It is not immediately obvious what the expected inputs to these fault
handlers is and how they calculate the number of unset bytes. Having
stared deeply at this in order to fix some corner cases, add some
comments to assist those who follow.
Signed-off-by: Matt Redfearn <matt.redfearn@mips.com>
Signed-off-by: Paul Burton <paul.burton@mips.com>
Patchwork: https://patchwork.linux-mips.org/patch/19339/
Cc: James Hogan <jhogan@kernel.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: <linux-mips@linux-mips.org>
Cc: <linux-kernel@vger.kernel.org>
The __clear_user function is defined to return the number of bytes that
could not be cleared. From the underlying memset / bzero implementation
this means setting register a2 to that number on return. Currently if a
page fault is triggered within the MIPSr6 version of setting of initial
unaligned bytes, the value loaded into a2 on return is meaningless.
During the MIPSr6 version of the initial unaligned bytes block, register
a2 contains the number of bytes to be set beyond the initial unaligned
bytes. The t0 register is initally set to the number of unaligned bytes
- STORSIZE, effectively a negative version of the number of unaligned
bytes. This is then incremented before each byte is saved.
The label .Lbyte_fixup\@ is jumped to on page fault. Currently the value
in a2 is incorrectly replaced by 0 - t0 + 1, effectively the number of
unaligned bytes remaining. This leads to the failures being reported by
the following test code:
static int __init test_clear_user(void)
{
int j, k;
pr_info("\n\n\nTesting clear_user\n");
for (j = 0; j < 512; j++) {
if ((k = clear_user(NULL+3, j)) != j) {
pr_err("clear_user (NULL %d) returned %d\n", j, k);
}
}
return 0;
}
late_initcall(test_clear_user);
Which reports:
[ 3.965439] Testing clear_user
[ 3.973169] clear_user (NULL 8) returned 6
[ 3.976782] clear_user (NULL 9) returned 6
[ 3.980390] clear_user (NULL 10) returned 6
[ 3.984052] clear_user (NULL 11) returned 6
[ 3.987524] clear_user (NULL 12) returned 6
Fix this by subtracting t0 from a2 (rather than $0), effectivey giving:
unset_bytes = (#bytes - (#unaligned bytes)) - (-#unaligned bytes remaining + 1) + 1
a2 = a2 - t0 + 1
This fixes the value returned from __clear user when the number of bytes
to set is > LONGSIZE and the address is invalid and unaligned.
Unfortunately, this breaks the fixup handling for unaligned bytes after
the final long, where register a2 still contains the number of bytes
remaining to be set and the t0 register is to 0 - the number of
unaligned bytes remaining.
Because t0 is now is now subtracted from a2 rather than 0, the number of
bytes unset is reported incorrectly:
static int __init test_clear_user(void)
{
char *test;
int j, k;
pr_info("\n\n\nTesting clear_user\n");
test = vmalloc(PAGE_SIZE);
for (j = 256; j < 512; j++) {
if ((k = clear_user(test + PAGE_SIZE - 254, j)) != j - 254) {
pr_err("clear_user (%px %d) returned %d\n",
test + PAGE_SIZE - 254, j, k);
}
}
return 0;
}
late_initcall(test_clear_user);
[ 3.976775] clear_user (c00000000000df02 256) returned 4
[ 3.981957] clear_user (c00000000000df02 257) returned 6
[ 3.986425] clear_user (c00000000000df02 258) returned 8
[ 3.990850] clear_user (c00000000000df02 259) returned 10
[ 3.995332] clear_user (c00000000000df02 260) returned 12
[ 3.999815] clear_user (c00000000000df02 261) returned 14
Fix this by ensuring that a2 is set to 0 during the set of final
unaligned bytes.
Signed-off-by: Matt Redfearn <matt.redfearn@mips.com>
Signed-off-by: Paul Burton <paul.burton@mips.com>
Fixes: 8c56208aff ("MIPS: lib: memset: Add MIPS R6 support")
Patchwork: https://patchwork.linux-mips.org/patch/19338/
Cc: James Hogan <jhogan@kernel.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Cc: stable@vger.kernel.org # v4.0+
Assembly language within the MIPS kernel conventionally indents
instructions which are in a branch delay slot to make them easier to
see. Commit 8483b14aaa ("MIPS: lib: memset: Whitespace fixes") rather
inexplicably removed all of these indentations from memset.S. Reinstate
the convention for all instructions in a branch delay slot. This
effectively reverts the above commit, plus other locations introduced
with MIPSR6 support.
Signed-off-by: Matt Redfearn <matt.redfearn@mips.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/19111/
Signed-off-by: James Hogan <jhogan@kernel.org>
The label .Llast_fixup\@ is jumped to on page fault within the final
byte set loop of memset (on < MIPSR6 architectures). For some reason, in
this fault handler, the v1 register is randomly set to a2 & STORMASK.
This clobbers v1 for the calling function. This can be observed with the
following test code:
static int __init __attribute__((optimize("O0"))) test_clear_user(void)
{
register int t asm("v1");
char *test;
int j, k;
pr_info("\n\n\nTesting clear_user\n");
test = vmalloc(PAGE_SIZE);
for (j = 256; j < 512; j++) {
t = 0xa5a5a5a5;
if ((k = clear_user(test + PAGE_SIZE - 256, j)) != j - 256) {
pr_err("clear_user (%px %d) returned %d\n", test + PAGE_SIZE - 256, j, k);
}
if (t != 0xa5a5a5a5) {
pr_err("v1 was clobbered to 0x%x!\n", t);
}
}
return 0;
}
late_initcall(test_clear_user);
Which demonstrates that v1 is indeed clobbered (MIPS64):
Testing clear_user
v1 was clobbered to 0x1!
v1 was clobbered to 0x2!
v1 was clobbered to 0x3!
v1 was clobbered to 0x4!
v1 was clobbered to 0x5!
v1 was clobbered to 0x6!
v1 was clobbered to 0x7!
Since the number of bytes that could not be set is already contained in
a2, the andi placing a value in v1 is not necessary and actively
harmful in clobbering v1.
Reported-by: James Hogan <jhogan@kernel.org>
Signed-off-by: Matt Redfearn <matt.redfearn@mips.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: stable@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/19109/
Signed-off-by: James Hogan <jhogan@kernel.org>
The __clear_user function is defined to return the number of bytes that
could not be cleared. From the underlying memset / bzero implementation
this means setting register a2 to that number on return. Currently if a
page fault is triggered within the memset_partial block, the value
loaded into a2 on return is meaningless.
The label .Lpartial_fixup\@ is jumped to on page fault. In order to work
out how many bytes failed to copy, the exception handler should find how
many bytes left in the partial block (andi a2, STORMASK), add that to
the partial block end address (a2), and subtract the faulting address to
get the remainder. Currently it incorrectly subtracts the partial block
start address (t1), which has additionally been clobbered to generate a
jump target in memset_partial. Fix this by adding the block end address
instead.
This issue was found with the following test code:
int j, k;
for (j = 0; j < 512; j++) {
if ((k = clear_user(NULL, j)) != j) {
pr_err("clear_user (NULL %d) returned %d\n", j, k);
}
}
Which now passes on Creator Ci40 (MIPS32) and Cavium Octeon II (MIPS64).
Suggested-by: James Hogan <jhogan@kernel.org>
Signed-off-by: Matt Redfearn <matt.redfearn@mips.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: stable@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/19108/
Signed-off-by: James Hogan <jhogan@kernel.org>
The MIPS kernel memset / bzero implementation includes a small_memset
branch which is used when the region to be set is smaller than a long (4
bytes on 32bit, 8 bytes on 64bit). The current small_memset
implementation uses a simple store byte loop to write the destination.
There are 2 issues with this implementation:
1. When EVA mode is active, user and kernel address spaces may overlap.
Currently the use of the sb instruction means kernel mode addressing is
always used and an intended write to userspace may actually overwrite
some critical kernel data.
2. If the write triggers a page fault, for example by calling
__clear_user(NULL, 2), instead of gracefully handling the fault, an OOPS
is triggered.
Fix these issues by replacing the sb instruction with the EX() macro,
which will emit EVA compatible instuctions as required. Additionally
implement a fault fixup for small_memset which sets a2 to the number of
bytes that could not be cleared (as defined by __clear_user).
Reported-by: Chuanhua Lei <chuanhua.lei@intel.com>
Signed-off-by: Matt Redfearn <matt.redfearn@mips.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: stable@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/18975/
Signed-off-by: James Hogan <jhogan@kernel.org>
Now that EXPORT_SYMBOL can be used from assembly source, move the
EXPORT_SYMBOL invocations for the memcpy & memset functions & variants
thereof to be alongside their definitions.
Signed-off-by: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14514/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
__clear_user() (and clear_user() which uses it), always access the user
mode address space, which results in EVA store instructions when EVA is
enabled even if the current user address limit is KERNEL_DS.
Fix this by adding a new symbol __bzero_kernel for the normal kernel
address space bzero in EVA mode, and call that from __clear_user() if
eva_kernel_access().
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Markos Chandras <markos.chandras@imgtec.com>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: Leonid Yegoshin <leonid.yegoshin@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/10844/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
MIPS R6 dropped the unaligned load and store instructions so
we need to re-write this part of the code for R6 to store
one byte at a time.
Signed-off-by: Leonid Yegoshin <Leonid.Yegoshin@imgtec.com>
Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
The toolchain defines exactly one of __MIPSEB__ and
__MIPSEL__. As a result, simplify the ifdefery a little bit.
Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/8522/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Build the __bzero function using the EVA load/store instructions
when operating in the EVA mode. This function is only used when
accessing user code so there is no need to build two distinct symbols
for user and kernel operations respectively.
Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
Build the __bzero symbol using a macor. In EVA mode we will
need to use similar code to do the userspace load operations so
it is better if we use a macro to avoid code duplications.
Signed-off-by: Markos Chandras <markos.chandras@imgtec.com>
Optimise 'memset' to use microMIPS instructions and/or optimisations
for binary size reduction. When the microMIPS ISA is not being used,
the library function compiles to the original binary code.
Signed-off-by: Steven J. Hill <Steven.Hill@imgtec.com>
Having received another series of whitespace patches I decided to do this
once and for all rather than dealing with this kind of patches trickling
in forever.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
partial_fixup is used in noreorder block.
Separating two consecutive loads can save one cycle on processors with
GPR intrelock and can fix load-use on processors that need a load delay slot.
Also do so for fwd_fixup.
[Ralf: Only R2000/R3000 class processors are lacking the the load-user
interlock and even some of those got it retrofitted. With R2000/R3000
being fairly uncommon these days the impact of this bug should be minor.]
Signed-off-by: Tony Wu <tung7970@gmail.com>
To: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/1768/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
These symbols appear in oprofile output, stacktraces and similar but only
make the output harder to read. Many identical symbol names such as
"both_aligned" were also being used in multiple source files making it
impossible to see which file actually was meant. So let's get rid of them.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
IP28 needs special treatment to avoid speculative accesses. gcc
takes care for .c code, but for assembly code we need to do it
manually.
This is taken from Peter Fuersts IP28 patches.
Signed-off-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
This complements the generic R4000/R4400 errata workaround code and adds
bits for the daddiu problem. In most places it just modifies handwritten
assembly code so that the assembler is allowed to use a temporary register
as daddiu may now be treated as a macro that expands to a sequence of li
and daddu. It is the AT register or, where AT is unavailable or used
explicitly for another purpose, an explicitly-named register is selected,
using the .set at=<reg> feature added recently to gas. This feature is
only used if CONFIG_CPU_DADDI_WORKAROUNDS has been set, so if the
workaround remains disabled, the required version of binutils stays
unchanged.
Similarly, daddiu instructions put in branch delay slots in noreorder
fragments are now taken out of them and the assembler is allowed to
reorder them itself as possible (which it does making the whole idea of
scheduling them into delay slots manually questionable).
Also in the very few places where such a simple conversion was not
possible, a handcoded longer sequence is implemented.
Other than that there are changes to code responsible for building the
TLB fault and page clear/copy handlers to avoid daddiu as appropriate.
These are only effective if the erratum is verified to be present at the
run time.
Finally there is a trivial update to __delay(), because it uses daddiu in
a branch delay slot.
Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The 32-bit version and 64-bit version are almost equal. Unify them.
This makes further improvements (for example, supporting CDEX, etc.)
easier.
Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
