Power-On-Self-Test support in U-Boot ------------------------------------ This project is to support Power-On-Self-Test (POST) in U-Boot. 1. High-level requirements The key requirements for this project are as follows: 1) The project shall develop a flexible framework for implementing and running Power-On-Self-Test in U-Boot. This framework shall possess the following features: o) Extensibility The framework shall allow adding/removing/replacing POST tests. Also, standalone POST tests shall be supported. o) Configurability The framework shall allow run-time configuration of the lists of tests running on normal/power-fail booting. o) Controllability The framework shall support manual running of the POST tests. 2) The results of tests shall be saved so that it will be possible to retrieve them from Linux. 3) The following POST tests shall be developed for MPC823E-based boards: o) CPU test o) Cache test o) Memory test o) Ethernet test o) Serial channels test o) Watchdog timer test o) RTC test o) I2C test o) SPI test o) USB test 4) The LWMON board shall be used for reference. 2. Design This section details the key points of the design for the project. The whole project can be divided into two independent tasks: enhancing U-Boot/Linux to provide a common framework for running POST tests and developing such tests for particular hardware. 2.1. Hardware-independent POST layer A new optional module will be added to U-Boot, which will run POST tests and collect their results at boot time. Also, U-Boot will support running POST tests manually at any time by executing a special command from the system console. The list of available POST tests will be configured at U-Boot build time. The POST layer will allow the developer to add any custom POST tests. All POST tests will be divided into the following groups: 1) Tests running on power-on booting only This group will contain those tests that run only once on power-on reset (e.g. watchdog test) 2) Tests running on normal booting only This group will contain those tests that do not take much time and can be run on the regular basis (e.g. CPU test) 3) Tests running in special "slow test mode" only This group will contain POST tests that consume much time and cannot be run regularly (e.g. strong memory test, I2C test) 4) Manually executed tests This group will contain those tests that can be run manually. If necessary, some tests may belong to several groups simultaneously. For example, SDRAM test may run in both normal and "slow test" mode. In normal mode, SDRAM test may perform a fast superficial memory test only, while running in slow test mode it may perform a full memory check-up. Also, all tests will be discriminated by the moment they run at. Specifically, the following groups will be singled out: 1) Tests running before relocating to RAM These tests will run immediately after initializing RAM as to enable modifying it without taking care of its contents. Basically, this group will contain memory tests only. 2) Tests running after relocating to RAM These tests will run immediately before entering the main loop as to guarantee full hardware initialization. The POST layer will also distinguish a special group of tests that may cause system rebooting (e.g. watchdog test). For such tests, the layer will automatically detect rebooting and will notify the test about it. 2.1.1. POST layer interfaces This section details the interfaces between the POST layer and the rest of U-Boot. The following flags will be defined: #define POST_POWERON 0x01 /* test runs on power-on booting */ #define POST_NORMAL 0x02 /* test runs on normal booting */ #define POST_SLOWTEST 0x04 /* test is slow, enabled by key press */ #define POST_POWERTEST 0x08 /* test runs after watchdog reset */ #define POST_ROM 0x100 /* test runs in ROM */ #define POST_RAM 0x200 /* test runs in RAM */ #define POST_MANUAL 0x400 /* test can be executed manually */ #define POST_REBOOT 0x800 /* test may cause rebooting */ #define POST_PREREL 0x1000 /* test runs before relocation */ The POST layer will export the following interface routines: o) int post_run(bd_t *bd, char *name, int flags); This routine will run the test (or the group of tests) specified by the name and flag arguments. More specifically, if the name argument is not NULL, the test with this name will be performed, otherwise all tests running in ROM/RAM (depending on the flag argument) will be executed. This routine will be called at least twice with name set to NULL, once from board_init_f() and once from board_init_r(). The flags argument will also specify the mode the test is executed in (power-on, normal, power-fail, manual). o) void post_reloc(ulong offset); This routine will be called from board_init_r() and will relocate the POST test table. o) int post_info(char *name); This routine will print the list of all POST tests that can be executed manually if name is NULL, and the description of a particular test if name is not NULL. o) int post_log(char *format, ...); This routine will be called from POST tests to log their results. Basically, this routine will print the results to stderr. The format of the arguments and the return value will be identical to the printf() routine. Also, the following board-specific routines will be called from the U-Boot common code: o) int board_power_mode(void) This routine will return the mode the system is running in (POST_POWERON, POST_NORMAL or POST_SHUTDOWN). o) void board_poweroff(void) This routine will turn off the power supply of the board. It will be called on power-fail booting after running all POST tests. o) int post_hotkeys_pressed(gd_t *gd) This routine will scan the keyboard to detect if a magic key combination has been pressed, or otherwise detect if the power-on long-running tests shall be executed or not ("normal" versus "slow" test mode). The list of available POST tests be kept in the post_tests array filled at U-Boot build time. The format of entry in this array will be as follows: struct post_test { char *name; char *cmd; char *desc; int flags; int (*test)(bd_t *bd, int flags); }; o) name This field will contain a short name of the test, which will be used in logs and on listing POST tests (e.g. CPU test). o) cmd This field will keep a name for identifying the test on manual testing (e.g. cpu). For more information, refer to section "Command line interface". o) desc This field will contain a detailed description of the test, which will be printed on user request. For more information, see section "Command line interface". o) flags This field will contain a combination of the bit flags described above, which will specify the mode the test is running in (power-on, normal, power-fail or manual mode), the moment it should be run at (before or after relocating to RAM), whether it can cause system rebooting or not. o) test This field will contain a pointer to the routine that will perform the test, which will take 2 arguments. The first argument will be a pointer to the board info structure, while the second will be a combination of bit flags specifying the mode the test is running in (POST_POWERON, POST_NORMAL, POST_SLOWTEST, POST_MANUAL) and whether the last execution of the test caused system rebooting (POST_REBOOT). The routine will return 0 on successful execution of the test, and 1 if the test failed. The lists of the POST tests that should be run at power-on/normal/ power-fail booting will be kept in the environment. Namely, the following environment variables will be used: post_poweron, powet_normal, post_slowtest. 2.1.2. Test results The results of tests will be collected by the POST layer. The POST log will have the following format: ... -------------------------------------------- START [PASSED|FAILED] -------------------------------------------- ... Basically, the results of tests will be printed to stderr. This feature may be enhanced in future to spool the log to a serial line, save it in non-volatile RAM (NVRAM), transfer it to a dedicated storage server and etc. 2.1.3. Integration issues All POST-related code will be #ifdef'ed with the CONFIG_POST macro. This macro will be defined in the config_.h file for those boards that need POST. The CONFIG_POST macro will contain the list of POST tests for the board. The macro will have the format of array composed of post_test structures: #define CONFIG_POST \ { "On-board peripherals test", "board", \ " This test performs full check-up of the " \ "on-board hardware.", \ POST_RAM | POST_SLOWTEST, \ &board_post_test \ } A new file, post.h, will be created in the include/ directory. This file will contain common POST declarations and will define a set of macros that will be reused for defining CONFIG_POST. As an example, the following macro may be defined: #define POST_CACHE \ { "Cache test", "cache", \ " This test verifies the CPU cache operation.", \ POST_RAM | POST_NORMAL, \ &cache_post_test \ } A new subdirectory will be created in the U-Boot root directory. It will contain the source code of the POST layer and most of POST tests. Each POST test in this directory will be placed into a separate file (it will be needed for building standalone tests). Some POST tests (mainly those for testing peripheral devices) will be located in the source files of the drivers for those devices. This way will be used only if the test subtantially uses the driver. 2.1.4. Standalone tests The POST framework will allow to develop and run standalone tests. A user-space library will be developed to provide the POST interface functions to standalone tests. 2.1.5. Command line interface A new command, diag, will be added to U-Boot. This command will be used for listing all available hardware tests, getting detailed descriptions of them and running these tests. More specifically, being run without any arguments, this command will print the list of all available hardware tests: => diag Available hardware tests: cache - cache test cpu - CPU test enet - SCC/FCC ethernet test Use 'diag [ []] ... ' to get more info. Use 'diag run [ []] ... ' to run tests. => If the first argument to the diag command is not 'run', detailed descriptions of the specified tests will be printed: => diag cpu cache cpu - CPU test This test verifies the arithmetic logic unit of CPU. cache - cache test This test verifies the CPU cache operation. => If the first argument to diag is 'run', the specified tests will be executed. If no tests are specified, all available tests will be executed. It will be prohibited to execute tests running in ROM manually. The 'diag' command will not display such tests and/or run them. 2.1.6. Power failure handling The Linux kernel will be modified to detect power failures and automatically reboot the system in such cases. It will be assumed that the power failure causes a system interrupt. To perform correct system shutdown, the kernel will register a handler of the power-fail IRQ on booting. Being called, the handler will run /sbin/reboot using the call_usermodehelper() routine. /sbin/reboot will automatically bring the system down in a secure way. This feature will be configured in/out from the kernel configuration file. The POST layer of U-Boot will check whether the system runs in power-fail mode. If it does, the system will be powered off after executing all hardware tests. 2.1.7. Hazardous tests Some tests may cause system rebooting during their execution. For some tests, this will indicate a failure, while for the Watchdog test, this means successful operation of the timer. In order to support such tests, the following scheme will be implemented. All the tests that may cause system rebooting will have the POST_REBOOT bit flag set in the flag field of the correspondent post_test structure. Before starting tests marked with this bit flag, the POST layer will store an identification number of the test in a location in IMMR. On booting, the POST layer will check the value of this variable and if it is set will skip over the tests preceding the failed one. On second execution of the failed test, the POST_REBOOT bit flag will be set in the flag argument to the test routine. This will allow to detect system rebooting on the previous iteration. For example, the watchdog timer test may have the following declaration/body: ... #define POST_WATCHDOG \ { "Watchdog timer test", "watchdog", \ " This test checks the watchdog timer.", \ POST_RAM | POST_POWERON | POST_REBOOT, \ &watchdog_post_test \ } ... ... int watchdog_post_test(bd_t *bd, int flags) { unsigned long start_time; if (flags & POST_REBOOT) { /* Test passed */ return 0; } else { /* disable interrupts */ disable_interrupts(); /* 10-second delay */ ... /* if we've reached this, the watchdog timer does not work */ enable_interrupts(); return 1; } } ... 2.2. Hardware-specific details This project will also develop a set of POST tests for MPC8xx- based systems. This section provides technical details of how it will be done. 2.2.1. Generic PPC tests The following generic POST tests will be developed: o) CPU test This test will check the arithmetic logic unit (ALU) of CPU. The test will take several milliseconds and will run on normal booting. o) Cache test This test will verify the CPU cache (L1 cache). The test will run on normal booting. o) Memory test This test will examine RAM and check it for errors. The test will always run on booting. On normal booting, only a limited amount of RAM will be checked. On power-fail booting a fool memory check-up will be performed. 2.2.1.1. CPU test This test will verify the following ALU instructions: o) Condition register istructions This group will contain: mtcrf, mfcr, mcrxr, crand, crandc, cror, crorc, crxor, crnand, crnor, creqv, mcrf. The mtcrf/mfcr instructions will be tested by loading different values into the condition register (mtcrf), moving its value to a general-purpose register (mfcr) and comparing this value with the expected one. The mcrxr instruction will be tested by loading a fixed value into the XER register (mtspr), moving XER value to the condition register (mcrxr), moving it to a general-purpose register (mfcr) and comparing the value of this register with the expected one. The rest of instructions will be tested by loading a fixed value into the condition register (mtcrf), executing each instruction several times to modify all 4-bit condition fields, moving the value of the conditional register to a general-purpose register (mfcr) and comparing it with the expected one. o) Integer compare instructions This group will contain: cmp, cmpi, cmpl, cmpli. To verify these instructions the test will run them with different combinations of operands, read the condition register value and compare it with the expected one. More specifically, the test will contain a pre-built table containing the description of each test case: the instruction, the values of the operands, the condition field to save the result in and the expected result. o) Arithmetic instructions This group will contain: add, addc, adde, addme, addze, subf, subfc, subfe, subme, subze, mullw, mulhw, mulhwu, divw, divwu, extsb, extsh. The test will contain a pre-built table of instructions, operands, expected results and expected states of the condition register. For each table entry, the test will cyclically use different sets of operand registers and result registers. For example, for instructions that use 3 registers on the first iteration r0/r1 will be used as operands and r2 for result. On the second iteration, r1/r2 will be used as operands and r3 as for result and so on. This will enable to verify all general-purpose registers. o) Logic instructions This group will contain: and, andc, andi, andis, or, orc, ori, oris, xor, xori, xoris, nand, nor, neg, eqv, cntlzw. The test scheme will be identical to that from the previous point. o) Shift instructions This group will contain: slw, srw, sraw, srawi, rlwinm, rlwnm, rlwimi The test scheme will be identical to that from the previous point. o) Branch instructions This group will contain: b, bl, bc. The first 2 instructions (b, bl) will be verified by jumping to a fixed address and checking whether control was transfered to that very point. For the bl instruction the value of the link register will be checked as well (using mfspr). To verify the bc instruction various combinations of the BI/BO fields, the CTR and the condition register values will be checked. The list of such combinations will be pre-built and linked in U-Boot at build time. o) Load/store instructions This group will contain: lbz(x)(u), lhz(x)(u), lha(x)(u), lwz(x)(u), stb(x)(u), sth(x)(u), stw(x)(u). All operations will be performed on a 16-byte array. The array will be 4-byte aligned. The base register will point to offset 8. The immediate offset (index register) will range in [-8 ... +7]. The test cases will be composed so that they will not cause alignment exceptions. The test will contain a pre-built table describing all test cases. For store instructions, the table entry will contain: the instruction opcode, the value of the index register and the value of the source register. After executing the instruction, the test will verify the contents of the array and the value of the base register (it must change for "store with update" instructions). For load instructions, the table entry will contain: the instruction opcode, the array contents, the value of the index register and the expected value of the destination register. After executing the instruction, the test will verify the value of the destination register and the value of the base register (it must change for "load with update" instructions). o) Load/store multiple/string instructions The CPU test will run in RAM in order to allow run-time modification of the code to reduce the memory footprint. 2.2.1.2 Special-Purpose Registers Tests TBD. 2.2.1.3. Cache test To verify the data cache operation the following test scenarios will be used: 1) Basic test #1 - turn on the data cache - switch the data cache to write-back or write-through mode - invalidate the data cache - write the negative pattern to a cached area - read the area The negative pattern must be read at the last step 2) Basic test #2 - turn on the data cache - switch the data cache to write-back or write-through mode - invalidate the data cache - write the zero pattern to a cached area - turn off the data cache - write the negative pattern to the area - turn on the data cache - read the area The negative pattern must be read at the last step 3) Write-through mode test - turn on the data cache - switch the data cache to write-through mode - invalidate the data cache - write the zero pattern to a cached area - flush the data cache - write the negative pattern to the area - turn off the data cache - read the area The negative pattern must be read at the last step 4) Write-back mode test - turn on the data cache - switch the data cache to write-back mode - invalidate the data cache - write the negative pattern to a cached area - flush the data cache - write the zero pattern to the area - invalidate the data cache - read the area The negative pattern must be read at the last step To verify the instruction cache operation the following test scenarios will be used: 1) Basic test #1 - turn on the instruction cache - unlock the entire instruction cache - invalidate the instruction cache - lock a branch instruction in the instruction cache - replace the branch instruction with "nop" - jump to the branch instruction - check that the branch instruction was executed 2) Basic test #2 - turn on the instruction cache - unlock the entire instruction cache - invalidate the instruction cache - jump to a branch instruction - check that the branch instruction was executed - replace the branch instruction with "nop" - invalidate the instruction cache - jump to the branch instruction - check that the "nop" instruction was executed The CPU test will run in RAM in order to allow run-time modification of the code. 2.2.1.4. Memory test The memory test will verify RAM using sequential writes and reads to/from RAM. Specifically, there will be several test cases that will use different patterns to verify RAM. Each test case will first fill a region of RAM with one pattern and then read the region back and compare its contents with the pattern. The following patterns will be used: 1) zero pattern (0x00000000) 2) negative pattern (0xffffffff) 3) checkerboard pattern (0x55555555, 0xaaaaaaaa) 4) bit-flip pattern ((1 << (offset % 32)), ~(1 << (offset % 32))) 5) address pattern (offset, ~offset) Patterns #1, #2 will help to find unstable bits. Patterns #3, #4 will be used to detect adherent bits, i.e. bits whose state may randomly change if adjacent bits are modified. The last pattern will be used to detect far-located errors, i.e. situations when writing to one location modifies an area located far from it. Also, usage of the last pattern will help to detect memory controller misconfigurations when RAM represents a cyclically repeated portion of a smaller size. Being run in normal mode, the test will verify only small 4Kb regions of RAM around each 1Mb boundary. For example, for 64Mb RAM the following areas will be verified: 0x00000000-0x00000800, 0x000ff800-0x00100800, 0x001ff800-0x00200800, ..., 0x03fff800- 0x04000000. If the test is run in power-fail mode, it will verify the whole RAM. The memory test will run in ROM before relocating U-Boot to RAM in order to allow RAM modification without saving its contents. 2.2.2. Common tests This section describes tests that are not based on any hardware peculiarities and use common U-Boot interfaces only. These tests do not need any modifications for porting them to another board/CPU. 2.2.2.1. I2C test For verifying the I2C bus, a full I2C bus scanning will be performed using the i2c_probe() routine. If any I2C device is found, the test will be considered as passed, otherwise failed. This particular way will be used because it provides the most common method of testing. For example, using the internal loopback mode of the CPM I2C controller for testing would not work on boards where the software I2C driver (also known as bit-banged driver) is used. 2.2.2.2. Watchdog timer test To test the watchdog timer the scheme mentioned above (refer to section "Hazardous tests") will be used. Namely, this test will be marked with the POST_REBOOT bit flag. On the first iteration, the test routine will make a 10-second delay. If the system does not reboot during this delay, the watchdog timer is not operational and the test fails. If the system reboots, on the second iteration the POST_REBOOT bit will be set in the flag argument to the test routine. The test routine will check this bit and report a success if it is set. 2.2.2.3. RTC test The RTC test will use the rtc_get()/rtc_set() routines. The following features will be verified: o) Time uniformity This will be verified by reading RTC in polling within a short period of time (5-10 seconds). o) Passing month boundaries This will be checked by setting RTC to a second before a month boundary and reading it after its passing the boundary. The test will be performed for both leap- and nonleap-years. 2.2.3. MPC8xx peripherals tests This project will develop a set of tests verifying the peripheral units of MPC8xx processors. Namely, the following controllers of the MPC8xx communication processor module (CPM) will be tested: o) Serial Management Controllers (SMC) o) Serial Communication Controllers (SCC) 2.2.3.1. Ethernet tests (SCC) The internal (local) loopback mode will be used to test SCC. To do that the controllers will be configured accordingly and several packets will be transmitted. These tests may be enhanced in future to use external loopback for testing. That will need appropriate reconfiguration of the physical interface chip. The test routines for the SCC ethernet tests will be located in arch/ppc/cpu/mpc8xx/scc.c. 2.2.3.2. UART tests (SMC/SCC) To perform these tests the internal (local) loopback mode will be used. The SMC/SCC controllers will be configured to connect the transmitter output to the receiver input. After that, several bytes will be transmitted. These tests may be enhanced to make to perform "external" loopback test using a loopback cable. In this case, the test will be executed manually. The test routine for the SMC/SCC UART tests will be located in arch/ppc/cpu/mpc8xx/serial.c. 2.2.3.3. USB test TBD 2.2.3.4. SPI test TBD