This implementation simply mimics that of BL31.
I did not implement the ENABLE_PIE support for BL2_IN_XIP_MEM=1 case.
It would make the linker script a bit uglier.
Change-Id: If3215abd99f2758dfb232e44b50320d04eba808b
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
When I tried ENABLE_PIE for my PLAT=uniphier platform, BL31 crashed
at its entry. When it is built with ENABLE_PIE=1, some sections are
inserted before the executable code.
$ make PLAT=uniphier CROSS_COMPILE=aarch64-linux-gnu- ENABLE_PIE=1 bl31
$ aarch64-linux-gnu-objdump -h build/uniphier/release/bl31/bl31.elf | head -n 13
build/uniphier/release/bl31/bl31.elf: file format elf64-littleaarch64
Sections:
Idx Name Size VMA LMA File off Algn
0 .dynsym 000002a0 0000000081000000 0000000081000000 00010000 2**3
CONTENTS, ALLOC, LOAD, READONLY, DATA
1 .dynstr 000002a0 00000000810002a0 00000000810002a0 000102a0 2**0
CONTENTS, ALLOC, LOAD, READONLY, DATA
2 .hash 00000124 0000000081000540 0000000081000540 00010540 2**3
CONTENTS, ALLOC, LOAD, READONLY, DATA
3 ro 0000699c 0000000081000664 0000000081000664 00010664 2**11
CONTENTS, ALLOC, LOAD, CODE
The previous stage loader generally jumps over to the base address of
BL31, where no valid instruction exists.
I checked the linker script of Linux (arch/arm64/kernel/vmlinux.lds.S)
and U-Boot (arch/arm/cpu/armv8/u-boot.lds), both of which support
relocation. They simply discard those sections.
Do similar in TF-A too.
Change-Id: I6c33e9143856765d4ffa24f3924b0ab51a17cde9
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
docs/getting_started/build-options.rst clearly says ENABLE_PIE is
currently only supported in BL31, but in fact, it has a stronger
limitation:
Defining ENABLE_PIE may corrupt BL1 and BL2. So, ENABLE_PIE is
supported only for platforms where BL31 is the only image built
in the TF-A tree.
Currently, ENABLE_PIE is enabled by two platforms,
plat/arm/common/arm_common.mk and ti/k3/common/plat_common.mk,
both of which enable ENABLE_PIE together with RESET_TO_BL31.
For platforms with the full boot sequence, ENABLE_PIE may break earlier
BL stages. For example, if I build PLAT=qemu with ENABLE_PIE=1, it
fails in BL1.
When ENABLE_PIE is enabled, PIE options are added to TF_CFLAGS and
TF_LDFLAGS, so all BL images are affected. It is problematic because
currently only the BL31 linker script handles it. Even if BL1/BL2
works, the image size would increase needlessly, at least.
Pass the PIE options only to BL images that support it.
Change-Id: I550e95148aa3c63571c8ad2081082c554a848f57
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
make_helpers/build_macros.mk supports per-BL CFLAGS. For example,
you can pass compiler flags only to BL31 by using BL31_CFLAGS.
This commit adds per-BL LDFLAGS support, which is useful as well.
My main motivation of this addition is to use it for ENABLE_PIE.
When ENABLE_PIE is enabled, some linker flags are added to TF_LDFLAGS,
which affects all the TF images. It will make more sense to pass the
relevant options only to BL images that support it.
Change-Id: I203acaab0091db5ae0ea6e66460ee7dc8d9c4d75
Signed-off-by: Masahiro Yamada <yamada.masahiro@socionext.com>
* changes:
Tegra194: mce: remove unused NVG functions
Tegra194: support for NVG interface v6.6
Tegra194: smmu: add PCIE0R1 mc reg to system suspend save list
Tegra194: enable driver for general purpose DMA engine
Tegra194: access XUSB_PADCTL registers on Si/FPGA platforms
Tegra194: organize the memory/mmio map to make it linear
Tegra194: memctrl: enable mc sid OVERRIDE for PCIE0R1
Tegra194: support for boot params wider than 32-bits
Tegra194: memctrl: set reorder depth limit for PCIE blocks
Tegra194: memctrl: program MC_TXN_OVERRIDE reg for PTCR, MPCORE and MIU
Tegra194: memctrl: set CGID_TAG_ADR instead of CGID_TAG_DEFAULT
Tegra194: memctrl: update mss reprogramming as HW PROD settings
Tegra194: memctrl: Disable PVARDC coalescer
Tegra194: memctrl: force seswr/rd transactions as passsthru & coherent
Tegra194: Request CG7 from last core in cluster
Tegra194: toggle SE clock during context save/restore
Tegra: bpmp: fix header file paths
* changes:
intel: Add function to check fpga readiness
intel: Add bridge control for FPGA reconfig
intel: FPGA config_isdone() status query
intel: System Manager refactoring
intel: Refactor reset manager driver
intel: Enable bridge access in Intel platform
intel: Modify non secure access function
PCIE0R1 security and override registers need to be preserved across
system suspend. Adding them to system suspend save register list.
Due to addition of above registers, increasing context save memory
by 2 bytes.
Change-Id: I1b3a56aee31f3c11e3edc2fb0a6da146eec1a30d
Signed-off-by: Pritesh Raithatha <praithatha@nvidia.com>
This patch enables the GPCDMA for all Tegra194 platforms to help
accelerate all the memory copy operations.
Change-Id: I8cbec99be6ebe4da74221245668b321ba9693479
Signed-off-by: Varun Wadekar <vwadekar@nvidia.com>
Many simulation/emulation platforms do not support this hardware block
leading to SErrors during register accesses.
This patch conditionally accesses the registers from this block only
on actual Si and FPGA platforms.
Change-Id: Ic22817a8c9f81978ba88c5362bfd734a0040d35d
Signed-off-by: Varun Wadekar <vwadekar@nvidia.com>
This patch organizes the platform memory/mmio map, so that the base
addresses for the apertures line up in ascending order. This makes
it easier for the xlat_tables_v2 library to create mappings for each
mmap_add_region call.
Change-Id: Ie1938ba043820625c9fea904009a3d2ccd29f7b3
Signed-off-by: Varun Wadekar <vwadekar@nvidia.com>
PCIE0R1 does not program stream IDs, so allow the stream ID to be
overriden by the MC.
Change-Id: I4dbd71e1ce24b11e646de421ef68c762818c2667
Signed-off-by: Pritesh Raithatha <praithatha@nvidia.com>
The previous bootloader is not able to pass boot params wider than
32-bits due to an oversight in the scratch register being used. A
new secure scratch register #75 has been assigned to pass the higher
bits.
This patch adds support to parse the higher bits from scratch #75
and use them in calculating the base address for the location of
the boot params.
Scratch #75 format
====================
31:16 - bl31_plat_params high address
15:0 - bl31_params high address
Change-Id: Id53c45f70a9cb370c776ed7c82ad3f2258576a80
Signed-off-by: Steven Kao <skao@nvidia.com>
HW bug in third party PCIE IP - PCIE datapath hangs when there are
more than 28 outstanding requests on data backbone for x1 controller.
Suggested SW WAR is to limit reorder_depth_limit to 16 for
PCIE 1W/2AW/3W clients.
Change-Id: Id5448251c35d2a93f66a8b5835ae4044f5cef067
Signed-off-by: Puneet Saxena <puneets@nvidia.com>
-PTCR is ISO client so setting it to FORCE_NON_COHERENT.
-MPCORER, MPCOREW and MIU0R/W to MIU7R/W clients itself will provide
ordering so no need to override from mc.
-MIU0R/W to MIU7R/W clients registers are not implemented in tegrasim
so skipping it for simulation.
-All the clients need to set CGID_TAG_ADR to maintain request ordering
within a 4K boundary.
Change-Id: Iaa3189a1f3e40fb4cef28be36bc4baeb5ac8f9ca
Signed-off-by: Pritesh Raithatha <praithatha@nvidia.com>
- All SoC clients should use CGID_TAG_ADR to improve perf
- Remove tegra194_txn_override_cfgs array that is not getting used.
Change-Id: I9130ef5ae8659ed5f9d843ab9a0ecf58b5ce9c74
Signed-off-by: Pritesh Raithatha <praithatha@nvidia.com>
Memory clients are divided in to ISO/NonISO/Order/Unordered/Low
BW/High BW. Based on the client types, HW team recommends, different
memory ordering settings, IO coherency settings and SMMU register settings
for optimized performance of the MC clients.
For example ordered ISO clients should be set as strongly ordered and
should bypass SCF and directly access MC hence set as
FORCE_NON_COHERENT. Like this there are multiple recommendations
for all of the MC clients.
This change sets all these MC registers as per HW spec file.
Change-Id: I8a8a0887cd86bf6fe8ac7835df6c888855738cd9
Signed-off-by: Puneet Saxena <puneets@nvidia.com>
Signed-off-by: Varun Wadekar <vwadekar@nvidia.com>
Due to a hardware bug PVA may perform memory transactions which
cause coalescer faults. This change works around the issue by
disabling coalescer for PVA0RDC and PVA1RDC.
Change-Id: I27d1f6e7bc819fb303dae98079d9277fa346a1d3
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Force memory transactions from seswr and sesrd as coherent_snoop from
no-override. This is necessary as niso clients should use coherent
path.
Presently its set as FORCE_COHERENT_SNOOP. Once SE+TZ is enabled
with SMMU, this needs to be replaced by FORCE_COHERENT.
Change-Id: I8b50722de743b9028129b4715769ef93deab73b5
Signed-off-by: Puneet Saxena <puneets@nvidia.com>
- SC7 requires all the cluster groups to be in CG7 state, else
is_sc7_allowed will get denied
- As a WAR while requesting CC6, request CG7 as well
- CG7 request will not be honored if it is not last core in Cluster
group
- This is just to satisfy MCE for now as CG7 is going to be defeatured
Change-Id: Ibf2f8a365a2e46bd427abd563da772b6b618350f
Signed-off-by: Vignesh Radhakrishnan <vigneshr@nvidia.com>
Masahiro Yamada
Soby Mathew
Deepika Bhavnani
Manish Pandey
Alexei Fedorov
Varun Wadekar
Pritesh Raithatha
Steven Kao
Puneet Saxena
Arto Merilainen
Vignesh Radhakrishnan