The addr32 prefix is required by binutils, because even when
given an explicit address which is greater than 64KiB, it will
throw a warning about truncation, and stupidly emit the opcode
with a 16-bit addressing mode and the wrong address.
However, in the case of LLVM, this doesn't happen, and is happy
to just use 32-bit addressing whenever it may require it. This
means that LLVM never really needs an explicit addr32 prefix to
use 32-bit addressing in 16-bit mode.
Change-Id: Ia160d3f7da6653ea24c8229dc26f265e5f15aabb
Also-by: Damien Zammit <damien@zamaudio.com>
Signed-off-by: Damien Zammit <damien@zamaudio.com>
Reviewed-on: https://review.coreboot.org/21219
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
LLVM AS doesn't support as much GNU junk extensions, data16/32
is almost never needed in truth if we just use the correct op
suffix. So do that here, fixes clang/llvm builds with the
integrated-as toggled on.
Change-Id: I6095d03d0289b418a49a10f135de5eb0e117cae0
Also-by: Damien Zammit <damien@zamaudio.com>
Signed-off-by: Damien Zammit <damien@zamaudio.com>
Reviewed-on: https://review.coreboot.org/21218
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Fail at build-time if one of the following happens:
Platform includes SMI handler setup function smm_init()
in the build when configuration has HAVE_SMI_HANDLER=n.
Platform does not implement smm_init_completion() when
HAVE_SMI_HANDLER=y.
Change-Id: I7d61c155d2b7c2d71987980db4c25d520452dabf
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/21097
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-by: Arthur Heymans <arthur@aheymans.xyz>
Split `i2c.h` into three pieces to ease reuse of the generic defi-
nitions. No code is changed.
* `i2c.h` - keeps the generic definitions
* `i2c_simple.h` - holds the current, limited to one controller driver
per board, devicetree independent I2C interface
* `i2c_bus.h` - will become the devicetree compatible interface for
native I2C (e.g. non-SMBus) controllers
Change-Id: I382d45c70f9314588663e1284f264f877469c74d
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: https://review.coreboot.org/20845
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Due to low-memory corruptions S3 support has now been
(at least temporarily) removed from AGESA platfroms.
Should we bring it back one day, CAR teardown on S3 path
will happen with an empty stack so ugly backup/recovery
of the stack will no longer be used.
If S3 feature is brought back, resume path code for FCH
will also see partial rewrite and agesawrapper.c file
will not be part of that.
Change-Id: Ib38c04d0e74f600e0b719940d5e2530f4c726cfd
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/20899
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Martin Roth <martinroth@google.com>
Our current struct for I2C segments `i2c_seg` was close to being compa-
tible to the Linux version `i2c_msg`, close to being compatible to SMBus
and close to being readable (e.g. what was `chip` supposed to mean?) but
turned out to be hard to fix.
Instead of extending it in a backwards compatible way (and not touching
current controller drivers), replace it with a Linux source compatible
`struct i2c_msg` and patch all the drivers and users with Coccinelle.
The new `struct i2c_msg` should ease porting drivers from Linux and help
to write SMBus compatible controller drivers.
Beside integer type changes, the field `read` is replaced with a generic
field `flags` and `chip` is renamed to `slave`.
Patched with Coccinelle using the clumsy spatch below and some manual
changes:
* Nested struct initializers and one field access skipped by Coccinelle.
* Removed assumption in the code that I2C_M_RD is 1.
* In `i2c.h`, changed all occurences of `chip` to `slave`.
@@ @@
-struct i2c_seg
+struct i2c_msg
@@ identifier msg; expression e; @@
(
struct i2c_msg msg = {
- .read = 0,
+ .flags = 0,
};
|
struct i2c_msg msg = {
- .read = 1,
+ .flags = I2C_M_RD,
};
|
struct i2c_msg msg = {
- .chip = e,
+ .slave = e,
};
)
@@ struct i2c_msg msg; statement S1, S2; @@
(
-if (msg.read)
+if (msg.flags & I2C_M_RD)
S1 else S2
|
-if (msg.read)
+if (msg.flags & I2C_M_RD)
S1
)
@@ struct i2c_msg *msg; statement S1, S2; @@
(
-if (msg->read)
+if (msg->flags & I2C_M_RD)
S1 else S2
|
-if (msg->read)
+if (msg->flags & I2C_M_RD)
S1
)
@@ struct i2c_msg msg; expression e; @@
(
-msg.read = 0;
+msg.flags = 0;
|
-msg.read = 1;
+msg.flags = I2C_M_RD;
|
-msg.read = e;
+msg.flags = e ? I2C_M_RD : 0;
|
-!!(msg.read)
+(msg.flags & I2C_M_RD)
|
-(msg.read)
+(msg.flags & I2C_M_RD)
)
@@ struct i2c_msg *msg; expression e; @@
(
-msg->read = 0;
+msg->flags = 0;
|
-msg->read = 1;
+msg->flags = I2C_M_RD;
|
-msg->read = e;
+msg->flags = e ? I2C_M_RD : 0;
|
-!!(msg->read)
+(msg->flags & I2C_M_RD)
|
-(msg->read)
+(msg->flags & I2C_M_RD)
)
@@ struct i2c_msg msg; @@
-msg.chip
+msg.slave
@@ struct i2c_msg *msg; expression e; @@
-msg[e].chip
+msg[e].slave
@ slave disable ptr_to_array @ struct i2c_msg *msg; @@
-msg->chip
+msg->slave
Change-Id: Ifd7cabf0a18ffd7a1def25d1d7059b713d0b7ea9
Signed-off-by: Nico Huber <nico.huber@secunet.com>
Reviewed-on: https://review.coreboot.org/20542
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-by: Werner Zeh <werner.zeh@siemens.com>
Reviewed-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
We define BINARYPI_LEGACY_WRAPPER a method of calling AGESA
via functions in agesawrapper.c file. The approach implemented
there makes it very inconvenient to do board-specific
customisation or present common platform-specific features.
Seems like it also causes assertion errors on AGESA side.
The flag is applied here to all boards and then individually
removed one at a time, as things get tested.
New method is not to call AGESA internal functions directly,
but via the dispatcher. AGESA call parameters are routed to
hooks in both platform and board -directories, to allow for
easy capture or modification as needed.
For each AGESA dispatcher call made, eventlog entries are
replayed to the console log. Also relocations of AGESA heap
that took place are recorded.
New method is expected to be compatible with binaryPI.
Change-Id: I2900249e60f21a13dc231f4a8a04835e090109d5
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/19272
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Boards without AGESA_LEGACY_WRAPPER gain EARLY_CBMEM_INIT.
This does not apply to family12 and family14 just yet, as
they do invalidate without write-back on CAR teardown.
Change-Id: I008356efa2bc3df0ed1f0720e225ecc7e9995127
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/19329
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
In AGESA specification AmdInitEnv() is to be called once
host memory allocator has started. In coreboot context this
could mean either availability of CBMEM or malloc heap.
As for AmdS3LateRestore(), there is no requirement to have
it run as part of the romstage either.
Change-Id: Icc8d97b82df89e2480e601d5c2e094de0365b0a5
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/18888
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
We define AGESA_LEGACY_WRAPPER a method of calling AGESA
via functions in agesawrapper.c file. The approach implemented
there makes it very inconvenient to do board-specific
customisation or present common platform-specific features.
Seems like it also causes assertion errors on AGESA side.
The flag is applied here to all boards and then individually
removed one at a time, as things get tested.
New method is not to call AGESA internal functions directly,
but via the dispatcher. AGESA call parameters are routed to
hooks in both platform and board -directories, to allow for
easy capture or modification as needed.
For each AGESA dispatcher call made, eventlog entries are
replayed to the console log. Also relocations of AGESA heap
that took place are recorded.
New method is expected to be compatible with binaryPI.
Change-Id: Iac3d7f8b0354e9f02c2625576f36fe06b05eb4ce
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/18628
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Fix remaining space prohibited between function name and open
parenthesis, line over 80 characters, unnecessary braces for single
statement blocks, space required before open brace errors and warnings
in subdirectories of src/cpu/amd/pi
Change-Id: I177ffe98a3674bd700a39eb8073db34adf9499b4
Signed-off-by: Evelyn Huang <evhuang@google.com>
Reviewed-on: https://review.coreboot.org/20098
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
At a process _start, the stack is expected to be aligned to a
16-byte boundary. Upon entry to any function the stack frame
must have the end of any arguments also aligned. In other words
the value of %esp+4 or %rsp+8 is always a multiple of 16 (1).
Align the stack down inside _secondary_start and preserve proper
alignment for the call to secondary_cpu_init.
Although 4-byte alignment is the minimum requirement for i386,
some AMD platforms use SSE instructions which expect 16-byte.
1) http://wiki.osdev.org/System_V_ABI
See "Initial Stack and Register State" and "The Stack Frame"
in the supplements.
BUG=chrome-os-partner:62841664
Change-Id: I72b7a474013e5caf67aedfabeb8d8d2553499b73
Signed-off-by: Marshall Dawson <marshalldawson3rd@gmail.com>
Reviewed-on: https://review.coreboot.org/20537
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Calling disable_cache_as_ram() with valuables in stack is not
a stable solution, as per documentation AMD_DISABLE_STACK
should destroy stack in cache.
While we still preserve cache contents (there is wbinvd deep
inside AMD_DISABLE_STACK macro), we now actually do a stack
switch and much more closely meet the specification of CAR
teardown sequence in AGESA specifications.
We now somewhat incorrectly include files from agesa/ tree,
but the whole agesawrapper.c file removal will address the
issue of overall directory layout.
Change-Id: I2bac098099c1caffea181356c63924f4b5a93b54
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/18525
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Patrick Georgi <pgeorgi@google.com>
Reviewed-by: Aaron Durbin <adurbin@chromium.org>
Change register preservations and fix comments about register
usage accordingly. Do this to avoid use of %mm0-2 registers inside
macros defined in gcccar.inc, as future implementation of
C_BOOTBLOCK_ENVIRONMENT will use them as well.
Adjust caller side accordingly.
Change-Id: Ic76fcc31ae714baf5259d17c41b62a3610aa947b
Signed-off-by: Marshall Dawson <marshalldawson3rd@gmail.com>
Signed-off-by: Kyösti Mälkki <kyosti.malkki@gmail.com>
Reviewed-on: https://review.coreboot.org/20579
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Marc Jones <marc@marcjonesconsulting.com>
Fix undefined behavior found by clang's -Wshift-sign-overflow. Left
shifting an int where the right operand is >= the width of the type
is undefined. Add UL suffix since it's safe for unsigned types.
Change-Id: Ieacf83d052bf4abfad639ef8e592bd8de17d16e6
Signed-off-by: Ryan Salsamendi <rsalsamendi@hotmail.com>
Reviewed-on: https://review.coreboot.org/20467
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Martin Roth <martinroth@google.com>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Philipp Deppenwiese <zaolin.daisuki@gmail.com>
Fixes report found by undefined behavior sanitizer. Dereferencing a
pointer that's not aligned to the size of access is undefined behavior.
Remove unnecessary memset().
Change-Id: I1362a3eb8c97f5c7e848d75f8d1a219968a7ef9e
Signed-off-by: Ryan Salsamendi <rsalsamendi@hotmail.com>
Reviewed-on: https://review.coreboot.org/20452
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Nico Huber <nico.h@gmx.de>
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
This requires to also unify the calling convention for
AGESA functions from
AGESA_STATUS (*agesa_func)(UINT32 Func, UINT32 Data, VOID *ConfigPtr)
to
AGESA_STATUS (*agesa_func)(UINT32 Func, UINTN Data, VOID *ConfigPtr)
On systems running 32bit x86 code this will not make a difference as
UINTN is uintptr_t which is 32bit on these machines.
Change-Id: I095ec2273c18a9fda11712654e290ebc41b27bd9
Signed-off-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
Reviewed-on: https://review.coreboot.org/20380
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Reviewed-by: Marshall Dawson <marshalldawson3rd@gmail.com>
If SSE instructions are enabled in the build assume the SMM
modules are compiled with SSE instructions. As such enable
the SSE instructions in SMM mode by setting up the cr4 register.
In addition, provide a place to save and restore the SSE state
in both the relocation handler and permanent handler.
Change-Id: Ifa16876b57544919fde88fba5b8f18e4ca286841
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: https://review.coreboot.org/20244
Reviewed-by: Furquan Shaikh <furquan@google.com>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
