/*
* ppb_save_config_regs
*
* This routine saves the state of the configuration registers of all
* the child nodes of each PBM.
*
* used by: ppb_detach() on suspends
*
* return value: none
*/
static void
ppb_save_config_regs(ppb_devstate_t *ppb_p)
{
int i;
dev_info_t *dip;
ddi_acc_handle_t config_handle;
for (i = 0, dip = ddi_get_child(ppb_p->dip); dip != NULL;
i++, dip = ddi_get_next_sibling(dip)) {
if (pci_config_setup(dip, &config_handle) != DDI_SUCCESS) {
cmn_err(CE_WARN, "%s%d: can't config space for %s%d\n",
ddi_driver_name(ppb_p->dip),
ddi_get_instance(ppb_p->dip),
ddi_driver_name(dip),
ddi_get_instance(dip));
continue;
}
ppb_p->config_state[i].dip = dip;
ppb_p->config_state[i].command =
pci_config_get16(config_handle, PCI_CONF_COMM);
pci_config_teardown(&config_handle);
}
ppb_p->config_state_index = i;
}
void
npe_enable_htmsi_children(dev_info_t *dip)
{
dev_info_t *cdip = ddi_get_child(dip);
ddi_acc_handle_t cfg_hdl;
if (!npe_enable_htmsi_flag)
return;
/*
* Hypertransport MSI remapping only applies to AMD CPUs using
* Hypertransport (K8 and above) and not other platforms with non-AMD
* CPUs that may be using Hypertransport internally in the chipset(s)
*/
if (!(cpuid_getvendor(CPU) == X86_VENDOR_AMD &&
cpuid_getfamily(CPU) >= 0xf))
return;
for (; cdip != NULL; cdip = ddi_get_next_sibling(cdip)) {
if (pci_config_setup(cdip, &cfg_hdl) != DDI_SUCCESS) {
cmn_err(CE_NOTE, "!npe_enable_htmsi_children: "
"pci_config_setup failed for %s",
ddi_node_name(cdip));
return;
}
(void) npe_enable_htmsi(cfg_hdl);
pci_config_teardown(&cfg_hdl);
}
}
static int
fco_peer(dev_info_t *ap, fco_handle_t rp, fc_ci_t *cp)
{
fc_phandle_t h;
dev_info_t *dip;
if (fc_cell2int(cp->nargs) != 1)
return (fc_syntax_error(cp, "nargs must be 1"));
if (fc_cell2int(cp->nresults) < 1)
return (fc_syntax_error(cp, "nresults must be > 0"));
/*
* Make sure this is a handle we gave out ...
*/
h = fc_cell2phandle(fc_arg(cp, 0));
if ((dip = fc_phandle_to_dip(fc_handle_to_phandle_head(rp), h)) == NULL)
return (fc_priv_error(cp, "unknown handle"));
/*
* Find the child and if there is one, return it ...
*/
dip = ddi_get_next_sibling(dip);
h = 0;
if (dip != NULL)
h = fc_dip_to_phandle(fc_handle_to_phandle_head(rp), dip);
cp->nresults = fc_int2cell(1);
fc_result(cp, 0) = fc_phandle2cell(h);
return (fc_success_op(ap, rp, cp));
}
/*
* Starting from the root node suspend all devices in the device tree.
* Assumes that all devices have already been marked busy.
*/
static int
sbdp_suspend_devices_(dev_info_t *dip, sbdp_sr_handle_t *srh)
{
major_t major;
char *dname;
for (; dip != NULL; dip = ddi_get_next_sibling(dip)) {
char d_name[40], d_alias[40], *d_info;
if (sbdp_suspend_devices_(ddi_get_child(dip), srh)) {
return (ENXIO);
}
if (!sbdp_is_real_device(dip))
continue;
major = (major_t)-1;
if ((dname = DEVI(dip)->devi_binding_name) != NULL)
major = ddi_name_to_major(dname);
#ifdef DEBUG
if (sbdp_bypass_device(dname)) {
SBDP_DBG_QR("bypassed suspend of %s (major# %d)\n",
dname, major);
continue;
}
#endif
if ((d_info = ddi_get_name_addr(dip)) == NULL)
d_info = "<null>";
d_name[0] = 0;
if (sbdp_resolve_devname(dip, d_name, d_alias) == 0) {
if (d_alias[0] != 0) {
SBDP_DBG_QR("\tsuspending %s@%s (aka %s)\n",
d_name, d_info, d_alias);
} else {
SBDP_DBG_QR("\tsuspending %s@%s\n",
d_name, d_info);
}
} else {
SBDP_DBG_QR("\tsuspending %s@%s\n", dname, d_info);
}
if (devi_detach(dip, DDI_SUSPEND) != DDI_SUCCESS) {
(void) sprintf(sbdp_get_err_buf(&srh->sep),
"%d", major);
sbdp_set_err(&srh->sep, ESGT_SUSPEND, NULL);
ndi_hold_devi(dip);
SR_FAILED_DIP(srh) = dip;
return (DDI_FAILURE);
}
}
return (DDI_SUCCESS);
}
void
pcmu_child_cfg_save(dev_info_t *dip)
{
dev_info_t *cdip;
int ret = DDI_SUCCESS;
/*
* Save the state of the configuration headers of child
* nodes.
*/
for (cdip = ddi_get_child(dip); cdip != NULL;
cdip = ddi_get_next_sibling(cdip)) {
/*
* Not interested in children who are not already
* init'ed. They will be set up in pcmu_init_child().
*/
if (i_ddi_node_state(cdip) < DS_INITIALIZED) {
PCMU_DBG2(PCMU_DBG_DETACH, dip, "DDI_SUSPEND: skipping "
"%s%d not in CF1\n", ddi_driver_name(cdip),
ddi_get_instance(cdip));
continue;
}
/*
* Only save config registers if not already saved by child.
*/
if (ddi_prop_exists(DDI_DEV_T_ANY, cdip, DDI_PROP_DONTPASS,
SAVED_CONFIG_REGS) == 1) {
continue;
}
/*
* The nexus needs to save config registers. Create a property
* so it knows to restore on resume.
*/
ret = ndi_prop_create_boolean(DDI_DEV_T_NONE, cdip,
"nexus-saved-config-regs");
if (ret != DDI_PROP_SUCCESS) {
cmn_err(CE_WARN, "%s%d can't update prop %s",
ddi_driver_name(cdip), ddi_get_instance(cdip),
"nexus-saved-config-regs");
}
(void) pci_save_config_regs(cdip);
}
}
void
pcmu_child_cfg_restore(dev_info_t *dip)
{
dev_info_t *cdip;
/*
* Restore config registers for children that did not save
* their own registers. Children pwr states are UNKNOWN after
* a resume since it is possible for the PM framework to call
* resume without an actual power cycle. (ie if suspend fails).
*/
for (cdip = ddi_get_child(dip); cdip != NULL;
cdip = ddi_get_next_sibling(cdip)) {
/*
* Not interested in children who are not already
* init'ed. They will be set up by pcmu_init_child().
*/
if (i_ddi_node_state(cdip) < DS_INITIALIZED) {
PCMU_DBG2(PCMU_DBG_DETACH, dip,
"DDI_RESUME: skipping %s%d not in CF1\n",
ddi_driver_name(cdip), ddi_get_instance(cdip));
continue;
}
/*
* Only restore config registers if saved by nexus.
*/
if (ddi_prop_exists(DDI_DEV_T_ANY, cdip, DDI_PROP_DONTPASS,
"nexus-saved-config-regs") == 1) {
(void) pci_restore_config_regs(cdip);
PCMU_DBG2(PCMU_DBG_PWR, dip,
"DDI_RESUME: nexus restoring %s%d config regs\n",
ddi_driver_name(cdip), ddi_get_instance(cdip));
if (ndi_prop_remove(DDI_DEV_T_NONE, cdip,
"nexus-saved-config-regs") != DDI_PROP_SUCCESS) {
cmn_err(CE_WARN, "%s%d can't remove prop %s",
ddi_driver_name(cdip),
ddi_get_instance(cdip),
"nexus-saved-config-regs");
}
}
}
}
void
npe_enable_htmsi_children(dev_info_t *dip)
{
dev_info_t *cdip = ddi_get_child(dip);
ddi_acc_handle_t cfg_hdl;
for (; cdip != NULL; cdip = ddi_get_next_sibling(cdip)) {
if (pci_config_setup(cdip, &cfg_hdl) != DDI_SUCCESS) {
cmn_err(CE_NOTE, "!npe_enable_htmsi_children: "
"pci_config_setup failed for %s",
ddi_node_name(cdip));
}
(void) npe_enable_htmsi(cfg_hdl);
pci_config_teardown(&cfg_hdl);
}
}
void
load_platform_drivers(void)
{
dev_info_t *dip; /* dip of the isa driver */
int simba_present = 0;
dev_info_t *root_child_node;
/*
* Install Isa driver. This is required for the southbridge IDE
* workaround - to reset the IDE channel during IDE bus reset.
* Panic the system in case ISA driver could not be loaded or
* any problem in accessing its pci config space. Since the register
* to reset the channel for IDE is in ISA config space!.
*/
root_child_node = ddi_get_child(ddi_root_node());
while (root_child_node != NULL) {
if (strcmp(ddi_node_name(root_child_node), "pci") == 0) {
root_child_node = ddi_get_child(root_child_node);
if (strcmp(ddi_node_name(root_child_node), "pci") == 0)
simba_present = 1;
break;
}
root_child_node = ddi_get_next_sibling(root_child_node);
}
if (simba_present)
dip = e_ddi_hold_devi_by_path(PLATFORM_ISA_PATHNAME_WITH_SIMBA, 0);
else
dip = e_ddi_hold_devi_by_path(PLATFORM_ISA_PATHNAME, 0);
if (dip == NULL) {
cmn_err(CE_PANIC, "Could not install the isa driver\n");
return;
}
if (pci_config_setup(dip, &platform_isa_handle) != DDI_SUCCESS) {
cmn_err(CE_PANIC, "Could not get the config space of isa\n");
return;
}
}
int
npe_restore_htconfig_children(dev_info_t *dip)
{
dev_info_t *cdip = ddi_get_child(dip);
int rval = DDI_SUCCESS;
for (; cdip != NULL; cdip = ddi_get_next_sibling(cdip)) {
if (ddi_prop_get_int(DDI_DEV_T_ANY, cdip, DDI_PROP_DONTPASS,
"htconfig-saved", 0) == 0)
continue;
if (pci_restore_config_regs(cdip) != DDI_SUCCESS) {
cmn_err(CE_WARN, "Failed to restore HT config "
"regs for %s\n", ddi_node_name(cdip));
rval = DDI_FAILURE;
}
}
return (rval);
}
/*
* save config regs for HyperTransport devices without drivers of classes:
* memory controller and hostbridge
*/
int
npe_save_htconfig_children(dev_info_t *dip)
{
dev_info_t *cdip = ddi_get_child(dip);
ddi_acc_handle_t cfg_hdl;
uint16_t ptr;
int rval = DDI_SUCCESS;
uint8_t cl, scl;
for (; cdip != NULL; cdip = ddi_get_next_sibling(cdip)) {
if (ddi_driver_major(cdip) != DDI_MAJOR_T_NONE)
continue;
if (pci_config_setup(cdip, &cfg_hdl) != DDI_SUCCESS)
return (DDI_FAILURE);
cl = pci_config_get8(cfg_hdl, PCI_CONF_BASCLASS);
scl = pci_config_get8(cfg_hdl, PCI_CONF_SUBCLASS);
if (((cl == PCI_CLASS_MEM && scl == PCI_MEM_RAM) ||
(cl == PCI_CLASS_BRIDGE && scl == PCI_BRIDGE_HOST)) &&
pci_htcap_locate(cfg_hdl, 0, 0, &ptr) == DDI_SUCCESS) {
if (pci_save_config_regs(cdip) != DDI_SUCCESS) {
cmn_err(CE_WARN, "Failed to save HT config "
"regs for %s\n", ddi_node_name(cdip));
rval = DDI_FAILURE;
} else if (ddi_prop_update_int(DDI_DEV_T_NONE, cdip,
"htconfig-saved", 1) != DDI_SUCCESS) {
cmn_err(CE_WARN, "Failed to set htconfig-saved "
"property for %s\n", ddi_node_name(cdip));
rval = DDI_FAILURE;
}
}
pci_config_teardown(&cfg_hdl);
}
return (rval);
}
/*
* check for a possible substitute node. This routine searches the
* children of parent_dip, looking for a node that:
* 1. is a prom node
* 2. binds to the same major number
* 3. there is no need to verify that the unit-address information
* match since it is likely that the substitute node
* will have none (e.g. disk) - this would be the reason the
* framework rejected it in the first place.
*
* assumes parent_dip is held
*/
static dev_info_t *
find_alternate_node(dev_info_t *parent_dip, major_t major)
{
int circ;
dev_info_t *child_dip;
/* lock down parent to keep children from being removed */
ndi_devi_enter(parent_dip, &circ);
for (child_dip = ddi_get_child(parent_dip); child_dip != NULL;
child_dip = ddi_get_next_sibling(child_dip)) {
/* look for obp node with matching major */
if ((ndi_dev_is_prom_node(child_dip) != 0) &&
(ddi_driver_major(child_dip) == major)) {
ndi_hold_devi(child_dip);
break;
}
}
ndi_devi_exit(parent_dip, circ);
return (child_dip);
}
int
xen_suspend_devices(dev_info_t *dip)
{
int error;
char buf[XPV_BUFSIZE];
SUSPEND_DEBUG("xen_suspend_devices\n");
for (; dip != NULL; dip = ddi_get_next_sibling(dip)) {
if (xen_suspend_devices(ddi_get_child(dip)))
return (ENXIO);
if (ddi_get_driver(dip) == NULL)
continue;
SUSPEND_DEBUG("Suspending device %s\n", ddi_deviname(dip, buf));
ASSERT((DEVI(dip)->devi_cpr_flags & DCF_CPR_SUSPENDED) == 0);
if (!i_ddi_devi_attached(dip)) {
error = DDI_FAILURE;
} else {
error = devi_detach(dip, DDI_SUSPEND);
}
if (error == DDI_SUCCESS) {
DEVI(dip)->devi_cpr_flags |= DCF_CPR_SUSPENDED;
} else {
SUSPEND_DEBUG("WARNING: Unable to suspend device %s\n",
ddi_deviname(dip, buf));
cmn_err(CE_WARN, "Unable to suspend device %s.",
ddi_deviname(dip, buf));
cmn_err(CE_WARN, "Device is busy or does not "
"support suspend/resume.");
return (ENXIO);
}
}
return (0);
}
/*
* cnex_find_chan_dip -- Find the dip of a device that is corresponding
* to the specific channel. Below are the details on how the dip
* is derived.
*
* - In the MD, the cfg-handle is expected to be unique for
* virtual-device nodes that have the same 'name' property value.
* This value is expected to be the same as that of "reg" property
* of the corresponding OBP device node.
*
* - The value of the 'name' property of a virtual-device node
* in the MD is expected to be the same for the corresponding
* OBP device node.
*
* - Find the virtual-device node corresponding to a channel-endpoint
* by walking backwards. Then obtain the values for the 'name' and
* 'cfg-handle' properties.
*
* - Walk all the children of the cnex, find a matching dip which
* has the same 'name' and 'reg' property values.
*
* - The channels that have no corresponding device driver are
* treated as if they correspond to the cnex driver,
* that is, return cnex dip for them. This means, the
* cnex acts as an umbrella device driver. Note, this is
* for 'intrstat' statistics purposes only. As a result of this,
* the 'intrstat' shows cnex as the device that is servicing the
* interrupts corresponding to these channels.
*
* For now, only one such case is known, that is, the channels that
* are used by the "domain-services".
*/
static dev_info_t *
cnex_find_chan_dip(dev_info_t *dip, uint64_t chan_id,
md_t *mdp, mde_cookie_t mde)
{
int listsz;
int num_nodes;
int num_devs;
uint64_t cfghdl;
char *md_name;
mde_cookie_t *listp;
dev_info_t *cdip = NULL;
num_nodes = md_node_count(mdp);
ASSERT(num_nodes > 0);
listsz = num_nodes * sizeof (mde_cookie_t);
listp = (mde_cookie_t *)kmem_zalloc(listsz, KM_SLEEP);
num_devs = md_scan_dag(mdp, mde, md_find_name(mdp, "virtual-device"),
md_find_name(mdp, "back"), listp);
ASSERT(num_devs <= 1);
if (num_devs <= 0) {
DWARN("cnex_find_chan_dip:channel(0x%llx): "
"No virtual-device found\n", chan_id);
goto fdip_exit;
}
if (md_get_prop_str(mdp, listp[0], "name", &md_name) != 0) {
DWARN("cnex_find_chan_dip:channel(0x%llx): "
"name property not found\n", chan_id);
goto fdip_exit;
}
D1("cnex_find_chan_dip: channel(0x%llx): virtual-device "
"name property value = %s\n", chan_id, md_name);
if (md_get_prop_val(mdp, listp[0], "cfg-handle", &cfghdl) != 0) {
DWARN("cnex_find_chan_dip:channel(0x%llx): virtual-device's "
"cfg-handle property not found\n", chan_id);
goto fdip_exit;
}
D1("cnex_find_chan_dip:channel(0x%llx): virtual-device cfg-handle "
" property value = 0x%x\n", chan_id, cfghdl);
for (cdip = ddi_get_child(dip); cdip != NULL;
cdip = ddi_get_next_sibling(cdip)) {
int *cnex_regspec;
uint32_t reglen;
char *dev_name;
if (ddi_prop_lookup_string(DDI_DEV_T_ANY, cdip,
DDI_PROP_DONTPASS, "name",
&dev_name) != DDI_PROP_SUCCESS) {
DWARN("cnex_find_chan_dip: name property not"
" found for dip(0x%p)\n", cdip);
continue;
}
if (strcmp(md_name, dev_name) != 0) {
ddi_prop_free(dev_name);
continue;
}
ddi_prop_free(dev_name);
if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, cdip,
DDI_PROP_DONTPASS, "reg",
&cnex_regspec, ®len) != DDI_SUCCESS) {
DWARN("cnex_find_chan_dip: reg property not"
" found for dip(0x%p)\n", cdip);
continue;
}
if (*cnex_regspec == cfghdl) {
D1("cnex_find_chan_dip:channel(0x%llx): found "
"dip(0x%p) drvname=%s\n", chan_id, cdip,
ddi_driver_name(cdip));
ddi_prop_free(cnex_regspec);
break;
}
ddi_prop_free(cnex_regspec);
}
fdip_exit:
if (cdip == NULL) {
/*
* If a virtual-device node exists but no dip found,
* then for now print a DEBUG error message only.
*/
if (num_devs > 0) {
DERR("cnex_find_chan_dip:channel(0x%llx): "
"No device found\n", chan_id);
}
/* If no dip was found, return cnex device's dip. */
cdip = dip;
}
kmem_free(listp, listsz);
D1("cnex_find_chan_dip:channel(0x%llx): returning dip=0x%p\n",
chan_id, cdip);
return (cdip);
}
/*
* Build the reserved ISA irq list, and store it in the table pointed to by
* reserved_irqs_table. The caller is responsible for allocating this table
* with a minimum of MAX_ISA_IRQ + 1 entries.
*
* The routine looks in the device tree at the subtree rooted at /isa
* for each of the devices under that node, if an interrupts property
* is present, its values are used to "reserve" irqs so that later ACPI
* configuration won't choose those irqs.
*
* In addition, if acpi_irq_check_elcr is set, will use ELCR register
* to identify reserved IRQs.
*/
void
build_reserved_irqlist(uchar_t *reserved_irqs_table)
{
dev_info_t *isanode = ddi_find_devinfo("isa", -1, 0);
dev_info_t *isa_child = 0;
int i;
uint_t elcrval;
/* Initialize the reserved ISA IRQs: */
for (i = 0; i <= MAX_ISA_IRQ; i++)
reserved_irqs_table[i] = 0;
if (acpi_irq_check_elcr) {
elcrval = (inb(ELCR_PORT2) << 8) | (inb(ELCR_PORT1));
if (ELCR_EDGE(elcrval, 0) && ELCR_EDGE(elcrval, 1) &&
ELCR_EDGE(elcrval, 2) && ELCR_EDGE(elcrval, 8) &&
ELCR_EDGE(elcrval, 13)) {
/* valid ELCR */
for (i = 0; i <= MAX_ISA_IRQ; i++)
if (!ELCR_LEVEL(elcrval, i))
reserved_irqs_table[i] = 1;
}
}
/* always check the isa devinfo nodes */
if (isanode != 0) { /* Found ISA */
uint_t intcnt; /* Interrupt count */
int *intrs; /* Interrupt values */
/* Load first child: */
isa_child = ddi_get_child(isanode);
while (isa_child != 0) { /* Iterate over /isa children */
/* if child has any interrupts, save them */
if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, isa_child,
DDI_PROP_DONTPASS, "interrupts", &intrs, &intcnt)
== DDI_PROP_SUCCESS) {
/*
* iterate over child interrupt list, adding
* them to the reserved irq list
*/
while (intcnt-- > 0) {
/*
* Each value MUST be <= MAX_ISA_IRQ
*/
if ((intrs[intcnt] > MAX_ISA_IRQ) ||
(intrs[intcnt] < 0))
continue;
reserved_irqs_table[intrs[intcnt]] = 1;
}
ddi_prop_free(intrs);
}
isa_child = ddi_get_next_sibling(isa_child);
}
/* The isa node was held by ddi_find_devinfo, so release it */
ndi_rele_devi(isanode);
}
/*
* Reserve IRQ14 & IRQ15 for IDE. It shouldn't be hard-coded
* here but there's no other way to find the irqs for
* legacy-mode ata (since it's hard-coded in pci-ide also).
*/
reserved_irqs_table[14] = 1;
reserved_irqs_table[15] = 1;
}
static void
sbdp_resume_devices(dev_info_t *start, sbdp_sr_handle_t *srh)
{
int circ;
dev_info_t *dip, *next, *last = NULL;
char *bn;
sbd_error_t *sep;
sep = &srh->sep;
/* attach in reverse device tree order */
while (last != start) {
dip = start;
next = ddi_get_next_sibling(dip);
while (next != last && dip != SR_FAILED_DIP(srh)) {
dip = next;
next = ddi_get_next_sibling(dip);
}
if (dip == SR_FAILED_DIP(srh)) {
/* Release hold acquired in sbdp_suspend_devices() */
ndi_rele_devi(dip);
SR_FAILED_DIP(srh) = NULL;
} else if (sbdp_is_real_device(dip) &&
SR_FAILED_DIP(srh) == NULL) {
if (DEVI(dip)->devi_binding_name != NULL) {
bn = ddi_binding_name(dip);
}
#ifdef DEBUG
if (!sbdp_bypass_device(bn)) {
#else
{
#endif
char d_name[40], d_alias[40], *d_info;
d_name[0] = 0;
d_info = ddi_get_name_addr(dip);
if (d_info == NULL)
d_info = "<null>";
if (!sbdp_resolve_devname(dip, d_name,
d_alias)) {
if (d_alias[0] != 0) {
SBDP_DBG_QR("\tresuming "
"%s@%s (aka %s)\n",
d_name, d_info,
d_alias);
} else {
SBDP_DBG_QR("\tresuming "
"%s@%s\n",
d_name, d_info);
}
} else {
SBDP_DBG_QR("\tresuming %s@%s\n",
bn, d_info);
}
if (devi_attach(dip, DDI_RESUME) !=
DDI_SUCCESS) {
/*
* Print a console warning,
* set an errno of ESGT_RESUME,
* and save the driver major
* number in the e_str.
*/
(void) sprintf(sbdp_get_err_buf(sep),
"%s@%s",
d_name[0] ? d_name : bn, d_info);
SBDP_DBG_QR("\tFAILED to resume "
"%s\n", sbdp_get_err_buf(sep));
sbdp_set_err(sep,
ESGT_RESUME, NULL);
}
}
}
ndi_devi_enter(dip, &circ);
sbdp_resume_devices(ddi_get_child(dip), srh);
ndi_devi_exit(dip, circ);
last = dip;
}
}
/*
* True if thread is virtually stopped. Similar to CPR_VSTOPPED
* but from DR point of view. These user threads are waiting in
* the kernel. Once they return from kernel, they will process
* the stop signal and stop.
*/
#define SBDP_VSTOPPED(t) \
((t)->t_state == TS_SLEEP && \
(t)->t_wchan != NULL && \
(t)->t_astflag && \
((t)->t_proc_flag & TP_CHKPT))
static int
sbdp_stop_user_threads(sbdp_sr_handle_t *srh)
{
int count;
char cache_psargs[PSARGSZ];
kthread_id_t cache_tp;
uint_t cache_t_state;
int bailout;
sbd_error_t *sep;
kthread_id_t tp;
extern void add_one_utstop();
extern void utstop_timedwait(clock_t);
extern void utstop_init(void);
#define SBDP_UTSTOP_RETRY 4
#define SBDP_UTSTOP_WAIT hz
if (sbdp_skip_user_threads)
return (DDI_SUCCESS);
sep = &srh->sep;
ASSERT(sep);
utstop_init();
/* we need to try a few times to get past fork, etc. */
for (count = 0; count < SBDP_UTSTOP_RETRY; count++) {
/* walk the entire threadlist */
mutex_enter(&pidlock);
for (tp = curthread->t_next; tp != curthread; tp = tp->t_next) {
proc_t *p = ttoproc(tp);
/* handle kernel threads separately */
if (p->p_as == &kas || p->p_stat == SZOMB)
continue;
mutex_enter(&p->p_lock);
thread_lock(tp);
if (tp->t_state == TS_STOPPED) {
/* add another reason to stop this thread */
tp->t_schedflag &= ~TS_RESUME;
} else {
tp->t_proc_flag |= TP_CHKPT;
thread_unlock(tp);
mutex_exit(&p->p_lock);
add_one_utstop();
mutex_enter(&p->p_lock);
thread_lock(tp);
aston(tp);
if (ISWAKEABLE(tp) || ISWAITING(tp)) {
setrun_locked(tp);
}
}
/* grab thread if needed */
if (tp->t_state == TS_ONPROC && tp->t_cpu != CPU)
poke_cpu(tp->t_cpu->cpu_id);
thread_unlock(tp);
mutex_exit(&p->p_lock);
}
mutex_exit(&pidlock);
/* let everything catch up */
utstop_timedwait(count * count * SBDP_UTSTOP_WAIT);
/* now, walk the threadlist again to see if we are done */
mutex_enter(&pidlock);
for (tp = curthread->t_next, bailout = 0;
tp != curthread; tp = tp->t_next) {
proc_t *p = ttoproc(tp);
/* handle kernel threads separately */
if (p->p_as == &kas || p->p_stat == SZOMB)
continue;
/*
* If this thread didn't stop, and we don't allow
* unstopped blocked threads, bail.
*/
thread_lock(tp);
if (!CPR_ISTOPPED(tp) &&
!(sbdp_allow_blocked_threads &&
SBDP_VSTOPPED(tp))) {
/* nope, cache the details for later */
bcopy(p->p_user.u_psargs, cache_psargs,
sizeof (cache_psargs));
cache_tp = tp;
cache_t_state = tp->t_state;
bailout = 1;
}
thread_unlock(tp);
}
mutex_exit(&pidlock);
/* were all the threads stopped? */
if (!bailout)
break;
}
/* were we unable to stop all threads after a few tries? */
if (bailout) {
cmn_err(CE_NOTE, "process: %s id: %p state: %x\n",
cache_psargs, cache_tp, cache_t_state);
(void) sprintf(sbdp_get_err_buf(sep), "%s", cache_psargs);
sbdp_set_err(sep, ESGT_UTHREAD, NULL);
return (ESRCH);
}
return (DDI_SUCCESS);
}
int
xen_resume_devices(dev_info_t *start, int resume_failed)
{
dev_info_t *dip, *next, *last = NULL;
int did_suspend;
int error = resume_failed;
char buf[XPV_BUFSIZE];
SUSPEND_DEBUG("xen_resume_devices\n");
while (last != start) {
dip = start;
next = ddi_get_next_sibling(dip);
while (next != last) {
dip = next;
next = ddi_get_next_sibling(dip);
}
/*
* cpr is the only one that uses this field and the device
* itself hasn't resumed yet, there is no need to use a
* lock, even though kernel threads are active by now.
*/
did_suspend = DEVI(dip)->devi_cpr_flags & DCF_CPR_SUSPENDED;
if (did_suspend)
DEVI(dip)->devi_cpr_flags &= ~DCF_CPR_SUSPENDED;
/*
* There may be background attaches happening on devices
* that were not originally suspended by cpr, so resume
* only devices that were suspended by cpr. Also, stop
* resuming after the first resume failure, but traverse
* the entire tree to clear the suspend flag.
*/
if (did_suspend && !error) {
SUSPEND_DEBUG("Resuming device %s\n",
ddi_deviname(dip, buf));
/*
* If a device suspended by cpr gets detached during
* the resume process (for example, due to hotplugging)
* before cpr gets around to issuing it a DDI_RESUME,
* we'll have problems.
*/
if (!i_ddi_devi_attached(dip)) {
cmn_err(CE_WARN, "Skipping %s, device "
"not ready for resume",
ddi_deviname(dip, buf));
} else {
if (devi_attach(dip, DDI_RESUME) !=
DDI_SUCCESS) {
error = ENXIO;
}
}
}
if (error == ENXIO) {
cmn_err(CE_WARN, "Unable to resume device %s",
ddi_deviname(dip, buf));
}
error = xen_resume_devices(ddi_get_child(dip), error);
last = dip;
}
return (error);
}
uint64_t *
get_intr_mapping_reg(int upaid, int slave)
{
int affin_upaid;
dev_info_t *affin_dip;
uint64_t *addr = intr_map_reg[upaid];
/* If we're a UPA master, or we have a valid mapping register. */
if (!slave || addr != NULL)
return (addr);
/*
* We only get here if we're a UPA slave only device whose interrupt
* mapping register has not been set.
* We need to try and install the nexus whose physical address
* space is where the slaves mapping register resides. They
* should call set_intr_mapping_reg() in their xxattach() to register
* the mapping register with the system.
*/
/*
* We don't know if a single- or multi-interrupt proxy is fielding
* our UPA slave interrupt, we must check both cases.
* Start out by assuming the multi-interrupt case.
* We assume that single- and multi- interrupters are not
* overlapping in UPA portid space.
*/
affin_upaid = upaid | 3;
/*
* We start looking for the multi-interrupter affinity node.
* We know it's ONLY a child of the root node since the root
* node defines UPA space.
*/
for (affin_dip = ddi_get_child(ddi_root_node()); affin_dip;
affin_dip = ddi_get_next_sibling(affin_dip))
if (ddi_prop_get_int(DDI_DEV_T_ANY, affin_dip,
DDI_PROP_DONTPASS, "upa-portid", -1) == affin_upaid)
break;
if (affin_dip) {
if (i_ddi_attach_node_hierarchy(affin_dip) == DDI_SUCCESS) {
/* try again to get the mapping register. */
addr = intr_map_reg[upaid];
}
}
/*
* If we still don't have a mapping register try single -interrupter
* case.
*/
if (addr == NULL) {
affin_upaid = upaid | 1;
for (affin_dip = ddi_get_child(ddi_root_node()); affin_dip;
affin_dip = ddi_get_next_sibling(affin_dip))
if (ddi_prop_get_int(DDI_DEV_T_ANY, affin_dip,
DDI_PROP_DONTPASS, "upa-portid", -1) == affin_upaid)
break;
if (affin_dip) {
if (i_ddi_attach_node_hierarchy(affin_dip)
== DDI_SUCCESS) {
/* try again to get the mapping register. */
addr = intr_map_reg[upaid];
}
}
}
return (addr);
}
void
load_platform_drivers(void)
{
extern int watchdog_available;
extern int watchdog_enable;
dev_info_t *dip; /* dip of the isa driver */
int simba_present = 0;
dev_info_t *root_child_node;
major_t major;
if (ddi_install_driver("power") != DDI_SUCCESS)
cmn_err(CE_WARN, "Failed to install \"power\" driver.");
/*
* Install Isa driver. This is required for the southbridge IDE
* workaround - to reset the IDE channel during IDE bus reset.
* Panic the system in case ISA driver could not be loaded or
* any problem in accessing its pci config space. Since the register
* to reset the channel for IDE is in ISA config space!.
*/
root_child_node = ddi_get_child(ddi_root_node());
while (root_child_node != NULL) {
if (strcmp(ddi_node_name(root_child_node), "pci") == 0) {
root_child_node = ddi_get_child(root_child_node);
if (strcmp(ddi_node_name(root_child_node), "pci") == 0)
simba_present = 1;
break;
}
root_child_node = ddi_get_next_sibling(root_child_node);
}
if (simba_present)
dip = e_ddi_hold_devi_by_path(PLATFORM_ISA_PATHNAME_WITH_SIMBA,
0);
else
dip = e_ddi_hold_devi_by_path(PLATFORM_ISA_PATHNAME, 0);
if (dip == NULL) {
cmn_err(CE_PANIC, "Could not install the isa driver\n");
return;
}
if (pci_config_setup(dip, &platform_isa_handle) != DDI_SUCCESS) {
cmn_err(CE_PANIC, "Could not get the config space of isa\n");
return;
}
/*
* Load the blade support chip driver.
*
*/
if (((major = ddi_name_to_major(BSC_DRV)) == -1) ||
(ddi_hold_installed_driver(major) == NULL)) {
cmn_err(CE_WARN, "%s: failed to load", BSC_DRV);
} else {
bsc_drv_func_ptr = (void (*)(struct bscv_idi_info *))
modgetsymvalue(BSC_DRV_FUNC, 0);
if (bsc_drv_func_ptr == NULL) {
cmn_err(CE_WARN, "load_platform_defaults: %s()"
" not found; signatures will not be updated\n",
BSC_DRV_FUNC);
watchdog_available = 0;
if (watchdog_enable) {
cmn_err(CE_WARN, "load_platform_defaults: %s()"
" not found; BSC OS watchdog service not available\n",
BSC_DRV_FUNC);
}
}
}
}
static void
dr_resume_devices(dev_info_t *start, dr_sr_handle_t *srh)
{
dr_handle_t *handle;
dev_info_t *dip, *next, *last = NULL;
major_t major;
char *bn;
int circ;
major = (major_t)-1;
/* attach in reverse device tree order */
while (last != start) {
dip = start;
next = ddi_get_next_sibling(dip);
while (next != last && dip != srh->sr_failed_dip) {
dip = next;
next = ddi_get_next_sibling(dip);
}
if (dip == srh->sr_failed_dip) {
/* release hold acquired in dr_suspend_devices() */
srh->sr_failed_dip = NULL;
ndi_rele_devi(dip);
} else if (dr_is_real_device(dip) &&
srh->sr_failed_dip == NULL) {
if ((bn = ddi_binding_name(dip)) != NULL) {
major = ddi_name_to_major(bn);
} else {
bn = "<null>";
}
if (!dr_bypass_device(bn) &&
!drmach_verify_sr(dip, 0)) {
char d_name[40], d_alias[40], *d_info;
d_name[0] = 0;
d_info = ddi_get_name_addr(dip);
if (d_info == NULL)
d_info = "<null>";
if (!dr_resolve_devname(dip, d_name,
d_alias)) {
if (d_alias[0] != 0) {
prom_printf("\tresuming "
"%s@%s (aka %s)\n",
d_name, d_info,
d_alias);
} else {
prom_printf("\tresuming "
"%s@%s\n",
d_name, d_info);
}
} else {
prom_printf("\tresuming %s@%s\n",
bn, d_info);
}
if (devi_attach(dip, DDI_RESUME) !=
DDI_SUCCESS) {
/*
* Print a console warning,
* set an e_code of ESBD_RESUME,
* and save the driver major
* number in the e_rsc.
*/
prom_printf("\tFAILED to resume %s@%s",
d_name[0] ? d_name : bn, d_info);
srh->sr_err_idx =
dr_add_int(srh->sr_err_ints,
srh->sr_err_idx, DR_MAX_ERR_INT,
(uint64_t)major);
handle = srh->sr_dr_handlep;
dr_op_err(CE_IGNORE, handle,
ESBD_RESUME, "%s@%s",
d_name[0] ? d_name : bn, d_info);
}
}
}
/* Hold parent busy while walking its children */
ndi_devi_enter(dip, &circ);
dr_resume_devices(ddi_get_child(dip), srh);
ndi_devi_exit(dip, circ);
last = dip;
}
}
/*
* The "dip" argument's parent (if it exists) must be held busy.
*/
static int
dr_suspend_devices(dev_info_t *dip, dr_sr_handle_t *srh)
{
dr_handle_t *handle;
major_t major;
char *dname;
int circ;
/*
* If dip is the root node, it has no siblings and it is
* always held. If dip is not the root node, dr_suspend_devices()
* will be invoked with the parent held busy.
*/
for (; dip != NULL; dip = ddi_get_next_sibling(dip)) {
char d_name[40], d_alias[40], *d_info;
ndi_devi_enter(dip, &circ);
if (dr_suspend_devices(ddi_get_child(dip), srh)) {
ndi_devi_exit(dip, circ);
return (ENXIO);
}
ndi_devi_exit(dip, circ);
if (!dr_is_real_device(dip))
continue;
major = (major_t)-1;
if ((dname = ddi_binding_name(dip)) != NULL)
major = ddi_name_to_major(dname);
if (dr_bypass_device(dname)) {
PR_QR(" bypassed suspend of %s (major# %d)\n", dname,
major);
continue;
}
if (drmach_verify_sr(dip, 1)) {
PR_QR(" bypassed suspend of %s (major# %d)\n", dname,
major);
continue;
}
if ((d_info = ddi_get_name_addr(dip)) == NULL)
d_info = "<null>";
d_name[0] = 0;
if (dr_resolve_devname(dip, d_name, d_alias) == 0) {
if (d_alias[0] != 0) {
prom_printf("\tsuspending %s@%s (aka %s)\n",
d_name, d_info, d_alias);
} else {
prom_printf("\tsuspending %s@%s\n",
d_name, d_info);
}
} else {
prom_printf("\tsuspending %s@%s\n", dname, d_info);
}
if (devi_detach(dip, DDI_SUSPEND) != DDI_SUCCESS) {
prom_printf("\tFAILED to suspend %s@%s\n",
d_name[0] ? d_name : dname, d_info);
srh->sr_err_idx = dr_add_int(srh->sr_err_ints,
srh->sr_err_idx, DR_MAX_ERR_INT,
(uint64_t)major);
ndi_hold_devi(dip);
srh->sr_failed_dip = dip;
handle = srh->sr_dr_handlep;
dr_op_err(CE_IGNORE, handle, ESBD_SUSPEND, "%s@%s",
d_name[0] ? d_name : dname, d_info);
return (DDI_FAILURE);
}
}
return (DDI_SUCCESS);
}
