Implement invalidation of specific pairs of GVA page table entries in
one or both of the GVA page tables. This is used when existing mappings
are replaced in the guest TLB by emulated TLBWI/TLBWR instructions. Due
to the sharing of page tables in the host kernel range, we should be
careful not to allow host pages to be invalidated.
Add a helper kvm_mips_walk_pgd() which can be used when walking of
either GPA (future patches) or GVA page tables is needed, optionally
with allocation of page tables along the way when they don't exist.
GPA page table walking will need to be protected by the kvm->mmu_lock,
so we also add a small MMU page cache in each KVM VCPU, like that found
for other architectures but smaller. This allows enough pages to be
pre-allocated to handle a single fault without holding the lock,
allowing the helper to run with the lock held without having to handle
allocation failures.
Using the same mechanism for GVA allows the same code to be used, and
allows it to use the same cache of allocated pages if the GPA walk
didn't need to allocate any new tables.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Implement invalidation of large ranges of virtual addresses from GVA
page tables in response to a guest ASID change (immediately for guest
kernel page table, lazily for guest user page table).
We iterate through a range of page tables invalidating entries and
freeing fully invalidated tables. To minimise overhead the exact ranges
invalidated depends on the flags argument to kvm_mips_flush_gva_pt(),
which also allows it to be used in future KVM_CAP_SYNC_MMU patches in
response to GPA changes, which unlike guest TLB mapping changes affects
guest KSeg0 mappings.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Refactor kvm_mips_host_tlb_inv() to also be able to invalidate any
matching TLB entry in the kernel ASID rather than assuming only the TLB
entries in the user ASID can change. Two new bool user/kernel arguments
allow the caller to indicate whether the mapping should affect each of
the ASIDs for guest user/kernel mode.
- kvm_mips_invalidate_guest_tlb() (used by TLBWI/TLBWR emulation) can
now invalidate any corresponding TLB entry in both the kernel ASID
(guest kernel may have accessed any guest mapping), and the user ASID
if the entry being replaced is in guest USeg (where guest user may
also have accessed it).
- The tlbmod fault handler (and the KSeg0 / TLB mapped / commpage fault
handlers in later patches) can now invalidate the corresponding TLB
entry in whichever ASID is currently active, since only a single page
table will have been updated anyway.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
kvm_mips_host_tlb_inv() uses the TLBP instruction to probe the host TLB
for an entry matching the given guest virtual address, and determines
whether a match was found based on whether CP0_Index > 0. This is
technically incorrect as an index of 0 (with the high bit clear) is a
perfectly valid TLB index.
This is harmless at the moment due to the use of at least 1 wired TLB
entry for the KVM commpage, however we will soon be ridding ourselves of
that particular wired entry so lets fix the condition in case the entry
needing invalidation does land at TLB index 0.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
tlbex.c uses the implementation dependent $22 CP0 register group on
NetLogic cores, with the help of the c0_kscratch() helper. Allow these
registers to be allocated by the KVM entry code too instead of assuming
KScratch registers are all $31, which will also allow pgd_reg to be
handled since it is allocated that way.
We also drop the masking of kscratch_mask with 0xfc, as it is redundant
for the standard KScratch registers (Config4.KScrExist won't have the
low 2 bits set anyway), and apparently not necessary for NetLogic.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Set init_mm as the active_mm and update mm_cpumask(current->mm) to
reflect that it isn't active when in guest context. This prevents cache
management code from attempting cache flushes on host virtual addresses
while in guest context, for example due to a cache management IPIs or
later when writing of dynamically translated code hits copy on write.
We do this using helpers in static kernel code to avoid having to export
init_mm to modules.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
We only need the guest ASID loaded while in guest context, i.e. while
running guest code and while handling guest exits. We load the guest
ASID when entering the guest, however we restore the host ASID later
than necessary, when the VCPU state is saved i.e. vcpu_put() or slightly
earlier if preempted after returning to the host.
This mismatch is both unpleasant and causes redundant host ASID restores
in kvm_trap_emul_vcpu_put(). Lets explicitly restore the host ASID when
returning to the host, and don't bother restoring the host ASID on
context switch in unless we're already in guest context.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Add implementation callbacks for entering the guest (vcpu_run()) and
reentering the guest (vcpu_reenter()), allowing implementation specific
operations to be performed before entering the guest or after returning
to the host without cluttering kvm_arch_vcpu_ioctl_run().
This allows the T&E specific lazy user GVA flush to be moved into
trap_emul.c, along with disabling of the HTW. We also move
kvm_mips_deliver_interrupts() as VZ will need to restore the guest timer
state prior to delivering interrupts.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
The kvm_vcpu_arch structure contains both mm_structs for allocating MMU
contexts (primarily the ASID) but it also copies the resulting ASIDs
into guest_{user,kernel}_asid[] arrays which are referenced from uasm
generated code.
This duplication doesn't seem to serve any purpose, and it gets in the
way of generalising the ASID handling across guest kernel/user modes, so
lets just extract the ASID straight out of the mm_struct on demand, and
in fact there are convenient cpu_context() and cpu_asid() macros for
doing so.
To reduce the verbosity of this code we do also add kern_mm and user_mm
local variables where the kernel and user mm_structs are used.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
The MIPS KVM host and guest GVA ASIDs may need regenerating when
scheduling a process in guest context, which is done from the
kvm_arch_vcpu_load() / kvm_arch_vcpu_put() functions in mmu.c.
However this is a fairly implementation specific detail. VZ for example
may use GuestIDs instead of normal ASIDs to distinguish mappings
belonging to different guests, and even on VZ without GuestID the root
TLB will be used differently to trap & emulate.
Trap & emulate GVA ASIDs only relate to the user part of the full
address space, so can be left active during guest exit handling (guest
context) to allow guest instructions to be easily read and translated.
VZ root ASIDs however are for GPA mappings so can't be left active
during normal kernel code. They also aren't useful for accessing guest
virtual memory, and we should have CP0_BadInstr[P] registers available
to provide encodings of trapping guest instructions anyway.
Therefore move the ASID preemption handling into the implementation
callback.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Convert the get_regs() and set_regs() callbacks to vcpu_load() and
vcpu_put(), which provide a cpu argument and more closely match the
kvm_arch_vcpu_load() / kvm_arch_vcpu_put() that they are called by.
This is in preparation for moving ASID management into the
implementations.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
KVM T&E uses an ASID for guest kernel mode and an ASID for guest user
mode. The current ASID is saved when the guest is scheduled out, and
restored when scheduling back in, with checks for whether the ASID needs
to be regenerated.
This isn't really necessary as the ASID can be easily determined by the
current guest mode, so lets simplify it to just read the required ASID
from guest_kernel_asid or guest_user_asid even if the ASID hasn't been
regenerated.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
MIPS incompletely implements the KVM_NMI ioctl to supposedly perform a
CPU reset, but all it actually does is invalidate the ASIDs. It doesn't
expose the KVM_CAP_USER_NMI capability which is supposed to indicate the
presence of the KVM_NMI ioctl, and no user software actually uses it on
MIPS.
Since this is dead code that would technically need updating for GVA
page table handling in upcoming patches, remove it now. If we wanted to
implement NMI injection later it can always be done properly along with
the KVM_CAP_USER_NMI capability, and if we wanted to implement a proper
CPU reset it would be better done with a separate ioctl.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Merge in MIPS prerequisites from GVA page tables and GPA page tables
series. The same branch can also merge into the MIPS tree.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
The protected cache ops contain no out of line fixup code to return an
error code in the event of a fault, with the cache op being skipped in
that case. For KVM however we'd like to detect this case as page
faulting will be disabled so it could happen during normal operation if
the GVA page tables were flushed, and need to be handled by the caller.
Add the out-of-line fixup code to load the error value -EFAULT into the
return variable, and adapt the protected cache line functions to pass
the error back to the caller.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
Export pmd_init(), invalid_pmd_table and tlbmiss_handler_setup_pgd to
GPL kernel modules so that MIPS KVM can use the inline page table
management functions and switch between page tables:
- pmd_init() will be used directly by KVM to initialise newly allocated
pmd tables with invalid lower level table pointers.
- invalid_pmd_table is used by pud_present(), pud_none(), and
pud_clear(), which KVM will use to test and clear pud entries.
- tlbmiss_handler_setup_pgd() will be called by KVM entry code to switch
to the appropriate GVA page tables.
Signed-off-by: James Hogan <james.hogan@imgtec.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: "Radim Krčmář" <rkrcmar@redhat.com>
Cc: linux-mips@linux-mips.org
Cc: kvm@vger.kernel.org
* Return directly after a call of the function "copy_from_user" failed
in a case block.
* Delete the jump label "out" which became unnecessary with
this refactoring.
Signed-off-by: Markus Elfring <elfring@users.sourceforge.net>
Reviewed-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: James Hogan <james.hogan@imgtec.com>
The DMA controller channel and port configuration is changed by
selecting the port or channel in one register and then update the
configuration in other registers. This has to be done in an atomic
operation. Previously only the local interrupts were deactivated which
works for single CPU systems. If the system supports SMP a better
locking is needed, use spinlocks instead.
On more recent SoCs (at least xrx200 and later) there are two memory
regions to change the configuration, there we could use one area for
each CPU and do not have to synchronize between the CPUs and more.
Signed-off-by: Hauke Mehrtens <hauke@hauke-m.de>
Cc: john@phrozen.org
Cc: linux-mips@linux-mips.org
Patchwork: https://patchwork.linux-mips.org/patch/14912/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
gcc warns about nonstandard declarations:
arch/mips/sgi-ip32/ip32-irq.c:31:1: error: 'inline' is not at beginning of declaration [-Werror=old-style-declaration]
arch/mips/sgi-ip32/ip32-irq.c:36:1: error: 'inline' is not at beginning of declaration [-Werror=old-style-declaration]
arch/mips/sgi-ip27/ip27-klnuma.c: In function 'replicate_kernel_text':
arch/mips/sgi-ip27/ip27-klnuma.c:85:116: error: old-style function definition [-Werror=old-style-definition]
Moving 'inline' before the return type, and adding argument types
shuts up the warning here. This patch affects several platforms,
but all in a trivial way. I'm fixing up all instances I found in
any of the 'defconfig' builds.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/15050/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
kernelci reports a failure of the ip28_defconfig build after upgrading its
gcc version:
arch/mips/sgi-ip22/Platform:29: *** gcc doesn't support needed option -mr10k-cache-barrier=store. Stop.
The problem apparently is that the -mr10k-cache-barrier=store option is now
rejected for CPUs other than r10k. Explicitly including the CPU in the
check fixes this and is safe because both options were introduced in
gcc-4.4.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/15049/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
A new gcc warning shows up for this old code with gcc-6:
arch/mips/loongson64/common/dma-swiotlb.c: In function 'loongson_dma_alloc_coherent':
arch/mips/loongson64/common/dma-swiotlb.c:35:2: error: suggest braces around empty body in an 'else' statement [-Werror=empty-body]
The code can be easily restructured to look more readable
and avoid the warning at the same time.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Robin Murphy <robin.murphy@arm.com>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/15047/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
request_mem_region returns a NULL pointer on error, comparing it
against a number results in a warning:
arch/mips/ralink/of.c: In function 'plat_of_remap_node':
arch/mips/ralink/of.c:45:15: error: ordered comparison of pointer with integer zero [-Werror=extra]
arch/mips/ralink/irq.c: In function 'intc_of_init':
arch/mips/ralink/irq.c:167:15: error: ordered comparison of pointer with integer zero [-Werror=extra]
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: John Crispin <john@phrozen.org>
Cc: Tobias Wolf <dev-NTEO@vplace.de>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/15045/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Hauke already fixed a couple of them, but one instance remains
that checks for a negative integer when it should check
for a NULL pointer:
arch/mips/lantiq/xway/sysctrl.c: In function 'ltq_soc_init':
arch/mips/lantiq/xway/sysctrl.c:473:19: error: ordered comparison of pointer with integer zero [-Werror=extra]
Fixes: 6e80785267 ("MIPS: Lantiq: Fix check for return value of request_mem_region()")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: John Crispin <john@phrozen.org>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/15043/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
vdso.h includes <spaces.h> implicitly after defining CONFIG_32BITS.
This defeats the override in mach-ip27/spaces.h, leading to
a build error that shows up in kernelci.org:
In file included from arch/mips/include/asm/mach-ip27/spaces.h:29:0,
from arch/mips/include/asm/page.h:12,
from arch/mips/vdso/vdso.h:26,
from arch/mips/vdso/gettimeofday.c:11:
arch/mips/include/asm/mach-generic/spaces.h:28:0: error: "CAC_BASE" redefined [-Werror]
#define CAC_BASE _AC(0x80000000, UL)
An earlier patch tried to make the second definition conditional,
but that patch had the #ifdef in the wrong place, and would lead
to another warning:
arch/mips/include/asm/io.h: In function 'phys_to_virt':
arch/mips/include/asm/io.h:138:9: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast]
For all I can tell, there is no other reason than vdso32 to ever
include this file with CONFIG_32BITS set, and the vdso itself should
never refer to the base addresses as it is running in user space,
so adding an #ifdef here is safe.
Link: https://patchwork.kernel.org/patch/9418187/
Fixes: 3ffc17d876 ("MIPS: Adjust MIPS64 CAC_BASE to reflect Config.K0")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Paul Burton <paul.burton@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/15039/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
kernelci.org reports tons of build warnings for linux-next:
35 WARNING: "memcpy" [fs/fat/msdos.ko] has no CRC!
35 WARNING: "__copy_user" [fs/fat/fat.ko] has no CRC!
32 WARNING: EXPORT symbol "memset" [vmlinux] version generation failed, symbol will not be versioned.
32 WARNING: EXPORT symbol "copy_page" [vmlinux] version generation failed, symbol will not be versioned.
32 WARNING: EXPORT symbol "clear_page" [vmlinux] version generation failed, symbol will not be versioned.
32 WARNING: EXPORT symbol "__strncpy_from_user_nocheck_asm" [vmlinux] version generation failed, symbol will not be versioned.
The problem here is mainly the missing asm/asm-prototypes.h header file
that is supposed to include the prototypes for each symbol that is exported
from an assembler file.
A second problem is that the asm/uaccess.h header contains some but not
all the necessary declarations for the user access helpers.
Finally, the vdso build is broken once we add asm/asm-prototypes.h, so
we have to fix this at the same time by changing the vdso header. My
approach here is to just not look for exported symbols in the VDSO
assembler files, as the symbols cannot be exported anyway.
Fixes: 576a2f0c5c ("MIPS: Export memcpy & memset functions alongside their definitions")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: James Hogan <james.hogan@imgtec.com>
Cc: Maciej W. Rozycki <macro@imgtec.com>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/15038/
Patchwork: https://patchwork.linux-mips.org/patch/15069/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
In linux-4.10-rc, NF_CT_PROTO_UDPLITE and NF_CT_PROTO_DCCP are bool
symbols instead of tristate, and kernelci.org reports a bunch of
warnings for this, like:
arch/mips/configs/malta_kvm_guest_defconfig:63:warning: symbol value 'm' invalid for NF_CT_PROTO_UDPLITE
arch/mips/configs/malta_defconfig:62:warning: symbol value 'm' invalid for NF_CT_PROTO_DCCP
arch/mips/configs/malta_defconfig:63:warning: symbol value 'm' invalid for NF_CT_PROTO_UDPLITE
arch/mips/configs/ip22_defconfig:70:warning: symbol value 'm' invalid for NF_CT_PROTO_DCCP
arch/mips/configs/ip22_defconfig:71:warning: symbol value 'm' invalid for NF_CT_PROTO_UDPLITE
This changes all the MIPS defconfigs with these symbols to have them
built-in.
Fixes: 9b91c96c5d ("netfilter: conntrack: built-in support for UDPlite")
Fixes: c51d39010a ("netfilter: conntrack: built-in support for DCCP")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Cc: linux-mips@linux-mips.org
Cc: linux-kernel@vger.kernel.org
Patchwork: https://patchwork.linux-mips.org/patch/14999/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
