[ Upstream commit cef397038167ac15d085914493d6c86385773709 ]
Stefan Agner reported a bug when using zsram on 32-bit Arm machines
with RAM above the 4GB address boundary:
Unable to handle kernel NULL pointer dereference at virtual address 00000000
pgd = a27bd01c
[00000000] *pgd=236a0003, *pmd=1ffa64003
Internal error: Oops: 207 [#1] SMP ARM
Modules linked in: mdio_bcm_unimac(+) brcmfmac cfg80211 brcmutil raspberrypi_hwmon hci_uart crc32_arm_ce bcm2711_thermal phy_generic genet
CPU: 0 PID: 123 Comm: mkfs.ext4 Not tainted 5.9.6 #1
Hardware name: BCM2711
PC is at zs_map_object+0x94/0x338
LR is at zram_bvec_rw.constprop.0+0x330/0xa64
pc : [<c0602b38>] lr : [<c0bda6a0>] psr: 60000013
sp : e376bbe0 ip : 00000000 fp : c1e2921c
r10: 00000002 r9 : c1dda730 r8 : 00000000
r7 : e8ff7a00 r6 : 00000000 r5 : 02f9ffa0 r4 : e3710000
r3 : 000fdffe r2 : c1e0ce80 r1 : ebf979a0 r0 : 00000000
Flags: nZCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment user
Control: 30c5383d Table: 235c2a80 DAC: fffffffd
Process mkfs.ext4 (pid: 123, stack limit = 0x495a22e6)
Stack: (0xe376bbe0 to 0xe376c000)
As it turns out, zsram needs to know the maximum memory size, which
is defined in MAX_PHYSMEM_BITS when CONFIG_SPARSEMEM is set, or in
MAX_POSSIBLE_PHYSMEM_BITS on the x86 architecture.
The same problem will be hit on all 32-bit architectures that have a
physical address space larger than 4GB and happen to not enable sparsemem
and include asm/sparsemem.h from asm/pgtable.h.
After the initial discussion, I suggested just always defining
MAX_POSSIBLE_PHYSMEM_BITS whenever CONFIG_PHYS_ADDR_T_64BIT is
set, or provoking a build error otherwise. This addresses all
configurations that can currently have this runtime bug, but
leaves all other configurations unchanged.
I looked up the possible number of bits in source code and
datasheets, here is what I found:
- on ARC, CONFIG_ARC_HAS_PAE40 controls whether 32 or 40 bits are used
- on ARM, CONFIG_LPAE enables 40 bit addressing, without it we never
support more than 32 bits, even though supersections in theory allow
up to 40 bits as well.
- on MIPS, some MIPS32r1 or later chips support 36 bits, and MIPS32r5
XPA supports up to 60 bits in theory, but 40 bits are more than
anyone will ever ship
- On PowerPC, there are three different implementations of 36 bit
addressing, but 32-bit is used without CONFIG_PTE_64BIT
- On RISC-V, the normal page table format can support 34 bit
addressing. There is no highmem support on RISC-V, so anything
above 2GB is unused, but it might be useful to eventually support
CONFIG_ZRAM for high pages.
Fixes: 61989a80fb3a ("staging: zsmalloc: zsmalloc memory allocation library")
Fixes: 02390b87a945 ("mm/zsmalloc: Prepare to variable MAX_PHYSMEM_BITS")
Acked-by: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Reviewed-by: Stefan Agner <stefan@agner.ch>
Tested-by: Stefan Agner <stefan@agner.ch>
Acked-by: Mike Rapoport <rppt@linux.ibm.com>
Link: https://lore.kernel.org/linux-mm/bdfa44bf1c570b05d6c70898e2bbb0acf234ecdf.1604762181.git.stefan@agner.ch/
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit b5fca7c55f9fbab5ad732c3bce00f31af6ba5cfa ]
AT_VECTOR_SIZE_ARCH should be defined with the maximum number of
NEW_AUX_ENT entries that ARCH_DLINFO can contain, but it wasn't defined
for RISC-V at all even though ARCH_DLINFO will contain one NEW_AUX_ENT
for the VDSO address.
Signed-off-by: Zong Li <zong.li@sifive.com>
Reviewed-by: Palmer Dabbelt <palmerdabbelt@google.com>
Reviewed-by: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 66d18dbda8469a944dfec6c49d26d5946efba218 ]
Without this we get lockdep failures. They're spurious failures as SMP isn't
up when ftrace_init_nop() is called. As far as I can tell the easiest fix is
to just take the lock, which also seems like the safest fix.
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Acked-by: Guo Ren <guoren@kernel.org>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 38b7c2a3ffb1fce8358ddc6006cfe5c038ff9963 ]
While digging through the recent mmiowb preemption issue it came up that
we aren't actually preventing IO from crossing a scheduling boundary.
While it's a bit ugly to overload smp_mb__after_spinlock() with this
behavior, it's what PowerPC is doing so there's some precedent.
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit 0cac21b02ba5f3095fd2dcc77c26a25a0b2432ed upstream.
With the current 8KB stack size there are frequent overflows in a 64-bit
configuration. We may split IRQ stacks off in the future, but this fixes a
number of issues right now.
Signed-off-by: Andreas Schwab <schwab@suse.de>
Reviewed-by: Anup Patel <anup@brainfault.org>
[Palmer: mention irqstack in the commit text]
Fixes: 7db91e57a0ac ("RISC-V: Task implementation")
Cc: stable@vger.kernel.org
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit 6c58f25e6938c073198af8b1e1832f83f8f0df33 ]
The argument passed to cmpxchg is not guaranteed to be sign
extended, but lr.w sign extends on RV64I. This makes cmpxchg
fail on clang built kernels when __old is negative.
To fix this, we just cast __old to long which sign extends on
RV64I. With this fix, clang built RISC-V kernels now boot.
Link: https://github.com/ClangBuiltLinux/linux/issues/867
Signed-off-by: Nathan Huckleberry <nhuck@google.com>
Signed-off-by: Palmer Dabbelt <palmerdabbelt@google.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
Arnd Bergmann