Linus Walleij | bc58177 | 2009-09-15 17:30:37 +0100 | [diff] [blame^] | 1 | ARM TCM (Tightly-Coupled Memory) handling in Linux |
| 2 | ---- |
| 3 | Written by Linus Walleij <linus.walleij@stericsson.com> |
| 4 | |
| 5 | Some ARM SoC:s have a so-called TCM (Tightly-Coupled Memory). |
| 6 | This is usually just a few (4-64) KiB of RAM inside the ARM |
| 7 | processor. |
| 8 | |
| 9 | Due to being embedded inside the CPU The TCM has a |
| 10 | Harvard-architecture, so there is an ITCM (instruction TCM) |
| 11 | and a DTCM (data TCM). The DTCM can not contain any |
| 12 | instructions, but the ITCM can actually contain data. |
| 13 | The size of DTCM or ITCM is minimum 4KiB so the typical |
| 14 | minimum configuration is 4KiB ITCM and 4KiB DTCM. |
| 15 | |
| 16 | ARM CPU:s have special registers to read out status, physical |
| 17 | location and size of TCM memories. arch/arm/include/asm/cputype.h |
| 18 | defines a CPUID_TCM register that you can read out from the |
| 19 | system control coprocessor. Documentation from ARM can be found |
| 20 | at http://infocenter.arm.com, search for "TCM Status Register" |
| 21 | to see documents for all CPUs. Reading this register you can |
| 22 | determine if ITCM (bit 0) and/or DTCM (bit 16) is present in the |
| 23 | machine. |
| 24 | |
| 25 | There is further a TCM region register (search for "TCM Region |
| 26 | Registers" at the ARM site) that can report and modify the location |
| 27 | size of TCM memories at runtime. This is used to read out and modify |
| 28 | TCM location and size. Notice that this is not a MMU table: you |
| 29 | actually move the physical location of the TCM around. At the |
| 30 | place you put it, it will mask any underlying RAM from the |
| 31 | CPU so it is usually wise not to overlap any physical RAM with |
| 32 | the TCM. The TCM memory exists totally outside the MMU and will |
| 33 | override any MMU mappings. |
| 34 | |
| 35 | Code executing inside the ITCM does not "see" any MMU mappings |
| 36 | and e.g. register accesses must be made to physical addresses. |
| 37 | |
| 38 | TCM is used for a few things: |
| 39 | |
| 40 | - FIQ and other interrupt handlers that need deterministic |
| 41 | timing and cannot wait for cache misses. |
| 42 | |
| 43 | - Idle loops where all external RAM is set to self-refresh |
| 44 | retention mode, so only on-chip RAM is accessible by |
| 45 | the CPU and then we hang inside ITCM waiting for an |
| 46 | interrupt. |
| 47 | |
| 48 | - Other operations which implies shutting off or reconfiguring |
| 49 | the external RAM controller. |
| 50 | |
| 51 | There is an interface for using TCM on the ARM architecture |
| 52 | in <asm/tcm.h>. Using this interface it is possible to: |
| 53 | |
| 54 | - Define the physical address and size of ITCM and DTCM. |
| 55 | |
| 56 | - Tag functions to be compiled into ITCM. |
| 57 | |
| 58 | - Tag data and constants to be allocated to DTCM and ITCM. |
| 59 | |
| 60 | - Have the remaining TCM RAM added to a special |
| 61 | allocation pool with gen_pool_create() and gen_pool_add() |
| 62 | and provice tcm_alloc() and tcm_free() for this |
| 63 | memory. Such a heap is great for things like saving |
| 64 | device state when shutting off device power domains. |
| 65 | |
| 66 | A machine that has TCM memory shall select HAVE_TCM in |
| 67 | arch/arm/Kconfig for itself, and then the |
| 68 | rest of the functionality will depend on the physical |
| 69 | location and size of ITCM and DTCM to be defined in |
| 70 | mach/memory.h for the machine. Code that needs to use |
| 71 | TCM shall #include <asm/tcm.h> If the TCM is not located |
| 72 | at the place given in memory.h it will be moved using |
| 73 | the TCM Region registers. |
| 74 | |
| 75 | Functions to go into itcm can be tagged like this: |
| 76 | int __tcmfunc foo(int bar); |
| 77 | |
| 78 | Variables to go into dtcm can be tagged like this: |
| 79 | int __tcmdata foo; |
| 80 | |
| 81 | Constants can be tagged like this: |
| 82 | int __tcmconst foo; |
| 83 | |
| 84 | To put assembler into TCM just use |
| 85 | .section ".tcm.text" or .section ".tcm.data" |
| 86 | respectively. |
| 87 | |
| 88 | Example code: |
| 89 | |
| 90 | #include <asm/tcm.h> |
| 91 | |
| 92 | /* Uninitialized data */ |
| 93 | static u32 __tcmdata tcmvar; |
| 94 | /* Initialized data */ |
| 95 | static u32 __tcmdata tcmassigned = 0x2BADBABEU; |
| 96 | /* Constant */ |
| 97 | static const u32 __tcmconst tcmconst = 0xCAFEBABEU; |
| 98 | |
| 99 | static void __tcmlocalfunc tcm_to_tcm(void) |
| 100 | { |
| 101 | int i; |
| 102 | for (i = 0; i < 100; i++) |
| 103 | tcmvar ++; |
| 104 | } |
| 105 | |
| 106 | static void __tcmfunc hello_tcm(void) |
| 107 | { |
| 108 | /* Some abstract code that runs in ITCM */ |
| 109 | int i; |
| 110 | for (i = 0; i < 100; i++) { |
| 111 | tcmvar ++; |
| 112 | } |
| 113 | tcm_to_tcm(); |
| 114 | } |
| 115 | |
| 116 | static void __init test_tcm(void) |
| 117 | { |
| 118 | u32 *tcmem; |
| 119 | int i; |
| 120 | |
| 121 | hello_tcm(); |
| 122 | printk("Hello TCM executed from ITCM RAM\n"); |
| 123 | |
| 124 | printk("TCM variable from testrun: %u @ %p\n", tcmvar, &tcmvar); |
| 125 | tcmvar = 0xDEADBEEFU; |
| 126 | printk("TCM variable: 0x%x @ %p\n", tcmvar, &tcmvar); |
| 127 | |
| 128 | printk("TCM assigned variable: 0x%x @ %p\n", tcmassigned, &tcmassigned); |
| 129 | |
| 130 | printk("TCM constant: 0x%x @ %p\n", tcmconst, &tcmconst); |
| 131 | |
| 132 | /* Allocate some TCM memory from the pool */ |
| 133 | tcmem = tcm_alloc(20); |
| 134 | if (tcmem) { |
| 135 | printk("TCM Allocated 20 bytes of TCM @ %p\n", tcmem); |
| 136 | tcmem[0] = 0xDEADBEEFU; |
| 137 | tcmem[1] = 0x2BADBABEU; |
| 138 | tcmem[2] = 0xCAFEBABEU; |
| 139 | tcmem[3] = 0xDEADBEEFU; |
| 140 | tcmem[4] = 0x2BADBABEU; |
| 141 | for (i = 0; i < 5; i++) |
| 142 | printk("TCM tcmem[%d] = %08x\n", i, tcmem[i]); |
| 143 | tcm_free(tcmem, 20); |
| 144 | } |
| 145 | } |