L2CTLR read/write functions are common to armv7 so, move
them in to include/asm/armv7.h and use them where ever it need.
Cc: Tom Warren <twarren@nvidia.com>
Signed-off-by: Jagan Teki <jagan@amarulasolutions.com>
Acked-by: Philipp Tomsich <philipp.tomsich@theobroma-systems.com>
Reviewed-by: Philipp Tomsich <philipp.tomsich@theobroma-systems.com>
[Backed out the change to arch/arm/mach-tegra/cache.c:]
Signed-off-by: Philipp Tomsich <philipp.tomsich@theobroma-systems.com>
Until now we've been using memory beyond psci_text_end as stack space
for the secure monitor or PSCI implementation, even if space was not
allocated for it.
This was partially fixed in ("ARM: allocate extra space for PSCI stack
in secure section during link phase"). However, calculating stack space
from psci_text_end in one place, while allocating the space in another
is error prone.
This patch adds a separate empty secure stack section, with space for
CONFIG_ARMV7_PSCI_NR_CPUS stacks, each 1 KB. There's also
__secure_stack_start and __secure_stack_end symbols. The linker script
handles calculating the correct VMAs for the stack section. For
platforms that relocate/copy the secure monitor before using it, the
space is not allocated in the executable, saving space.
For platforms that do not define CONFIG_ARMV7_PSCI_NR_CPUS, a whole page
of stack space for 4 CPUs is allocated, matching the previous behavior.
Signed-off-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Commit bfb33f0bc4 ("sunxi: mctl_mem_matches: Add missing memory
barrier") broke compilation for the Pine64, as dram_helper.c now
includes <asm/armv7.h>, which does not compile on arm64.
Fix this by moving all barrier instructions into a separate header
file, which can easily be shared between arm and arm64.
Also extend the inline assembly to take the "sy" argument, which is
optional for ARMv7, but mandatory for v8.
This fixes compilation for 64-bit sunxi boards (Pine64).
Acked-by: Ian Campbell <ijc@hellion.org.uk>
Signed-off-by: Andre Przywara <andre.przywara@arm.com>
In this case the secure code lives in RAM, and hence the memory node in
the device tree needs to be adjusted. This avoids that the OS will map
and possibly access the reservation.
Add support for setting CONFIG_ARMV7_SECURE_RESERVE_SIZE to carve out
such a region. We only support cutting off memory from the beginning or
the end of a RAM bank as we do not want to increase their number (which
would happen if punching a hole) for simplicity reasons
This will be used in a subsequent patch for Jetson-TK1.
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Tom Warren <twarren@nvidia.com>
CONFIG_ARMV7_VIRT depends on CONFIG_ARMV7_NONSEC, thus doesn't need to
be taken into account additionally. CONFIG_ARMV7_PSCI is only set on
boards that support CONFIG_ARMV7_NONSEC, and it only works on those.
CC: Tang Yuantian <Yuantian.Tang@freescale.com>
CC: York Sun <yorksun@freescale.com>
CC: Steve Rae <srae@broadcom.com>
CC: Andre Przywara <andre.przywara@linaro.org>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Tested-by: Alison Wang <alison.wang@freescale.com>
Signed-off-by: Tom Warren <twarren@nvidia.com>
This enables ARMv7 barrier operations support when
march=armv7-a is enabled.
Using CP15 barriers causes U-Boot bootm command crash when
transferring control to the loaded image on Renesas R8A7794 Cortex A7 CPU.
Using ARMv7 barrier operations instead of the deprecated CP15 barriers
helps to avoid these issues.
Signed-off-by: Valentine Barshak <valentine.barshak+renesas@cogentembedded.com>
Signed-off-by: Vladimir Barinov <vladimir.barinov+renesas@cogentembedded.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
Reviewed-by: Nobuhiro Iwamatsu <nobuhiro.iwamatsu.yj@renesas.com>
454179: Stale prediction may inhibit target address misprediction on
next predicted taken branch
Impacts: Every Cortex-A8 processors with revision lower than r2p1
Work around: Set IBE and disable branch size mispredict to 1
Also provide a hook for SoC specific handling to take place if needed.
Based on ARM errata Document revision 20.0 (13 Nov 2010)
Signed-off-by: Nishanth Menon <nm@ti.com>
Tested-by: Matt Porter <mporter@konsulko.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
Add workaround for Cortex-A15 ARM erratum 798870 which says
"If back-to-back speculative cache line fills (fill A and fill B) are
issued from the L1 data cache of a CPU to the L2 cache, the second
request (fill B) is then cancelled, and the second request would have
detected a hazard against a recent write or eviction (write B) to the
same cache line as fill B then the L2 logic might deadlock."
Implementations for SoC families such as Exynos, OMAP5/DRA7 etc
will be widely different.
Every SoC has slightly different manner of setting up access to L2ACLR
and similar registers since the Secure Monitor handling of Secure
Monitor Call(smc) is diverse. Hence an weak function is introduced
which may be overriden to implement SoC specific accessor implementation.
Based on ARM errata Document revision 18.0 (22 Nov 2013)
Signed-off-by: Nishanth Menon <nm@ti.com>
Tested-by: Matt Porter <mporter@konsulko.com>
Reviewed-by: Tom Rini <trini@konsulko.com>
The way the PSCI DT update happens currently means we pull in
<asm/armv7.h> everywhere, including on ARMv8 and that in turn brings in
<asm/io.h> for some non-PSCI related things that header needs to deal
with.
To fix this, we rework the hook slightly. A good portion of
arch/arm/cpu/armv7/virt-dt.c is common looking and I hope that when PSCI
is needed on ARMv8 we can re-use this by and large. So rename the
current hook to psci_update_dt(), move the prototype to <asm/psci.h> and
add an #ifdef that will make re-use later easier.
Reported-by: York Sun <yorksun@freescale.com>
Cc: Marc Zyngier <marc.zyngier@arm.com>
Cc: York Sun <yorksun@freescale.com>
Cc: Ian Campbell <ijc@hellion.org.uk>
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Albert ARIBAUD <albert.u.boot@aribaud.net>
Signed-off-by: Tom Rini <trini@konsulko.com>
Acked-by: York Sun <yorksun@freescale.com>
With a389531 we now call readl() from this file so add <asm/io.h> so
that we have a prototype for the function.
Signed-off-by: Tom Rini <trini@konsulko.com>
This patch adds workaround for the ARM errata 799270 which says
"If the L2 cache logic clock is stopped because of L2 inactivity,
setting or clearing the ACTLR.SMP bit might not be effective. The bit is
modified in the ACTLR, meaning a read of the register returns the
updated value. However the logic that uses that bit retains the previous
value."
Signed-off-by: Kimoon Kim <kimoon.kim@samsung.com>
Signed-off-by: Akshay Saraswat <akshay.s@samsung.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
This patch adds workaround for ARM errata 798870 which says
"If back-to-back speculative cache line fills (fill A and fill B) are
issued from the L1 data cache of a CPU to the L2 cache, the second
request (fill B) is then cancelled, and the second request would have
detected a hazard against a recent write or eviction (write B) to the
same cache line as fill B then the L2 logic might deadlock."
Signed-off-by: Kimoon Kim <kimoon.kim@samsung.com>
Signed-off-by: Akshay Saraswat <akshay.s@samsung.com>
Reviewed-by: Simon Glass <sjg@chromium.org>
Tested-by: Simon Glass <sjg@chromium.org>
Signed-off-by: Minkyu Kang <mk7.kang@samsung.com>
Commit 8bc347e2ec "ARM: bootm: Allow booting in secure mode on hyp capable
systems" added the capability to select nonsec vs sec mode boot via an
environment var.
There is a subtle gotcha with this functionality, which is that the PSCI nodes
are still created in the fdt (via armv7_update_dt->fdt_psci) even when booting
in secure mode. Which means that if the kernel is PSCI aware then it will fail
to boot because it will try and do PSCI from secure world, which won't work.
This likely didn't get noticed before because the original purpose was to
support booting the legacy linux-sunxi kernels which don't understand PSCI.
To fix expose boot_nonsec (renaming with armv7_ prefix) outside of bootm.c and
use from the virt-dt code.
As well as avoiding the creation of the PSCI nodes we should also avoid
reserving the secure RAM, so do so.
Signed-off-by: Ian Campbell <ijc@hellion.org.uk>
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Albert ARIBAUD <albert.u.boot@aribaud.net>
Cc: Tom Rini <trini@ti.com>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Generate the PSCI node in the device tree.
Also add a reserve section for the "secure" code that lives in
in normal RAM, so that the kernel knows it'd better not trip on
it.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Ian Campbell <ijc@hellion.org.uk>
The current non-sec switching code suffers from one major issue:
it cannot run in secure RAM, as a large part of u-boot still needs
to be run while we're switched to non-secure.
This patch reworks the whole HYP/non-secure strategy by:
- making sure the secure code is the *last* thing u-boot executes
before entering the payload
- performing an exception return from secure mode directly into
the payload
- allowing the code to be dynamically relocated to secure RAM
before switching to non-secure.
This involves quite a bit of horrible code, specially as u-boot
relocation is quite primitive.
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Ian Campbell <ijc@hellion.org.uk>
For the KVM and XEN hypervisors to be usable, we need to enter the
kernel in HYP mode. Now that we already are in non-secure state,
HYP mode switching is within short reach.
While doing the non-secure switch, we have to enable the HVC
instruction and setup the HYP mode HVBAR (while still secure).
The actual switch is done by dropping back from a HYP mode handler
without actually leaving HYP mode, so we introduce a new handler
routine in our new secure exception vector table.
In the assembly switching routine we save and restore the banked LR
and SP registers around the hypercall to do the actual HYP mode
switch.
The C routine first checks whether we are in HYP mode already and
also whether the virtualization extensions are available. It also
checks whether the HYP mode switch was finally successful.
The bootm command part only calls the new function after the
non-secure switch.
Signed-off-by: Andre Przywara <andre.przywara@linaro.org>
Currently the non-secure switch is only done for the boot processor.
To enable full SMP support, we have to switch all secondary cores
into non-secure state also.
So we add an entry point for secondary CPUs coming out of low-power
state and make sure we put them into WFI again after having switched
to non-secure state.
For this we acknowledge and EOI the wake-up IPI, then go into WFI.
Once being kicked out of it later, we sanity check that the start
address has actually been changed (since another attempt to switch
to non-secure would block the core) and jump to the new address.
The actual CPU kick is done by sending an inter-processor interrupt
via the GIC to all CPU interfaces except the requesting processor.
The secondary cores will then setup their respective GIC CPU
interface.
While this approach is pretty universal across several ARMv7 boards,
we make this function weak in case someone needs to tweak this for
a specific board.
The way of setting the secondary's start address is board specific,
but mostly different only in the actual SMP pen address, so we also
provide a weak default implementation and just depend on the proper
address to be set in the config file.
Signed-off-by: Andre Przywara <andre.przywara@linaro.org>
The core specific part of the work is done in the assembly routine
in nonsec_virt.S, introduced with the previous patch, but for the full
glory we need to setup the GIC distributor interface once for the
whole system, which is done in C here.
The routine is placed in arch/arm/cpu/armv7 to allow easy access from
other ARMv7 boards.
We check the availability of the security extensions first.
Since we need a safe way to access the GIC, we use the PERIPHBASE
registers on Cortex-A15 and A7 CPUs and do some sanity checks.
Boards not implementing the CBAR can override this value via a
configuration file variable.
Then we actually do the GIC enablement:
a) enable the GIC distributor, both for non-secure and secure state
(GICD_CTLR[1:0] = 11b)
b) allow all interrupts to be handled from non-secure state
(GICD_IGROUPRn = 0xFFFFFFFF)
The core specific GIC setup is then done in the assembly routine.
Signed-off-by: Andre Przywara <andre.przywara@linaro.org>
While actually switching to non-secure state is one thing, another
part of this process is to make sure that we still have full access
to the interrupt controller (GIC).
The GIC is fully aware of secure vs. non-secure state, some
registers are banked, others may be configured to be accessible from
secure state only.
To be as generic as possible, we get the GIC memory mapped address
based on the PERIPHBASE value in the CBAR register. Since this
register is not architecturally defined, we check the MIDR before to
be from an A15 or A7.
For CPUs not having the CBAR or boards with wrong information herein
we allow providing the base address as a configuration variable.
Now that we know the GIC address, we:
a) allow private interrupts to be delivered to the core
(GICD_IGROUPR0 = 0xFFFFFFFF)
b) enable the CPU interface (GICC_CTLR[0] = 1)
c) set the priority filter to allow non-secure interrupts
(GICC_PMR = 0xFF)
Also we allow access to all coprocessor interfaces from non-secure
state by writing the appropriate bits in the NSACR register.
The generic timer base frequency register is only accessible from
secure state, so we have to program it now. Actually this should be
done from primary firmware before, but some boards seems to omit
this, so if needed we do this here with a board specific value.
The Versatile Express board does not need this, so we remove the
frequency from the configuration file here.
After having switched to non-secure state, we also enable the
non-secure GIC CPU interface, since this register is banked.
Since we need to call this routine also directly from the smp_pen
later (where we don't have any stack), we can only use caller saved
registers r0-r3 and r12 to not mess with the compiler.
Signed-off-by: Andre Przywara <andre.przywara@linaro.org>
armv7.h contains some useful constants, but also C prototypes.
To include it also in assembly files, protect the non-assembly
part appropriately.
Signed-off-by: Andre Przywara <andre.przywara@linaro.org>
Adding the CPU detection suport for OMAP5430 and
OMAP5432 ES2.0 SOCs.
Signed-off-by: R Sricharan <r.sricharan@ti.com>
Cc: Tom Rini <trini@ti.com>
Cc: Nishanth Menon <nm@ti.com>
This patch adds the minimal support for OMAP5. The platform and machine
specific headers and sources updated for OMAP5430.
OMAP5430 is Texas Instrument's SOC based on ARM Cortex-A15 SMP architecture.
It's a dual core SOC with GIC used for interrupt handling and SCU for cache
coherency.
Also moved some part of code from the basic platform support that can be made
common for OMAP4/5. Rest is kept out seperately. The same approach is followed
for clocks and emif support in the subsequent patches.
Signed-off-by: sricharan <r.sricharan@ti.com>
Signed-off-by: Sandeep Paulraj <s-paulraj@ti.com>
- Add a framework for layered cache maintenance
- separate out SOC specific outer cache maintenance from
maintenance of caches known to CPU
- Add generic ARMv7 cache maintenance operations that affect all
caches known to ARMv7 CPUs. For instance in Cortex-A8 these
opertions will affect both L1 and L2 caches. In Cortex-A9
these will affect only L1 cache
- D-cache operations supported:
- Invalidate entire D-cache
- Invalidate D-cache range
- Flush(clean & invalidate) entire D-cache
- Flush D-cache range
- I-cache operations supported:
- Invalidate entire I-cache
- Add maintenance functions for TLB, branch predictor array etc.
- Enable -march=armv7-a so that armv7 assembly instructions can be
used
Signed-off-by: Aneesh V <aneesh@ti.com>
Jagan Teki
Chen-Yu Tsai
Andre Przywara
Jan Kiszka
Valentine Barshak
Nishanth Menon
Tom Rini
Akshay Saraswat
Ian Campbell
Marc Zyngier
Andre Przywara
Wolfgang Denk
SRICHARAN R
Aneesh V