Messages in this thread Patch in this message | | | Date | Tue, 17 Mar 2009 20:54:34 +0800 | From | Lai Jiangshan <> | Subject | [PATCH 1/2] ftrace: protect running nmi |
| |
Steven Rostedt wrote: >> atomic_inc(&nmi_running); >> /* Must have nmi_running seen before reading write flag */ >> smp_mb(); >> - if (mod_code_write) { >> + if (!atomic_read(&nmi_executing) && mod_code_write) { >> ftrace_mod_code(); >> atomic_inc(&nmi_update_count); >> } > > Here we have another race window. If cpu#1 has that NMI and right here > we get a SMI (something to make the race window bigger). cpu#2 could > have set the mod_code_write and cpu#3 could have another NMI that sees > it but does not see the nmi_executing flag. Now we are in the same > scenario as you nicely described up above.
I missed this window.
> void ftrace_nmi_enter(void) > { > if (atomic_inc_return(&mod_code_write) > 10000) { > ftrace_mod_code(); > atomic_inc(&nmi_update_count); > } > smp_mb(); > } > > void ftrace_nmi_exit(void) > { > smp_mb(); > atomic_dec(&mod_code_write); > } > > Then in do_ftrace_mod_code ... > > > while (atomic_cmpxchg(&mod_code_write, 0, 10001) != 0) > ; > > [...] > > > while (atomic_cmpxchg(&mode_code_write, 10001, 0) != 10001) > ; > > > Does this look like it would solve the issue? >
It's very nice. The write-flag and the counter are put into an atomic together. The write-flag is changed only when there is no running NMI. So any NMI sees this flag, all other running NMIs also see this flag when them were entering.
Subject: [PATCH 1/2] ftrace: protect running nmi (V2)
When I review the sensitive code ftrace_nmi_enter(), I found the atomic variable nmi_running does protect NMI VS do_ftrace_mod_code(), but it can not protects NMI(entered nmi) VS NMI(ftrace_nmi_enter()).
cpu#1 | cpu#2 | cpu#3 ftrace_nmi_enter() | do_ftrace_mod_code() | not modify | | ------------------------|-----------------------|-- executing | set mod_code_write = 1| executing --|-----------------------|-------------------- executing | | ftrace_nmi_enter() executing | | do modify ------------------------|-----------------------|----------------- ftrace_nmi_exit() | |
cpu#3 may be being modified the code which is still being executed on cpu#1, it will have undefined results and possibly take a GPF, this patch prevents it occurred.
Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com> --- diff --git a/arch/x86/kernel/ftrace.c b/arch/x86/kernel/ftrace.c index 1d0d7f4..699a1c0 100644 --- a/arch/x86/kernel/ftrace.c +++ b/arch/x86/kernel/ftrace.c @@ -79,11 +79,11 @@ static unsigned char *ftrace_call_replace(unsigned long ip, unsigned long addr) * * 1) Put the instruction pointer into the IP buffer * and the new code into the "code" buffer. - * 2) Set a flag that says we are modifying code - * 3) Wait for any running NMIs to finish. - * 4) Write the code - * 5) clear the flag. - * 6) Wait for any running NMIs to finish. + * 2) Wait for any running NMIs to finish and set a flag that says + * we are modifying code, it is done in an atomic operation. + * 3) Write the code + * 4) clear the flag and wait for any running NMIs to finish, + * it is also done in an atomic operation. * * If an NMI is executed, the first thing it does is to call * "ftrace_nmi_enter". This will check if the flag is set to write @@ -95,9 +95,9 @@ static unsigned char *ftrace_call_replace(unsigned long ip, unsigned long addr) * are the same as what exists. */ +#define MOD_CODE_WRITE_FLAG (1 << 31) /* set when NMI should do the write */ static atomic_t nmi_running = ATOMIC_INIT(0); static int mod_code_status; /* holds return value of text write */ -static int mod_code_write; /* set when NMI should do the write */ static void *mod_code_ip; /* holds the IP to write to */ static void *mod_code_newcode; /* holds the text to write to the IP */ @@ -124,40 +124,36 @@ static void ftrace_mod_code(void) */ mod_code_status = probe_kernel_write(mod_code_ip, mod_code_newcode, MCOUNT_INSN_SIZE); - - /* if we fail, then kill any new writers */ - if (mod_code_status) - mod_code_write = 0; } void ftrace_nmi_enter(void) { - atomic_inc(&nmi_running); - /* Must have nmi_running seen before reading write flag */ - smp_mb(); - if (mod_code_write) { + if (atomic_inc_return(&nmi_running) & MOD_CODE_WRITE_FLAG) { + smp_rmb(); ftrace_mod_code(); atomic_inc(&nmi_update_count); } + /* Must have previous changes seen before executions */ + smp_mb(); } void ftrace_nmi_exit(void) { /* Finish all executions before clearing nmi_running */ - smp_wmb(); + smp_mb(); atomic_dec(&nmi_running); } -static void wait_for_nmi(void) +static void wait_and_set(int wait_val, int set_val) { - if (!atomic_read(&nmi_running)) - return; + int wait = 0; - do { + while (atomic_cmpxchg(&nmi_running, wait_val, set_val) != wait_val) { + wait = 1; cpu_relax(); - } while (atomic_read(&nmi_running)); + } - nmi_wait_count++; + nmi_wait_count += wait; } static int @@ -166,15 +162,13 @@ do_ftrace_mod_code(unsigned long ip, void *new_code) mod_code_ip = (void *)ip; mod_code_newcode = new_code; - /* The buffers need to be visible before we let NMIs write them */ - smp_wmb(); - - mod_code_write = 1; - - /* Make sure write bit is visible before we wait on NMIs */ + /* + * The previous variables need to be visible before NMIs sees + * the MOD_CODE_WRITE_FLAG. + */ smp_mb(); - wait_for_nmi(); + wait_and_set(0, MOD_CODE_WRITE_FLAG); /* Make sure all running NMIs have finished before we write the code */ smp_mb(); @@ -182,14 +176,9 @@ do_ftrace_mod_code(unsigned long ip, void *new_code) ftrace_mod_code(); /* Make sure the write happens before clearing the bit */ - smp_wmb(); - - mod_code_write = 0; - - /* make sure NMIs see the cleared bit */ smp_mb(); - wait_for_nmi(); + wait_and_set(MOD_CODE_WRITE_FLAG, 0); return mod_code_status; }
| |