coreboot-kgpe-d16/src/soc/rockchip/rk3288/bootblock.c

26 lines
652 B
C
Raw Normal View History

/* SPDX-License-Identifier: GPL-2.0-only */
rk3288: Add early SRAM mapping Solving the DACR bug will mean that XN bits suddenly become enforced on non-LPAE systems, and we will no longer be able to execute out of a region mapped DCACHE_OFF. When we enable the MMU in romstage we are still executing out of SRAM, so we would instantly kill ourselves. Solve this issue by enabling the MMU earlier (in the bootblock) and mapping the SRAM regions as DCACHE_WRITETHROUGH. They should really be DCACHE_WRITEBACK, but it looks like there might be hardware limitations in the Cortex-A12 cache architecture that prevent us from doing so. Write-through mappings are equivalent to normal non-cacheable on the A12 anyway, and by using this attribute we don't need to introduce a new DCACHE_OFF_BUT_WITHOUT_XN_BIT type in our API. (Also, using normal non-cacheable might still have a slight speed advantage over strongly ordered since it should fetch whole cache lines at once if the processor finds enough accesses it can combine.) CQ-DEPEND=CL:223783 BUG=chrome-os-partner:32118 TEST=None (depends on follow-up CL) Change-Id: I1e5127421f82177ca11af892b1539538b379625e Signed-off-by: Patrick Georgi <pgeorgi@chromium.org> Original-Commit-Id: e7b079f4b6a69449f3c7cc18ef0e1704f2006847 Original-Change-Id: I53e827d95acc2db909f1251de78d65e295eceaa7 Original-Signed-off-by: Julius Werner <jwerner@chromium.org> Original-Reviewed-on: https://chromium-review.googlesource.com/223782 Original-Reviewed-by: Aaron Durbin <adurbin@chromium.org> Reviewed-on: http://review.coreboot.org/9342 Tested-by: build bot (Jenkins) Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
2014-10-16 03:50:45 +02:00
#include <arch/cache.h>
#include <bootblock_common.h>
rk3288: Change all SoC headers to <soc/headername.h> system This patch is the start of a series to change all non-x86 SoC-specific headers to be included as <soc/header.h> instead of the old <soc/vendor/chip/header.h> or "header.h". It will add an include/soc/ directory under every src/soc/vendor/chip/ and append the .../include/ part of that to the global include path. This matches the usage of <arch/header.h> for architecture-specific headers and had already been done for some headers on Tegra. It has the advantage that a source file which does not know the specific SoC used (e.g. Tegra files common for multiple chips, or a global include file) can still include SoC-specific headers and access macros/types defined there. It also makes the includes for mainboard files more readable, and reduces the chance to pull in a wrong header when copying mainboard sources to use a different-related SoC (e.g. using a Tegra124 mainboard as template for a Tegra132 one). For easier maintainability, every SoC family is modified individually. This patch starts out by changing Rk3288. Also alphabetized headers in affected files since we touch them anyway. BUG=None TEST=Whole series: compared binary images for Daisy, Nyan_Blaze, Rush_Ryu, Storm, Urara and Veyron_Pinky. Confirmed that they are byte-for-byte identical except for timestamps, hashes, and __LINE__ macro replacements. Compile-tested individual patches. Change-Id: I4d74a0c56be278e591a9cf43f93e9900e41f4319 Signed-off-by: Patrick Georgi <pgeorgi@chromium.org> Original-Commit-Id: 4ad8b6d2e0280428aa9742f0f7b723c00857334a Original-Change-Id: I415b8dbe735e572d4ae2cb1df62d66bcce386fff Original-Signed-off-by: Julius Werner <jwerner@chromium.org> Original-Reviewed-on: https://chromium-review.googlesource.com/222025 Reviewed-on: http://review.coreboot.org/9349 Tested-by: build bot (Jenkins) Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
2014-10-20 22:14:55 +02:00
#include <soc/addressmap.h>
#include <soc/clock.h>
#include <soc/grf.h>
#include <soc/timer.h>
rk3288: Add early SRAM mapping Solving the DACR bug will mean that XN bits suddenly become enforced on non-LPAE systems, and we will no longer be able to execute out of a region mapped DCACHE_OFF. When we enable the MMU in romstage we are still executing out of SRAM, so we would instantly kill ourselves. Solve this issue by enabling the MMU earlier (in the bootblock) and mapping the SRAM regions as DCACHE_WRITETHROUGH. They should really be DCACHE_WRITEBACK, but it looks like there might be hardware limitations in the Cortex-A12 cache architecture that prevent us from doing so. Write-through mappings are equivalent to normal non-cacheable on the A12 anyway, and by using this attribute we don't need to introduce a new DCACHE_OFF_BUT_WITHOUT_XN_BIT type in our API. (Also, using normal non-cacheable might still have a slight speed advantage over strongly ordered since it should fetch whole cache lines at once if the processor finds enough accesses it can combine.) CQ-DEPEND=CL:223783 BUG=chrome-os-partner:32118 TEST=None (depends on follow-up CL) Change-Id: I1e5127421f82177ca11af892b1539538b379625e Signed-off-by: Patrick Georgi <pgeorgi@chromium.org> Original-Commit-Id: e7b079f4b6a69449f3c7cc18ef0e1704f2006847 Original-Change-Id: I53e827d95acc2db909f1251de78d65e295eceaa7 Original-Signed-off-by: Julius Werner <jwerner@chromium.org> Original-Reviewed-on: https://chromium-review.googlesource.com/223782 Original-Reviewed-by: Aaron Durbin <adurbin@chromium.org> Reviewed-on: http://review.coreboot.org/9342 Tested-by: build bot (Jenkins) Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
2014-10-16 03:50:45 +02:00
#include <symbols.h>
void bootblock_soc_init(void)
{
rkclk_init();
rk3288: Add early SRAM mapping Solving the DACR bug will mean that XN bits suddenly become enforced on non-LPAE systems, and we will no longer be able to execute out of a region mapped DCACHE_OFF. When we enable the MMU in romstage we are still executing out of SRAM, so we would instantly kill ourselves. Solve this issue by enabling the MMU earlier (in the bootblock) and mapping the SRAM regions as DCACHE_WRITETHROUGH. They should really be DCACHE_WRITEBACK, but it looks like there might be hardware limitations in the Cortex-A12 cache architecture that prevent us from doing so. Write-through mappings are equivalent to normal non-cacheable on the A12 anyway, and by using this attribute we don't need to introduce a new DCACHE_OFF_BUT_WITHOUT_XN_BIT type in our API. (Also, using normal non-cacheable might still have a slight speed advantage over strongly ordered since it should fetch whole cache lines at once if the processor finds enough accesses it can combine.) CQ-DEPEND=CL:223783 BUG=chrome-os-partner:32118 TEST=None (depends on follow-up CL) Change-Id: I1e5127421f82177ca11af892b1539538b379625e Signed-off-by: Patrick Georgi <pgeorgi@chromium.org> Original-Commit-Id: e7b079f4b6a69449f3c7cc18ef0e1704f2006847 Original-Change-Id: I53e827d95acc2db909f1251de78d65e295eceaa7 Original-Signed-off-by: Julius Werner <jwerner@chromium.org> Original-Reviewed-on: https://chromium-review.googlesource.com/223782 Original-Reviewed-by: Aaron Durbin <adurbin@chromium.org> Reviewed-on: http://review.coreboot.org/9342 Tested-by: build bot (Jenkins) Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
2014-10-16 03:50:45 +02:00
mmu_init();
/* Start with a clean slate. */
mmu_config_range(0, 4096, DCACHE_OFF);
/* SRAM is tightly wedged between registers, need to use subtables. Map
* write-through as equivalent for non-cacheable without XN on A17. */
mmu_config_range_kb((uintptr_t)_sram/KiB,
REGION_SIZE(sram)/KiB, DCACHE_WRITETHROUGH);
rk3288: Add early SRAM mapping Solving the DACR bug will mean that XN bits suddenly become enforced on non-LPAE systems, and we will no longer be able to execute out of a region mapped DCACHE_OFF. When we enable the MMU in romstage we are still executing out of SRAM, so we would instantly kill ourselves. Solve this issue by enabling the MMU earlier (in the bootblock) and mapping the SRAM regions as DCACHE_WRITETHROUGH. They should really be DCACHE_WRITEBACK, but it looks like there might be hardware limitations in the Cortex-A12 cache architecture that prevent us from doing so. Write-through mappings are equivalent to normal non-cacheable on the A12 anyway, and by using this attribute we don't need to introduce a new DCACHE_OFF_BUT_WITHOUT_XN_BIT type in our API. (Also, using normal non-cacheable might still have a slight speed advantage over strongly ordered since it should fetch whole cache lines at once if the processor finds enough accesses it can combine.) CQ-DEPEND=CL:223783 BUG=chrome-os-partner:32118 TEST=None (depends on follow-up CL) Change-Id: I1e5127421f82177ca11af892b1539538b379625e Signed-off-by: Patrick Georgi <pgeorgi@chromium.org> Original-Commit-Id: e7b079f4b6a69449f3c7cc18ef0e1704f2006847 Original-Change-Id: I53e827d95acc2db909f1251de78d65e295eceaa7 Original-Signed-off-by: Julius Werner <jwerner@chromium.org> Original-Reviewed-on: https://chromium-review.googlesource.com/223782 Original-Reviewed-by: Aaron Durbin <adurbin@chromium.org> Reviewed-on: http://review.coreboot.org/9342 Tested-by: build bot (Jenkins) Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
2014-10-16 03:50:45 +02:00
dcache_mmu_enable();
rkclk_configure_crypto(148500*KHz);
}