1An ad-hoc collection of notes on IA64 MCA and INIT processing.  Feel
   2free to update it with notes about any area that is not clear.
   6MCA/INIT are completely asynchronous.  They can occur at any time, when
   7the OS is in any state.  Including when one of the cpus is already
   8holding a spinlock.  Trying to get any lock from MCA/INIT state is
   9asking for deadlock.  Also the state of structures that are protected
  10by locks is indeterminate, including linked lists.
  14The complicated ia64 MCA process.  All of this is mandated by Intel's
  15specification for ia64 SAL, error recovery and unwind, it is not as
  16if we have a choice here.
  18* MCA occurs on one cpu, usually due to a double bit memory error.
  19  This is the monarch cpu.
  21* SAL sends an MCA rendezvous interrupt (which is a normal interrupt)
  22  to all the other cpus, the slaves.
  24* Slave cpus that receive the MCA interrupt call down into SAL, they
  25  end up spinning disabled while the MCA is being serviced.
  27* If any slave cpu was already spinning disabled when the MCA occurred
  28  then it cannot service the MCA interrupt.  SAL waits ~20 seconds then
  29  sends an unmaskable INIT event to the slave cpus that have not
  30  already rendezvoused.
  32* Because MCA/INIT can be delivered at any time, including when the cpu
  33  is down in PAL in physical mode, the registers at the time of the
  34  event are _completely_ undefined.  In particular the MCA/INIT
  35  handlers cannot rely on the thread pointer, PAL physical mode can
  36  (and does) modify TP.  It is allowed to do that as long as it resets
  37  TP on return.  However MCA/INIT events expose us to these PAL
  38  internal TP changes.  Hence curr_task().
  40* If an MCA/INIT event occurs while the kernel was running (not user
  41  space) and the kernel has called PAL then the MCA/INIT handler cannot
  42  assume that the kernel stack is in a fit state to be used.  Mainly
  43  because PAL may or may not maintain the stack pointer internally.
  44  Because the MCA/INIT handlers cannot trust the kernel stack, they
  45  have to use their own, per-cpu stacks.  The MCA/INIT stacks are
  46  preformatted with just enough task state to let the relevant handlers
  47  do their job.
  49* Unlike most other architectures, the ia64 struct task is embedded in
  50  the kernel stack[1].  So switching to a new kernel stack means that
  51  we switch to a new task as well.  Because various bits of the kernel
  52  assume that current points into the struct task, switching to a new
  53  stack also means a new value for current.
  55* Once all slaves have rendezvoused and are spinning disabled, the
  56  monarch is entered.  The monarch now tries to diagnose the problem
  57  and decide if it can recover or not.
  59* Part of the monarch's job is to look at the state of all the other
  60  tasks.  The only way to do that on ia64 is to call the unwinder,
  61  as mandated by Intel.
  63* The starting point for the unwind depends on whether a task is
  64  running or not.  That is, whether it is on a cpu or is blocked.  The
  65  monarch has to determine whether or not a task is on a cpu before it
  66  knows how to start unwinding it.  The tasks that received an MCA or
  67  INIT event are no longer running, they have been converted to blocked
  68  tasks.  But (and its a big but), the cpus that received the MCA
  69  rendezvous interrupt are still running on their normal kernel stacks!
  71* To distinguish between these two cases, the monarch must know which
  72  tasks are on a cpu and which are not.  Hence each slave cpu that
  73  switches to an MCA/INIT stack, registers its new stack using
  74  set_curr_task(), so the monarch can tell that the _original_ task is
  75  no longer running on that cpu.  That gives us a decent chance of
  76  getting a valid backtrace of the _original_ task.
  78* MCA/INIT can be nested, to a depth of 2 on any cpu.  In the case of a
  79  nested error, we want diagnostics on the MCA/INIT handler that
  80  failed, not on the task that was originally running.  Again this
  81  requires set_curr_task() so the MCA/INIT handlers can register their
  82  own stack as running on that cpu.  Then a recursive error gets a
  83  trace of the failing handler's "task".
  85[1] My (Keith Owens) original design called for ia64 to separate its
  86    struct task and the kernel stacks.  Then the MCA/INIT data would be
  87    chained stacks like i386 interrupt stacks.  But that required
  88    radical surgery on the rest of ia64, plus extra hard wired TLB
  89    entries with its associated performance degradation.  David
  90    Mosberger vetoed that approach.  Which meant that separate kernel
  91    stacks meant separate "tasks" for the MCA/INIT handlers.
  95INIT is less complicated than MCA.  Pressing the nmi button or using
  96the equivalent command on the management console sends INIT to all
  97cpus.  SAL picks one of the cpus as the monarch and the rest are
  98slaves.  All the OS INIT handlers are entered at approximately the same
  99time.  The OS monarch prints the state of all tasks and returns, after
 100which the slaves return and the system resumes.
 102At least that is what is supposed to happen.  Alas there are broken
 103versions of SAL out there.  Some drive all the cpus as monarchs.  Some
 104drive them all as slaves.  Some drive one cpu as monarch, wait for that
 105cpu to return from the OS then drive the rest as slaves.  Some versions
 106of SAL cannot even cope with returning from the OS, they spin inside
 107SAL on resume.  The OS INIT code has workarounds for some of these
 108broken SAL symptoms, but some simply cannot be fixed from the OS side.
 112The scheduler hooks used by ia64 (curr_task, set_curr_task) are layer
 113violations.  Unfortunately MCA/INIT start off as massive layer
 114violations (can occur at _any_ time) and they build from there.
 116At least ia64 makes an attempt at recovering from hardware errors, but
 117it is a difficult problem because of the asynchronous nature of these
 118errors.  When processing an unmaskable interrupt we sometimes need
 119special code to cope with our inability to take any locks.
 123How is ia64 MCA/INIT different from x86 NMI?
 125* x86 NMI typically gets delivered to one cpu.  MCA/INIT gets sent to
 126  all cpus.
 128* x86 NMI cannot be nested.  MCA/INIT can be nested, to a depth of 2
 129  per cpu.
 131* x86 has a separate struct task which points to one of multiple kernel
 132  stacks.  ia64 has the struct task embedded in the single kernel
 133  stack, so switching stack means switching task.
 135* x86 does not call the BIOS so the NMI handler does not have to worry
 136  about any registers having changed.  MCA/INIT can occur while the cpu
 137  is in PAL in physical mode, with undefined registers and an undefined
 138  kernel stack.
 140* i386 backtrace is not very sensitive to whether a process is running
 141  or not.  ia64 unwind is very, very sensitive to whether a process is
 142  running or not.
 146What happens when MCA/INIT is delivered what a cpu is running user
 147space code?
 149The user mode registers are stored in the RSE area of the MCA/INIT on
 150entry to the OS and are restored from there on return to SAL, so user
 151mode registers are preserved across a recoverable MCA/INIT.  Since the
 152OS has no idea what unwind data is available for the user space stack,
 153MCA/INIT never tries to backtrace user space.  Which means that the OS
 154does not bother making the user space process look like a blocked task,
 155i.e. the OS does not copy pt_regs and switch_stack to the user space
 156stack.  Also the OS has no idea how big the user space RSE and memory
 157stacks are, which makes it too risky to copy the saved state to a user
 158mode stack.
 162How do we get a backtrace on the tasks that were running when MCA/INIT
 163was delivered?
 165mca.c:::ia64_mca_modify_original_stack().  That identifies and
 166verifies the original kernel stack, copies the dirty registers from
 167the MCA/INIT stack's RSE to the original stack's RSE, copies the
 168skeleton struct pt_regs and switch_stack to the original stack, fills
 169in the skeleton structures from the PAL minstate area and updates the
 170original stack's thread.ksp.  That makes the original stack look
 171exactly like any other blocked task, i.e. it now appears to be
 172sleeping.  To get a backtrace, just start with thread.ksp for the
 173original task and unwind like any other sleeping task.
 177How do we identify the tasks that were running when MCA/INIT was
 180If the previous task has been verified and converted to a blocked
 181state, then sos->prev_task on the MCA/INIT stack is updated to point to
 182the previous task.  You can look at that field in dumps or debuggers.
 183To help distinguish between the handler and the original tasks,
 184handlers have _TIF_MCA_INIT set in thread_info.flags.
 186The sos data is always in the MCA/INIT handler stack, at offset
 187MCA_SOS_OFFSET.  You can get that value from mca_asm.h or calculate it
 188as KERNEL_STACK_SIZE - sizeof(struct pt_regs) - sizeof(struct
 189ia64_sal_os_state), with 16 byte alignment for all structures.
 191Also the comm field of the MCA/INIT task is modified to include the pid
 192of the original task, for humans to use.  For example, a comm field of
 193'MCA 12159' means that pid 12159 was running when the MCA was