All of lore.kernel.org
 help / color / mirror / Atom feed
From: Michael Roth <mdroth@linux.vnet.ibm.com>
To: Alexey Kardashevskiy <aik@ozlabs.ru>, qemu-devel@nongnu.org
Cc: ncmike@ncultra.org, nfont@linux.vnet.ibm.com,
	qemu-ppc@nongnu.org, agraf@suse.de, tyreld@linux.vnet.ibm.com
Subject: Re: [Qemu-devel] [PATCH 01/12] spapr: populate DRC entries for root dt node
Date: Tue, 26 Aug 2014 09:56:51 -0500	[thread overview]
Message-ID: <20140826145651.21832.79008@loki> (raw)
In-Reply-To: <53FC3D59.1020702@ozlabs.ru>

Quoting Alexey Kardashevskiy (2014-08-26 02:55:05)
> On 08/19/2014 10:21 AM, Michael Roth wrote:
> > From: Nathan Fontenot <nfont@linux.vnet.ibm.com>
> > 
> > This add entries to the root OF node to advertise our PHBs as being
> > DR-capable in according with PAPR specification.
> > 
> > Each PHB is given a name of PHB<bus#>, advertised as a PHB type,
> > and associated with a power domain of -1 (indicating to guests that
> > power management is handled automatically by hardware).
> > 
> > We currently allocate entries for up to 32 DR-capable PHBs, though
> > this limit can be increased later.
> > 
> > DrcEntry objects to track the state of the DR-connector associated
> > with each PHB are stored in a 32-entry array, and each DrcEntry has
> > in turn have a dynamically-sized number of child DR-connectors,
> > which we will use later to track the state of DR-connectors
> > associated with a PHB's physical slots.
> > 
> > Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com>
> > Signed-off-by: Michael Roth <mdroth@linux.vnet.ibm.com>
> > ---
> >  hw/ppc/spapr.c         | 143 +++++++++++++++++++++++++++++++++++++++++++++++++
> >  hw/ppc/spapr_pci.c     |   1 +
> >  include/hw/ppc/spapr.h |  35 ++++++++++++
> >  3 files changed, 179 insertions(+)
> > 
> > diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> > index 5c92707..d5e46c3 100644
> > --- a/hw/ppc/spapr.c
> > +++ b/hw/ppc/spapr.c
> > @@ -296,6 +296,143 @@ static hwaddr spapr_node0_size(void)
> >      return ram_size;
> >  }
> >  
> > +sPAPRDrcEntry *spapr_phb_to_drc_entry(uint64_t buid)
> > +{
> > +    int i;
> > +
> > +    for (i = 0; i < SPAPR_DRC_TABLE_SIZE; i++) {
> > +        if (spapr->drc_table[i].phb_buid == buid) {
> > +            return &spapr->drc_table[i];
> > +        }
> > +     }
> > +
> > +     return NULL;
> > +}
> > +
> > +static void spapr_init_drc_table(void)
> > +{
> > +    int i;
> > +
> > +    memset(spapr->drc_table, 0, sizeof(spapr->drc_table));
> > +
> > +    /* For now we only care about PHB entries */
> > +    for (i = 0; i < SPAPR_DRC_TABLE_SIZE; i++) {
> > +        spapr->drc_table[i].drc_index = 0x2000001 + i;
> > +    }
> > +}
> > +
> > +sPAPRDrcEntry *spapr_add_phb_to_drc_table(uint64_t buid, uint32_t state)
> > +{
> > +    sPAPRDrcEntry *empty_drc = NULL;
> > +    sPAPRDrcEntry *found_drc = NULL;
> > +    int i, phb_index;
> > +
> > +    for (i = 0; i < SPAPR_DRC_TABLE_SIZE; i++) {
> > +        if (spapr->drc_table[i].phb_buid == 0) {
> > +            empty_drc = &spapr->drc_table[i];
> > +        }
> > +
> > +        if (spapr->drc_table[i].phb_buid == buid) {
> > +            found_drc = &spapr->drc_table[i];
> 
> return &spapr->drc_table[i];
> ?

That makes sense. Looking at this again though I think maybe
I should drop the found_drc stuff completely, or maybe even
assert if we attempt to re-add a phb. Current callers already
do spapr_phb_to_drc_entry beforehand anyway, which should
cover the case where it's already there. So maybe something
like this?

sPAPRDrcEntry *spapr_add_phb_to_drc_table(uint64_t buid, uint32_t state)
{
    sPAPRDrcEntry *empty_drc = NULL;
    int i, phb_index;

    for (i = 0; i < SPAPR_DRC_TABLE_SIZE; i++) {
        g_assert(spapr->drc_table[i].phb_buid != buid);
        if (spapr->drc_table[i].phb_buid == 0) {
            empty_drc = &spapr->drc_table[i];
            break;
        }
    }

    if (!empty_drc) {
        return NULL;
    }

    empty_drc->phb_buid = buid;
    empty_drc->state = state;
    empty_drc->cc_state.fdt = NULL;
    empty_drc->cc_state.offset = 0;
    empty_drc->cc_state.depth = 0;
    empty_drc->cc_state.state = CC_STATE_IDLE;
    empty_drc->child_entries =
        g_malloc0(sizeof(sPAPRDrcEntry) * SPAPR_DRC_PHB_SLOT_MAX);
    phb_index = buid - SPAPR_PCI_BASE_BUID;
    for (i = 0; i < SPAPR_DRC_PHB_SLOT_MAX; i++) {
        empty_drc->child_entries[i].drc_index =
            SPAPR_DRC_DEV_ID_BASE + (phb_index << 8) + (i << 3);
    }

    return empty_drc;
}

> 
> 
> > +            break;
> > +        }
> > +    }
> > +
> > +    if (found_drc) {
> > +        return found_drc;
> > +    }
> 
>    if (!empty_drc) {
>         return NULL;
>    }
> 
> ?
> 
> 
> > +
> > +    if (empty_drc) {
> 
> and no need in this :)
> 
> 
> > +        empty_drc->phb_buid = buid;
> > +        empty_drc->state = state;
> > +        empty_drc->cc_state.fdt = NULL;
> > +        empty_drc->cc_state.offset = 0;
> > +        empty_drc->cc_state.depth = 0;
> > +        empty_drc->cc_state.state = CC_STATE_IDLE;
> > +        empty_drc->child_entries =
> > +            g_malloc0(sizeof(sPAPRDrcEntry) * SPAPR_DRC_PHB_SLOT_MAX);
> > +        phb_index = buid - SPAPR_PCI_BASE_BUID;
> > +        for (i = 0; i < SPAPR_DRC_PHB_SLOT_MAX; i++) {
> > +            empty_drc->child_entries[i].drc_index =
> > +                SPAPR_DRC_DEV_ID_BASE + (phb_index << 8) + (i << 3);
> > +        }
> > +        return empty_drc;
> > +    }
> > +
> > +    return NULL;
> > +}
> > +
> > +static void spapr_create_drc_dt_entries(void *fdt)
> > +{
> > +    char char_buf[1024];
> > +    uint32_t int_buf[SPAPR_DRC_TABLE_SIZE + 1];
> > +    uint32_t *entries;
> > +    int offset, fdt_offset;
> > +    int i, ret;
> > +
> > +    fdt_offset = fdt_path_offset(fdt, "/");
> > +
> > +    /* ibm,drc-indexes */
> > +    memset(int_buf, 0, sizeof(int_buf));
> > +    int_buf[0] = SPAPR_DRC_TABLE_SIZE;
> > +
> > +    for (i = 1; i <= SPAPR_DRC_TABLE_SIZE; i++) {
> > +        int_buf[i] = spapr->drc_table[i-1].drc_index;
> > +    }
> > +
> > +    ret = fdt_setprop(fdt, fdt_offset, "ibm,drc-indexes", int_buf,
> > +                      sizeof(int_buf));
> > +    if (ret) {
> > +        fprintf(stderr, "Couldn't finalize ibm,drc-indexes property\n");
> 
> return here and below in the same error cases?

Yah, that's clearly more sensible. I suppose if we introduce an error exit case
this should stop being a void function as well.

> 
> > +    }
> > +
> > +    /* ibm,drc-power-domains */
> > +    memset(int_buf, 0, sizeof(int_buf));
> > +    int_buf[0] = SPAPR_DRC_TABLE_SIZE;
> > +
> > +    for (i = 1; i <= SPAPR_DRC_TABLE_SIZE; i++) {
> > +        int_buf[i] = 0xffffffff;
> > +    }
> > +
> > +    ret = fdt_setprop(fdt, fdt_offset, "ibm,drc-power-domains", int_buf,
> > +                      sizeof(int_buf));
> > +    if (ret) {
> > +        fprintf(stderr, "Couldn't finalize ibm,drc-power-domains property\n");
> > +    }
> > +
> > +    /* ibm,drc-names */
> > +    memset(char_buf, 0, sizeof(char_buf));
> > +    entries = (uint32_t *)&char_buf[0];
> > +    *entries = SPAPR_DRC_TABLE_SIZE;
> > +    offset = sizeof(*entries);
> > +
> > +    for (i = 0; i < SPAPR_DRC_TABLE_SIZE; i++) {
> > +        offset += sprintf(char_buf + offset, "PHB %d", i + 1);
> > +        char_buf[offset++] = '\0';
> > +    }
> > +
> > +    ret = fdt_setprop(fdt, fdt_offset, "ibm,drc-names", char_buf, offset);
> > +    if (ret) {
> > +        fprintf(stderr, "Couldn't finalize ibm,drc-names property\n");
> > +    }
> > +
> > +    /* ibm,drc-types */
> > +    memset(char_buf, 0, sizeof(char_buf));
> > +    entries = (uint32_t *)&char_buf[0];
> > +    *entries = SPAPR_DRC_TABLE_SIZE;
> > +    offset = sizeof(*entries);
> > +
> > +    for (i = 0; i < SPAPR_DRC_TABLE_SIZE; i++) {
> > +        offset += sprintf(char_buf + offset, "PHB");
> > +        char_buf[offset++] = '\0';
> > +    }
> > +
> > +    ret = fdt_setprop(fdt, fdt_offset, "ibm,drc-types", char_buf, offset);
> > +    if (ret) {
> > +        fprintf(stderr, "Couldn't finalize ibm,drc-types property\n");
> > +    }
> > +}
> > +
> >  #define _FDT(exp) \
> >      do { \
> >          int ret = (exp);                                           \
> > @@ -731,6 +868,7 @@ static void spapr_finalize_fdt(sPAPREnvironment *spapr,
> >      char *bootlist;
> >      void *fdt;
> >      sPAPRPHBState *phb;
> > +    sPAPRDrcEntry *drc_entry;
> >  
> >      fdt = g_malloc(FDT_MAX_SIZE);
> >  
> > @@ -750,6 +888,8 @@ static void spapr_finalize_fdt(sPAPREnvironment *spapr,
> >      }
> >  
> >      QLIST_FOREACH(phb, &spapr->phbs, list) {
> > +        drc_entry = spapr_phb_to_drc_entry(phb->buid);
> > +        g_assert(drc_entry);
> >          ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
> >      }
> >  
> > @@ -789,6 +929,8 @@ static void spapr_finalize_fdt(sPAPREnvironment *spapr,
> >          spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
> >      }
> >  
> > +    spapr_create_drc_dt_entries(fdt);
> > +
> >      _FDT((fdt_pack(fdt)));
> >  
> >      if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
> > @@ -1443,6 +1585,7 @@ static void ppc_spapr_init(MachineState *machine)
> >      spapr_pci_msi_init(spapr, SPAPR_PCI_MSI_WINDOW);
> >      spapr_pci_rtas_init();
> >  
> > +    spapr_init_drc_table();
> >      phb = spapr_create_phb(spapr, 0);
> >  
> >      for (i = 0; i < nb_nics; i++) {
> > diff --git a/hw/ppc/spapr_pci.c b/hw/ppc/spapr_pci.c
> > index 9ed39a9..e85134f 100644
> > --- a/hw/ppc/spapr_pci.c
> > +++ b/hw/ppc/spapr_pci.c
> > @@ -531,6 +531,7 @@ static void spapr_phb_realize(DeviceState *dev, Error **errp)
> >              + sphb->index * SPAPR_PCI_WINDOW_SPACING;
> >          sphb->mem_win_addr = windows_base + SPAPR_PCI_MMIO_WIN_OFF;
> >          sphb->io_win_addr = windows_base + SPAPR_PCI_IO_WIN_OFF;
> > +        spapr_add_phb_to_drc_table(sphb->buid, 2 /* Unusable */);
> 
> 
> What exactly does "unusable" mean here? Macro?

That's the 9003/"DR-entity-sense" for the DRC entry corresponding to the PHB
itself (as opposed to the sensors for each of its slots). In the case of the
slots, we advertise them as 'physical' DR resources in the PHB's "ibm,drc-types"
property. In the case of the PHBs we advertise them as 'logical'/dlpar DR
resources in the root DT node's "ibm,drc-types" property. We do not actually
support 'logical' DR operations in this implementation though (though we may
in the future to support PHB hotplug), so we've set this to 'unusable' for
now.

But according to PAPR 2.7 13.5.3.3 this corresponds to:

"Returned for logical DR entities when the DR entity is not currently available
to the OS, but may possibly be made available to the OS by calling set-indicator
with the allocation-state indicator, setting that indicator to usable."

So maybe it makes more sense to just set it to present/(1)?

I don't think the PHB sensors will actually get read unless we attempt dlpar
operations in the guest (as opposed to pci hp), so it's probably mostly unused
now, but seems like a more sensible default. Will test and change it if it
doesn't break anything.

Macros make sense... we actually already have:

#define INDICATOR_ENTITY_SENSE_EMPTY    0
#define INDICATOR_ENTITY_SENSE_PRESENT  1

so not adding 'unused' was an oversight. I'll add it either way for
completeness.

> 
> 
> 
> >      }
> >  
> >      if (sphb->buid == -1) {
> > diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
> > index 36e8e51..c93794b 100644
> > --- a/include/hw/ppc/spapr.h
> > +++ b/include/hw/ppc/spapr.h
> > @@ -10,6 +10,36 @@ struct sPAPRNVRAM;
> >  
> >  #define HPTE64_V_HPTE_DIRTY     0x0000000000000040ULL
> >  
> > +/* For dlparable/hotpluggable slots */
> > +#define SPAPR_DRC_TABLE_SIZE    32
> > +#define SPAPR_DRC_PHB_SLOT_MAX  32
> > +#define SPAPR_DRC_DEV_ID_BASE   0x40000000
> 
> 
> Is this SPAPR_DRC_DEV_ID_BASE really necessary (can it be zero)? Is that
> global id or per PCI bus or per PHB?
> 
> 
> > +
> > +typedef struct sPAPRConfigureConnectorState {
> > +    void *fdt;
> > +    int offset_start;
> > +    int offset;
> > +    int depth;
> > +    PCIDevice *dev;
> > +    enum {
> > +        CC_STATE_IDLE = 0,
> > +        CC_STATE_PENDING = 1,
> > +        CC_STATE_ACTIVE,
> > +    } state;
> > +} sPAPRConfigureConnectorState;
> > +
> > +typedef struct sPAPRDrcEntry sPAPRDrcEntry;
> > +
> > +struct sPAPRDrcEntry {
> > +    uint32_t drc_index;
> > +    uint64_t phb_buid;
> > +    void *fdt;
> > +    int fdt_offset;
> > +    uint32_t state;
> > +    sPAPRConfigureConnectorState cc_state;
> > +    sPAPRDrcEntry *child_entries;
> > +};
> > +
> >  typedef struct sPAPREnvironment {
> >      struct VIOsPAPRBus *vio_bus;
> >      QLIST_HEAD(, sPAPRPHBState) phbs;
> > @@ -39,6 +69,9 @@ typedef struct sPAPREnvironment {
> >      int htab_save_index;
> >      bool htab_first_pass;
> >      int htab_fd;
> > +
> > +    /* state for Dynamic Reconfiguration Connectors */
> > +    sPAPRDrcEntry drc_table[SPAPR_DRC_TABLE_SIZE];
> >  } sPAPREnvironment;
> >  
> >  #define H_SUCCESS         0
> > @@ -481,5 +514,7 @@ int spapr_dma_dt(void *fdt, int node_off, const char *propname,
> >                   uint32_t liobn, uint64_t window, uint32_t size);
> >  int spapr_tcet_dma_dt(void *fdt, int node_off, const char *propname,
> >                        sPAPRTCETable *tcet);
> > +sPAPRDrcEntry *spapr_add_phb_to_drc_table(uint64_t buid, uint32_t state);
> > +sPAPRDrcEntry *spapr_phb_to_drc_entry(uint64_t buid);
> >  
> >  #endif /* !defined (__HW_SPAPR_H__) */
> > 
> 
> 
> -- 
> Alexey

  parent reply	other threads:[~2014-08-26 14:57 UTC|newest]

Thread overview: 69+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-08-19  0:21 [Qemu-devel] [PATCH v3 00/12] spapr: add support for pci hotplug Michael Roth
2014-08-19  0:21 ` [Qemu-devel] [PATCH 01/12] spapr: populate DRC entries for root dt node Michael Roth
2014-08-26  7:55   ` Alexey Kardashevskiy
2014-08-26  8:24     ` Alexey Kardashevskiy
2014-08-26 15:25       ` Michael Roth
2014-08-26 15:41         ` Michael Roth
2014-08-29 18:27         ` Tyrel Datwyler
2014-08-29 23:15           ` Alexander Graf
2014-08-26 14:56     ` Michael Roth [this message]
2014-09-05  0:31     ` [Qemu-devel] [Qemu-ppc] " Tyrel Datwyler
2014-08-26 11:11   ` [Qemu-devel] " Alexander Graf
2014-08-26 16:47     ` Michael Roth
2014-08-26 17:16       ` Alexander Graf
2014-09-03  5:55   ` Bharata B Rao
2014-09-05 22:00   ` Tyrel Datwyler
2014-08-19  0:21 ` [Qemu-devel] [PATCH 02/12] spapr_pci: populate DRC dt entries for PHBs Michael Roth
2014-08-26  8:32   ` Alexey Kardashevskiy
2014-08-26 17:16     ` Michael Roth
2014-08-26  9:09   ` Alexey Kardashevskiy
2014-08-26 17:52     ` Michael Roth
2014-08-26 11:29   ` Alexander Graf
2014-08-26 18:30     ` Michael Roth
2014-08-19  0:21 ` [Qemu-devel] [PATCH 03/12] spapr: add helper to retrieve a PHB/device DrcEntry Michael Roth
2014-08-19  0:21 ` [Qemu-devel] [PATCH 04/12] spapr_pci: add set-indicator RTAS interface Michael Roth
2014-08-26 11:36   ` Alexander Graf
2014-09-05  2:55     ` Nathan Fontenot
2014-09-30 22:08     ` Michael Roth
2014-10-01 14:30       ` Alexander Graf
2014-11-26  4:51         ` Bharata B Rao
2014-11-26  4:54         ` Bharata B Rao
2014-11-26  6:27           ` Michael Roth
2014-12-01  4:57             ` Bharata B Rao
2014-12-23 15:12               ` Michael Roth
2015-01-01  6:35                 ` Bharata B Rao
2014-08-19  0:21 ` [Qemu-devel] [PATCH 05/12] spapr_pci: add get/set-power-level RTAS interfaces Michael Roth
2014-08-19  0:21 ` [Qemu-devel] [PATCH 06/12] spapr_pci: add get-sensor-state RTAS interface Michael Roth
2014-09-05  0:34   ` Tyrel Datwyler
2014-08-19  0:21 ` [Qemu-devel] [PATCH 07/12] spapr_pci: add ibm, configure-connector " Michael Roth
2014-08-26  9:12   ` Alexey Kardashevskiy
2014-09-05  3:03     ` Nathan Fontenot
2014-08-26 11:39   ` Alexander Graf
2014-08-19  0:21 ` [Qemu-devel] [PATCH 08/12] pci: allow 0 address for PCI IO regions Michael Roth
2014-08-26  9:14   ` Alexey Kardashevskiy
2014-08-26 11:55     ` Peter Maydell
2014-08-26 18:34     ` Michael Roth
2014-08-26 11:41   ` Alexander Graf
2014-08-27 13:47   ` Michael S. Tsirkin
2014-08-28 21:21     ` Michael Roth
2014-08-28 21:33       ` Peter Maydell
2014-08-28 21:46         ` Michael S. Tsirkin
2014-08-19  0:21 ` [Qemu-devel] [PATCH 09/12] spapr_pci: enable basic hotplug operations Michael Roth
2014-08-26  9:40   ` Alexey Kardashevskiy
2014-08-26 12:30   ` Alexander Graf
2014-09-03 10:33   ` Bharata B Rao
2014-09-03 23:03     ` Michael Roth
2014-09-04 15:08       ` Bharata B Rao
2014-09-04 16:12         ` Michael Roth
2014-09-04 16:34           ` Michael Roth
2014-09-05  3:10             ` Nathan Fontenot
2014-09-05 17:17               ` [Qemu-devel] [Qemu-ppc] " Tyrel Datwyler
2014-08-19  0:21 ` [Qemu-devel] [PATCH 10/12] spapr_events: re-use EPOW event infrastructure for hotplug events Michael Roth
2014-08-26  9:28   ` Alexey Kardashevskiy
2014-08-19  0:21 ` [Qemu-devel] [PATCH 11/12] spapr_events: event-scan RTAS interface Michael Roth
2014-08-26  9:30   ` Alexey Kardashevskiy
2014-08-29 18:43     ` Tyrel Datwyler
2014-08-19  0:21 ` [Qemu-devel] [PATCH 12/12] spapr_pci: emit hotplug add/remove events during hotplug Michael Roth
2014-08-26  9:35   ` Alexey Kardashevskiy
2014-08-26 12:36   ` Alexander Graf
2014-08-26  9:24 ` [Qemu-devel] [PATCH v3 00/12] spapr: add support for pci hotplug Alexey Kardashevskiy

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20140826145651.21832.79008@loki \
    --to=mdroth@linux.vnet.ibm.com \
    --cc=agraf@suse.de \
    --cc=aik@ozlabs.ru \
    --cc=ncmike@ncultra.org \
    --cc=nfont@linux.vnet.ibm.com \
    --cc=qemu-devel@nongnu.org \
    --cc=qemu-ppc@nongnu.org \
    --cc=tyreld@linux.vnet.ibm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.