Skip to content

GitLab

Switch branch/tag
  1. 28 Sep, 2018 1 commit
  3. 22 Aug, 2018 2 commits
  5. 10 Aug, 2018 1 commit
    • Antonio Nino Diaz's avatar
      xlat v2: Support the EL2 translation regime · 1a92a0e0
      Antonio Nino Diaz authored 
      

The translation library is useful elsewhere. Even though this repository
doesn't exercise the EL2 support of the library, it is better to have it
here as well to make it easier to maintain.

enable_mmu_secure() and enable_mmu_direct() have been deprecated. The
functions are still present, but they are behind ERROR_DEPRECATED and
they call the new functions enable_mmu_svc_mon() and
enable_mmu_direct_svc_mon().

Change-Id: I13ad10cd048d9cc2d55e0fff9a5133671b67dcba
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      1a92a0e0
  7. 07 Aug, 2018 1 commit
    • Antonio Nino Diaz's avatar
      xlat v2: Flush xlat tables after being modified · 3e318e40
      Antonio Nino Diaz authored 
      During cold boot, the initial translation tables are created with data
caches disabled, so all modifications go to memory directly. After the
MMU is enabled and data cache is enabled, any modification to the tables
goes to data cache, and eventually may get flushed to memory.

If CPU0 modifies the tables while CPU1 is off, CPU0 will have the
modified tables in its data cache. When CPU1 is powered on, the MMU is
enabled, then it enables coherency, and then it enables the data cache.
Until this is done, CPU1 isn't in coherency, and the translation tables
it sees can be outdated if CPU0 still has some modified entries in its
data cache.

This can be a problem in some cases. For example, the warm boot code
uses only the tables mapped during cold boot, which don't normally
change. However, if they are modified (and a RO page is made RW, or a XN
page is made executable) the CPU will see the old attributes and crash
when it tries to access it.

This doesn't happen in systems with HW_ASSISTED_COHERENCY or
WARMBOOT_ENABLE_DCACHE_EARLY. In these systems, the data cache is
enabled at the same time as the MMU. As soon as this happens, the CPU is
in coherency.

There was an attempt of a fix in psci_helpers.S, but it didn't solve the
problem. That code has been deleted. The code was introduced in commit
<26441030

> ("Invalidate TLB entries during warm boot").

Now, during a map or unmap operation, the memory associated to each
modified table is flushed. Traversing a table will also flush it's
memory, as there is no way to tell in the current implementation if the
table that has been traversed has also been modified.

Change-Id: I4b520bca27502f1018878061bc5fb82af740bb92
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      3e318e40
  9. 06 Aug, 2018 1 commit
  11. 02 Aug, 2018 1 commit
  13. 30 Jul, 2018 1 commit
  15. 15 Jul, 2018 2 commits
  17. 13 Jul, 2018 4 commits
    • Antonio Nino Diaz's avatar
      xlat v2: Turn MMU parameters into 64-bit values · 6563c0be
      Antonio Nino Diaz authored 
      

Most registers are 64-bit wide, even in AArch32 mode:

- MAIR_ELx is equivalent to MAIR0 and MAIR1.
- TTBR is 64 bit in both AArch64 and AArch32.

The only difference is the TCR register, which is 32 bit in AArch32 and
in EL3 in AArch64. For consistency with the rest of ELs in AArch64, it
makes sense to also have it as a 64-bit value.

Change-Id: I2274d66a28876702e7085df5f8aad0e7ec139da9
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      6563c0be
    • Antonio Nino Diaz's avatar
      xlat v2: Remove IMAGE_EL define · aa1d5f60
      Antonio Nino Diaz authored 
      

The Exception Level is now detected at runtime. This means that it is not
needed to hardcode the EL used by each image.

This doesn't result in a substantial increase of the image size because
the initialization functions that aren't used are garbage-collected by
the linker.

In AArch32 the current EL has been changed from EL3 to EL1 because the
the AArch32 PL1&0 translation regime behaves more like the AArch64 EL1&0
translation regime than the EL3 one.

Change-Id: I941404299ebe7666ca17619207c923b49a55cb73
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      aa1d5f60
    • Antonio Nino Diaz's avatar
      xlat v2: Remove unused tlbi helper · 8d164bc6
      Antonio Nino Diaz authored 
      

xlat_arch_tlbi_va_regime() isn't used, so it has been renamed to
xlat_arch_tlbi_va() and the previous implementation has been removed.

Change-Id: Ic118bed3fb68234748d86b2e9e95b25650289276
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      8d164bc6
    • Antonio Nino Diaz's avatar
      xlat v2: Remove architectural headers · 468e2382
      Antonio Nino Diaz authored 
      

They only contained one function that is easily integrated in the private
library header and the existing architectural C files.

This also helps making the library more portable, as the Makefile of the
library now doesn't use the variable INCLUDES, which is specific to this
codebase and doesn't respect the namespace of the library.

Change-Id: I22228e6a97e9b4f346f5cd8947609263e8df71d8
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      468e2382
  19. 03 Jul, 2018 3 commits
    • Antonio Nino Diaz's avatar
      xlat v2: Make get/set attrs functions less verbose · 6a086061
      Antonio Nino Diaz authored 
      

It is useful to have LOG_LEVEL_VERBOSE because it prints the memory map
of each image, but that also means that the change_mem_attributes and
get_mem_attributes functions have verbose prints, and generate a too
long text output that hides other useful information.

As they were mostly there for debug purposes, this patch removes them.

Change-Id: I2986537377d1f78be2b79cc8a6cf230c380bdb55
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      6a086061
    • Antonio Nino Diaz's avatar
      xlat v2: Clean debug xlat tables descriptor print · f9d58d17
      Antonio Nino Diaz authored 
      

The previous debug output for EL1&0 translation regimes was too verbose,
which makes it hard to read and hides the intent behind the parameters
assigned to each region. This patch simplifies this output and makes the
outputs for EL3 and EL1&0 mostly the same. The difference is that in
EL1&0 it is specified whether the region is exclusively accessible from
EL1 (PRIV) or both EL0 and EL1 (USER).

For example:

    MEM-RW(PRIV)-NOACCESS(USER)-XN(PRIV)-XN(USER)-S
    MEM-RO(PRIV)-NOACCESS(USER)-EXEC(PRIV)-EXEC(USER)-S

After the change, it becomes this:

    MEM-RW-XN-PRIV-S
    MEM-RO-EXEC-PRIV-S

Change-Id: I15f4b99058429d42107fbf89e15f4838a9b559a5
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      f9d58d17
    • Antonio Nino Diaz's avatar
      xlat v2: Split code into separate files · fd2299e6
      Antonio Nino Diaz authored 
      

Instead of having one big file with all the code, it's better to have
a few smaller files that are more manageable:

- xlat_tables_core.c: Code related to the core functionality of the
  library (map and unmap regions, initialize xlat context).
- xlat_tables_context.c: Instantiation of the active image context
  as well as APIs to manipulate it.
- xlat_tables_utils.c: Helper code that isn't part of the core
  functionality (change attributes, debug print messages).

Change-Id: I3ea956fc1afd7473c0bb5e7c6aab3b2e5d88c711
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      fd2299e6
  21. 27 Jun, 2018 1 commit
    • Jeenu Viswambharan's avatar
      xlat v2: Split MMU setup and enable · 0cc7aa89
      Jeenu Viswambharan authored 
      

At present, the function provided by the translation library to enable
MMU constructs appropriate values for translation library, and programs
them to the right registers. The construction of initial values,
however, is only required once as both the primary and secondaries
program the same values.

Additionally, the MMU-enabling function is written in C, which means
there's an active stack at the time of enabling MMU. On some systems,
like Arm DynamIQ, having active stack while enabling MMU during warm
boot might lead to coherency problems.

This patch addresses both the above problems by:

  - Splitting the MMU-enabling function into two: one that sets up
    values to be programmed into the registers, and another one that
    takes the pre-computed values and writes to the appropriate
    registers. With this, the primary effectively calls both functions
    to have the MMU enabled, but secondaries only need to call the
    latter.

  - Rewriting the function that enables MMU in assembly so that it
    doesn't use stack.

This patch fixes a bunch of MISRA issues on the way.

Change-Id: I0faca97263a970ffe765f0e731a1417e43fbfc45
Signed-off-by: default avatarJeenu Viswambharan <jeenu.viswambharan@arm.com>
      0cc7aa89
  23. 22 Jun, 2018 1 commit
    • Antonio Nino Diaz's avatar
      xlat: Remove mmap_attr_t enum type · 3a1b7b10
      Antonio Nino Diaz authored 
      

The values defined in this type are used in logical operations, which
goes against MISRA Rule 10.1: "Operands shall not be of an inappropriate
essential type".

Now, `unsigned int` is used instead. This also allows us to move the
dynamic mapping bit from 30 to 31. It was an undefined behaviour in the
past because an enum is signed by default, and bit 31 corresponds to the
sign bit. It is undefined behaviour to modify the sign bit. Now, bit 31
is free to use as it was originally meant to be.

mmap_attr_t is now defined as an `unsigned int` for backwards
compatibility.

Change-Id: I6b31218c14b9c7fdabebe432de7fae6e90a97f34
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      3a1b7b10
  25. 21 Jun, 2018 1 commit
  27. 14 Jun, 2018 1 commit
    • Yann Gautier's avatar
      xlat_v2: add a check on mm_cursor->size to avoid infinite loop · 75df6269
      Yann Gautier authored 
      

The issue can occur if end_va is equal to the max architecture address,
and when mm_cursor point to the last entry of mmap_region_t table: {0}.
The first line of the while will then be true, e.g. on AARCH32, we have:
mm_cursor->base_va (=0) + mm_cursor->size (=0) - 1 == end_va (=0xFFFFFFFF)
And the mm_cursor->size = 0 will be lesser than mm->size

A check on mm_cursor->size != 0 should be done as in the previous while,
to avoid such kind of infinite loop.

fixes arm-software/tf-issues#594
Signed-off-by: default avatarYann Gautier <yann.gautier@st.com>
      75df6269
  29. 13 Jun, 2018 1 commit
    • Antonio Nino Diaz's avatar
      xlat v2: Introduce xlat granule size helpers · a0b9bb79
      Antonio Nino Diaz authored 
      

The function xlat_arch_is_granule_size_supported() can be used to check
if a specific granule size is supported. In Armv8, AArch32 only supports
4 KiB pages. AArch64 supports 4 KiB, 16 KiB or 64 KiB depending on the
implementation, which is detected at runtime.

The function xlat_arch_get_max_supported_granule_size() returns the max
granule size supported by the implementation.

Even though right now they are only used by SPM, they may be useful in
other places in the future. This patch moves the code currently in SPM
to the xlat tables lib so that it can be reused.

Change-Id: If54624a5ecf20b9b9b7f38861b56383a03bbc8a4
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      a0b9bb79
  31. 11 Jun, 2018 1 commit
    • John Tsichritzis's avatar
      MISRA rule 21.15 fix · bdcd33a8
      John Tsichritzis authored 
      

    Rule 21.15: The pointer arguments to the Standard Library functions
    memcpy, memmove and memcmp shall be pointers to qualified or unqualified
    versions of compatible types.

    Basically that means that both pointer arguments must be of the same
    type. However, even if the pointers passed as arguments to the above
    functions are of the same type, Coverity still thinks it's a violation
    if we do pointer arithmetics directly at the function call. Thus the
    pointer arithmetic operations were moved outside of the function
    argument.

    First detected on the following configuration
            make PLAT=fvp LOG_LEVEL=50

    Change-Id: I8b912ec1bfa6f2d60857cb1bd453981fd7001b94
Signed-off-by: default avatarJohn Tsichritzis <john.tsichritzis@arm.com>
      bdcd33a8
  33. 09 May, 2018 1 commit
  35. 02 May, 2018 1 commit
    • Antonio Nino Diaz's avatar
      xlat: Have all values of PARange for 8.x architectures · d3c4487c
      Antonio Nino Diaz authored 
      

In AArch64, the field ID_AA64MMFR0_EL1.PARange has a different set of
allowed values depending on the architecture version.

Previously, we only compiled the Trusted Firmware with the values that
were allowed by the architecture. However, given that this field is
read-only, it is easier to compile the code with all values regardless
of the target architecture.

Change-Id: I57597ed103dd0189b1fb738a9ec5497391c10dd1
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      d3c4487c
  37. 26 Apr, 2018 1 commit
    • Antonio Nino Diaz's avatar
      xlat: Set AP[1] to 1 when it is RES1 · 01c0a38e
      Antonio Nino Diaz authored 
      

According to the ARMv8 ARM issue C.a:

    AP[1] is valid only for stage 1 of a translation regime that can
    support two VA ranges. It is RES 1 when stage 1 translations can
    support only one VA range.

This means that, even though this bit is ignored, it should be set to 1
in the EL3 and EL2 translation regimes.

For translation regimes consisting on EL0 and a higher regime this bit
selects between control at EL0 or at the higher Exception level. The
regimes that support two VA ranges are EL1&0 and EL2&0 (the later one
is only available since ARMv8.1).

This fix has to be applied to both versions of the translation tables
library.

Change-Id: If19aaf588551bac7aeb6e9a686cf0c2068e7c181
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      01c0a38e
  39. 17 Apr, 2018 1 commit
  41. 09 Apr, 2018 1 commit
    • Varun Wadekar's avatar
      lib: xlat_tables_v2: reduce time required to add a mmap region · 0ed32232
      Varun Wadekar authored 
      

The last entry in the mapping table is not necessarily the same as the
end of the table. This patch loops through the table to find the last
entry marker, on every new mmap addition. The memove operation then
has to only move the memory between current entry and the last entry.
For platforms that arrange their MMIO map properly, this opearation
turns out to be a NOP.

The previous implementation added significant overhead per mmap
addition as the memmove operation always moved the difference between
the current mmap entry and the end of the table.

Tested on Tegra platforms and this new approach improves the memory
mapping time by ~75%, thus significantly reducing boot time on some
platforms.

Change-Id: Ie3478fa5942379282ef58bee2085da799137e2ca
Signed-off-by: default avatarVarun Wadekar <vwadekar@nvidia.com>
      0ed32232
  43. 28 Feb, 2018 2 commits
  45. 27 Feb, 2018 1 commit
  47. 10 Jan, 2018 1 commit
    • Antonio Nino Diaz's avatar
      xlat v2: Correctly unmap regions on map error · 96abc22b
      Antonio Nino Diaz authored 
      

`mm_cursor` doesn't have the needed data because the `memmove()` that
is called right before it overwrites that information. In order to get
the information of the region that was being mapped, `mm` has to be used
instead (like it is done to fill the fields of `unmap_mm`).

If the incorrect information is read, this check isn't reliable and
`xlat_tables_unmap_region` may be requested to unmap memory that isn't
mapped at all, triggering assertions.

Change-Id: I602d4ac83095d4e5dac9deb34aa5d00d00e6c289
Signed-off-by: default avatarAntonio Nino Diaz <antonio.ninodiaz@arm.com>
      96abc22b
  49. 17 Nov, 2017 1 commit
  51. 08 Nov, 2017 2 commits
  53. 17 Oct, 2017 2 commits
  55. 05 Oct, 2017 3 commits