Djam/arm-trusted-firmware
1
0
Fork 0
mirror of https://github.com/ARM-software/arm-trusted-firmware.git synced 2025-04-10 22:54:13 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
arm-trusted-firmware/lib/cpus/aarch64/cortex_a510.S
Boyan Karatotev 89dba82dfa perf(cpus): make reset errata do fewer branches 
Errata application is painful for performance. For a start, it's done
when the core has just come out of reset, which means branch predictors
and caches will be empty so a branch to a workaround function must be
fetched from memory and that round trip is very slow. Then it also runs
with the I-cache off, which means that the loop to iterate over the
workarounds must also be fetched from memory on each iteration.

We can remove both branches. First, we can simply apply every erratum
directly instead of defining a workaround function and jumping to it.
Currently, no errata that need to be applied at both reset and runtime,
with the same workaround function, exist. If the need arose in future,
this should be achievable with a reset + runtime wrapper combo.

Then, we can construct a function that applies each erratum linearly
instead of looping over the list. If this function is part of the reset
function, then the only "far" branches at reset will be for the checker
functions. Importantly, this mitigates the slowdown even when an erratum
is disabled.

The result is ~50% speedup on N1SDP and ~20% on AArch64 Juno on wakeup
from PSCI calls that end in powerdown. This is roughly back to the
baseline of v2.9, before the errata framework regressed on performance
(or a little better). It is important to note that there are other
slowdowns since then that remain unknown.

Change-Id: Ie4d5288a331b11fd648e5c4a0b652b74160b07b9
Signed-off-by: Boyan Karatotev <boyan.karatotev@arm.com>
2025-02-24 09:36:11 +00:00

235 lines
8 KiB
ArmAsm
Raw Blame History

/*
* Copyright (c) 2023-2025, Arm Limited. All rights reserved.
*
* SPDX-License-Identifier: BSD-3-Clause
*/
#include <arch.h>
#include <asm_macros.S>
#include <common/bl_common.h>
#include <cortex_a510.h>
#include <cpu_macros.S>
#include <dsu_macros.S>
#include <plat_macros.S>
/* Hardware handled coherency */
#if HW_ASSISTED_COHERENCY == 0
#error "Cortex-A510 must be compiled with HW_ASSISTED_COHERENCY enabled"
#endif
/* 64-bit only core */
#if CTX_INCLUDE_AARCH32_REGS == 1
#error "Cortex-A510 supports only AArch64. Compile with CTX_INCLUDE_AARCH32_REGS=0"
#endif
cpu_reset_prologue cortex_a510
workaround_reset_start cortex_a510, ERRATUM(1922240), ERRATA_A510_1922240
/* Apply the workaround by setting IMP_CMPXACTLR_EL1[11:10] = 0b11. */
sysreg_bitfield_insert CORTEX_A510_CMPXACTLR_EL1, CORTEX_A510_CMPXACTLR_EL1_SNPPREFERUNIQUE_DISABLE, \
CORTEX_A510_CMPXACTLR_EL1_SNPPREFERUNIQUE_SHIFT, CORTEX_A510_CMPXACTLR_EL1_SNPPREFERUNIQUE_WIDTH
workaround_reset_end cortex_a510, ERRATUM(1922240)
check_erratum_ls cortex_a510, ERRATUM(1922240), CPU_REV(0, 0)
workaround_reset_start cortex_a510, ERRATUM(2041909), ERRATA_A510_2041909
/* Apply workaround */
mov x0, xzr
msr S3_6_C15_C4_0, x0
isb
mov x0, #0x8500000
msr S3_6_C15_C4_2, x0
mov x0, #0x1F700000
movk x0, #0x8, lsl #32
msr S3_6_C15_C4_3, x0
mov x0, #0x3F1
movk x0, #0x110, lsl #16
msr S3_6_C15_C4_1, x0
workaround_reset_end cortex_a510, ERRATUM(2041909)
check_erratum_range cortex_a510, ERRATUM(2041909), CPU_REV(0, 2), CPU_REV(0, 2)
workaround_reset_start cortex_a510, ERRATUM(2042739), ERRATA_A510_2042739
/* Apply the workaround by disabling ReadPreferUnique. */
sysreg_bitfield_insert CORTEX_A510_CPUECTLR_EL1, CORTEX_A510_CPUECTLR_EL1_READPREFERUNIQUE_DISABLE, \
CORTEX_A510_CPUECTLR_EL1_READPREFERUNIQUE_SHIFT, CORTEX_A510_CPUECTLR_EL1_READPREFERUNIQUE_WIDTH
workaround_reset_end cortex_a510, ERRATUM(2042739)
check_erratum_ls cortex_a510, ERRATUM(2042739), CPU_REV(0, 2)
workaround_reset_start cortex_a510, ERRATUM(2080326), ERRATA_A510_2080326
/* Apply workaround */
mov x0, #1
msr S3_6_C15_C4_0, x0
isb
mov x0, #0x0100
movk x0, #0x0E08, lsl #16
msr S3_6_C15_C4_2, x0
mov x0, #0x0300
movk x0, #0x0F1F, lsl #16
movk x0, #0x0008, lsl #32
msr S3_6_C15_C4_3, x0
mov x0, #0x03F1
movk x0, #0x00C0, lsl #16
msr S3_6_C15_C4_1, x0
isb
workaround_reset_end cortex_a510, ERRATUM(2080326)
check_erratum_range cortex_a510, ERRATUM(2080326), CPU_REV(0, 2), CPU_REV(0, 2)
workaround_reset_start cortex_a510, ERRATUM(2172148), ERRATA_A510_2172148
/*
* Force L2 allocation of transient lines by setting
* CPUECTLR_EL1.RSCTL=0b01 and CPUECTLR_EL1.NTCTL=0b01.
*/
mrs x0, CORTEX_A510_CPUECTLR_EL1
mov x1, #1
bfi x0, x1, #CORTEX_A510_CPUECTLR_EL1_RSCTL_SHIFT, #2
bfi x0, x1, #CORTEX_A510_CPUECTLR_EL1_NTCTL_SHIFT, #2
msr CORTEX_A510_CPUECTLR_EL1, x0
workaround_reset_end cortex_a510, ERRATUM(2172148)
check_erratum_ls cortex_a510, ERRATUM(2172148), CPU_REV(1, 0)
workaround_reset_start cortex_a510, ERRATUM(2218950), ERRATA_A510_2218950
/* Set bit 18 in CPUACTLR_EL1 */
sysreg_bitfield_insert CORTEX_A510_CPUACTLR_EL1, CORTEX_A510_CPUACTLR_EL1_ALIAS_LOADSTORE_DISABLE, \
CORTEX_A510_CPUACTLR_EL1_ALIAS_LOADSTORE_SHIFT, CORTEX_A510_CPUACTLR_EL1_ALIAS_LOADSTORE_WIDTH
/* Set bit 25 in CMPXACTLR_EL1 */
sysreg_bitfield_insert CORTEX_A510_CMPXACTLR_EL1, CORTEX_A510_CMPXACTLR_EL1_ALIAS_LOADSTORE_DISABLE, \
CORTEX_A510_CMPXACTLR_EL1_ALIAS_LOADSTORE_SHIFT, CORTEX_A510_CMPXACTLR_EL1_ALIAS_LOADSTORE_WIDTH
workaround_reset_end cortex_a510, ERRATUM(2218950)
check_erratum_ls cortex_a510, ERRATUM(2218950), CPU_REV(1, 0)
/* --------------------------------------------------
* This workaround is not a typical errata fix. MPMM
* is disabled here, but this conflicts with the BL31
* MPMM support. So in addition to simply disabling
* the feature, a flag is set in the MPMM library
* indicating that it should not be enabled even if
* ENABLE_MPMM=1.
* --------------------------------------------------
*/
workaround_reset_start cortex_a510, ERRATUM(2250311), ERRATA_A510_2250311
/* Disable MPMM */
mrs x0, CPUMPMMCR_EL3
bfm x0, xzr, #0, #0 /* bfc instruction does not work in GCC */
msr CPUMPMMCR_EL3, x0
#if ENABLE_MPMM && IMAGE_BL31
/* If ENABLE_MPMM is set, tell the runtime lib to skip enabling it. */
bl mpmm_errata_disable
#endif
workaround_reset_end cortex_a510, ERRATUM(2250311)
check_erratum_ls cortex_a510, ERRATUM(2250311), CPU_REV(1, 0)
workaround_reset_start cortex_a510, ERRATUM(2288014), ERRATA_A510_2288014
/* Apply the workaround by setting IMP_CPUACTLR_EL1[18] = 0b1. */
sysreg_bitfield_insert CORTEX_A510_CPUACTLR_EL1, CORTEX_A510_CPUACTLR_EL1_DATA_CORRUPT_DISABLE, \
CORTEX_A510_CPUACTLR_EL1_DATA_CORRUPT_SHIFT, CORTEX_A510_CPUACTLR_EL1_DATA_CORRUPT_WIDTH
workaround_reset_end cortex_a510, ERRATUM(2288014)
check_erratum_ls cortex_a510, ERRATUM(2288014), CPU_REV(1, 0)
workaround_reset_start cortex_a510, ERRATUM(2347730), ERRATA_A510_2347730
/*
* Set CPUACTLR_EL1[17] to 1'b1, which disables
* specific microarchitectural clock gating
* behaviour.
*/
sysreg_bit_set CORTEX_A510_CPUACTLR_EL1, CORTEX_A510_CPUACTLR_EL1_BIT_17
workaround_reset_end cortex_a510, ERRATUM(2347730)
check_erratum_ls cortex_a510, ERRATUM(2347730), CPU_REV(1, 1)
workaround_reset_start cortex_a510, ERRATUM(2371937), ERRATA_A510_2371937
/*
* Cacheable atomic operations can be forced
* to be executed near by setting
* IMP_CPUECTLR_EL1.ATOM=0b010. ATOM is found
* in [40:38] of CPUECTLR_EL1.
*/
sysreg_bitfield_insert CORTEX_A510_CPUECTLR_EL1, CORTEX_A510_CPUECTLR_EL1_ATOM_EXECALLINSTRNEAR, \
CORTEX_A510_CPUECTLR_EL1_ATOM_SHIFT, CORTEX_A510_CPUECTLR_EL1_ATOM_WIDTH
workaround_reset_end cortex_a510, ERRATUM(2371937)
check_erratum_ls cortex_a510, ERRATUM(2371937), CPU_REV(1, 1)
workaround_reset_start cortex_a510, ERRATUM(2666669), ERRATA_A510_2666669
sysreg_bit_set CORTEX_A510_CPUACTLR_EL1, CORTEX_A510_CPUACTLR_EL1_BIT_38
workaround_reset_end cortex_a510, ERRATUM(2666669)
check_erratum_ls cortex_a510, ERRATUM(2666669), CPU_REV(1, 1)
.global erratum_cortex_a510_2684597_wa
workaround_runtime_start cortex_a510, ERRATUM(2684597), ERRATA_A510_2684597, CORTEX_A510_MIDR
/*
* Many assemblers do not yet understand the "tsb csync" mnemonic,
* so use the equivalent hint instruction.
*/
hint #18 /* tsb csync */
workaround_runtime_end cortex_a510, ERRATUM(2684597)
check_erratum_ls cortex_a510, ERRATUM(2684597), CPU_REV(1, 2)
workaround_reset_start cortex_a510, ERRATUM(2313941), ERRATA_DSU_2313941
errata_dsu_2313941_wa_impl
workaround_reset_end cortex_a510, ERRATUM(2313941)
check_erratum_custom_start cortex_a510, ERRATUM(2313941)
check_errata_dsu_2313941_impl
ret
check_erratum_custom_end cortex_a510, ERRATUM(2313941)
/* ----------------------------------------------------
* HW will do the cache maintenance while powering down
* ----------------------------------------------------
*/
func cortex_a510_core_pwr_dwn
/* ---------------------------------------------------
* Enable CPU power down bit in power control register
* ---------------------------------------------------
*/
sysreg_bit_set CORTEX_A510_CPUPWRCTLR_EL1, CORTEX_A510_CPUPWRCTLR_EL1_CORE_PWRDN_BIT
isb
ret
endfunc cortex_a510_core_pwr_dwn
cpu_reset_func_start cortex_a510
/* Disable speculative loads */
msr SSBS, xzr
cpu_reset_func_end cortex_a510
/* ---------------------------------------------
* This function provides Cortex-A510 specific
* register information for crash reporting.
* It needs to return with x6 pointing to
* a list of register names in ascii and
* x8 - x15 having values of registers to be
* reported.
* ---------------------------------------------
*/
.section .rodata.cortex_a510_regs, "aS"
cortex_a510_regs: /* The ascii list of register names to be reported */
.asciz "cpuectlr_el1", ""
func cortex_a510_cpu_reg_dump
adr x6, cortex_a510_regs
mrs x8, CORTEX_A510_CPUECTLR_EL1
ret
endfunc cortex_a510_cpu_reg_dump
declare_cpu_ops cortex_a510, CORTEX_A510_MIDR, \
cortex_a510_reset_func, \
cortex_a510_core_pwr_dwn
Reference in a new issue View git blame Copy permalink