This patch reworks the manner in which the M,A, C, SA, I, WXN & EE bits of
SCTLR_EL3 & SCTLR_EL1 are managed. The EE bit is cleared immediately after reset
in EL3. The I, A and SA bits are set next in EL3 and immediately upon entry in
S-EL1. These bits are no longer managed in the blX_arch_setup() functions. They
do not have to be saved and restored either. The M, WXN and optionally the C
bit are set in the enable_mmu_elX() function. This is done during both the warm
and cold boot paths.
Achin Gupta [Thu, 26 Jun 2014 10:12:37 +0000 (11:12 +0100)]
Remove the concept of coherent stacks
This patch removes the allocation of memory for coherent stacks, associated
accessor function and some dead code which called the accessor function. It also
updates the porting guide to remove the concept and the motivation behind using
stacks allocated in coherent memory.
Achin Gupta [Thu, 26 Jun 2014 08:58:52 +0000 (09:58 +0100)]
Remove coherent stack usage from the warm boot path
This patch uses stacks allocated in normal memory to enable the MMU early in the
warm boot path thus removing the dependency on stacks allocated in coherent
memory. Necessary cache and stack maintenance is performed when a cpu is being
powered down and up. This avoids any coherency issues that can arise from
reading speculatively fetched stale stack memory from another CPUs cache. These
changes affect the warm boot path in both BL3-1 and BL3-2.
The EL3 system registers responsible for preserving the MMU state are not saved
and restored any longer. Static values are used to program these system
registers when a cpu is powered on or resumed from suspend.
Achin Gupta [Thu, 26 Jun 2014 07:59:07 +0000 (08:59 +0100)]
Make enablement of the MMU more flexible
This patch adds a 'flags' parameter to each exception level specific function
responsible for enabling the MMU. At present only a single flag which indicates
whether the data cache should also be enabled is implemented. Subsequent patches
will use this flag when enabling the MMU in the warm boot paths.
Achin Gupta [Wed, 25 Jun 2014 18:26:22 +0000 (19:26 +0100)]
Remove coherent stack usage from the cold boot path
This patch reworks the cold boot path across the BL1, BL2, BL3-1 and BL3-2 boot
loader stages to not use stacks allocated in coherent memory for early platform
setup and enabling the MMU. Stacks allocated in normal memory are used instead.
Attributes for stack memory change from nGnRnE when the MMU is disabled to
Normal WBWA Inner-shareable when the MMU and data cache are enabled. It is
possible for the CPU to read stale stack memory after the MMU is enabled from
another CPUs cache. Hence, it is unsafe to turn on the MMU and data cache while
using normal stacks when multiple CPUs are a part of the same coherency
domain. It is safe to do so in the cold boot path as only the primary cpu
executes it. The secondary cpus are in a quiescent state.
This patch does not remove the allocation of coherent stack memory. That is done
in a subsequent patch.
Update the documentation about the memory layout on FVP
Update the "Memory layout on FVP platforms" section in the Firmware
Design document to reflect the overlaying of BL1 and BL2 images
by BL3-1 and BL3-2.
Also update the Porting Guide document to mention the
BL31_PROGBITS_LIMIT and BL32_PROGBITS_LIMIT constants.
- Add support for loading a BL3-0 image in BL2. Information about
memory extents is populated by platform-specific code. Subsequent
handling of BL3-0 is also platform specific.
The BL2 main function has been broken down to improve readability.
The BL3-2 image is now loaded before the BL3-3 image to align with
the boot flow.
- Build system: Add support for specifying a BL3-0 image that will be
included into the FIP image.
- IO FIP driver: Add support for identifying a BL3-0 image inside a
FIP image.
- Update the documentation to reflect the above changes.
fvp: Reuse BL1 and BL2 memory through image overlaying
This patch re-organizes the memory layout on FVP as to give the
BL3-2 image as much memory as possible.
Considering these two facts:
- not all images need to live in memory at the same time. Once
in BL3-1, the memory used by BL1 and BL2 can be reclaimed.
- when BL2 loads the BL3-1 and BL3-2 images, it only considers the
PROGBITS sections of those 2 images. The memory occupied by the
NOBITS sections will be touched only at execution of the BL3-x
images;
Then it is possible to choose the different base addresses such that
the NOBITS sections of BL3-1 and BL3-2 overlay BL1 and BL2.
On FVP we choose to put:
- BL1 and BL3-1 at the top of the Trusted RAM, with BL3-1 NOBITS
sections overlaying BL1;
- BL3-2 at the bottom of the Trusted RAM, with its NOBITS sections
overlaying BL2;
New platform-specific constants have been introduced to easily check
at link time that BL3-1 and BL3-2 PROGBITS sections don't overwrite
BL1 and BL2. These are optional and the platform code is free to define
them or not. If not defined, the linker won't attempt to check
image overlaying.
Dan Handley [Fri, 20 Jun 2014 08:43:15 +0000 (09:43 +0100)]
Refactor fvp gic code to be a generic driver
Refactor the FVP gic code in plat/fvp/fvp_gic.c to be a generic ARM
GIC driver in drivers/arm/gic/arm_gic.c. Provide the platform
specific inputs in the arm_gic_setup() function so that the driver
has no explicit dependency on platform code.
Provide weak implementations of the platform interrupt controller
API in a new file, plat/common/plat_gic.c. These simply call through
to the ARM GIC driver.
Move the only remaining FVP GIC function, fvp_gic_init() to
plat/fvp/aarch64/fvp_common.c and remove plat/fvp/fvp_gic.c
Dan Handley [Fri, 20 Jun 2014 11:02:01 +0000 (12:02 +0100)]
Refactor fvp_config into common platform header
Changed the fvp_config array in fvp_common.c into a struct and
moved into a new optional common platform header,
include/plat/common/plat_config.h. Removed the config definitions
in fvp_def.h and updated all references to the platform config.
This makes the interface to the platform config cleaner and uses
a little less RAM.
Lin Ma [Fri, 27 Jun 2014 23:56:30 +0000 (16:56 -0700)]
Calculate TCR bits based on VA and PA
Currently the TCR bits are hardcoded in xlat_tables.c. In order to
map higher physical address into low virtual address, the TCR bits
need to be configured accordingly.
This patch is to save the max VA and PA and calculate the TCR.PS/IPS
and t0sz bits in init_xlat_tables function.
There was already a rudimentary mechanism to detect whether BL1
R/W data was loaded at the top or bottom of memory. Basically,
- either BL1 was loaded at the very end of the trusted RAM
- in all other cases BL1 was considered sitting at the bottom of
the memory and the memory usage structure was updated accordingly,
potentially resulting in critical memory waste.
For instance, if BL1 R/W base address was set to
(TZRAM_END - 4096 - bl1_size), it would virtually occupy the whole
memory.
This patch improves the mechanism to detect the location of BL1
to avoid such scenarios.
This concept is no longer required since we now support loading of
images at fixed addresses only.
The image loader now automatically detects the position of the image
inside the current memory layout and updates the layout such that
memory fragmentation is minimised.
The 'attr' field of the meminfo data structure, which used to hold
the bottom/top loading information, has been removed. Also the 'next'
field has been removed as it wasn't used anywhere.
The 'init_bl2_mem_layout()' function has been moved out of common
code and put in BL1-specific code. It has also been renamed into
'bl1_init_bl2_mem_layout'.
Andrew Thoelke [Thu, 26 Jun 2014 13:27:26 +0000 (14:27 +0100)]
Support later revisions of the Foundation FVP
The code in the FVP port which checks the platform type and
revision information in the SYS_ID register strictly supported
only the first revision of the Base and Foundation FVPs.
The current check also does not reflect the fact that the
board revision field is 'local' to the board type (HBI field).
Support for a new Foundation model is required now, and the
checking code is relaxed to allow execution (with a diagnostic)
on unrecognised revisions of the Base and Foundation FVP.
Andrew Thoelke [Mon, 9 Jun 2014 11:44:21 +0000 (12:44 +0100)]
Remove current CPU mpidr from PSCI common code
Many of the interfaces internal to PSCI pass the current CPU
MPIDR_EL1 value from function to function. This is not required,
and with inline access to the system registers is less efficient
than requiring the code to read that register whenever required.
This patch remove the mpidr parameter from the affected interfaces
and reduces code in FVP BL3-1 size by 160 bytes.
Vikram Kanigiri [Mon, 2 Jun 2014 13:59:00 +0000 (14:59 +0100)]
Simplify entry point information generation code on FVP
This patch reworks FVP specific code responsible for determining
the entry point information for BL3-2 and BL3-3 stages when BL3-1
is configured as the reset handler.
fvp: Fix register name in 'plat_print_gic_regs' macro
The 'plat_print_gic_regs' macro was accessing the GICC_CTLR register
using the GICD_CTLR offset. This still generates the right code in
the end because GICD_CTLR == GICC_CTLR but this patch fixes it for
the logic of the code.
Andrew Thoelke [Mon, 9 Jun 2014 11:54:15 +0000 (12:54 +0100)]
Remove calling CPU mpidr from bakery lock API
The bakery lock code currently expects the calling code to pass
the MPIDR_EL1 of the current CPU.
This is not always done correctly. Also the change to provide
inline access to system registers makes it more efficient for the
bakery lock code to obtain the MPIDR_EL1 directly.
This change removes the mpidr parameter from the bakery lock
interface, and results in a code reduction of 160 bytes for the
ARM FVP port.
Andrew Thoelke [Thu, 19 Jun 2014 23:36:14 +0000 (00:36 +0100)]
Correctly dimension the PSCI aff_map_node array
The array of affinity nodes is currently allocated for 32 entries
with the PSCI_NUM_AFFS value defined in psci.h. This is not enough
for large systems, and will substantially over allocate the array
for small systems.
This patch introduces an optional platform definition
PLATFORM_NUM_AFFS to platform_def.h. If defined this value is
used for PSCI_NUM_AFFS, otherwise a value of two times the number
of CPU cores is used.
The FVP port defines PLATFORM_NUM_AFFS to be 10 which saves
nearly 1.5KB of memory.
Andrew Thoelke [Thu, 19 Jun 2014 23:38:03 +0000 (00:38 +0100)]
Eliminate psci_suspend_context array
psci_suspend_context is an array of cache-line aligned structures
containing the single power_state integer per cpu. This array is
the only structure indexed by the aff_map_node.data integer.
This patch saves 2KB of BL3-1 memory by placing the CPU
power_state value directly in the aff_map_node structure. As a
result, this value is now never cached and the cache clean when
writing the value is no longer required.
Andrew Thoelke [Wed, 4 Jun 2014 20:10:52 +0000 (21:10 +0100)]
Initialise CPU contexts from entry_point_info
Consolidate all BL3-1 CPU context initialization for cold boot, PSCI
and SPDs into two functions:
* The first uses entry_point_info to initialize the relevant
cpu_context for first entry into a lower exception level on a CPU
* The second populates the EL1 and EL2 system registers as needed
from the cpu_context to ensure correct entry into the lower EL
This patch alters the way that BL3-1 determines which exception level
is used when first entering EL1 or EL2 during cold boot - this is now
fully determined by the SPSR value in the entry_point_info for BL3-3,
as set up by the platform code in BL2 (or otherwise provided to BL3-1).
In the situation that EL1 (or svc mode) is selected for a processor
that supports EL2, the context management code will now configure all
essential EL2 register state to ensure correct execution of EL1. This
allows the platform code to run non-secure EL1 payloads directly
without requiring a small EL2 stub or OS loader.
Add the '-Wmissing-include-dirs' flag to the CFLAGS and ASFLAGS
to make the build fail if the compiler or the assembler is given
a nonexistant directory in the list of directories to be searched
for header files.
Also remove 'include/bl1' and 'include/bl2' directories from the
search path for header files as they don't exist anymore.
Soby Mathew [Fri, 6 Jun 2014 09:18:52 +0000 (10:18 +0100)]
Remove re-initialisation of system timers after warm boot for FVP
This patch removes the reinitialisation of memory mapped system timer
registers after a warm boot for the FVP. The system timers in FVP are
in the 'Always ON' power domain which meant the reinitialisation was
redundant and it could have conflicted with the setup the normal
world has done.
The programming of CNTACR(x) and CNTNSAR, the system timer registers,
are removed from the warm boot path with this patch.
Andrew Thoelke [Mon, 2 Jun 2014 11:38:12 +0000 (12:38 +0100)]
Remove early_exceptions from BL3-1
The crash reporting support and early initialisation of the
cpu_data allow the runtime_exception vectors to be used from
the start in BL3-1, removing the need for the additional
early_exception vectors and 2KB of code from BL3-1.
Andrew Thoelke [Mon, 2 Jun 2014 10:40:35 +0000 (11:40 +0100)]
Per-cpu data cache restructuring
This patch prepares the per-cpu pointer cache for wider use by:
* renaming the structure to cpu_data and placing in new header
* providing accessors for this CPU, or other CPUs
* splitting the initialization of the TPIDR pointer from the
initialization of the cpu_data content
* moving the crash stack initialization to a crash stack function
* setting the TPIDR pointer very early during boot
Andrew Thoelke [Tue, 3 Jun 2014 10:50:53 +0000 (11:50 +0100)]
Make the BL3-1 crash reporting optional
This patch makes the console crash dump of processor register
state optional based on the CRASH_REPORTING make variable.
This defaults to only being enabled for DEBUG builds. This can
be overridden by setting a different value in the platform
makefile or on the make command line.
Andrew Thoelke [Wed, 14 May 2014 16:09:32 +0000 (17:09 +0100)]
Provide cm_get/set_context() for current CPU
All callers of cm_get_context() pass the calling CPU MPIDR to the
function. Providing a specialised version for the current
CPU results in a reduction in code size and better readability.
The current function has been renamed to cm_get_context_by_mpidr()
and the existing name is now used for the current-CPU version.
The same treatment has been done to cm_set_context(), although
only both forms are used at present in the PSCI and TSPD code.
Andrew Thoelke [Tue, 10 Jun 2014 15:37:37 +0000 (16:37 +0100)]
PSCI SMC handler improvements
The SMC handler for PSCI was not correctly handling calls from
secure states, or from AArch32.
This patch completes the handler implementation to correctly
detect secure callers and to clear the top bits in parameters from
AArch32 callers.
The patch also reorganises the switch statement to separate SMC64 and
SMC32 function IDs which allows the compiler to generate much smaller
code for the function.
Make the entry point argument optional in load_image()
There are cases where the entry point information is useless to the
caller, e.g. when an image just needs to be loaded in memory but won't
ever be executed.
This patch allows load_image() function to take a NULL pointer as the
entry point argument. In this case, it won't be populated.
Before attempting to write a character, the PL011 driver polls
the PL011_UARTFR_TXFF bit to know whether the UART FIFO is full.
However, the comparison with 1 was incorrect because
PL011_UARTFR_TXFF is not at bit 0. This patch fixes it.
Include 'platform_def.h' header file in 'crash_reporting.S'
'crash_reporting.S' needs to include 'platform_def.h' to get the
definition of PLATFORM_CORE_COUNT.
Note: On FVP it was compiling because 'platform_def.h' gets included
through 'plat/fvp/include/plat_macros.S' but we don't want to rely on
that for other platforms.
Achin Gupta [Mon, 2 Jun 2014 21:27:36 +0000 (22:27 +0100)]
Document design of the Interrupt Mangement Framework
This patch adds documentation that describes the design of the Interrupt
management framework in the ARM Trusted Firmware. The porting-guide.md has also
been updated to describe the interface that should be implemented by each
platform to support this framework.
Lin Ma [Mon, 2 Jun 2014 18:45:36 +0000 (11:45 -0700)]
Enable mapping higher physical address
Current ATF uses a direct physical-to-virtual mapping, that is, a physical
address is mapped to the same address in the virtual space. For example,
physical address 0x8000_0000 is mapped to 0x8000_0000 virtual. This
approach works fine for FVP as all its physical addresses fall into 0 to
4GB range. But for other platform where all I/O addresses are 48-bit long,
If we follow the same direct mapping, we would need virtual address range
from 0 to 0x8fff_ffff_ffff, which is about 144TB. This requires a
significant amount of memory for MMU tables and it is not necessary to use
that much virtual space in ATF.
The patch is to enable mapping a physical address range to an arbitrary
virtual address range (instead of flat mapping)
Changed "base" to "base_va" and added "base_pa" in mmap_region_t and
modified functions such as mmap_add_region and init_xlation_table etc.
Fixes ARM-software/tf-issues#158
Dan Handley [Thu, 29 May 2014 11:30:24 +0000 (12:30 +0100)]
Fix porting guide references to platform.h
Following recent refactoring changes to platform.h, this commit updates
porting-guide.md to correctly refer to platform.h and platform_def.h where
appropriate.
Soby Mathew [Tue, 27 May 2014 15:54:31 +0000 (16:54 +0100)]
Pass 'cookie' parameter to interrupt handler in BL3-1
The interrupt handling routine in BL3-1 expects a cookie as its last
parameter which was not being passed when invoking the interrupt
handler in BL3-1. This patch fixes that by passing a dummy cookie
parameter in the x3 register.
Andrew Thoelke [Wed, 28 May 2014 21:22:55 +0000 (22:22 +0100)]
Allow platform parameter X1 to be passed to BL3-1
bl2_main() was overwriting any platform set X1 parameter for BL3-1
with the value zero.
This patch ensure that any platform set value is correctly passed
to BL3-1. The FVP port adds a check to verify this parameter is
being passed correctly.
Dan Handley [Tue, 27 May 2014 15:17:21 +0000 (16:17 +0100)]
Further renames of platform porting functions
Rename the ic_* platform porting functions to plat_ic_* to be
consistent with the other functions in platform.h. Also rename
bl31_get_next_image_info() to bl31_plat_get_next_image_ep_info()
and remove the duplicate declaration in bl31.h.
Dan Handley [Fri, 16 May 2014 13:08:45 +0000 (14:08 +0100)]
Add enable mmu platform porting interfaces
Previously, the enable_mmu_elX() functions were implicitly part of
the platform porting layer since they were included by generic
code. These functions have been placed behind 2 new platform
functions, bl31_plat_enable_mmu() and bl32_plat_enable_mmu().
These are weakly defined so that they can be optionally overridden
by platform ports.
Also, the enable_mmu_elX() functions have been moved to
lib/aarch64/xlat_tables.c for optional re-use by platform ports.
These functions are tightly coupled with the translation table
initialization code.
Dan Handley [Thu, 15 May 2014 13:53:30 +0000 (14:53 +0100)]
Rename FVP specific files and functions
FVP specific files and functions containing the word "plat" have been
renamed to use the word "fvp" to distinguish them from the common
platform functionality and porting functions.
Dan Handley [Thu, 15 May 2014 13:11:36 +0000 (14:11 +0100)]
Move BL porting functions into platform.h
Some platform porting functions were in BL specific header files.
These have been moved to platform.h so that all porting functions
are in the same place. The functions are now grouped by BL.
Obsolete BL headers files have been removed.
Also, the weak declaration of the init_bl2_mem_layout() function
has been moved out the header file and into the source file
(bl_common.c) using the more succinct #pragma syntax. This
mitigates the risk of 2 weak definitions being created and the
wrong one being picked up by the compiler.
projects / arm-tf.git / log