ARCv2 MMU is software walked and Linux implements 2 levels of paging: pgd/pte.
Forthcoming hw will have multiple levels, so this change preps mm code
for same. It is also fun to try multi levels even on soft-walked code to
ensure generic mm code is robust to handle.
overview
________
2 levels {pgd, pte} : pmd is folded but pmd_* macros are valid and operate on pgd
3 levels {pgd, pmd, pte}:
- pud is folded and pud_* macros point to pgd
- pmd_* macros operate on actual pmd
code changes
____________
1. #include <asm-generic/pgtable-nopud.h>
2. Define CONFIG_PGTABLE_LEVELS 3
3a. Define PMD_SHIFT, PMD_SIZE, PMD_MASK, pmd_t
3b. Define pmd_val() which actually deals with pmd
(pmd_offset(), pmd_index() are provided by generic code)
3c. pmd_alloc_one()/pmd_free() also provided by generic code
(pmd_populate/pmd_free already exist)
4. Define pud_none(), pud_bad() macros based on generic pud_val() which
internally pertains to pgd now.
4b. define pud_populate() to just setup pgd
Acked-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Vineet Gupta <vgupta@kernel.org>
So far ARC pgtable_t has not been struct page based to avoid extra
page_address() calls involved. However the differences are down to
noise and get in the way of using generic code, hence this patch.
This also allows us to reuse generic THP depost/withdraw code.
There's some additional consideration for PGDIR_SHIFT in 4K page config.
Now due to page tables being PAGE_SIZE deep only, the address split
can't be really arbitrary.
Tested-by: kernel test robot <lkp@intel.com>
Suggested-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Vineet Gupta <vgupta@kernel.org>
MMU SCRATCH_DATA0 register is intended to cache task pgd. However in
ARC700 SMP port, it has to be repurposed for re-entrant interrupt
handling, while UP port doesn't. We currently handle these use-cases
using a fabricated #define which has usual issues of dependency nesting
and obvious ugliness.
So clean this up: for ARC700 don't use to cache pgd (even in UP) and do
the opposite for ARCv2.
And while here, switch to canonical pgd_offset().
Acked-by: Mike Rapoport <rppt@linux.ibm.com>
Signed-off-by: Vineet Gupta <vgupta@kernel.org>
There's no known/active customer using them with latest kernels anyways.
Removal helps cleanup code and remove the hack for
MMU_VER to MMU_V[3-4] conversion
Signed-off-by: Vineet Gupta <vgupta@kernel.org>
32-bit PAGE_MASK can not be used as a mask for physical addresses
when PAE is enabled. PAGE_MASK_PHYS must be used for physical
addresses instead of PAGE_MASK.
Without this, init gets SIGSEGV if pte_modify was called:
| potentially unexpected fatal signal 11.
| Path: /bin/busybox
| CPU: 0 PID: 1 Comm: init Not tainted 5.12.0-rc5-00003-g1e43c377a79f-dirty
| Insn could not be fetched
| @No matching VMA found
| ECR: 0x00040000 EFA: 0x00000000 ERET: 0x00000000
| STAT: 0x80080082 [IE U ] BTA: 0x00000000
| SP: 0x5f9ffe44 FP: 0x00000000 BLK: 0xaf3d4
| LPS: 0x000d093e LPE: 0x000d0950 LPC: 0x00000000
| r00: 0x00000002 r01: 0x5f9fff14 r02: 0x5f9fff20
| ...
| Kernel panic - not syncing: Attempted to kill init! exitcode=0x0000000b
Signed-off-by: Vladimir Isaev <isaev@synopsys.com>
Reported-by: kernel test robot <lkp@intel.com>
Cc: Vineet Gupta <vgupta@synopsys.com>
Cc: stable@vger.kernel.org
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
For MMUv3 (and prior) the flush_tlb_{range,mm,page} API use the MMU
TLBWrite cmd which already nukes the entire uTLB, so NO need for
additional IVUTLB cmd from utlb_invalidate() - hence this patch
local_flush_tlb_all() is special since it uses a weaker TLBWriteNI
cmd (prec commit) to shoot down JTLB, hence we retain the explicit
uTLB flush
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
The unconditional full TLB flush (on say ASID rollover) iterates over each
entry and uses TLBWrite to zero it out. TLBWrite by design also invalidates
the uTLBs thus we end up invalidating it as many times as numbe rof
entries (512 or 1k)
Optimize this by using a weaker TLBWriteNI cmd in loop, which doesn't
tinker with uTLBs and an explicit one time IVUTLB, outside the loop to
invalidate them all once.
And given the optimiztion, the IVUTLB is now needed on MMUv4 too where
the uTLBs and JTLBs are otherwise coherent given the TLBInsertEntry /
TLBDeleteEntry commands
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
ARC700 exception (and intr handling) didn't have auto stack switching
thus had to rely on stashing a reg temporarily (to free it up) at a
known place in memory, allowing to code up the low level stack switching.
This however was not re-entrant in SMP which thus had to repurpose the
per-cpu MMU SCRATCH DATA register otherwise used to "cache" the task pdg
pointer (vs. reading it from mm struct)
The newer HS cores do have auto-stack switching and thus even SMP builds
can use the MMU SCRATCH reg as originally intended.
This patch fixes the restriction to ARC700 SMP builds only
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Based on 2 normalized pattern(s):
this program is free software you can redistribute it and or modify
it under the terms of the gnu general public license version 2 as
published by the free software foundation
this program is free software you can redistribute it and or modify
it under the terms of the gnu general public license version 2 as
published by the free software foundation #
extracted by the scancode license scanner the SPDX license identifier
GPL-2.0-only
has been chosen to replace the boilerplate/reference in 4122 file(s).
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Enrico Weigelt <info@metux.net>
Reviewed-by: Kate Stewart <kstewart@linuxfoundation.org>
Reviewed-by: Allison Randal <allison@lohutok.net>
Cc: linux-spdx@vger.kernel.org
Link: https://lkml.kernel.org/r/20190604081206.933168790@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
| arch/arc/mm/tlb.c:914:2: warning: variable length array 'pd0' is used [-Wvla]
| arch/arc/include/asm/cmpxchg.h:95:29: warning: value computed is not used [-Wunused-value]
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
HS48 cpus will have a new MMUv5, although Linux is currently not
explicitly supporting the newer features (so remains at V4).
The existing software/hardware version check is very tight and causes
boot abort. Given that the MMUv5 hardware is backwards compatible,
relax the boot check to allow current kernel support level to work
with new hardware.
Also while at it, move the ancient MMU related code to under ARCompact
builds as baseline MMU for HS cpus is v4.
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
I recently came upon a scenario where I would get a double fault
machine check exception tiriggered by a kernel module.
However the ensuing crash stacktrace (ksym lookup) was not working
correctly.
Turns out that machine check auto-disables MMU while modules are allocated
in kernel vaddr spapce.
This patch re-enables the MMU before start printing the stacktrace
making stacktracing of modules work upon a fatal exception.
Cc: stable@kernel.org
Signed-off-by: Jose Abreu <joabreu@synopsys.com>
Reviewed-by: Alexey Brodkin <abrodkin@synopsys.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
[vgupta: moved code into low level handler to avoid in 2 places]
Some of the boot printing code had printk() w/o explicit log level.
This patch introduces consistency allowing platforms to switch to less
verbose console logging using cmdline.
NPS400 with 4K CPUs needs to avoid the cpu info printing for faster
bootup.
Signed-off-by: Noam Camus <noamca@mellanox.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
PAE40 confiuration in hardware extends some of the address registers
for TLB/cache ops to 2 words.
So far kernel was NOT setting the higher word if feature was not enabled
in software which is wrong. Those need to be set to 0 in such case.
Normally this would be done in the cache flush / tlb ops, however since
these registers only exist conditionally, this would have to be
conditional to a flag being set on boot which is expensive/ugly -
specially for the more common case of PAE exists but not in use.
Optimize that by zero'ing them once at boot - nobody will write to
them afterwards
Cc: stable@vger.kernel.org #4.4+
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Update code that relied on sched.h including various MM types for them.
This will allow us to remove the <linux/mm_types.h> include from <linux/sched.h>.
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
NPS use special mapping right below TASK_SIZE.
Hence we need to lower STACK_TOP so that user stack won't
overlap NPS special mapping.
Signed-off-by: Noam Camus <noamc@ezchip.com>
Acked-by: Vineet Gupta <vgupta@synopsys.com>
On ARC, lower 2G of address space is translated and used for
- user vaddr space (region 0 to 5)
- unused kernel-user gutter (region 6)
- kernel vaddr space (region 7)
where each region simply represents 256MB of address space.
The kernel vaddr space of 256MB is used to implement vmalloc, modules
So far this was enough, but not on EZChip system with 4K CPUs (given
that per cpu mechanism uses vmalloc for allocating chunks)
So allow VMALLOC_SIZE to be configurable by expanding down into the unused
kernel-user gutter region which at default 256M was excessive anyways.
Also use _BITUL() to fix a build error since PGDIR_SIZE cannot use "1UL"
as called from assembly code in mm/tlbex.S
Signed-off-by: Noam Camus <noamc@ezchip.com>
[vgupta: rewrote changelog, debugged bootup crash due to int vs. hex]
Acked-by: Vineet Gupta <vgupta@synopsys.com>
This is the first working implementation of 40-bit physical address
extension on ARCv2.
Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
That way a single flip of phys_addr_t to 64 bit ensures all places
dealing with physical addresses get correct data
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
- Move the verbosity knob from .data to .bss by using inverted logic
- No need to readout PD1 descriptor
- clip the non pfn bits of PD0 to avoid clipping inside the loop
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
This frees up some bits to hold more high level info such as PAE being
present, w/o increasing the size of already bloated cpuinfo struct
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Implement the TLB flush routine to evict a sepcific Super TLB entry,
vs. moving to a new ASID on every such flush.
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
MMUv4 in HS38x cores supports Super Pages which are basis for Linux THP
support.
Normal and Super pages can co-exist (ofcourse not overlap) in TLB with a
new bit "SZ" in TLB page desciptor to distinguish between them.
Super Page size is configurable in hardware (4K to 16M), but fixed once
RTL builds.
The exact THP size a Linx configuration will support is a function of:
- MMU page size (typical 8K, RTL fixed)
- software page walker address split between PGD:PTE:PFN (typical
11:8:13, but can be changed with 1 line)
So for above default, THP size supported is 8K * 256 = 2M
Default Page Walker is 2 levels, PGD:PTE:PFN, which in THP regime
reduces to 1 level (as PTE is folded into PGD and canonically referred
to as PMD).
Thus thp PMD accessors are implemented in terms of PTE (just like sparc)
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
- Add mm_cpumask setting (aggregating only, unlike some other arches)
used to restrict the TLB flush cross-calling
- cross-calling versions of TLB flush routines (thanks to Noam)
Signed-off-by: Noam Camus <noamc@ezchip.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
------------------>8----------------------
arch/arc/mm/tlb.c: In function ‘do_tlb_overlap_fault’:
arch/arc/mm/tlb.c:688:13: warning: array subscript is above array bounds
[-Warray-bounds]
(pd0[n] & PAGE_MASK)) {
^
------------------>8----------------------
While at it, remove the usless last iteration of outer loop when reading
a TLB SET for duplicate entries.
Suggested-by: Mischa Jonker <mjonker@synopsys.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
This helps remove asid-to-mm reverse map
While mm->context.id contains the ASID assigned to a process, our ASID
allocator also used asid_mm_map[] reverse map. In a new allocation
cycle (mm->ASID >= @asid_cache), the Round Robin ASID allocator used this
to check if new @asid_cache belonged to some mm2 (from prev cycle).
If so, it could locate that mm using the ASID reverse map, and mark that
mm as unallocated ASID, to force it to refresh at the time of switch_mm()
However, for SMP, the reverse map has to be maintained per CPU, so
becomes 2 dimensional, hence got rid of it.
With reverse map gone, it is NOT possible to reach out to current
assignee. So we track the ASID allocation generation/cycle and
on every switch_mm(), check if the current generation of CPU ASID is
same as mm's ASID; If not it is refreshed.
(Based loosely on arch/sh implementation)
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
ASID allocation changes/1
This patch does 2 things:
(1) get_new_mmu_context() NOW moves mm->ASID to a new value ONLY if it
was from a prev allocation cycle/generation OR if mm had no ASID
allocated (vs. before would unconditionally moving to a new ASID)
Callers desiring unconditional update of ASID, e.g.local_flush_tlb_mm()
(for parent's address space invalidation at fork) need to first force
the parent to an unallocated ASID.
(2) get_new_mmu_context() always sets the MMU PID reg with unchanged/new
ASID value.
The gains are:
- consolidation of all asid alloc logic into get_new_mmu_context()
- avoiding code duplication in switch_mm() for PID reg setting
- Enables future change to fold activate_mm() into switch_mm()
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
-Asm code already has values of SW and HW ASID values, so they can be
passed to the printing routine.
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
This reorganizes the current TLB operations into psuedo-ops to better
pair with MMUv4's native Insert/Delete operations
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
With previous commit freeing up PTE bits, reassign them so as to:
- Match the bit to H/w counterpart where possible
(e.g. MMUv2 GLOBAL/PRESENT, this avoids a shift in create_tlb())
- Avoid holes in _PAGE_xxx definitions
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
The current ARC VM code has 13 flags in Page Table entry: some software
(accesed/dirty/non-linear-maps) and rest hardware specific. With 8k MMU
page, we need 19 bits for addressing page frame so remaining 13 bits is
just about enough to accomodate the current flags.
In MMUv4 there are 2 additional flags, SZ (normal or super page) and WT
(cache access mode write-thru) - and additionally PFN is 20 bits (vs. 19
before for 8k). Thus these can't be held in current PTE w/o making each
entry 64bit wide.
It seems there is some scope of compressing the current PTE flags (and
freeing up a few bits). Currently PTE contains fully orthogonal distinct
access permissions for kernel and user mode (Kr, Kw, Kx; Ur, Uw, Ux)
which can be folded into one set (R, W, X). The translation of 3 PTE
bits into 6 TLB bits (when programming the MMU) can be done based on
following pre-requites/assumptions:
1. For kernel-mode-only translations (vmalloc: 0x7000_0000 to
0x7FFF_FFFF), PTE additionally has PAGE_GLOBAL flag set (and user
space entries can never be global). Thus such a PTE can translate
to Kr, Kw, Kx (as appropriate) and zero for User mode counterparts.
2. For non global entries, the PTE flags can be used to create mirrored
K and U TLB bits. This is true after commit a950549c67
"ARC: copy_(to|from)_user() to honor usermode-access permissions"
which ensured that user-space translations _MUST_ have same access
permissions for both U/K mode accesses so that copy_{to,from}_user()
play fair with fault based CoW break and such...
There is no such thing as free lunch - the cost is slightly infalted
TLB-Miss Handlers.
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
The __cpuinit type of throwaway sections might have made sense
some time ago when RAM was more constrained, but now the savings
do not offset the cost and complications. For example, the fix in
commit 5e427ec2d0 ("x86: Fix bit corruption at CPU resume time")
is a good example of the nasty type of bugs that can be created
with improper use of the various __init prefixes.
After a discussion on LKML[1] it was decided that cpuinit should go
the way of devinit and be phased out. Once all the users are gone,
we can then finally remove the macros themselves from linux/init.h.
Note that some harmless section mismatch warnings may result, since
notify_cpu_starting() and cpu_up() are arch independent (kernel/cpu.c)
are flagged as __cpuinit -- so if we remove the __cpuinit from
arch specific callers, we will also get section mismatch warnings.
As an intermediate step, we intend to turn the linux/init.h cpuinit
content into no-ops as early as possible, since that will get rid
of these warnings. In any case, they are temporary and harmless.
This removes all the arch/arc uses of the __cpuinit macros from
all C files. Currently arc does not have any __CPUINIT used in
assembly files.
[1] https://lkml.org/lkml/2013/5/20/589
Cc: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Vineet Gupta