Deller <deller@gmx.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

commit cb22f247f371bd206a88cf0e0c05d80b8b62fb26 upstream.

The following testcase exposed a problem with our read access checks
in get_user() and raw_copy_from_user():

#include <stdint.h>
#include <stddef.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <errno.h>
#include <sys/mman.h>
#include <sys/types.h>

int main(int argc, char **argv)
{
  unsigned long page_size = sysconf(_SC_PAGESIZE);
  char *p = malloc(3 * page_size);
  char *p_aligned;

  /* initialize memory region. If not initialized, write syscall below will correctly return EFAULT. */
  if (1)
	memset(p, 'X', 3 * page_size);

  p_aligned = (char *) ((((uintptr_t) p) + (2*page_size - 1)) & ~(page_size - 1));
  /* Drop PROT_READ protection. Kernel and userspace should fault when accessing that memory region */
  mprotect(p_aligned, page_size, PROT_NONE);

  /* the following write() should return EFAULT, since PROT_READ was dropped by previous mprotect() */
  int ret = write(2, p_aligned, 1);
  if (!ret || errno != EFAULT)
	printf("\n FAILURE: write() did not returned expected EFAULT value\n");

  return 0;
}

Because of the way _PAGE_READ is handled, kernel code never generates
a read access fault when it access a page as the kernel privilege level
is always less than PL1 in the PTE.

This patch reworks the comments in the make_insert_tlb macro to try
to make this clearer.

Signed-off-by: John David Anglin <dave.anglin@bell.net>
Signed-off-by: Helge Deller <deller@gmx.de>
Cc: stable@vger.kernel.org # v5.12+
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

For an itlb miss when executing code above 4 Gb on ILP64 adjust the
iasq/iaoq in the same way isr/ior was adjusted.  This fixes signal
delivery for the 64-bit static test program from
http://ftp.parisc-linux.org/src/64bit.tar.gz.  Note that signals are
handled by the signal trampoline code in the 64-bit VDSO which is mapped
into high userspace memory region above 4GB for 64-bit processes.

Signed-off-by: Helge Deller <deller@gmx.de>
Cc: stable@vger.kernel.org	# v4.19+

When calculating the pfn for the iitlbt/idtlbt instruction, do not
drop the upper 5 address bits. This doesn't seem to have an effect
on physical hardware which uses less physical address bits, but in
qemu the missing bits are visible.

Signed-off-by: Helge Deller <deller@gmx.de>
Cc:  <stable@vger.kernel.org>

An excerpt from the PA8800 ERS states:

* The PA8800 violates the seven instruction pipeline rule when performing
  TLB inserts or PxTLBE instructions with the PSW C bit on. The instruction
  will take effect by the 12th instruction after the insert or purge.

I believe we have a problem with handling TLB misses. We don't fill
the pipeline following TLB inserts. As a result, we likely fault again
after returning from the interruption.

The above statement indicates that we need at least seven instructions
after the insert on pre PA8800 processors and we need 12 instructions
on PA8800/PA8900 processors.

Here we add macros and code to provide the required number instructions
after a TLB insert.

Signed-off-by: John David Anglin <dave.anglin@bell.net>
Suggested-by: Helge Deller <deller@gmx.de>
Cc: stable@vger.kernel.org
Signed-off-by: Helge Deller <deller@gmx.de>

For the TLB_PTLOCK checks we used an optimization to store the spc
register into the spinlock to unlock it. This optimization works as
long as the lightweight spinlock checks (CONFIG_LIGHTWEIGHT_SPINLOCK_CHECK)
aren't enabled, because they really check if the lock word is zero or
__ARCH_SPIN_LOCK_UNLOCKED_VAL and abort with a kernel crash
("Spinlock was trashed") otherwise.

Drop that optimization to make it possible to activate both checks
at the same time.

Noticed-by: Sam James <sam@gentoo.org>
Signed-off-by: Helge Deller <deller@gmx.de>
Tested-by: Sam James <sam@gentoo.org>
Cc: stable@vger.kernel.org # v6.4+
Fixes: 15e64ef6520e ("parisc: Add lightweight spinlock checks")

Commit df24e1783e6e ("parisc: Add vDSO support") introduced the vDSO
support, for which a _PAGE_SPECIAL page table flag was needed.  Since we
wanted to keep every page table entry in 32-bits, this patch re-used the
existing - but yet unused - _PAGE_DMB flag (which triggers a hardware break
if a page is accessed) to store the special bit.

But when graphics card memory is mmapped into userspace, the kernel uses
vm_iomap_memory() which sets the the special flag. So, with the DMB bit
set, every access to the graphics memory now triggered a hardware
exception and segfaulted the userspace program.

Fix this breakage by dropping the DMB bit when writing the page
protection bits to the CPU TLB.

In addition this patch adds a small optimization: if huge pages aren't
configured (which is at least the case for 32-bit kernels), then the
special bit is stored in the hpage (HUGE PAGE) bit instead. That way we
can skip to reset the DMB bit.

Fixes: df24e1783e6e ("parisc: Add vDSO support")
Cc: <stable@vger.kernel.org> # 5.18+
Signed-off-by: Helge Deller <deller@gmx.de>

Instead of converting the physical address of the tmpalias mapping to
the tlb insert format inside all the various tmpalias functions, move
this conversion over to the DTLB miss handler. The physical address is
already in %r26 (or will be calculated into %r23), so there are no
additional steps needed in the functions themselves.

Additionally use the dep_safe() and depi_safe() macros to avoid
differentiating between 32- and 64-bit builds and as such make the code
much more readable.

The check if "ldil L%(TMPALIAS_MAP_START)" will sign extend into the
upper 32 bits can be dropped, because we added a compile time check in
an earlier patch.

Signed-off-by: Helge Deller <deller@gmx.de>

Remove the hardcoded bit definitions in the tmpalias assembly code.
This makes it easy to change the size of the tmpalias region.

The alignment of the tmpalias region is reduced from 16 MB to 8 MB.

Signed-off-by: John David Anglin <dave.anglin@bell.net>
Signed-off-by: Helge Deller <deller@gmx.de>