This is part III of ASI RFC v4. Please refer to the cover letter of part I for an overview the ASI RFC. https://lore.kernel.org/lkml/20200504144939.11318-1-alexandre.chartre@xxxxxxxxxx/ This part provides a driver and a CLI for testing and introspecting an ASI. The driver creates a test ASI and provide the ability to run some test sequences on the test ASI such as: - a simple enter/exit ASI - using printk with ASI - access a map or unmapped buffer while in ASI - receive an interrupt or NMI while in ASI - scheduling in/out a task using ASI The driver reports if the ASI remains active or not at the end of the test sequence. It also provides the ability to: - query the ASI fault (i.e. places where the ASI had a fault) - clear the ASI fault log - toggle the reporting of stack trace on an ASI fault - query ASI mappings (i.e. VA ranges mapped in the ASI) - add an ASI mapping - clear an ASI mapping The asicmd user CLI is provided to interact with the ASI test driver. More details and examples are provided below for using the asicmd CLI. Thanks, alex. ===== ASI Test Driver and CLI Usage ----------------------------- The ASI test driver is built as a loadable module (asi.ko) in the drivers/staging/asi/ directory. You can load it with modprobe or insmod. The ASI test CLI (asicmd) is built in the drivers/staging/asi/ directory. The ASI test driver should be loaded before using the asicmd CLI. Usage: ------ Running the asicmd without arguments will show the different commands available: # ./asicmd Usage: asicmd (<cmd>|<test>...) Commands: all - run all tests fault - list ASI faults fltclr - clear ASI faults stkon - show stack on ASI fault stkoff - do not show stack on ASI fault map - list ASI mappings mapadd - add ASI mappings mapclr - clear ASI mapping Tests: nop - enter/exit ASI and nothing else mem - enter ASI and accessed an unmapped buffer memmap - enter ASI and accessed a mapped buffer intr - receive an interruption while running with ASI nmi - receive a NMI while running with ASI intrnmi - receive a NMI in an interrupt received while running with ASI sched - call schedule() while running with ASI printk - call printk() while running with ASI Running a Test Sequence: ------------------------ Just specify the test sequence to run. For example, to run the "nop" test sequence, run "asicmd nop": # ./asicmd nop Test nop (sequence 0) - rv = 0 ; result = 0 ; asi active - expect = asi active TEST OK The command reports that the ASI was active at the end of the test sequence and that this was the expected result. So the test was successful. Another example with the "mem" test sequence: # ./asicmd mem Test mem (sequence 2) - rv = 0 ; result = 0 ; asi inactive - expect = asi inactive TEST OK This tests sequence access a buffer which is not mapped in ASI so it causes ASI to exit. The test effectively reports that the ASI was inactive at the end of the test sequence, but this is the expected result so the test is successful. All test sequences can be run with the "all" command: asicmd all Managing ASI Faults: -------------------- The asicmd CLI has commands to manage ASI faults i.e. faults that occurred while running ASI because ASI is accessing an unmapped address: - "asicmd fault" reports all ASI faults. It shows the address where the fault has occurred and the number of time the fault has occurred. - "asicmd fltclr" clears the ASI fault log. For example, the "mem" will cause an ASI fault and we can see it with the "fault" command: # ./asicmd fltclr # ./asicmd fault ASI has no fault # ./asicmd mem Test mem (sequence 2) - rv = 0 ; result = 0 ; asi inactive - expect = asi inactive TEST OK # ./asicmd fault ADDRESS COUNT SYMBOL 0xffffffffc0515037 1 asidrv_mem_run+0x37/0x7f [asi] Managing ASI Mappings: ---------------------- The asicmd CLI has commands to manage ASI mappings i.e. address ranges which are mapped in the ASI page-table: - "asicmd map" reports all ASI mappings. It shows the address, size and page-table level of the mappings. - "asicmd mapadd" adds a mapping to the ASI page-table. The syntax is: asicmd mapadd [percpu:]<addr>:<size>[:<level>] . <addr> is the start address of the mapping and <size> is its size. . <level> is the page-table level of the mapping, either "pte", "pmd", "pud", "p4d" or "pgd". By default, the level is "pte". . "percpu" should be prepended when <addr> is the offset of a percpu buffer - "asicmd" mapclr" clears a mapping in the ASI page-table. The syntax is: asicmd mapclr [percpu:]<addr> . <addr> is the start address of the mapping. The system automatically figures out the size and page-table level of the mapping. . "percpu" should be prepended when <addr> is the offset of a percpu buffer Example: # ./asicmd map ADDRESS SIZE LEVEL 0xffff88806946f000 0x838 PTE 0xffff88806820a000 0x60 PTE 0xffffffffc0515000 0x7000 PTE 0xffff88807ddac4c0 0xa0 PTE 0xffff88807dd2c4c0 0xa0 PTE 0xffff88807dcac4c0 0xa0 PTE 0xffff88807dc2c4c0 0xa0 PTE 0xffff88807dd98bc0 0x8 PTE 0xffff88807dd18bc0 0x8 PTE 0xffff88807dc98bc0 0x8 PTE 0xffff88807dc18bc0 0x8 PTE 0xffff88807dd80000 0x30 PTE 0xffff88807dd00000 0x30 PTE 0xffff88807dc80000 0x30 PTE 0xffff88807dc00000 0x30 PTE 0xfffffe0000001000 0x8000000000 P4D 0xffffffff80000000 0x40000000 PTE This shows the default core mappings for ASI. These mappings shouldn't be removed. If the ASI faults then you can add the missing mapping. For example with the "printk" test sequence: # ./asicmd fltclr # ./asicmd printk Test printk (sequence 1) - rv = 0 ; result = 0 ; asi inactive - expect = asi active TEST FAILED - unexpected ASI state The "printk" test sequence failed because of an ASI fault: # ./asicmd fault ADDRESS COUNT SYMBOL 0xffffffff8110b5a1 1 vprintk_func+0x11/0xbc With the source code or debugger, we can figure out that the fault is due to printk_context not being mapped in the ASI. 0xffffffff8110b5a1 <+17>: mov %gs:0x7ef106f8(%rip),%eax # 0x1bca0 <printk_context> So we can add the missing mapping into the ASI with "mapadd"; note that printk_context is a percpu buffer and that sizeof(printk_context) is 4: # ./asicmd mapadd percpu:0x1bca0:4 mapadd 1bca0/4/0 percpu If we re-run the "printk" test then we got: # ./asicmd fltclr # ./asicmd printk Test printk (sequence 1) - rv = 0 ; result = 0 ; asi inactive - expect = asi active TEST FAILED - unexpected ASI state # ./asicmd fault ADDRESS COUNT SYMBOL 0xffffffff811081f3 1 log_store.constprop.27+0x1f3/0x280 We still see a new fault but at a difference address (this time because cpu_number is not mapped). ----- Alexandre Chartre (14): mm/asi: Define the test ASI type asidrv: Introduce the ASI driver asidrv: Introduce the ASIDRV_IOCTL_RUN_SEQUENCE ioctl asidrv: Sequence to test ASI access to mapped/unmapped memory asidrv: Sequence to test interrupt on ASI asidrv: Sequence to test NMI on ASI asidrv: Sequence to test interrupt+NMI on ASI asidrv: Sequence to test scheduling in/out with ASI asidrv: Add ioctls to manage ASI page faults asidrv: Add ioctls to manage ASI mapped VA ranges asidrv/asicmd: Introduce the asicmd command asidrv/asicmd: Add more test sequences for testing ASI asidrv/asicmd: Add options to manage ASI page faults asidrv/asicmd: Add options to manage ASI mapped VA ranges arch/x86/include/asm/asi.h | 2 + arch/x86/mm/asi.c | 1 + drivers/staging/Makefile | 1 + drivers/staging/asi/Makefile | 13 + drivers/staging/asi/asicmd.c | 439 ++++++++++++++ drivers/staging/asi/asidrv.c | 1077 ++++++++++++++++++++++++++++++++++ drivers/staging/asi/asidrv.h | 102 ++++ 7 files changed, 1635 insertions(+) create mode 100644 drivers/staging/asi/Makefile create mode 100644 drivers/staging/asi/asicmd.c create mode 100644 drivers/staging/asi/asidrv.c create mode 100644 drivers/staging/asi/asidrv.h -- 2.18.2
