| Age | Commit message (Collapse) | Author |
|---|
|
The cooperative multitasking support allows the boot state machine
to be ran cooperatively with other threads of work. The main thread
still continues to run the boot state machine
(src/lib/hardwaremain.c). All callbacks from the state machine are
still ran synchronously from within the main thread's context.
Without any other code added the only change to the boot sequence
when cooperative multitasking is enabled is the queueing of an idlle
thread. The idle thread is responsible for ensuring progress is made
by calling timer callbacks.
The main thread can yield to any other threads in the system. That
means that anyone that spins up a thread must ensure no shared
resources are used from 2 or more execution contexts. The support
is originally intentioned to allow for long work itesm with busy
loops to occur in parallel during a boot.
Note that the intention on when to yield a thread will be on
calls to udelay().
Change-Id: Ia4d67a38665b12ce2643474843a93babd8a40c77
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3206
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
it has been unused since 9 years or so, hence drop it.
Change-Id: I0706feb7b3f2ada8ecb92176a94f6a8df53eaaa1
Signed-off-by: Stefan Reinauer <reinauer@google.com>
Reviewed-on: http://review.coreboot.org/3212
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Tested-by: build bot (Jenkins)
|
|
Internally there were states that had an attribute to
indicate that the timers needed to be drained. Now that
there is a way to block state transitions rely on this
ability instead of draining timers. The timers will
drain themselves when a state is blocked.
Change-Id: I59be9a71b2fd5a17310854d2f91c2a8957aafc28
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3205
Tested-by: build bot (Jenkins)
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
|
|
In order to properly sequence the boot state machine it's
important that outside code can block the transition from
one state to the next. When timers are not involved there's
no reason for any of the existing code to block a state
transition. However, if there is a timer callback that needs to
complete by a certain point in the boot sequence it is necessary
to place a block for the given state.
To that end, 4 new functions are added to provide the API for
blocking a state.
1. boot_state_block(boot_state_t state, boot_state_sequence_t seq);
2. boot_state_unblock(boot_state_t state, boot_state_sequence_t seq);
3. boot_state_current_block(void);
4. boot_state_current_unblock(void);
Change-Id: Ieb37050ff652fd85a6b1e0e2f81a1a2807bab8e0
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3204
Tested-by: build bot (Jenkins)
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
|
|
When TIMER_QUEUE is configured on call the timer callbacks on
entry into a state but before its entry callbacks. In addition
provide a barrier to the following states so that timers are drained
before proceeding. This allows for blocking state traversal for key
components of boot.
BS_OS_RESUME
BS_WRITE_TABLES
BS_PAYLOAD_LOAD
BS_PAYLOAD_BOOT
Future functionality consists of evaluating the timer callbacks within
the device tree. One example is dev_initialize() as that seems state
seems to take 90% of the boot time. The timer callbacks could then be
ran in a more granular manner.
Change-Id: Idb549ea17c5ec38eb57b4f6f366a1c2183f4a6dd
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3159
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
When the MONOTONIC_TIMER is available track the entry, run, and exit
times for each state. It should be noted that the times for states that
vector to OS or a payload do not have their times reported.
Change-Id: I6af23fe011609e0b1e019f35ee40f1fbebd59c9d
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3156
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
The notion of loading a payload in the current boot state
machine isn't actually loading the payload. The reason is
that cbfs is just walked to find the payload. The actual
loading and booting were occuring in selfboot(). Change this
balance so that loading occurs in one function and actual
booting happens in another. This allows for ample opportunity
to delay work until just before booting.
Change-Id: Ic91ed6050fc5d8bb90c8c33a44eea3b1ec84e32d
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3139
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
The cbmem_post_handling() function was implemented by 2
chipsets in order to save memory configuration in flash. Convert
both of these chipsets to use the boot state machine callbacks
to perform the saving of the memory configuration.
Change-Id: I697e5c946281b85a71d8533437802d7913135af3
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3137
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
There were previously 2 functions, init_cbmem_pre_device() and
init_cbmem_post_device(), where the 2 cbmem implementations
implemented one or the other. These 2 functions are no longer
needed to be called in the boot flow once the boot state callbacks
are utilized.
Change-Id: Ida71f1187bdcc640ae600705ddb3517e1410a80d
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3136
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
Utilize the static boot state callback scheduling to initialize
and tear down the coverage infrastructure at the appropriate points.
The coverage initialization is performed at BS_PRE_DEVICE which is the
earliest point a callback can be called. The tear down occurs at the
2 exit points of ramstage: OS resume and payload boot.
Change-Id: Ie5ee51268e1f473f98fa517710a266e38dc01b6d
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3135
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
It's helpful to provide a distinct state that affirmatively
describes that OS resume will occur. The previous code included
the check and the actual resuming in one function. Because of this
grouping one had to annotate the innards of the ACPI resume
path to perform specific actions before OS resume. By providing
a distinct state in the boot state machine the necessary actions
can be scheduled accordingly without modifying the ACPI code.
Change-Id: I8b00aacaf820cbfbb21cb851c422a143371878bd
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3134
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
Many of the boot state callbacks can be scheduled at compile time.
Therefore, provide a way for a compilation unit to inform the
boot state machine when its callbacks should be called. Each C
module can export the callbacks and their scheduling requirements
without changing the shared boot flow code.
Change-Id: Ibc4cea4bd5ad45b2149c2d4aa91cbea652ed93ed
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3133
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
The boot flow currently has a fixed ordering. The ordering
is dictated by the device tree and on x86 the PCI device ordering
for when actions are performed. Many of the new machines and
configurations have dependencies that do not follow the device
ordering.
In order to be more flexible the concept of a boot state machine
is introduced. At the boundaries (entry and exit) of each state there
is opportunity to run callbacks. This ability allows one to schedule
actions to be performed without adding board-specific code to
the shared boot flow.
Change-Id: I757f406c97445f6d9b69c003bb9610b16b132aa6
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/3132
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
Dynamic cbmem is now a requirement for relocatable ramstage.
This patch replaces the reserve_* fields in the romstage_handoff
structure by using the dynamic cbmem library.
The haswell code is not moved over in this commit, but it should be
safe because there is a hard requirement for DYNAMIC_CBMEM when using
a reloctable ramstage.
Change-Id: I59ab4552c3ae8c2c3982df458cd81a4a9b712cc2
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/2849
Tested-by: build bot (Jenkins)
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
|
|
This patch adds a parallel implementation of cbmem that supports
dynamic sizing. The original implementation relied on reserving
a fixed-size block of memory for adding cbmem entries. In order to
allow for more flexibility for adding cbmem allocations the dynamic
cbmem infrastructure was developed as an alternative to the fixed block
approach. Also, the amount of memory to reserve for cbmem allocations
does not need to be known prior to the first allocation.
The dynamic cbmem code implements the same API as the existing cbmem
code except for cbmem_init() and cbmem_reinit(). The add and find
routines behave the same way. The dynamic cbmem infrastructure
uses a top down allocator that starts allocating from a board/chipset
defined function cbmem_top(). A root pointer lives just below
cbmem_top(). In turn that pointer points to the root block which
contains the entries for all the large alloctations. The corresponding
block for each large allocation falls just below the previous entry.
It should be noted that this implementation rounds all allocations
up to a 4096 byte granularity. Though a packing allocator could
be written for small allocations it was deemed OK to just fragment
the memory as there shouldn't be that many small allocations. The
result is less code with a tradeoff of some wasted memory.
+----------------------+ <- cbmem_top()
| +----| root pointer |
| | +----------------------+
| | | |--------+
| +--->| root block |-----+ |
| +----------------------+ | |
| | | | |
| | | | |
| | alloc N |<----+ |
| +----------------------+ |
| | | |
| | | |
\|/ | alloc N + 1 |<-------+
v +----------------------+
In addition to preserving the previous cbmem API, the dynamic
cbmem API allows for removing blocks from cbmem. This allows for
the boot process to allocate memory that can be discarded after
it's been used for performing more complex boot tasks in romstage.
In order to plumb this support in there were some issues to work
around regarding writing of coreboot tables. There were a few
assumptions to how cbmem was layed out which dictated some ifdef
guarding and other runtime checks so as not to incorrectly
tag the e820 and coreboot memory tables.
The example shown below is using dynamic cbmem infrastructure.
The reserved memory for cbmem is less than 512KiB.
coreboot memory table:
0. 0000000000000000-0000000000000fff: CONFIGURATION TABLES
1. 0000000000001000-000000000002ffff: RAM
2. 0000000000030000-000000000003ffff: RESERVED
3. 0000000000040000-000000000009ffff: RAM
4. 00000000000a0000-00000000000fffff: RESERVED
5. 0000000000100000-0000000000efffff: RAM
6. 0000000000f00000-0000000000ffffff: RESERVED
7. 0000000001000000-000000007bf80fff: RAM
8. 000000007bf81000-000000007bffffff: CONFIGURATION TABLES
9. 000000007c000000-000000007e9fffff: RESERVED
10. 00000000f0000000-00000000f3ffffff: RESERVED
11. 00000000fed10000-00000000fed19fff: RESERVED
12. 00000000fed84000-00000000fed84fff: RESERVED
13. 0000000100000000-00000001005fffff: RAM
Wrote coreboot table at: 7bf81000, 0x39c bytes, checksum f5bf
coreboot table: 948 bytes.
CBMEM ROOT 0. 7bfff000 00001000
MRC DATA 1. 7bffe000 00001000
ROMSTAGE 2. 7bffd000 00001000
TIME STAMP 3. 7bffc000 00001000
ROMSTG STCK 4. 7bff7000 00005000
CONSOLE 5. 7bfe7000 00010000
VBOOT 6. 7bfe6000 00001000
RAMSTAGE 7. 7bf98000 0004e000
GDT 8. 7bf97000 00001000
ACPI 9. 7bf8b000 0000c000
ACPI GNVS 10. 7bf8a000 00001000
SMBIOS 11. 7bf89000 00001000
COREBOOT 12. 7bf81000 00008000
And the corresponding e820 entries:
BIOS-e820: [mem 0x0000000000000000-0x0000000000000fff] type 16
BIOS-e820: [mem 0x0000000000001000-0x000000000002ffff] usable
BIOS-e820: [mem 0x0000000000030000-0x000000000003ffff] reserved
BIOS-e820: [mem 0x0000000000040000-0x000000000009ffff] usable
BIOS-e820: [mem 0x00000000000a0000-0x00000000000fffff] reserved
BIOS-e820: [mem 0x0000000000100000-0x0000000000efffff] usable
BIOS-e820: [mem 0x0000000000f00000-0x0000000000ffffff] reserved
BIOS-e820: [mem 0x0000000001000000-0x000000007bf80fff] usable
BIOS-e820: [mem 0x000000007bf81000-0x000000007bffffff] type 16
BIOS-e820: [mem 0x000000007c000000-0x000000007e9fffff] reserved
BIOS-e820: [mem 0x00000000f0000000-0x00000000f3ffffff] reserved
BIOS-e820: [mem 0x00000000fed10000-0x00000000fed19fff] reserved
BIOS-e820: [mem 0x00000000fed84000-0x00000000fed84fff] reserved
BIOS-e820: [mem 0x0000000100000000-0x00000001005fffff] usable
Change-Id: Ie3bca52211800a8652a77ca684140cfc9b3b9a6b
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/2848
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
When CONFIG_EARLY_CBMEM_INIT is selected romstage is supposed to have
initialized cbmem. Therefore provide a weak function for the chipset
to implement named cbmem_get_table_location(). When
CONFIG_EARLY_CBMEM_INIT is selected cbmem_get_table_location() will be
called to get the cbmem location and size. After that cbmem_initialize()
is called.
Change-Id: Idc45a95f9d4b1d83eb3c6d4977f7a8c80c1ffe76
Signed-off-by: Aaron Durbin <adurbin@chromium.org>
Reviewed-on: http://review.coreboot.org/2797
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
It's been on for all boards per default since several years now
and the old code path probably doesn't even work anymore. Let's
just have one consistent way of doing things.
Change-Id: I58da7fe9b89a648d9a7165d37e0e35c88c06ac7e
Signed-off-by: Stefan Reinauer <reinauer@google.com>
Reviewed-on: http://review.coreboot.org/2547
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
Summary:
Isolate CBFS underlying I/O to board/arch-specific implementations as
"media stream", to allow loading and booting romstage on non-x86.
CBFS functions now all take a new "media source" parameter; use
CBFS_DEFAULT_MEDIA if you simply want to load from main firmware.
API Changes:
cbfs_find => cbfs_get_file.
cbfs_find_file => cbfs_get_file_content.
cbfs_get_file => cbfs_get_file_content with correct type.
CBFS used to work only on memory-mapped ROM (all x86). For platforms like ARM,
the ROM may come from USB, UART, or SPI -- any serial devices and not available
for memory mapping.
To support these devices (and allowing CBFS to read from multiple source
at the same time), CBFS operations are now virtual-ized into "cbfs_media". To
simplify porting existing code, every media source must support both "reading
into pre-allocated memory (read)" and "read and return an allocated buffer
(map)". For devices without native memory-mapped ROM, "cbfs_simple_buffer*"
provides simple memory mapping simulation.
Every CBFS function now takes a cbfs_media* as parameter. CBFS_DEFAULT_MEDIA
is defined for CBFS functions to automatically initialize a per-board default
media (CBFS will internally calls init_default_cbfs_media). Also revised CBFS
function names relying on memory mapped backend (ex, "cbfs_find" => actually
loads files). Now we only have two getters:
struct cbfs_file *entry = cbfs_get_file(media, name);
void *data = cbfs_get_file_content(CBFS_DEFAULT_MEDIA, name, type);
Test results:
- Verified to work on x86/qemu.
- Compiles on ARM, and follow up commit will provide working SPI driver.
Change-Id: Iac911ded25a6f2feffbf3101a81364625bb07746
Signed-off-by: Hung-Te Lin <hungte@chromium.org>
Reviewed-on: http://review.coreboot.org/2182
Tested-by: build bot (Jenkins)
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
In order to provide some insight on what code is executed during
coreboot's run time and how well our test scenarios work, this
adds code coverage support to coreboot's ram stage. This should
be easily adaptable for payloads, and maybe even romstage.
See http://gcc.gnu.org/onlinedocs/gcc/Gcov.html for
more information.
To instrument coreboot, select CONFIG_COVERAGE ("Code coverage
support") in Kconfig, and recompile coreboot. coreboot will then
store its code coverage information into CBMEM, if possible.
Then, run "cbmem -CV" as root on the target system running the
instrumented coreboot binary. This will create a whole bunch of
.gcda files that contain coverage information. Tar them up, copy
them to your build system machine, and untar them. Then you can
use your favorite coverage utility (gcov, lcov, ...) to visualize
code coverage.
For a sneak peak of what will expect you, please take a look
at http://www.coreboot.org/~stepan/coreboot-coverage/
Change-Id: Ib287d8309878a1f5c4be770c38b1bc0bb3aa6ec7
Signed-off-by: Stefan Reinauer <reinauer@google.com>
Reviewed-on: http://review.coreboot.org/2052
Tested-by: build bot (Jenkins)
Reviewed-by: David Hendricks <dhendrix@chromium.org>
Reviewed-by: Martin Roth <martin@se-eng.com>
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
|
|
It only has two files, move them to src/lib
Change-Id: I17943db4c455aa3a934db1cf56e56e89c009679f
Signed-off-by: Stefan Reinauer <reinauer@google.com>
Reviewed-on: http://review.coreboot.org/1959
Reviewed-by: Ronald G. Minnich <rminnich@gmail.com>
Tested-by: build bot (Jenkins)
|
