Re: [Qemu-devel] [PATCH 08/11] spapr_events: add support for dedicated hotplug event source

[David Gibson <david@gibson.dropbear.id.au>]

[-- Attachment #1: Type: text/plain, Size: 16141 bytes --]

On Fri, Oct 14, 2016 at 01:44:17PM -0500, Michael Roth wrote:
> Quoting David Gibson (2016-10-13 23:56:43)
> > On Wed, Oct 12, 2016 at 06:13:56PM -0500, Michael Roth wrote:
> > > Hotplug events were previously delivered using an EPOW interrupt
> > > and were queued by linux guests into a circular buffer. For traditional
> > > EPOW events like shutdown/resets, this isn't an issue, but for hotplug
> > > events there are cases where this buffer can be exhausted, resulting
> > > in the loss of hotplug events, resets, etc.
> > > 
> > > Newer-style hotplug event are delivered using a dedicated event source.
> > > We enable this in supported guests by adding standard an additional
> > > event source in the guest device-tree via /event-sources, and, if
> > > the guest advertises support for the newer-style hotplug events,
> > > using the corresponding interrupt to signal the available of
> > > hotplug/unplug events.
> > > 
> > > Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
> > 
> > So.. are you saying that as well as allowing new event types, the new
> > special hotplug event souce effectively allows for a bigger queue?
> > 
> > Does that mean that we didn't even necessarily need the base+length
> > unplug events, because we could now have sent the many single-LMB
> > unplug requests that were necessary?  Or does it not increase the
> > effective queue enough for that?
> 
> I assume there are still some internal limits, but the events
> are now processed using a workqueue which should provide a bit more
> headroom than the RTAS event buffer used for EPOW events. Maybe
> John (on cc:) can provide more insight into what the actual limits
> 
> In either case, the possibility for huge memory hotplug/unplug
> situations still warrant the use of count+index I think, and since
> support for the new event delivery mechanism is negotiated using the
> same option bit as count+index, there's not really any reason not to
> use it when we can. For situations where we can't (CPU/PCI), it
> might give us a bit of breathing room there as well.

Ok, makes sense.  Thanks for the extra information.

> 
> > 
> > > ---
> > >  hw/ppc/spapr.c         |  10 ++--
> > >  hw/ppc/spapr_events.c  | 148 ++++++++++++++++++++++++++++++++++++++-----------
> > >  include/hw/ppc/spapr.h |   3 +-
> > >  3 files changed, 120 insertions(+), 41 deletions(-)
> > > 
> > > diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> > > index d80a6fa..2037222 100644
> > > --- a/hw/ppc/spapr.c
> > > +++ b/hw/ppc/spapr.c
> > > @@ -275,8 +275,7 @@ static void *spapr_create_fdt_skel(hwaddr initrd_base,
> > >                                     hwaddr initrd_size,
> > >                                     hwaddr kernel_size,
> > >                                     bool little_endian,
> > > -                                   const char *kernel_cmdline,
> > > -                                   uint32_t epow_irq)
> > > +                                   const char *kernel_cmdline)
> > >  {
> > >      void *fdt;
> > >      uint32_t start_prop = cpu_to_be32(initrd_base);
> > > @@ -437,7 +436,7 @@ static void *spapr_create_fdt_skel(hwaddr initrd_base,
> > >      _FDT((fdt_end_node(fdt)));
> > >  
> > >      /* event-sources */
> > > -    spapr_events_fdt_skel(fdt, epow_irq);
> > > +    spapr_events_fdt_skel(fdt);
> > >  
> > >      /* /hypervisor node */
> > >      if (kvm_enabled()) {
> > > @@ -1944,7 +1943,7 @@ static void ppc_spapr_init(MachineState *machine)
> > >      }
> > >      g_free(filename);
> > >  
> > > -    /* Set up EPOW events infrastructure */
> > > +    /* Set up RTAS event infrastructure */
> > >      spapr_events_init(spapr);
> > >  
> > >      /* Set up the RTC RTAS interfaces */
> > > @@ -2076,8 +2075,7 @@ static void ppc_spapr_init(MachineState *machine)
> > >      /* Prepare the device tree */
> > >      spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
> > >                                              kernel_size, kernel_le,
> > > -                                            kernel_cmdline,
> > > -                                            spapr->check_exception_irq);
> > > +                                            kernel_cmdline);
> > >      assert(spapr->fdt_skel != NULL);
> > >  
> > >      /* used by RTAS */
> > > diff --git a/hw/ppc/spapr_events.c b/hw/ppc/spapr_events.c
> > > index 4c7b6ae..f8bbec6 100644
> > > --- a/hw/ppc/spapr_events.c
> > > +++ b/hw/ppc/spapr_events.c
> > > @@ -40,6 +40,7 @@
> > >  #include "hw/ppc/spapr_drc.h"
> > >  #include "qemu/help_option.h"
> > >  #include "qemu/bcd.h"
> > > +#include "hw/ppc/spapr_ovec.h"
> > >  #include <libfdt.h>
> > >  
> > >  struct rtas_error_log {
> > > @@ -206,28 +207,104 @@ struct hp_log_full {
> > >      struct rtas_event_log_v6_hp hp;
> > >  } QEMU_PACKED;
> > >  
> > > -#define EVENT_MASK_INTERNAL_ERRORS           0x80000000
> > > -#define EVENT_MASK_EPOW                      0x40000000
> > > -#define EVENT_MASK_HOTPLUG                   0x10000000
> > > -#define EVENT_MASK_IO                        0x08000000
> > > +typedef enum EventClassIndex {
> > > +    EVENT_CLASS_INTERNAL_ERRORS     = 0,
> > > +    EVENT_CLASS_EPOW                = 1,
> > > +    EVENT_CLASS_RESERVED            = 2,
> > > +    EVENT_CLASS_HOT_PLUG            = 3,
> > > +    EVENT_CLASS_IO                  = 4,
> > > +    EVENT_CLASS_MAX
> > > +} EventClassIndex;
> > > +
> > > +#define EVENT_CLASS_MASK(index) (1 << (31 - index))
> > > +
> > > +typedef struct EventSource {
> > > +    const char *name;
> > > +    int irq;
> > > +    uint32_t mask;
> > > +    bool enabled;
> > > +} EventSource;
> > > +
> > > +static EventSource event_source[EVENT_CLASS_MAX] = {
> > > +    [EVENT_CLASS_INTERNAL_ERRORS]       = { .name = "internal-errors", },
> > > +    [EVENT_CLASS_EPOW]                  = { .name = "epow-events", },
> > > +    [EVENT_CLASS_HOT_PLUG]              = { .name = "hot-plug-events", },
> > > +    [EVENT_CLASS_IO]                    = { .name = "ibm,io-events", },
> > > +};
> > > +
> > > +static void rtas_event_source_register(EventClassIndex index, int irq)
> > > +{
> > > +    /* we only support 1 irq per event class at the moment */
> > > +    g_assert(!event_source[index].enabled);
> > > +    event_source[index].irq = irq;
> > > +    event_source[index].mask = EVENT_CLASS_MASK(index);
> > > +    event_source[index].enabled = true;
> > > +}
> > 
> > I really don't like adding a mutable global table.  This should
> > probably be under the sPAPRMachineState.
> 
> I think I started off going that route. Not sure what steered me in this
> direction, but will take another look at that approach.
> 
> > 
> > > -void spapr_events_fdt_skel(void *fdt, uint32_t check_exception_irq)
> > > +void spapr_events_fdt_skel(void *fdt)
> > >  {
> > > -    uint32_t irq_ranges[] = {cpu_to_be32(check_exception_irq), cpu_to_be32(1)};
> > > -    uint32_t interrupts[] = {cpu_to_be32(check_exception_irq), 0};
> > > +    uint32_t irq_ranges[EVENT_CLASS_MAX * 2];
> > > +    int i, count = 0;
> > >  
> > >      _FDT((fdt_begin_node(fdt, "event-sources")));
> > >  
> > > +    for (i = 0, count = 0; i < EVENT_CLASS_MAX; i++) {
> > > +        /* TODO: what does 0 entail? */
> > 
> > It's just part of the interrupt specifier format expected by the
> > event-sources binding.  It's not really important.
> > 
> > > +        uint32_t interrupts[] = { cpu_to_be32(event_source[i].irq), 0 };
> > > +
> > > +        if (!event_source[i].enabled) {
> > > +            continue;
> > > +        }
> > > +
> > > +        _FDT((fdt_begin_node(fdt, event_source[i].name)));
> > > +        _FDT((fdt_property(fdt, "interrupts", interrupts, sizeof(interrupts))));
> > > +        _FDT((fdt_end_node(fdt)));
> > > +
> > > +        irq_ranges[count++] = interrupts[0];
> > > +        irq_ranges[count++] = cpu_to_be32(1);
> > > +    }
> > > +
> > > +    /* TODO: confirm the count is the last expected element */
> > > +    irq_ranges[count] = cpu_to_be32(count);
> > > +    count++;
> > > +
> > >      _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
> > >      _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
> > >      _FDT((fdt_property(fdt, "interrupt-ranges",
> > > -                       irq_ranges, sizeof(irq_ranges))));
> > > +                       irq_ranges, count * sizeof(uint32_t))));
> > >  
> > > -    _FDT((fdt_begin_node(fdt, "epow-events")));
> > > -    _FDT((fdt_property(fdt, "interrupts", interrupts, sizeof(interrupts))));
> > >      _FDT((fdt_end_node(fdt)));
> > > +}
> > >  
> > > -    _FDT((fdt_end_node(fdt)));
> > > +static const EventSource *rtas_event_log_to_source(int log_type)
> > > +{
> > > +    const EventSource *source;
> > > +
> > > +    switch (log_type) {
> > > +        case RTAS_LOG_TYPE_HOTPLUG:
> > > +            source = &event_source[EVENT_CLASS_HOT_PLUG];
> > > +            if (event_source[EVENT_CLASS_HOT_PLUG].enabled) {
> > > +                break;
> > > +            }
> > 
> > This should probably be using the flag you already have in the
> > MachineState, rather than a global.
> > 
> > > +            /* fall back to epow for legacy hotplug interrupt source */
> > > +        case RTAS_LOG_TYPE_EPOW:
> > > +            source = &event_source[EVENT_CLASS_EPOW];
> > > +            break;
> > > +        default:
> > > +            source = NULL;
> > > +    }
> > > +
> > > +    return source;
> > > +}
> > > +
> > > +static int rtas_event_log_to_irq(int log_type)
> > > +{
> > > +    const EventSource *source = rtas_event_log_to_source(log_type);
> > > +
> > > +    g_assert(source);
> > > +    g_assert(source->enabled);
> > > +
> > > +    return source->irq;
> > >  }
> > >  
> > >  static void rtas_event_log_queue(int log_type, void *data, bool exception)
> > > @@ -248,19 +325,14 @@ static sPAPREventLogEntry *rtas_event_log_dequeue(uint32_t event_mask,
> > >      sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
> > >      sPAPREventLogEntry *entry = NULL;
> > >  
> > > -    /* we only queue EPOW events atm. */
> > > -    if ((event_mask & EVENT_MASK_EPOW) == 0) {
> > > -        return NULL;
> > > -    }
> > > -
> > >      QTAILQ_FOREACH(entry, &spapr->pending_events, next) {
> > > +        const EventSource *source = rtas_event_log_to_source(entry->log_type);
> > > +
> > >          if (entry->exception != exception) {
> > >              continue;
> > >          }
> > >  
> > > -        /* EPOW and hotplug events are surfaced in the same manner */
> > > -        if (entry->log_type == RTAS_LOG_TYPE_EPOW ||
> > > -            entry->log_type == RTAS_LOG_TYPE_HOTPLUG) {
> > > +        if (source->mask & event_mask) {
> > >              break;
> > >          }
> > >      }
> > > @@ -277,19 +349,14 @@ static bool rtas_event_log_contains(uint32_t event_mask, bool exception)
> > >      sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
> > >      sPAPREventLogEntry *entry = NULL;
> > >  
> > > -    /* we only queue EPOW events atm. */
> > > -    if ((event_mask & EVENT_MASK_EPOW) == 0) {
> > > -        return false;
> > > -    }
> > > -
> > >      QTAILQ_FOREACH(entry, &spapr->pending_events, next) {
> > > +        const EventSource *source = rtas_event_log_to_source(entry->log_type);
> > > +
> > >          if (entry->exception != exception) {
> > >              continue;
> > >          }
> > >  
> > > -        /* EPOW and hotplug events are surfaced in the same manner */
> > > -        if (entry->log_type == RTAS_LOG_TYPE_EPOW ||
> > > -            entry->log_type == RTAS_LOG_TYPE_HOTPLUG) {
> > > +        if (source->mask & event_mask) {
> > >              return true;
> > >          }
> > >      }
> > > @@ -377,7 +444,8 @@ static void spapr_powerdown_req(Notifier *n, void *opaque)
> > >  
> > >      rtas_event_log_queue(RTAS_LOG_TYPE_EPOW, new_epow, true);
> > >  
> > > -    qemu_irq_pulse(xics_get_qirq(spapr->xics, spapr->check_exception_irq));
> > > +    qemu_irq_pulse(xics_get_qirq(spapr->xics,
> > > +                                 rtas_event_log_to_irq(RTAS_LOG_TYPE_EPOW)));
> > >  }
> > >  
> > >  static void spapr_hotplug_set_signalled(uint32_t drc_index)
> > > @@ -459,7 +527,8 @@ static void spapr_hotplug_req_event(uint8_t hp_id, uint8_t hp_action,
> > >  
> > >      rtas_event_log_queue(RTAS_LOG_TYPE_HOTPLUG, new_hp, true);
> > >  
> > > -    qemu_irq_pulse(xics_get_qirq(spapr->xics, spapr->check_exception_irq));
> > > +    qemu_irq_pulse(xics_get_qirq(spapr->xics,
> > > +                                 rtas_event_log_to_irq(RTAS_LOG_TYPE_HOTPLUG)));
> > >  }
> > >  
> > >  void spapr_hotplug_req_add_by_index(sPAPRDRConnector *drc)
> > > @@ -505,6 +574,7 @@ static void check_exception(PowerPCCPU *cpu, sPAPRMachineState *spapr,
> > >      uint64_t xinfo;
> > >      sPAPREventLogEntry *event;
> > >      struct rtas_error_log *hdr;
> > > +    int i;
> > >  
> > >      if ((nargs < 6) || (nargs > 7) || nret != 1) {
> > >          rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
> > > @@ -541,8 +611,11 @@ static void check_exception(PowerPCCPU *cpu, sPAPRMachineState *spapr,
> > >       * do the latter here, since our code relies on edge-triggered
> > >       * interrupts.
> > >       */
> > > -    if (rtas_event_log_contains(mask, true)) {
> > > -        qemu_irq_pulse(xics_get_qirq(spapr->xics, spapr->check_exception_irq));
> > > +    for (i = 0; i < EVENT_CLASS_MAX; i++) {
> > > +        if (rtas_event_log_contains(EVENT_CLASS_MASK(i), true)) {
> > > +            g_assert(event_source[i].enabled);
> > > +            qemu_irq_pulse(xics_get_qirq(spapr->xics, event_source[i].irq));
> > > +        }
> > >      }
> > >  
> > >      return;
> > > @@ -594,8 +667,17 @@ out_no_events:
> > >  void spapr_events_init(sPAPRMachineState *spapr)
> > >  {
> > >      QTAILQ_INIT(&spapr->pending_events);
> > > -    spapr->check_exception_irq = xics_spapr_alloc(spapr->xics, 0, 0, false,
> > > -                                            &error_fatal);
> > > +
> > > +    rtas_event_source_register(EVENT_CLASS_EPOW,
> > > +                               xics_spapr_alloc(spapr->xics, 0, 0, false,
> > > +                                                &error_fatal));
> > > +
> > > +    if (spapr->use_hotplug_event_source) {
> > > +        rtas_event_source_register(EVENT_CLASS_HOT_PLUG,
> > > +                                   xics_spapr_alloc(spapr->xics, 0, 0, false,
> > > +                                                    &error_fatal));
> > > +    }
> > > +
> > >      spapr->epow_notifier.notify = spapr_powerdown_req;
> > >      qemu_register_powerdown_notifier(&spapr->epow_notifier);
> > >      spapr_rtas_register(RTAS_CHECK_EXCEPTION, "check-exception",
> > > diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
> > > index d1a4a14..2295ac6 100644
> > > --- a/include/hw/ppc/spapr.h
> > > +++ b/include/hw/ppc/spapr.h
> > > @@ -71,7 +71,6 @@ struct sPAPRMachineState {
> > >      sPAPROptionVector *ov5_cas;
> > >      bool cas_reboot;
> > >  
> > > -    uint32_t check_exception_irq;
> > >      Notifier epow_notifier;
> > >      QTAILQ_HEAD(, sPAPREventLogEntry) pending_events;
> > >      bool use_hotplug_event_source;
> > > @@ -579,7 +578,7 @@ struct sPAPREventLogEntry {
> > >  };
> > >  
> > >  void spapr_events_init(sPAPRMachineState *sm);
> > > -void spapr_events_fdt_skel(void *fdt, uint32_t epow_irq);
> > > +void spapr_events_fdt_skel(void *fdt);
> > >  int spapr_h_cas_compose_response(sPAPRMachineState *sm,
> > >                                   target_ulong addr, target_ulong size,
> > >                                   bool cpu_update,
> > 
> 

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 819 bytes --]