core: juno: workaround cortex-a57 errata 808870

Workaround errata 808870:
Unconditional VLDM instructions might cause an alignment fault even
though the address is aligned

Products Affected: Cortex

core: juno: workaround cortex-a57 errata 808870

Workaround errata 808870:
Unconditional VLDM instructions might cause an alignment fault even
though the address is aligned

Products Affected: Cortex-A57 MPCore.
Present in: r0p0

The workaround is to avoid generating the problematic instructions in
AArch32 TA.

Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

abort.c: arm32: assume VFP instr if undef

If an undefined instruction exception is raised from user mode assume it
is a VFP instruction unless VFP already is enabled.

This avoids reading user mode

abort.c: arm32: assume VFP instr if undef

If an undefined instruction exception is raised from user mode assume it
is a VFP instruction unless VFP already is enabled.

This avoids reading user mode memory while handling an abort which until
now has kept an undiscovered race where a page could become inaccessible
before the abort handler had the chance to read the instruction from the
page.

There is room for false positives. Those will be discovered the next
time the instruction is executed and still causes an undefined
instruction exception. Only this time VFP is already enabled so we know
it's not a VFP instruction. Enabling VFP in vain like this is harmless.

Reviewed-by: Jerome Forissier <jerome.forissier@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: prevent allocation when exception index tables are empty

This change prevent the core from allocating memory, mapping and
other resources to map exception index tables that are all empty.

Sig

core: prevent allocation when exception index tables are empty

This change prevent the core from allocating memory, mapping and
other resources to map exception index tables that are all empty.

Signed-off-by: Etienne Carriere <etienne.carriere@st.com>
Reviewed-by: Jerome Forissier <jerome.forissier@linaro.org>

show more ...

core: correct time conversion in delay support

The previous code may overflow in 32bit architectures. This change
fixes the issue by forcing 64bit computation during frequency
to counter conversion.

core: correct time conversion in delay support

The previous code may overflow in 32bit architectures. This change
fixes the issue by forcing 64bit computation during frequency
to counter conversion.

Signed-off-by: Etienne Carriere <etienne.carriere@st.com>
Reviewed-by: Jerome Forissier <jerome.forissier@linaro.org>

show more ...

thread: fix stack space to not be part of tee.bin

Commit 935364080364 ("thread: move stacks to separate sections") makes
stack space to be part of tee.bin which leads to approx. 130K increase
in siz

thread: fix stack space to not be part of tee.bin

Commit 935364080364 ("thread: move stacks to separate sections") makes
stack space to be part of tee.bin which leads to approx. 130K increase
in size of tee.bin for platform with 2 threads.
So this patch fixes the stack space to be in NOLOAD section only.

Fixes: 935364080364 ("thread: move stacks to separate sections")
Signed-off-by: Sumit Garg <sumit.garg@linaro.org>
Reviewed-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: arm32: fix saving vfp state

Prior to this patch TF-A in AArch32 didn't save the normal world VFP
data registers and thus always restored then as zeroes and causing
problems in normal world.

W

core: arm32: fix saving vfp state

Prior to this patch TF-A in AArch32 didn't save the normal world VFP
data registers and thus always restored then as zeroes and causing
problems in normal world.

With this patch if running with TF-A in AArch32 save the VFP state using
the same logic as in AArch64. Since TF-A saves and restores CPACR_EL1 we
cannot tell if normal world currently is using VFP or not so we have to
assume that it is and always save the VFP data registers if they are
about to be changed.

Reviewed-by: Jerome Forissier <jerome.forissier@linaro.org>
Tested-by: Jens Wiklander <jens.wiklander@linaro.org> (Hikey AArch32)
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: user_ta: set freed EXIDX pointer to NULL

Clear reference of user TA exception index table once it
is freed, otherwise it is freed a second time. This
issue occurs when set_exidx() fails for so

core: user_ta: set freed EXIDX pointer to NULL

Clear reference of user TA exception index table once it
is freed, otherwise it is freed a second time. This
issue occurs when set_exidx() fails for some reasons as
an out of memory issue. This change prevents core from
panicking when it occurs.

Signed-off-by: Etienne Carriere <etienne.carriere@linaro.org>
Reviewed-by: Jerome Forissier <jerome.forissier@linaro.org>

show more ...

core: default guard reg_shm mobjs

Add a special guard for registered shared memory MOBJs to make sure that
it's not possible to release MOBJs which aren't created by an explicit
registration from no

core: default guard reg_shm mobjs

Add a special guard for registered shared memory MOBJs to make sure that
it's not possible to release MOBJs which aren't created by an explicit
registration from normal world.

Acked-by: Etienne Carriere <etienne.carriere@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: mobj_reg_shm_free() using mobj_reg_shm_put()

Let reg_shm MOBJs be completely reference counted meaning that
mobj_free() on a reg_shm MOBJ only decrease the reference counter and
only free it i

core: mobj_reg_shm_free() using mobj_reg_shm_put()

Let reg_shm MOBJs be completely reference counted meaning that
mobj_free() on a reg_shm MOBJ only decrease the reference counter and
only free it if it reaches 0.

Reviewed-by: Etienne Carriere <etienne.carriere@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: remove mobj_reg_shm_{,un}map()

Removes mobj_reg_shm_map() and mobj_reg_shm_unmap(), they are replaced
by mobj_reg_shm_inc_map() and mobj_reg_shm_dec_map().

Reviewed-by: Etienne Carriere <etie

core: remove mobj_reg_shm_{,un}map()

Removes mobj_reg_shm_map() and mobj_reg_shm_unmap(), they are replaced
by mobj_reg_shm_inc_map() and mobj_reg_shm_dec_map().

Reviewed-by: Etienne Carriere <etienne.carriere@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: pta: use mobj_reg_shm_{inc,dec}_map()

Use mobj_reg_shm_inc_map() and mobj_reg_shm_dec_map() instead of
mobj_reg_shm_map() and mobj_reg_shm_unmap().

Acked-by: Etienne Carriere <etienne.carrier

core: pta: use mobj_reg_shm_{inc,dec}_map()

Use mobj_reg_shm_inc_map() and mobj_reg_shm_dec_map() instead of
mobj_reg_shm_map() and mobj_reg_shm_unmap().

Acked-by: Etienne Carriere <etienne.carriere@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: add mobj_reg_shm_{inc,dec}_map()

mobj_reg_shm_inc_map() used when a reg_shm mobj needs to be mapped and
mobj_reg_shm_dec_map() is called when the mapping isn't needed any
longer.

Reviewed-by:

core: add mobj_reg_shm_{inc,dec}_map()

mobj_reg_shm_inc_map() used when a reg_shm mobj needs to be mapped and
mobj_reg_shm_dec_map() is called when the mapping isn't needed any
longer.

Reviewed-by: Etienne Carriere <etienne.carriere@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: remove reg_shm_{put,free}_by_cookie()

Removes the two functions mobj_reg_shm_put_by_cookie() and
mobj_reg_shm_free_by_cookie(). mobj_reg_shm_put() and mobj_free() should
be used instead.

Revi

core: remove reg_shm_{put,free}_by_cookie()

Removes the two functions mobj_reg_shm_put_by_cookie() and
mobj_reg_shm_free_by_cookie(). mobj_reg_shm_put() and mobj_free() should
be used instead.

Reviewed-by: Etienne Carriere <etienne.carriere@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: entry_std: free/put by mobj in call cleanup

Use the mobj pointer instead of a shm_ref when cleaning up parameters
in cleanup_shm_refs() at the end of a call.

Reviewed-by: Etienne Carriere <e

core: entry_std: free/put by mobj in call cleanup

Use the mobj pointer instead of a shm_ref when cleaning up parameters
in cleanup_shm_refs() at the end of a call.

Reviewed-by: Etienne Carriere <etienne.carriere@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: add mobj_reg_shm_put()

Adds mobj_reg_shm_put() for reference counting without cookie.

Reviewed-by: Etienne Carriere <etienne.carriere@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander

core: add mobj_reg_shm_put()

Adds mobj_reg_shm_put() for reference counting without cookie.

Reviewed-by: Etienne Carriere <etienne.carriere@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: correct memory layout trace

Buffers that end at end of the available address range which
may happen on 32bit machine fail have an end address that of
computed as 0. This change uses a 64bit ad

core: correct memory layout trace

Buffers that end at end of the available address range which
may happen on 32bit machine fail have an end address that of
computed as 0. This change uses a 64bit address computation
to prevent the displayed end address being 0.

Signed-off-by: Etienne Carriere <etienne.carriere@linaro.org>
Reviewed-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: correct overflows in range overlap functions

Buffers that end at end of the available address range which
may happen on 32bit machine fail have an end address that of
computed as 0. This chang

core: correct overflows in range overlap functions

Buffers that end at end of the available address range which
may happen on 32bit machine fail have an end address that of
computed as 0. This change uses the computation already used
in _core_is_buffer_inside() to ensure functions
_core_is_buffer_outside() and _core_is_buffer_intersect()
return a reliable result.

Signed-off-by: Etienne Carriere <etienne.carriere@linaro.org>
Reviewed-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

abort.c: always save VFP state with thread context

Saving VFP state requires a thread context, if none is available print
abort info and panic().

Reviewed-by: Volodymyr Babchuk <volodymyr_babchuk@e

abort.c: always save VFP state with thread context

Saving VFP state requires a thread context, if none is available print
abort info and panic().

Reviewed-by: Volodymyr Babchuk <volodymyr_babchuk@epam.com>
Fixes: cfa34ec63699 ("abort.c: manipulate with VFP state only if thread is active")
Tested-by: Jens Wiklander <jens.wiklander@linaro.org> (Hikey AArch64 pager)
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: trace out size of the pager physical page pool

Add a nice info trace about the size of the pager physical
page pool.

Signed-off-by: Etienne Carriere <etienne.carriere@linaro.org>
Reviewed-by:

core: trace out size of the pager physical page pool

Add a nice info trace about the size of the pager physical
page pool.

Signed-off-by: Etienne Carriere <etienne.carriere@linaro.org>
Reviewed-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: correct unpaged constraint on GIC driver

Release of secondary boot cores on 32bit machine use SMC that
issue a SGI on secondary core. Since the interrupt is raised
from the monitor mode, the r

core: correct unpaged constraint on GIC driver

Release of secondary boot cores on 32bit machine use SMC that
issue a SGI on secondary core. Since the interrupt is raised
from the monitor mode, the related GIC driver resources must be
tagged as unpaged.

This change costs around 300 bytes of unpaged resident memory.

Signed-off-by: Etienne Carriere <etienne.carriere@linaro.org>
Reviewed-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: correct dt_find_compatible_driver() if no driver found

By convention the compatible array for driver probing ends with
a NULL compatible string ID reference.

This change ensures such NULL ref

core: correct dt_find_compatible_driver() if no driver found

By convention the compatible array for driver probing ends with
a NULL compatible string ID reference.

This change ensures such NULL reference is properly handled as
the end of the compatible array references.

Signed-off-by: Etienne Carriere <etienne.carriere@linaro.org>
Reviewed-by: Jerome Forissier <jerome.forissier@linaro.org>

show more ...

Allow mixed declaration and code

Removes the -Wdeclaration-after-statement compiler flag to allow mixed
declaration and code

Acked-by: Jerome Forissier <jerome.forissier@linaro.org>
Signed-off-by:

Allow mixed declaration and code

Removes the -Wdeclaration-after-statement compiler flag to allow mixed
declaration and code

Acked-by: Jerome Forissier <jerome.forissier@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

optee_msg.h: remove OPTEE_MSG_ATTR_FRAGMENT

OPTEE_MSG_ATTR_FRAGMENT isn't defined. Remove it from comments and replace
with OPTEE_MSG_ATTR_NONCONTIG where applicable.

Acked-by: Jerome Forissier <je

optee_msg.h: remove OPTEE_MSG_ATTR_FRAGMENT

OPTEE_MSG_ATTR_FRAGMENT isn't defined. Remove it from comments and replace
with OPTEE_MSG_ATTR_NONCONTIG where applicable.

Acked-by: Jerome Forissier <jerome.forissier@linaro.org>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

core: reference count registered shared memory

* Adds a reference counter to registered shared memory to make sure that
it's not unregistered while in use.
* Updates entry_std to use the reference

core: reference count registered shared memory

* Adds a reference counter to registered shared memory to make sure that
it's not unregistered while in use.
* Updates entry_std to use the reference counting functions

Reviewed-by: Volodymyr Babchuk <volodymyr_babchuk@epam.com>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...

entry_std: use READ_ONCE() to read from shared memory

Use the READ_ONCE() macro everywhere shared memory is read to make sure
that unexpected values can not be used.

Reviewed-by: Volodymyr Babchuk

entry_std: use READ_ONCE() to read from shared memory

Use the READ_ONCE() macro everywhere shared memory is read to make sure
that unexpected values can not be used.

Reviewed-by: Volodymyr Babchuk <volodymyr_babchuk@epam.com>
Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org>

show more ...