aboutsummaryrefslogtreecommitdiff
path: root/drivers/misc
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/misc')
-rw-r--r--drivers/misc/Kconfig4
-rw-r--r--drivers/misc/eeprom/at24.c67
-rw-r--r--drivers/misc/eeprom/at25.c58
-rw-r--r--drivers/misc/sgi-gru/Makefile2
-rw-r--r--drivers/misc/sgi-gru/gru_instructions.h22
-rw-r--r--drivers/misc/sgi-gru/grufault.c130
-rw-r--r--drivers/misc/sgi-gru/grufile.c36
-rw-r--r--drivers/misc/sgi-gru/gruhandles.c183
-rw-r--r--drivers/misc/sgi-gru/gruhandles.h178
-rw-r--r--drivers/misc/sgi-gru/grukservices.c131
-rw-r--r--drivers/misc/sgi-gru/grukservices.h33
-rw-r--r--drivers/misc/sgi-gru/grumain.c84
-rw-r--r--drivers/misc/sgi-gru/gruprocfs.c45
-rw-r--r--drivers/misc/sgi-gru/grutables.h41
-rw-r--r--drivers/misc/sgi-gru/grutlbpurge.c7
-rw-r--r--drivers/misc/sgi-xp/xpc.h33
-rw-r--r--drivers/misc/sgi-xp/xpc_channel.c8
-rw-r--r--drivers/misc/sgi-xp/xpc_main.c6
-rw-r--r--drivers/misc/sgi-xp/xpc_sn2.c20
-rw-r--r--drivers/misc/sgi-xp/xpc_uv.c229
20 files changed, 899 insertions, 418 deletions
diff --git a/drivers/misc/Kconfig b/drivers/misc/Kconfig
index 5f3bff43462..0b92b2f6ea6 100644
--- a/drivers/misc/Kconfig
+++ b/drivers/misc/Kconfig
@@ -165,7 +165,7 @@ config SGI_XP
depends on (IA64_GENERIC || IA64_SGI_SN2 || IA64_SGI_UV || X86_UV) && SMP
select IA64_UNCACHED_ALLOCATOR if IA64_GENERIC || IA64_SGI_SN2
select GENERIC_ALLOCATOR if IA64_GENERIC || IA64_SGI_SN2
- select SGI_GRU if (IA64_GENERIC || IA64_SGI_UV || X86_64) && SMP
+ select SGI_GRU if X86_64 && SMP
---help---
An SGI machine can be divided into multiple Single System
Images which act independently of each other and have
@@ -189,7 +189,7 @@ config HP_ILO
config SGI_GRU
tristate "SGI GRU driver"
- depends on (X86_UV || IA64_SGI_UV || IA64_GENERIC) && SMP
+ depends on X86_UV && SMP
default n
select MMU_NOTIFIER
---help---
diff --git a/drivers/misc/eeprom/at24.c b/drivers/misc/eeprom/at24.c
index d4775528abc..d184dfab963 100644
--- a/drivers/misc/eeprom/at24.c
+++ b/drivers/misc/eeprom/at24.c
@@ -53,6 +53,7 @@
struct at24_data {
struct at24_platform_data chip;
+ struct memory_accessor macc;
bool use_smbus;
/*
@@ -225,14 +226,11 @@ static ssize_t at24_eeprom_read(struct at24_data *at24, char *buf,
return status;
}
-static ssize_t at24_bin_read(struct kobject *kobj, struct bin_attribute *attr,
+static ssize_t at24_read(struct at24_data *at24,
char *buf, loff_t off, size_t count)
{
- struct at24_data *at24;
ssize_t retval = 0;
- at24 = dev_get_drvdata(container_of(kobj, struct device, kobj));
-
if (unlikely(!count))
return count;
@@ -262,12 +260,14 @@ static ssize_t at24_bin_read(struct kobject *kobj, struct bin_attribute *attr,
return retval;
}
+static ssize_t at24_bin_read(struct kobject *kobj, struct bin_attribute *attr,
+ char *buf, loff_t off, size_t count)
+{
+ struct at24_data *at24;
-/*
- * REVISIT: export at24_bin{read,write}() to let other kernel code use
- * eeprom data. For example, it might hold a board's Ethernet address, or
- * board-specific calibration data generated on the manufacturing floor.
- */
+ at24 = dev_get_drvdata(container_of(kobj, struct device, kobj));
+ return at24_read(at24, buf, off, count);
+}
/*
@@ -347,14 +347,11 @@ static ssize_t at24_eeprom_write(struct at24_data *at24, char *buf,
return -ETIMEDOUT;
}
-static ssize_t at24_bin_write(struct kobject *kobj, struct bin_attribute *attr,
+static ssize_t at24_write(struct at24_data *at24,
char *buf, loff_t off, size_t count)
{
- struct at24_data *at24;
ssize_t retval = 0;
- at24 = dev_get_drvdata(container_of(kobj, struct device, kobj));
-
if (unlikely(!count))
return count;
@@ -384,6 +381,39 @@ static ssize_t at24_bin_write(struct kobject *kobj, struct bin_attribute *attr,
return retval;
}
+static ssize_t at24_bin_write(struct kobject *kobj, struct bin_attribute *attr,
+ char *buf, loff_t off, size_t count)
+{
+ struct at24_data *at24;
+
+ at24 = dev_get_drvdata(container_of(kobj, struct device, kobj));
+ return at24_write(at24, buf, off, count);
+}
+
+/*-------------------------------------------------------------------------*/
+
+/*
+ * This lets other kernel code access the eeprom data. For example, it
+ * might hold a board's Ethernet address, or board-specific calibration
+ * data generated on the manufacturing floor.
+ */
+
+static ssize_t at24_macc_read(struct memory_accessor *macc, char *buf,
+ off_t offset, size_t count)
+{
+ struct at24_data *at24 = container_of(macc, struct at24_data, macc);
+
+ return at24_read(at24, buf, offset, count);
+}
+
+static ssize_t at24_macc_write(struct memory_accessor *macc, char *buf,
+ off_t offset, size_t count)
+{
+ struct at24_data *at24 = container_of(macc, struct at24_data, macc);
+
+ return at24_write(at24, buf, offset, count);
+}
+
/*-------------------------------------------------------------------------*/
static int at24_probe(struct i2c_client *client, const struct i2c_device_id *id)
@@ -413,6 +443,9 @@ static int at24_probe(struct i2c_client *client, const struct i2c_device_id *id)
* is recommended anyhow.
*/
chip.page_size = 1;
+
+ chip.setup = NULL;
+ chip.context = NULL;
}
if (!is_power_of_2(chip.byte_len))
@@ -463,6 +496,8 @@ static int at24_probe(struct i2c_client *client, const struct i2c_device_id *id)
at24->bin.read = at24_bin_read;
at24->bin.size = chip.byte_len;
+ at24->macc.read = at24_macc_read;
+
writable = !(chip.flags & AT24_FLAG_READONLY);
if (writable) {
if (!use_smbus || i2c_check_functionality(client->adapter,
@@ -470,6 +505,8 @@ static int at24_probe(struct i2c_client *client, const struct i2c_device_id *id)
unsigned write_max = chip.page_size;
+ at24->macc.write = at24_macc_write;
+
at24->bin.write = at24_bin_write;
at24->bin.attr.mode |= S_IWUSR;
@@ -520,6 +557,10 @@ static int at24_probe(struct i2c_client *client, const struct i2c_device_id *id)
at24->write_max,
use_smbus ? ", use_smbus" : "");
+ /* export data to kernel code */
+ if (chip.setup)
+ chip.setup(&at24->macc, chip.context);
+
return 0;
err_clients:
diff --git a/drivers/misc/eeprom/at25.c b/drivers/misc/eeprom/at25.c
index 290dbe99647..6bc0dac5c1e 100644
--- a/drivers/misc/eeprom/at25.c
+++ b/drivers/misc/eeprom/at25.c
@@ -30,6 +30,7 @@
struct at25_data {
struct spi_device *spi;
+ struct memory_accessor mem;
struct mutex lock;
struct spi_eeprom chip;
struct bin_attribute bin;
@@ -75,6 +76,13 @@ at25_ee_read(
struct spi_transfer t[2];
struct spi_message m;
+ if (unlikely(offset >= at25->bin.size))
+ return 0;
+ if ((offset + count) > at25->bin.size)
+ count = at25->bin.size - offset;
+ if (unlikely(!count))
+ return count;
+
cp = command;
*cp++ = AT25_READ;
@@ -127,13 +135,6 @@ at25_bin_read(struct kobject *kobj, struct bin_attribute *bin_attr,
dev = container_of(kobj, struct device, kobj);
at25 = dev_get_drvdata(dev);
- if (unlikely(off >= at25->bin.size))
- return 0;
- if ((off + count) > at25->bin.size)
- count = at25->bin.size - off;
- if (unlikely(!count))
- return count;
-
return at25_ee_read(at25, buf, off, count);
}
@@ -146,6 +147,13 @@ at25_ee_write(struct at25_data *at25, char *buf, loff_t off, size_t count)
unsigned buf_size;
u8 *bounce;
+ if (unlikely(off >= at25->bin.size))
+ return -EFBIG;
+ if ((off + count) > at25->bin.size)
+ count = at25->bin.size - off;
+ if (unlikely(!count))
+ return count;
+
/* Temp buffer starts with command and address */
buf_size = at25->chip.page_size;
if (buf_size > io_limit)
@@ -253,18 +261,31 @@ at25_bin_write(struct kobject *kobj, struct bin_attribute *bin_attr,
dev = container_of(kobj, struct device, kobj);
at25 = dev_get_drvdata(dev);
- if (unlikely(off >= at25->bin.size))
- return -EFBIG;
- if ((off + count) > at25->bin.size)
- count = at25->bin.size - off;
- if (unlikely(!count))
- return count;
-
return at25_ee_write(at25, buf, off, count);
}
/*-------------------------------------------------------------------------*/
+/* Let in-kernel code access the eeprom data. */
+
+static ssize_t at25_mem_read(struct memory_accessor *mem, char *buf,
+ off_t offset, size_t count)
+{
+ struct at25_data *at25 = container_of(mem, struct at25_data, mem);
+
+ return at25_ee_read(at25, buf, offset, count);
+}
+
+static ssize_t at25_mem_write(struct memory_accessor *mem, char *buf,
+ off_t offset, size_t count)
+{
+ struct at25_data *at25 = container_of(mem, struct at25_data, mem);
+
+ return at25_ee_write(at25, buf, offset, count);
+}
+
+/*-------------------------------------------------------------------------*/
+
static int at25_probe(struct spi_device *spi)
{
struct at25_data *at25 = NULL;
@@ -317,6 +338,10 @@ static int at25_probe(struct spi_device *spi)
at25->addrlen = addrlen;
/* Export the EEPROM bytes through sysfs, since that's convenient.
+ * And maybe to other kernel code; it might hold a board's Ethernet
+ * address, or board-specific calibration data generated on the
+ * manufacturing floor.
+ *
* Default to root-only access to the data; EEPROMs often hold data
* that's sensitive for read and/or write, like ethernet addresses,
* security codes, board-specific manufacturing calibrations, etc.
@@ -324,17 +349,22 @@ static int at25_probe(struct spi_device *spi)
at25->bin.attr.name = "eeprom";
at25->bin.attr.mode = S_IRUSR;
at25->bin.read = at25_bin_read;
+ at25->mem.read = at25_mem_read;
at25->bin.size = at25->chip.byte_len;
if (!(chip->flags & EE_READONLY)) {
at25->bin.write = at25_bin_write;
at25->bin.attr.mode |= S_IWUSR;
+ at25->mem.write = at25_mem_write;
}
err = sysfs_create_bin_file(&spi->dev.kobj, &at25->bin);
if (err)
goto fail;
+ if (chip->setup)
+ chip->setup(&at25->mem, chip->context);
+
dev_info(&spi->dev, "%Zd %s %s eeprom%s, pagesize %u\n",
(at25->bin.size < 1024)
? at25->bin.size
diff --git a/drivers/misc/sgi-gru/Makefile b/drivers/misc/sgi-gru/Makefile
index 9e9170b3599..bcd8136d2f9 100644
--- a/drivers/misc/sgi-gru/Makefile
+++ b/drivers/misc/sgi-gru/Makefile
@@ -3,5 +3,5 @@ ifdef CONFIG_SGI_GRU_DEBUG
endif
obj-$(CONFIG_SGI_GRU) := gru.o
-gru-y := grufile.o grumain.o grufault.o grutlbpurge.o gruprocfs.o grukservices.o
+gru-y := grufile.o grumain.o grufault.o grutlbpurge.o gruprocfs.o grukservices.o gruhandles.o
diff --git a/drivers/misc/sgi-gru/gru_instructions.h b/drivers/misc/sgi-gru/gru_instructions.h
index 48762e7b98b..3fde33c1e8f 100644
--- a/drivers/misc/sgi-gru/gru_instructions.h
+++ b/drivers/misc/sgi-gru/gru_instructions.h
@@ -19,8 +19,11 @@
#ifndef __GRU_INSTRUCTIONS_H__
#define __GRU_INSTRUCTIONS_H__
-#define gru_flush_cache_hook(p)
-#define gru_emulator_wait_hook(p, w)
+extern int gru_check_status_proc(void *cb);
+extern int gru_wait_proc(void *cb);
+extern void gru_wait_abort_proc(void *cb);
+
+
/*
* Architecture dependent functions
@@ -29,16 +32,16 @@
#if defined(CONFIG_IA64)
#include <linux/compiler.h>
#include <asm/intrinsics.h>
-#define __flush_cache(p) ia64_fc(p)
+#define __flush_cache(p) ia64_fc((unsigned long)p)
/* Use volatile on IA64 to ensure ordering via st4.rel */
-#define gru_ordered_store_int(p,v) \
+#define gru_ordered_store_int(p, v) \
do { \
barrier(); \
*((volatile int *)(p)) = v; /* force st.rel */ \
} while (0)
#elif defined(CONFIG_X86_64)
#define __flush_cache(p) clflush(p)
-#define gru_ordered_store_int(p,v) \
+#define gru_ordered_store_int(p, v) \
do { \
barrier(); \
*(int *)p = v; \
@@ -558,20 +561,19 @@ extern int gru_get_cb_exception_detail(void *cb,
#define GRU_EXC_STR_SIZE 256
-extern int gru_check_status_proc(void *cb);
-extern int gru_wait_proc(void *cb);
-extern void gru_wait_abort_proc(void *cb);
/*
* Control block definition for checking status
*/
struct gru_control_block_status {
unsigned int icmd :1;
- unsigned int unused1 :31;
+ unsigned int ima :3;
+ unsigned int reserved0 :4;
+ unsigned int unused1 :24;
unsigned int unused2 :24;
unsigned int istatus :2;
unsigned int isubstatus :4;
- unsigned int inused3 :2;
+ unsigned int unused3 :2;
};
/* Get CB status */
diff --git a/drivers/misc/sgi-gru/grufault.c b/drivers/misc/sgi-gru/grufault.c
index 3ee698ad859..ab118558552 100644
--- a/drivers/misc/sgi-gru/grufault.c
+++ b/drivers/misc/sgi-gru/grufault.c
@@ -32,6 +32,7 @@
#include <linux/device.h>
#include <linux/io.h>
#include <linux/uaccess.h>
+#include <linux/security.h>
#include <asm/pgtable.h>
#include "gru.h"
#include "grutables.h"
@@ -266,6 +267,44 @@ err:
return 1;
}
+static int gru_vtop(struct gru_thread_state *gts, unsigned long vaddr,
+ int write, int atomic, unsigned long *gpa, int *pageshift)
+{
+ struct mm_struct *mm = gts->ts_mm;
+ struct vm_area_struct *vma;
+ unsigned long paddr;
+ int ret, ps;
+
+ vma = find_vma(mm, vaddr);
+ if (!vma)
+ goto inval;
+
+ /*
+ * Atomic lookup is faster & usually works even if called in non-atomic
+ * context.
+ */
+ rmb(); /* Must/check ms_range_active before loading PTEs */
+ ret = atomic_pte_lookup(vma, vaddr, write, &paddr, &ps);
+ if (ret) {
+ if (atomic)
+ goto upm;
+ if (non_atomic_pte_lookup(vma, vaddr, write, &paddr, &ps))
+ goto inval;
+ }
+ if (is_gru_paddr(paddr))
+ goto inval;
+ paddr = paddr & ~((1UL << ps) - 1);
+ *gpa = uv_soc_phys_ram_to_gpa(paddr);
+ *pageshift = ps;
+ return 0;
+
+inval:
+ return -1;
+upm:
+ return -2;
+}
+
+
/*
* Drop a TLB entry into the GRU. The fault is described by info in an TFH.
* Input:
@@ -280,10 +319,8 @@ static int gru_try_dropin(struct gru_thread_state *gts,
struct gru_tlb_fault_handle *tfh,
unsigned long __user *cb)
{
- struct mm_struct *mm = gts->ts_mm;
- struct vm_area_struct *vma;
- int pageshift, asid, write, ret;
- unsigned long paddr, gpa, vaddr;
+ int pageshift = 0, asid, write, ret, atomic = !cb;
+ unsigned long gpa = 0, vaddr = 0;
/*
* NOTE: The GRU contains magic hardware that eliminates races between
@@ -317,28 +354,19 @@ static int gru_try_dropin(struct gru_thread_state *gts,
if (atomic_read(&gts->ts_gms->ms_range_active))
goto failactive;
- vma = find_vma(mm, vaddr);
- if (!vma)
+ ret = gru_vtop(gts, vaddr, write, atomic, &gpa, &pageshift);
+ if (ret == -1)
goto failinval;
+ if (ret == -2)
+ goto failupm;
- /*
- * Atomic lookup is faster & usually works even if called in non-atomic
- * context.
- */
- rmb(); /* Must/check ms_range_active before loading PTEs */
- ret = atomic_pte_lookup(vma, vaddr, write, &paddr, &pageshift);
- if (ret) {
- if (!cb)
+ if (!(gts->ts_sizeavail & GRU_SIZEAVAIL(pageshift))) {
+ gts->ts_sizeavail |= GRU_SIZEAVAIL(pageshift);
+ if (atomic || !gru_update_cch(gts, 0)) {
+ gts->ts_force_cch_reload = 1;
goto failupm;
- if (non_atomic_pte_lookup(vma, vaddr, write, &paddr,
- &pageshift))
- goto failinval;
+ }
}
- if (is_gru_paddr(paddr))
- goto failinval;
-
- paddr = paddr & ~((1UL << pageshift) - 1);
- gpa = uv_soc_phys_ram_to_gpa(paddr);
gru_cb_set_istatus_active(cb);
tfh_write_restart(tfh, gpa, GAA_RAM, vaddr, asid, write,
GRU_PAGESIZE(pageshift));
@@ -368,6 +396,7 @@ failupm:
failfmm:
/* FMM state on UPM call */
+ gru_flush_cache(tfh);
STAT(tlb_dropin_fail_fmm);
gru_dbg(grudev, "FAILED fmm tfh: 0x%p, state %d\n", tfh, tfh->state);
return 0;
@@ -448,6 +477,7 @@ irqreturn_t gru_intr(int irq, void *dev_id)
up_read(&gts->ts_mm->mmap_sem);
} else {
tfh_user_polling_mode(tfh);
+ STAT(intr_mm_lock_failed);
}
}
return IRQ_HANDLED;
@@ -497,10 +527,8 @@ int gru_handle_user_call_os(unsigned long cb)
if (!gts)
return -EINVAL;
- if (ucbnum >= gts->ts_cbr_au_count * GRU_CBR_AU_SIZE) {
- ret = -EINVAL;
+ if (ucbnum >= gts->ts_cbr_au_count * GRU_CBR_AU_SIZE)
goto exit;
- }
/*
* If force_unload is set, the UPM TLB fault is phony. The task
@@ -508,6 +536,20 @@ int gru_handle_user_call_os(unsigned long cb)
* unload the context. The task will page fault and assign a new
* context.
*/
+ if (gts->ts_tgid_owner == current->tgid && gts->ts_blade >= 0 &&
+ gts->ts_blade != uv_numa_blade_id()) {
+ STAT(call_os_offnode_reference);
+ gts->ts_force_unload = 1;
+ }
+
+ /*
+ * CCH may contain stale data if ts_force_cch_reload is set.
+ */
+ if (gts->ts_gru && gts->ts_force_cch_reload) {
+ gru_update_cch(gts, 0);
+ gts->ts_force_cch_reload = 0;
+ }
+
ret = -EAGAIN;
cbrnum = thread_cbr_number(gts, ucbnum);
if (gts->ts_force_unload) {
@@ -541,11 +583,13 @@ int gru_get_exception_detail(unsigned long arg)
if (!gts)
return -EINVAL;
- if (gts->ts_gru) {
- ucbnum = get_cb_number((void *)excdet.cb);
+ ucbnum = get_cb_number((void *)excdet.cb);
+ if (ucbnum >= gts->ts_cbr_au_count * GRU_CBR_AU_SIZE) {
+ ret = -EINVAL;
+ } else if (gts->ts_gru) {
cbrnum = thread_cbr_number(gts, ucbnum);
cbe = get_cbe_by_index(gts->ts_gru, cbrnum);
- prefetchw(cbe); /* Harmless on hardware, required for emulator */
+ prefetchw(cbe);/* Harmless on hardware, required for emulator */
excdet.opc = cbe->opccpy;
excdet.exopc = cbe->exopccpy;
excdet.ecause = cbe->ecause;
@@ -567,6 +611,31 @@ int gru_get_exception_detail(unsigned long arg)
/*
* User request to unload a context. Content is saved for possible reload.
*/
+static int gru_unload_all_contexts(void)
+{
+ struct gru_thread_state *gts;
+ struct gru_state *gru;
+ int gid, ctxnum;
+
+ if (!capable(CAP_SYS_ADMIN))
+ return -EPERM;
+ foreach_gid(gid) {
+ gru = GID_TO_GRU(gid);
+ spin_lock(&gru->gs_lock);
+ for (ctxnum = 0; ctxnum < GRU_NUM_CCH; ctxnum++) {
+ gts = gru->gs_gts[ctxnum];
+ if (gts && mutex_trylock(&gts->ts_ctxlock)) {
+ spin_unlock(&gru->gs_lock);
+ gru_unload_context(gts, 1);
+ gru_unlock_gts(gts);
+ spin_lock(&gru->gs_lock);
+ }
+ }
+ spin_unlock(&gru->gs_lock);
+ }
+ return 0;
+}
+
int gru_user_unload_context(unsigned long arg)
{
struct gru_thread_state *gts;
@@ -578,6 +647,9 @@ int gru_user_unload_context(unsigned long arg)
gru_dbg(grudev, "gseg 0x%lx\n", req.gseg);
+ if (!req.gseg)
+ return gru_unload_all_contexts();
+
gts = gru_find_lock_gts(req.gseg);
if (!gts)
return -EINVAL;
@@ -609,7 +681,7 @@ int gru_user_flush_tlb(unsigned long arg)
if (!gts)
return -EINVAL;
- gru_flush_tlb_range(gts->ts_gms, req.vaddr, req.vaddr + req.len);
+ gru_flush_tlb_range(gts->ts_gms, req.vaddr, req.len);
gru_unlock_gts(gts);
return 0;
diff --git a/drivers/misc/sgi-gru/grufile.c b/drivers/misc/sgi-gru/grufile.c
index c67e4e8bd62..3e6e42d2f01 100644
--- a/drivers/misc/sgi-gru/grufile.c
+++ b/drivers/misc/sgi-gru/grufile.c
@@ -45,7 +45,9 @@
#include <asm/uv/uv_mmrs.h>
struct gru_blade_state *gru_base[GRU_MAX_BLADES] __read_mostly;
-unsigned long gru_start_paddr, gru_end_paddr __read_mostly;
+unsigned long gru_start_paddr __read_mostly;
+unsigned long gru_end_paddr __read_mostly;
+unsigned int gru_max_gids __read_mostly;
struct gru_stats_s gru_stats;
/* Guaranteed user available resources on each node */
@@ -101,7 +103,7 @@ static int gru_file_mmap(struct file *file, struct vm_area_struct *vma)
return -EPERM;
if (vma->vm_start & (GRU_GSEG_PAGESIZE - 1) ||
- vma->vm_end & (GRU_GSEG_PAGESIZE - 1))
+ vma->vm_end & (GRU_GSEG_PAGESIZE - 1))
return -EINVAL;
vma->vm_flags |=
@@ -273,8 +275,11 @@ static void gru_init_chiplet(struct gru_state *gru, unsigned long paddr,
gru->gs_blade_id = bid;
gru->gs_cbr_map = (GRU_CBR_AU == 64) ? ~0 : (1UL << GRU_CBR_AU) - 1;
gru->gs_dsr_map = (1UL << GRU_DSR_AU) - 1;
+ gru->gs_asid_limit = MAX_ASID;
gru_tgh_flush_init(gru);
- gru_dbg(grudev, "bid %d, nid %d, gru %x, vaddr %p (0x%lx)\n",
+ if (gru->gs_gid >= gru_max_gids)
+ gru_max_gids = gru->gs_gid + 1;
+ gru_dbg(grudev, "bid %d, nid %d, gid %d, vaddr %p (0x%lx)\n",
bid, nid, gru->gs_gid, gru->gs_gru_base_vaddr,
gru->gs_gru_base_paddr);
gru_kservices_init(gru);
@@ -295,7 +300,7 @@ static int gru_init_tables(unsigned long gru_base_paddr, void *gru_base_vaddr)
for_each_online_node(nid) {
bid = uv_node_to_blade_id(nid);
pnode = uv_node_to_pnode(nid);
- if (gru_base[bid])
+ if (bid < 0 || gru_base[bid])
continue;
page = alloc_pages_node(nid, GFP_KERNEL, order);
if (!page)
@@ -308,11 +313,11 @@ static int gru_init_tables(unsigned long gru_base_paddr, void *gru_base_vaddr)
dsrbytes = 0;
cbrs = 0;
for (gru = gru_base[bid]->bs_grus, chip = 0;
- chip < GRU_CHIPLETS_PER_BLADE;
+ chip < GRU_CHIPLETS_PER_BLADE;
chip++, gru++) {
paddr = gru_chiplet_paddr(gru_base_paddr, pnode, chip);
vaddr = gru_chiplet_vaddr(gru_base_vaddr, pnode, chip);
- gru_init_chiplet(gru, paddr, vaddr, bid, nid, chip);
+ gru_init_chiplet(gru, paddr, vaddr, nid, bid, chip);
n = hweight64(gru->gs_cbr_map) * GRU_CBR_AU_SIZE;
cbrs = max(cbrs, n);
n = hweight64(gru->gs_dsr_map) * GRU_DSR_AU_BYTES;
@@ -370,26 +375,26 @@ static int __init gru_init(void)
void *gru_start_vaddr;
if (!is_uv_system())
- return 0;
+ return -ENODEV;
#if defined CONFIG_IA64
gru_start_paddr = 0xd000000000UL; /* ZZZZZZZZZZZZZZZZZZZ fixme */
#else
gru_start_paddr = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR) &
0x7fffffffffffUL;
-
#endif
gru_start_vaddr = __va(gru_start_paddr);
- gru_end_paddr = gru_start_paddr + MAX_NUMNODES * GRU_SIZE;
+ gru_end_paddr = gru_start_paddr + GRU_MAX_BLADES * GRU_SIZE;
printk(KERN_INFO "GRU space: 0x%lx - 0x%lx\n",
gru_start_paddr, gru_end_paddr);
irq = get_base_irq();
for (chip = 0; chip < GRU_CHIPLETS_PER_BLADE; chip++) {
ret = request_irq(irq + chip, gru_intr, 0, id, NULL);
- /* TODO: fix irq handling on x86. For now ignore failures because
+ /* TODO: fix irq handling on x86. For now ignore failure because
* interrupts are not required & not yet fully supported */
if (ret) {
- printk("!!!WARNING: GRU ignoring request failure!!!\n");
+ printk(KERN_WARNING
+ "!!!WARNING: GRU ignoring request failure!!!\n");
ret = 0;
}
if (ret) {
@@ -435,7 +440,7 @@ exit1:
static void __exit gru_exit(void)
{
- int i, bid;
+ int i, bid, gid;
int order = get_order(sizeof(struct gru_state) *
GRU_CHIPLETS_PER_BLADE);
@@ -445,6 +450,9 @@ static void __exit gru_exit(void)
for (i = 0; i < GRU_CHIPLETS_PER_BLADE; i++)
free_irq(IRQ_GRU + i, NULL);
+ foreach_gid(gid)
+ gru_kservices_exit(GID_TO_GRU(gid));
+
for (bid = 0; bid < GRU_MAX_BLADES; bid++)
free_pages((unsigned long)gru_base[bid], order);
@@ -469,7 +477,11 @@ struct vm_operations_struct gru_vm_ops = {
.fault = gru_fault,
};
+#ifndef MODULE
fs_initcall(gru_init);
+#else
+module_init(gru_init);
+#endif
module_exit(gru_exit);
module_param(gru_options, ulong, 0644);
diff --git a/drivers/misc/sgi-gru/gruhandles.c b/drivers/misc/sgi-gru/gruhandles.c
new file mode 100644
index 00000000000..9b7ccb32869
--- /dev/null
+++ b/drivers/misc/sgi-gru/gruhandles.c
@@ -0,0 +1,183 @@
+/*
+ * GRU KERNEL MCS INSTRUCTIONS
+ *
+ * Copyright (c) 2008 Silicon Graphics, Inc. All Rights Reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ */
+
+#include <linux/kernel.h>
+#include "gru.h"
+#include "grulib.h"
+#include "grutables.h"
+
+/* 10 sec */
+#ifdef CONFIG_IA64
+#include <asm/processor.h>
+#define GRU_OPERATION_TIMEOUT (((cycles_t) local_cpu_data->itc_freq)*10)
+#else
+#include <asm/tsc.h>
+#define GRU_OPERATION_TIMEOUT ((cycles_t) tsc_khz*10*1000)
+#endif
+
+/* Extract the status field from a kernel handle */
+#define GET_MSEG_HANDLE_STATUS(h) (((*(unsigned long *)(h)) >> 16) & 3)
+
+struct mcs_op_statistic mcs_op_statistics[mcsop_last];
+
+static void update_mcs_stats(enum mcs_op op, unsigned long clks)
+{
+ atomic_long_inc(&mcs_op_statistics[op].count);
+ atomic_long_add(clks, &mcs_op_statistics[op].total);
+ if (mcs_op_statistics[op].max < clks)
+ mcs_op_statistics[op].max = clks;
+}
+
+static void start_instruction(void *h)
+{
+ unsigned long *w0 = h;
+
+ wmb(); /* setting CMD bit must be last */
+ *w0 = *w0 | 1;
+ gru_flush_cache(h);
+}
+
+static int wait_instruction_complete(void *h, enum mcs_op opc)
+{
+ int status;
+ cycles_t start_time = get_cycles();
+
+ while (1) {
+ cpu_relax();
+ status = GET_MSEG_HANDLE_STATUS(h);
+ if (status != CCHSTATUS_ACTIVE)
+ break;
+ if (GRU_OPERATION_TIMEOUT < (get_cycles() - start_time))
+ panic("GRU %p is malfunctioning\n", h);
+ }
+ if (gru_options & OPT_STATS)
+ update_mcs_stats(opc, get_cycles() - start_time);
+ return status;
+}
+
+int cch_allocate(struct gru_context_configuration_handle *cch,
+ int asidval, int sizeavail, unsigned long cbrmap,
+ unsigned long dsrmap)
+{
+ int i;
+
+ for (i = 0; i < 8; i++) {
+ cch->asid[i] = (asidval++);
+ cch->sizeavail[i] = sizeavail;
+ }
+ cch->dsr_allocation_map = dsrmap;
+ cch->cbr_allocation_map = cbrmap;
+ cch->opc = CCHOP_ALLOCATE;
+ start_instruction(cch);
+ return wait_instruction_complete(cch, cchop_allocate);
+}
+
+int cch_start(struct gru_context_configuration_handle *cch)
+{
+ cch->opc = CCHOP_START;
+ start_instruction(cch);
+ return wait_instruction_complete(cch, cchop_start);
+}
+
+int cch_interrupt(struct gru_context_configuration_handle *cch)
+{
+ cch->opc = CCHOP_INTERRUPT;
+ start_instruction(cch);
+ return wait_instruction_complete(cch, cchop_interrupt);
+}
+
+int cch_deallocate(struct gru_context_configuration_handle *cch)
+{
+ cch->opc = CCHOP_DEALLOCATE;
+ start_instruction(cch);
+ return wait_instruction_complete(cch, cchop_deallocate);
+}
+
+int cch_interrupt_sync(struct gru_context_configuration_handle
+ *cch)
+{
+ cch->opc = CCHOP_INTERRUPT_SYNC;
+ start_instruction(cch);
+ return wait_instruction_complete(cch, cchop_interrupt_sync);
+}
+
+int tgh_invalidate(struct gru_tlb_global_handle *tgh,
+ unsigned long vaddr, unsigned long vaddrmask,
+ int asid, int pagesize, int global, int n,
+ unsigned short ctxbitmap)
+{
+ tgh->vaddr = vaddr;
+ tgh->asid = asid;
+ tgh->pagesize = pagesize;
+ tgh->n = n;
+ tgh->global = global;
+ tgh->vaddrmask = vaddrmask;
+ tgh->ctxbitmap = ctxbitmap;
+ tgh->opc = TGHOP_TLBINV;
+ start_instruction(tgh);
+ return wait_instruction_complete(tgh, tghop_invalidate);
+}
+
+void tfh_write_only(struct gru_tlb_fault_handle *tfh,
+ unsigned long pfn, unsigned long vaddr,
+ int asid, int dirty, int pagesize)
+{
+ tfh->fillasid = asid;
+ tfh->fillvaddr = vaddr;
+ tfh->pfn = pfn;
+ tfh->dirty = dirty;
+ tfh->pagesize = pagesize;
+ tfh->opc = TFHOP_WRITE_ONLY;
+ start_instruction(tfh);
+}
+
+void tfh_write_restart(struct gru_tlb_fault_handle *tfh,
+ unsigned long paddr, int gaa,
+ unsigned long vaddr, int asid, int dirty,
+ int pagesize)
+{
+ tfh->fillasid = asid;
+ tfh->fillvaddr = vaddr;
+ tfh->pfn = paddr >> GRU_PADDR_SHIFT;
+ tfh->gaa = gaa;
+ tfh->dirty = dirty;
+ tfh->pagesize = pagesize;
+ tfh->opc = TFHOP_WRITE_RESTART;
+ start_instruction(tfh);
+}
+
+void tfh_restart(struct gru_tlb_fault_handle *tfh)
+{
+ tfh->opc = TFHOP_RESTART;
+ start_instruction(tfh);
+}
+
+void tfh_user_polling_mode(struct gru_tlb_fault_handle *tfh)
+{
+ tfh->opc = TFHOP_USER_POLLING_MODE;
+ start_instruction(tfh);
+}
+
+void tfh_exception(struct gru_tlb_fault_handle *tfh)
+{
+ tfh->opc = TFHOP_EXCEPTION;
+ start_instruction(tfh);
+}
+
diff --git a/drivers/misc/sgi-gru/gruhandles.h b/drivers/misc/sgi-gru/gruhandles.h
index b63018d60fe..1ed74d7508c 100644
--- a/drivers/misc/sgi-gru/gruhandles.h
+++ b/drivers/misc/sgi-gru/gruhandles.h
@@ -489,170 +489,28 @@ enum gru_cbr_state {
* 64m 26 8
* ...
*/
-#define GRU_PAGESIZE(sh) ((((sh) > 20 ? (sh) + 2: (sh)) >> 1) - 6)
+#define GRU_PAGESIZE(sh) ((((sh) > 20 ? (sh) + 2 : (sh)) >> 1) - 6)
#define GRU_SIZEAVAIL(sh) (1UL << GRU_PAGESIZE(sh))
/* minimum TLB purge count to ensure a full purge */
#define GRUMAXINVAL 1024UL
-
-/* Extract the status field from a kernel handle */
-#define GET_MSEG_HANDLE_STATUS(h) (((*(unsigned long *)(h)) >> 16) & 3)
-
-static inline void start_instruction(void *h)
-{
- unsigned long *w0 = h;
-
- wmb(); /* setting CMD bit must be last */
- *w0 = *w0 | 1;
- gru_flush_cache(h);
-}
-
-static inline int wait_instruction_complete(void *h)
-{
- int status;
-
- do {
- cpu_relax();
- barrier();
- status = GET_MSEG_HANDLE_STATUS(h);
- } while (status == CCHSTATUS_ACTIVE);
- return status;
-}
-
-#if defined CONFIG_IA64
-static inline void cch_allocate_set_asids(
- struct gru_context_configuration_handle *cch, int asidval)
-{
- int i;
-
- for (i = 0; i <= RGN_HPAGE; i++) { /* assume HPAGE is last region */
- cch->asid[i] = (asidval++);
-#if 0
- /* ZZZ hugepages not supported yet */
- if (i == RGN_HPAGE)
- cch->sizeavail[i] = GRU_SIZEAVAIL(hpage_shift);
- else
-#endif
- cch->sizeavail[i] = GRU_SIZEAVAIL(PAGE_SHIFT);
- }
-}
-#elif defined CONFIG_X86_64
-static inline void cch_allocate_set_asids(
- struct gru_context_configuration_handle *cch, int asidval)
-{
- int i;
-
- for (i = 0; i < 8; i++) {
- cch->asid[i] = asidval++;
- cch->sizeavail[i] = GRU_SIZEAVAIL(PAGE_SHIFT) |
- GRU_SIZEAVAIL(21);
- }
-}
-#endif
-
-static inline int cch_allocate(struct gru_context_configuration_handle *cch,
- int asidval, unsigned long cbrmap,
- unsigned long dsrmap)
-{
- cch_allocate_set_asids(cch, asidval);
- cch->dsr_allocation_map = dsrmap;
- cch->cbr_allocation_map = cbrmap;
- cch->opc = CCHOP_ALLOCATE;
- start_instruction(cch);
- return wait_instruction_complete(cch);
-}
-
-static inline int cch_start(struct gru_context_configuration_handle *cch)
-{
- cch->opc = CCHOP_START;
- start_instruction(cch);
- return wait_instruction_complete(cch);
-}
-
-static inline int cch_interrupt(struct gru_context_configuration_handle *cch)
-{
- cch->opc = CCHOP_INTERRUPT;
- start_instruction(cch);
- return wait_instruction_complete(cch);
-}
-
-static inline int cch_deallocate(struct gru_context_configuration_handle *cch)
-{
- cch->opc = CCHOP_DEALLOCATE;
- start_instruction(cch);
- return wait_instruction_complete(cch);
-}
-
-static inline int cch_interrupt_sync(struct gru_context_configuration_handle
- *cch)
-{
- cch->opc = CCHOP_INTERRUPT_SYNC;
- start_instruction(cch);
- return wait_instruction_complete(cch);
-}
-
-static inline int tgh_invalidate(struct gru_tlb_global_handle *tgh,
- unsigned long vaddr, unsigned long vaddrmask,
- int asid, int pagesize, int global, int n,
- unsigned short ctxbitmap)
-{
- tgh->vaddr = vaddr;
- tgh->asid = asid;
- tgh->pagesize = pagesize;
- tgh->n = n;
- tgh->global = global;
- tgh->vaddrmask = vaddrmask;
- tgh->ctxbitmap = ctxbitmap;
- tgh->opc = TGHOP_TLBINV;
- start_instruction(tgh);
- return wait_instruction_complete(tgh);
-}
-
-static inline void tfh_write_only(struct gru_tlb_fault_handle *tfh,
- unsigned long pfn, unsigned long vaddr,
- int asid, int dirty, int pagesize)
-{
- tfh->fillasid = asid;
- tfh->fillvaddr = vaddr;
- tfh->pfn = pfn;
- tfh->dirty = dirty;
- tfh->pagesize = pagesize;
- tfh->opc = TFHOP_WRITE_ONLY;
- start_instruction(tfh);
-}
-
-static inline void tfh_write_restart(struct gru_tlb_fault_handle *tfh,
- unsigned long paddr, int gaa,
- unsigned long vaddr, int asid, int dirty,
- int pagesize)
-{
- tfh->fillasid = asid;
- tfh->fillvaddr = vaddr;
- tfh->pfn = paddr >> GRU_PADDR_SHIFT;
- tfh->gaa = gaa;
- tfh->dirty = dirty;
- tfh->pagesize = pagesize;
- tfh->opc = TFHOP_WRITE_RESTART;
- start_instruction(tfh);
-}
-
-static inline void tfh_restart(struct gru_tlb_fault_handle *tfh)
-{
- tfh->opc = TFHOP_RESTART;
- start_instruction(tfh);
-}
-
-static inline void tfh_user_polling_mode(struct gru_tlb_fault_handle *tfh)
-{
- tfh->opc = TFHOP_USER_POLLING_MODE;
- start_instruction(tfh);
-}
-
-static inline void tfh_exception(struct gru_tlb_fault_handle *tfh)
-{
- tfh->opc = TFHOP_EXCEPTION;
- start_instruction(tfh);
-}
+int cch_allocate(struct gru_context_configuration_handle *cch,
+ int asidval, int sizeavail, unsigned long cbrmap, unsigned long dsrmap);
+
+int cch_start(struct gru_context_configuration_handle *cch);
+int cch_interrupt(struct gru_context_configuration_handle *cch);
+int cch_deallocate(struct gru_context_configuration_handle *cch);
+int cch_interrupt_sync(struct gru_context_configuration_handle *cch);
+int tgh_invalidate(struct gru_tlb_global_handle *tgh, unsigned long vaddr,
+ unsigned long vaddrmask, int asid, int pagesize, int global, int n,
+ unsigned short ctxbitmap);
+void tfh_write_only(struct gru_tlb_fault_handle *tfh, unsigned long pfn,
+ unsigned long vaddr, int asid, int dirty, int pagesize);
+void tfh_write_restart(struct gru_tlb_fault_handle *tfh, unsigned long paddr,
+ int gaa, unsigned long vaddr, int asid, int dirty, int pagesize);
+void tfh_restart(struct gru_tlb_fault_handle *tfh);
+void tfh_user_polling_mode(struct gru_tlb_fault_handle *tfh);
+void tfh_exception(struct gru_tlb_fault_handle *tfh);
#endif /* __GRUHANDLES_H__ */
diff --git a/drivers/misc/sgi-gru/grukservices.c b/drivers/misc/sgi-gru/grukservices.c
index 880c55dfb66..d8bd7d84a7c 100644
--- a/drivers/misc/sgi-gru/grukservices.c
+++ b/drivers/misc/sgi-gru/grukservices.c
@@ -52,8 +52,10 @@
*/
/* Blade percpu resources PERMANENTLY reserved for kernel use */
-#define GRU_NUM_KERNEL_CBR 1
+#define GRU_NUM_KERNEL_CBR 1
#define GRU_NUM_KERNEL_DSR_BYTES 256
+#define GRU_NUM_KERNEL_DSR_CL (GRU_NUM_KERNEL_DSR_BYTES / \
+ GRU_CACHE_LINE_BYTES)
#define KERNEL_CTXNUM 15
/* GRU instruction attributes for all instructions */
@@ -94,7 +96,6 @@ struct message_header {
char fill;
};
-#define QLINES(mq) ((mq) + offsetof(struct message_queue, qlines))
#define HSTATUS(mq, h) ((mq) + offsetof(struct message_queue, hstatus[h]))
static int gru_get_cpu_resources(int dsr_bytes, void **cb, void **dsr)
@@ -122,7 +123,7 @@ int gru_get_cb_exception_detail(void *cb,
struct gru_control_block_extended *cbe;
cbe = get_cbe(GRUBASE(cb), get_cb_number(cb));
- prefetchw(cbe); /* Harmless on hardware, required for emulator */
+ prefetchw(cbe); /* Harmless on hardware, required for emulator */
excdet->opc = cbe->opccpy;
excdet->exopc = cbe->exopccpy;
excdet->ecause = cbe->ecause;
@@ -250,7 +251,8 @@ static inline void restore_present2(void *p, int val)
* Create a message queue.
* qlines - message queue size in cache lines. Includes 2-line header.
*/
-int gru_create_message_queue(void *p, unsigned int bytes)
+int gru_create_message_queue(struct gru_message_queue_desc *mqd,
+ void *p, unsigned int bytes, int nasid, int vector, int apicid)
{
struct message_queue *mq = p;
unsigned int qlines;
@@ -265,6 +267,12 @@ int gru_create_message_queue(void *p, unsigned int bytes)
mq->hstatus[0] = 0;
mq->hstatus[1] = 1;
mq->head = gru_mesq_head(2, qlines / 2 + 1);
+ mqd->mq = mq;
+ mqd->mq_gpa = uv_gpa(mq);
+ mqd->qlines = qlines;
+ mqd->interrupt_pnode = UV_NASID_TO_PNODE(nasid);
+ mqd->interrupt_vector = vector;
+ mqd->interrupt_apicid = apicid;
return 0;
}
EXPORT_SYMBOL_GPL(gru_create_message_queue);
@@ -277,8 +285,8 @@ EXPORT_SYMBOL_GPL(gru_create_message_queue);
* -1 - if mesq sent successfully but queue not full
* >0 - unexpected error. MQE_xxx returned
*/
-static int send_noop_message(void *cb,
- unsigned long mq, void *mesg)
+static int send_noop_message(void *cb, struct gru_message_queue_desc *mqd,
+ void *mesg)
{
const struct message_header noop_header = {
.present = MQS_NOOP, .lines = 1};
@@ -289,7 +297,7 @@ static int send_noop_message(void *cb,
STAT(mesq_noop);
save_mhdr = *mhdr;
*mhdr = noop_header;
- gru_mesq(cb, mq, gru_get_tri(mhdr), 1, IMA);
+ gru_mesq(cb, mqd->mq_gpa, gru_get_tri(mhdr), 1, IMA);
ret = gru_wait(cb);
if (ret) {
@@ -313,7 +321,7 @@ static int send_noop_message(void *cb,
break;
case CBSS_PUT_NACKED:
STAT(mesq_noop_put_nacked);
- m = mq + (gru_get_amo_value_head(cb) << 6);
+ m = mqd->mq_gpa + (gru_get_amo_value_head(cb) << 6);
gru_vstore(cb, m, gru_get_tri(mesg), XTYPE_CL, 1, 1,
IMA);
if (gru_wait(cb) == CBS_IDLE)
@@ -333,30 +341,20 @@ static int send_noop_message(void *cb,
/*
* Handle a gru_mesq full.
*/
-static int send_message_queue_full(void *cb,
- unsigned long mq, void *mesg, int lines)
+static int send_message_queue_full(void *cb, struct gru_message_queue_desc *mqd,
+ void *mesg, int lines)
{
union gru_mesqhead mqh;
unsigned int limit, head;
unsigned long avalue;
- int half, qlines, save;
+ int half, qlines;
/* Determine if switching to first/second half of q */
avalue = gru_get_amo_value(cb);
head = gru_get_amo_value_head(cb);
limit = gru_get_amo_value_limit(cb);
- /*
- * Fetch "qlines" from the queue header. Since the queue may be
- * in memory that can't be accessed using socket addresses, use
- * the GRU to access the data. Use DSR space from the message.
- */
- save = *(int *)mesg;
- gru_vload(cb, QLINES(mq), gru_get_tri(mesg), XTYPE_W, 1, 1, IMA);
- if (gru_wait(cb) != CBS_IDLE)
- goto cberr;
- qlines = *(int *)mesg;
- *(int *)mesg = save;
+ qlines = mqd->qlines;
half = (limit != qlines);
if (half)
@@ -365,7 +363,7 @@ static int send_message_queue_full(void *cb,
mqh = gru_mesq_head(2, qlines / 2 + 1);
/* Try to get lock for switching head pointer */
- gru_gamir(cb, EOP_IR_CLR, HSTATUS(mq, half), XTYPE_DW, IMA);
+ gru_gamir(cb, EOP_IR_CLR, HSTATUS(mqd->mq_gpa, half), XTYPE_DW, IMA);
if (gru_wait(cb) != CBS_IDLE)
goto cberr;
if (!gru_get_amo_value(cb)) {
@@ -375,8 +373,8 @@ static int send_message_queue_full(void *cb,
/* Got the lock. Send optional NOP if queue not full, */
if (head != limit) {
- if (send_noop_message(cb, mq, mesg)) {
- gru_gamir(cb, EOP_IR_INC, HSTATUS(mq, half),
+ if (send_noop_message(cb, mqd, mesg)) {
+ gru_gamir(cb, EOP_IR_INC, HSTATUS(mqd->mq_gpa, half),
XTYPE_DW, IMA);
if (gru_wait(cb) != CBS_IDLE)
goto cberr;
@@ -387,14 +385,16 @@ static int send_message_queue_full(void *cb,
}
/* Then flip queuehead to other half of queue. */
- gru_gamer(cb, EOP_ERR_CSWAP, mq, XTYPE_DW, mqh.val, avalue, IMA);
+ gru_gamer(cb, EOP_ERR_CSWAP, mqd->mq_gpa, XTYPE_DW, mqh.val, avalue,
+ IMA);
if (gru_wait(cb) != CBS_IDLE)
goto cberr;
/* If not successfully in swapping queue head, clear the hstatus lock */
if (gru_get_amo_value(cb) != avalue) {
STAT(mesq_qf_switch_head_failed);
- gru_gamir(cb, EOP_IR_INC, HSTATUS(mq, half), XTYPE_DW, IMA);
+ gru_gamir(cb, EOP_IR_INC, HSTATUS(mqd->mq_gpa, half), XTYPE_DW,
+ IMA);
if (gru_wait(cb) != CBS_IDLE)
goto cberr;
}
@@ -404,15 +404,25 @@ cberr:
return MQE_UNEXPECTED_CB_ERR;
}
+/*
+ * Send a cross-partition interrupt to the SSI that contains the target
+ * message queue. Normally, the interrupt is automatically delivered by hardware
+ * but some error conditions require explicit delivery.
+ */
+static void send_message_queue_interrupt(struct gru_message_queue_desc *mqd)
+{
+ if (mqd->interrupt_vector)
+ uv_hub_send_ipi(mqd->interrupt_pnode, mqd->interrupt_apicid,
+ mqd->interrupt_vector);
+}
+
/*
* Handle a gru_mesq failure. Some of these failures are software recoverable
* or retryable.
*/
-static int send_message_failure(void *cb,
- unsigned long mq,
- void *mesg,
- int lines)
+static int send_message_failure(void *cb, struct gru_message_queue_desc *mqd,
+ void *mesg, int lines)
{
int substatus, ret = 0;
unsigned long m;
@@ -429,7 +439,7 @@ static int send_message_failure(void *cb,
break;
case CBSS_QLIMIT_REACHED:
STAT(mesq_send_qlimit_reached);
- ret = send_message_queue_full(cb, mq, mesg, lines);
+ ret = send_message_queue_full(cb, mqd, mesg, lines);
break;
case CBSS_AMO_NACKED:
STAT(mesq_send_amo_nacked);
@@ -437,12 +447,14 @@ static int send_message_failure(void *cb,
break;
case CBSS_PUT_NACKED:
STAT(mesq_send_put_nacked);
- m =mq + (gru_get_amo_value_head(cb) << 6);
+ m = mqd->mq_gpa + (gru_get_amo_value_head(cb) << 6);
gru_vstore(cb, m, gru_get_tri(mesg), XTYPE_CL, lines, 1, IMA);
- if (gru_wait(cb) == CBS_IDLE)
+ if (gru_wait(cb) == CBS_IDLE) {
ret = MQE_OK;
- else
+ send_message_queue_interrupt(mqd);
+ } else {
ret = MQE_UNEXPECTED_CB_ERR;
+ }
break;
default:
BUG();
@@ -452,12 +464,12 @@ static int send_message_failure(void *cb,
/*
* Send a message to a message queue
- * cb GRU control block to use to send message
- * mq message queue
+ * mqd message queue descriptor
* mesg message. ust be vaddr within a GSEG
* bytes message size (<= 2 CL)
*/
-int gru_send_message_gpa(unsigned long mq, void *mesg, unsigned int bytes)
+int gru_send_message_gpa(struct gru_message_queue_desc *mqd, void *mesg,
+ unsigned int bytes)
{
struct message_header *mhdr;
void *cb;
@@ -481,10 +493,10 @@ int gru_send_message_gpa(unsigned long mq, void *mesg, unsigned int bytes)
do {
ret = MQE_OK;
- gru_mesq(cb, mq, gru_get_tri(mhdr), clines, IMA);
+ gru_mesq(cb, mqd->mq_gpa, gru_get_tri(mhdr), clines, IMA);
istatus = gru_wait(cb);
if (istatus != CBS_IDLE)
- ret = send_message_failure(cb, mq, dsr, clines);
+ ret = send_message_failure(cb, mqd, dsr, clines);
} while (ret == MQIE_AGAIN);
gru_free_cpu_resources(cb, dsr);
@@ -497,9 +509,9 @@ EXPORT_SYMBOL_GPL(gru_send_message_gpa);
/*
* Advance the receive pointer for the queue to the next message.
*/
-void gru_free_message(void *rmq, void *mesg)
+void gru_free_message(struct gru_message_queue_desc *mqd, void *mesg)
{
- struct message_queue *mq = rmq;
+ struct message_queue *mq = mqd->mq;
struct message_header *mhdr = mq->next;
void *next, *pnext;
int half = -1;
@@ -529,16 +541,16 @@ EXPORT_SYMBOL_GPL(gru_free_message);
* present. User must call next_message() to move to next message.
* rmq message queue
*/
-void *gru_get_next_message(void *rmq)
+void *gru_get_next_message(struct gru_message_queue_desc *mqd)
{
- struct message_queue *mq = rmq;
+ struct message_queue *mq = mqd->mq;
struct message_header *mhdr = mq->next;
int present = mhdr->present;
/* skip NOOP messages */
STAT(mesq_receive);
while (present == MQS_NOOP) {
- gru_free_message(rmq, mhdr);
+ gru_free_message(mqd, mhdr);
mhdr = mq->next;
present = mhdr->present;
}
@@ -576,7 +588,7 @@ int gru_copy_gpa(unsigned long dest_gpa, unsigned long src_gpa,
if (gru_get_cpu_resources(GRU_NUM_KERNEL_DSR_BYTES, &cb, &dsr))
return MQE_BUG_NO_RESOURCES;
gru_bcopy(cb, src_gpa, dest_gpa, gru_get_tri(dsr),
- XTYPE_B, bytes, GRU_NUM_KERNEL_DSR_BYTES, IMA);
+ XTYPE_B, bytes, GRU_NUM_KERNEL_DSR_CL, IMA);
ret = gru_wait(cb);
gru_free_cpu_resources(cb, dsr);
return ret;
@@ -611,7 +623,7 @@ static int quicktest(struct gru_state *gru)
if (word0 != word1 || word0 != MAGIC) {
printk
- ("GRU quicktest err: gru %d, found 0x%lx, expected 0x%lx\n",
+ ("GRU quicktest err: gid %d, found 0x%lx, expected 0x%lx\n",
gru->gs_gid, word1, MAGIC);
BUG(); /* ZZZ should not be fatal */
}
@@ -660,15 +672,15 @@ int gru_kservices_init(struct gru_state *gru)
cch->tlb_int_enable = 0;
cch->tfm_done_bit_enable = 0;
cch->unmap_enable = 1;
- err = cch_allocate(cch, 0, cbr_map, dsr_map);
+ err = cch_allocate(cch, 0, 0, cbr_map, dsr_map);
if (err) {
gru_dbg(grudev,
- "Unable to allocate kernel CCH: gru %d, err %d\n",
+ "Unable to allocate kernel CCH: gid %d, err %d\n",
gru->gs_gid, err);
BUG();
}
if (cch_start(cch)) {
- gru_dbg(grudev, "Unable to start kernel CCH: gru %d, err %d\n",
+ gru_dbg(grudev, "Unable to start kernel CCH: gid %d, err %d\n",
gru->gs_gid, err);
BUG();
}
@@ -678,3 +690,22 @@ int gru_kservices_init(struct gru_state *gru)
quicktest(gru);
return 0;
}
+
+void gru_kservices_exit(struct gru_state *gru)
+{
+ struct gru_context_configuration_handle *cch;
+ struct gru_blade_state *bs;
+
+ bs = gru->gs_blade;
+ if (gru != &bs->bs_grus[1])
+ return;
+
+ cch = get_cch(gru->gs_gru_base_vaddr, KERNEL_CTXNUM);
+ lock_cch_handle(cch);
+ if (cch_interrupt_sync(cch))
+ BUG();
+ if (cch_deallocate(cch))
+ BUG();
+ unlock_cch_handle(cch);
+}
+
diff --git a/drivers/misc/sgi-gru/grukservices.h b/drivers/misc/sgi-gru/grukservices.h
index eb17e0a3ac6..747ed315d56 100644
--- a/drivers/misc/sgi-gru/grukservices.h
+++ b/drivers/misc/sgi-gru/grukservices.h
@@ -41,6 +41,15 @@
* - gru_create_message_queue() needs interrupt vector info
*/
+struct gru_message_queue_desc {
+ void *mq; /* message queue vaddress */
+ unsigned long mq_gpa; /* global address of mq */
+ int qlines; /* queue size in CL */
+ int interrupt_vector; /* interrupt vector */
+ int interrupt_pnode; /* pnode for interrupt */
+ int interrupt_apicid; /* lapicid for interrupt */
+};
+
/*
* Initialize a user allocated chunk of memory to be used as
* a message queue. The caller must ensure that the queue is
@@ -51,14 +60,19 @@
* to manage the queue.
*
* Input:
- * p pointer to user allocated memory.
+ * mqd pointer to message queue descriptor
+ * p pointer to user allocated mesq memory.
* bytes size of message queue in bytes
+ * vector interrupt vector (zero if no interrupts)
+ * nasid nasid of blade where interrupt is delivered
+ * apicid apicid of cpu for interrupt
*
* Errors:
* 0 OK
* >0 error
*/
-extern int gru_create_message_queue(void *p, unsigned int bytes);
+extern int gru_create_message_queue(struct gru_message_queue_desc *mqd,
+ void *p, unsigned int bytes, int nasid, int vector, int apicid);
/*
* Send a message to a message queue.
@@ -68,7 +82,7 @@ extern int gru_create_message_queue(void *p, unsigned int bytes);
*
*
* Input:
- * xmq message queue - must be a UV global physical address
+ * mqd pointer to message queue descriptor
* mesg pointer to message. Must be 64-bit aligned
* bytes size of message in bytes
*
@@ -77,8 +91,8 @@ extern int gru_create_message_queue(void *p, unsigned int bytes);
* >0 Send failure - see error codes below
*
*/
-extern int gru_send_message_gpa(unsigned long mq_gpa, void *mesg,
- unsigned int bytes);
+extern int gru_send_message_gpa(struct gru_message_queue_desc *mqd,
+ void *mesg, unsigned int bytes);
/* Status values for gru_send_message() */
#define MQE_OK 0 /* message sent successfully */
@@ -94,10 +108,11 @@ extern int gru_send_message_gpa(unsigned long mq_gpa, void *mesg,
* API extensions may allow for out-of-order freeing.
*
* Input
- * mq message queue
+ * mqd pointer to message queue descriptor
* mesq message being freed
*/
-extern void gru_free_message(void *mq, void *mesq);
+extern void gru_free_message(struct gru_message_queue_desc *mqd,
+ void *mesq);
/*
* Get next message from message queue. Returns pointer to
@@ -106,13 +121,13 @@ extern void gru_free_message(void *mq, void *mesq);
* in order to move the queue pointers to next message.
*
* Input
- * mq message queue
+ * mqd pointer to message queue descriptor
*
* Output:
* p pointer to message
* NULL no message available
*/
-extern void *gru_get_next_message(void *mq);
+extern void *gru_get_next_message(struct gru_message_queue_desc *mqd);
/*
diff --git a/drivers/misc/sgi-gru/grumain.c b/drivers/misc/sgi-gru/grumain.c
index 3d2fc216bae..ec3f7a17d22 100644
--- a/drivers/misc/sgi-gru/grumain.c
+++ b/drivers/misc/sgi-gru/grumain.c
@@ -76,10 +76,9 @@ int gru_cpu_fault_map_id(void)
/* Hit the asid limit. Start over */
static int gru_wrap_asid(struct gru_state *gru)
{
- gru_dbg(grudev, "gru %p\n", gru);
+ gru_dbg(grudev, "gid %d\n", gru->gs_gid);
STAT(asid_wrap);
gru->gs_asid_gen++;
- gru_flush_all_tlb(gru);
return MIN_ASID;
}
@@ -88,19 +87,21 @@ static int gru_reset_asid_limit(struct gru_state *gru, int asid)
{
int i, gid, inuse_asid, limit;
- gru_dbg(grudev, "gru %p, asid 0x%x\n", gru, asid);
+ gru_dbg(grudev, "gid %d, asid 0x%x\n", gru->gs_gid, asid);
STAT(asid_next);
limit = MAX_ASID;
if (asid >= limit)
asid = gru_wrap_asid(gru);
+ gru_flush_all_tlb(gru);
gid = gru->gs_gid;
again:
for (i = 0; i < GRU_NUM_CCH; i++) {
if (!gru->gs_gts[i])
continue;
inuse_asid = gru->gs_gts[i]->ts_gms->ms_asids[gid].mt_asid;
- gru_dbg(grudev, "gru %p, inuse_asid 0x%x, cxtnum %d, gts %p\n",
- gru, inuse_asid, i, gru->gs_gts[i]);
+ gru_dbg(grudev, "gid %d, gts %p, gms %p, inuse 0x%x, cxt %d\n",
+ gru->gs_gid, gru->gs_gts[i], gru->gs_gts[i]->ts_gms,
+ inuse_asid, i);
if (inuse_asid == asid) {
asid += ASID_INC;
if (asid >= limit) {
@@ -120,8 +121,8 @@ again:
}
gru->gs_asid_limit = limit;
gru->gs_asid = asid;
- gru_dbg(grudev, "gru %p, new asid 0x%x, new_limit 0x%x\n", gru, asid,
- limit);
+ gru_dbg(grudev, "gid %d, new asid 0x%x, new_limit 0x%x\n", gru->gs_gid,
+ asid, limit);
return asid;
}
@@ -130,14 +131,12 @@ static int gru_assign_asid(struct gru_state *gru)
{
int asid;
- spin_lock(&gru->gs_asid_lock);
gru->gs_asid += ASID_INC;
asid = gru->gs_asid;
if (asid >= gru->gs_asid_limit)
asid = gru_reset_asid_limit(gru, asid);
- spin_unlock(&gru->gs_asid_lock);
- gru_dbg(grudev, "gru %p, asid 0x%x\n", gru, asid);
+ gru_dbg(grudev, "gid %d, asid 0x%x\n", gru->gs_gid, asid);
return asid;
}
@@ -215,17 +214,20 @@ static int check_gru_resources(struct gru_state *gru, int cbr_au_count,
* TLB manangment requires tracking all GRU chiplets that have loaded a GSEG
* context.
*/
-static int gru_load_mm_tracker(struct gru_state *gru, struct gru_mm_struct *gms,
- int ctxnum)
+static int gru_load_mm_tracker(struct gru_state *gru,
+ struct gru_thread_state *gts)
{
+ struct gru_mm_struct *gms = gts->ts_gms;
struct gru_mm_tracker *asids = &gms->ms_asids[gru->gs_gid];
- unsigned short ctxbitmap = (1 << ctxnum);
+ unsigned short ctxbitmap = (1 << gts->ts_ctxnum);
int asid;
spin_lock(&gms->ms_asid_lock);
asid = asids->mt_asid;
- if (asid == 0 || asids->mt_asid_gen != gru->gs_asid_gen) {
+ spin_lock(&gru->gs_asid_lock);
+ if (asid == 0 || (asids->mt_ctxbitmap == 0 && asids->mt_asid_gen !=
+ gru->gs_asid_gen)) {
asid = gru_assign_asid(gru);
asids->mt_asid = asid;
asids->mt_asid_gen = gru->gs_asid_gen;
@@ -233,6 +235,7 @@ static int gru_load_mm_tracker(struct gru_state *gru, struct gru_mm_struct *gms,
} else {
STAT(asid_reuse);
}
+ spin_unlock(&gru->gs_asid_lock);
BUG_ON(asids->mt_ctxbitmap & ctxbitmap);
asids->mt_ctxbitmap |= ctxbitmap;
@@ -241,24 +244,28 @@ static int gru_load_mm_tracker(struct gru_state *gru, struct gru_mm_struct *gms,
spin_unlock(&gms->ms_asid_lock);
gru_dbg(grudev,
- "gru %x, gms %p, ctxnum 0x%d, asid 0x%x, asidmap 0x%lx\n",
- gru->gs_gid, gms, ctxnum, asid, gms->ms_asidmap[0]);
+ "gid %d, gts %p, gms %p, ctxnum %d, asid 0x%x, asidmap 0x%lx\n",
+ gru->gs_gid, gts, gms, gts->ts_ctxnum, asid,
+ gms->ms_asidmap[0]);
return asid;
}
static void gru_unload_mm_tracker(struct gru_state *gru,
- struct gru_mm_struct *gms, int ctxnum)
+ struct gru_thread_state *gts)
{
+ struct gru_mm_struct *gms = gts->ts_gms;
struct gru_mm_tracker *asids;
unsigned short ctxbitmap;
asids = &gms->ms_asids[gru->gs_gid];
- ctxbitmap = (1 << ctxnum);
+ ctxbitmap = (1 << gts->ts_ctxnum);
spin_lock(&gms->ms_asid_lock);
+ spin_lock(&gru->gs_asid_lock);
BUG_ON((asids->mt_ctxbitmap & ctxbitmap) != ctxbitmap);
asids->mt_ctxbitmap ^= ctxbitmap;
- gru_dbg(grudev, "gru %x, gms %p, ctxnum 0x%d, asidmap 0x%lx\n",
- gru->gs_gid, gms, ctxnum, gms->ms_asidmap[0]);
+ gru_dbg(grudev, "gid %d, gts %p, gms %p, ctxnum 0x%d, asidmap 0x%lx\n",
+ gru->gs_gid, gts, gms, gts->ts_ctxnum, gms->ms_asidmap[0]);
+ spin_unlock(&gru->gs_asid_lock);
spin_unlock(&gms->ms_asid_lock);
}
@@ -319,6 +326,7 @@ static struct gru_thread_state *gru_alloc_gts(struct vm_area_struct *vma,
gts->ts_vma = vma;
gts->ts_tlb_int_select = -1;
gts->ts_gms = gru_register_mmu_notifier();
+ gts->ts_sizeavail = GRU_SIZEAVAIL(PAGE_SHIFT);
if (!gts->ts_gms)
goto err;
@@ -399,7 +407,7 @@ static void gru_free_gru_context(struct gru_thread_state *gts)
struct gru_state *gru;
gru = gts->ts_gru;
- gru_dbg(grudev, "gts %p, gru %p\n", gts, gru);
+ gru_dbg(grudev, "gts %p, gid %d\n", gts, gru->gs_gid);
spin_lock(&gru->gs_lock);
gru->gs_gts[gts->ts_ctxnum] = NULL;
@@ -408,6 +416,7 @@ static void gru_free_gru_context(struct gru_thread_state *gts)
__clear_bit(gts->ts_ctxnum, &gru->gs_context_map);
gts->ts_ctxnum = NULLCTX;
gts->ts_gru = NULL;
+ gts->ts_blade = -1;
spin_unlock(&gru->gs_lock);
gts_drop(gts);
@@ -432,8 +441,8 @@ static inline long gru_copy_handle(void *d, void *s)
return GRU_HANDLE_BYTES;
}
-static void gru_prefetch_context(void *gseg, void *cb, void *cbe, unsigned long cbrmap,
- unsigned long length)
+static void gru_prefetch_context(void *gseg, void *cb, void *cbe,
+ unsigned long cbrmap, unsigned long length)
{
int i, scr;
@@ -500,12 +509,12 @@ void gru_unload_context(struct gru_thread_state *gts, int savestate)
zap_vma_ptes(gts->ts_vma, UGRUADDR(gts), GRU_GSEG_PAGESIZE);
cch = get_cch(gru->gs_gru_base_vaddr, ctxnum);
+ gru_dbg(grudev, "gts %p\n", gts);
lock_cch_handle(cch);
if (cch_interrupt_sync(cch))
BUG();
- gru_dbg(grudev, "gts %p\n", gts);
- gru_unload_mm_tracker(gru, gts->ts_gms, gts->ts_ctxnum);
+ gru_unload_mm_tracker(gru, gts);
if (savestate)
gru_unload_context_data(gts->ts_gdata, gru->gs_gru_base_vaddr,
ctxnum, gts->ts_cbr_map,
@@ -534,7 +543,7 @@ static void gru_load_context(struct gru_thread_state *gts)
cch = get_cch(gru->gs_gru_base_vaddr, ctxnum);
lock_cch_handle(cch);
- asid = gru_load_mm_tracker(gru, gts->ts_gms, gts->ts_ctxnum);
+ asid = gru_load_mm_tracker(gru, gts);
cch->tfm_fault_bit_enable =
(gts->ts_user_options == GRU_OPT_MISS_FMM_POLL
|| gts->ts_user_options == GRU_OPT_MISS_FMM_INTR);
@@ -544,7 +553,8 @@ static void gru_load_context(struct gru_thread_state *gts)
cch->tlb_int_select = gts->ts_tlb_int_select;
}
cch->tfm_done_bit_enable = 0;
- err = cch_allocate(cch, asid, gts->ts_cbr_map, gts->ts_dsr_map);
+ err = cch_allocate(cch, asid, gts->ts_sizeavail, gts->ts_cbr_map,
+ gts->ts_dsr_map);
if (err) {
gru_dbg(grudev,
"err %d: cch %p, gts %p, cbr 0x%lx, dsr 0x%lx\n",
@@ -565,11 +575,12 @@ static void gru_load_context(struct gru_thread_state *gts)
/*
* Update fields in an active CCH:
* - retarget interrupts on local blade
+ * - update sizeavail mask
* - force a delayed context unload by clearing the CCH asids. This
* forces TLB misses for new GRU instructions. The context is unloaded
* when the next TLB miss occurs.
*/
-static int gru_update_cch(struct gru_thread_state *gts, int int_select)
+int gru_update_cch(struct gru_thread_state *gts, int force_unload)
{
struct gru_context_configuration_handle *cch;
struct gru_state *gru = gts->ts_gru;
@@ -583,9 +594,11 @@ static int gru_update_cch(struct gru_thread_state *gts, int int_select)
goto exit;
if (cch_interrupt(cch))
BUG();
- if (int_select >= 0) {
- gts->ts_tlb_int_select = int_select;
- cch->tlb_int_select = int_select;
+ if (!force_unload) {
+ for (i = 0; i < 8; i++)
+ cch->sizeavail[i] = gts->ts_sizeavail;
+ gts->ts_tlb_int_select = gru_cpu_fault_map_id();
+ cch->tlb_int_select = gru_cpu_fault_map_id();
} else {
for (i = 0; i < 8; i++)
cch->asid[i] = 0;
@@ -617,7 +630,7 @@ static int gru_retarget_intr(struct gru_thread_state *gts)
gru_dbg(grudev, "retarget from %d to %d\n", gts->ts_tlb_int_select,
gru_cpu_fault_map_id());
- return gru_update_cch(gts, gru_cpu_fault_map_id());
+ return gru_update_cch(gts, 0);
}
@@ -688,7 +701,7 @@ static void gru_steal_context(struct gru_thread_state *gts)
STAT(steal_context_failed);
}
gru_dbg(grudev,
- "stole gru %x, ctxnum %d from gts %p. Need cb %d, ds %d;"
+ "stole gid %d, ctxnum %d from gts %p. Need cb %d, ds %d;"
" avail cb %ld, ds %ld\n",
gru->gs_gid, ctxnum, ngts, cbr, dsr, hweight64(gru->gs_cbr_map),
hweight64(gru->gs_dsr_map));
@@ -727,6 +740,7 @@ again:
}
reserve_gru_resources(gru, gts);
gts->ts_gru = gru;
+ gts->ts_blade = gru->gs_blade_id;
gts->ts_ctxnum =
find_first_zero_bit(&gru->gs_context_map, GRU_NUM_CCH);
BUG_ON(gts->ts_ctxnum == GRU_NUM_CCH);
@@ -737,7 +751,7 @@ again:
STAT(assign_context);
gru_dbg(grudev,
- "gseg %p, gts %p, gru %x, ctx %d, cbr %d, dsr %d\n",
+ "gseg %p, gts %p, gid %d, ctx %d, cbr %d, dsr %d\n",
gseg_virtual_address(gts->ts_gru, gts->ts_ctxnum), gts,
gts->ts_gru->gs_gid, gts->ts_ctxnum,
gts->ts_cbr_au_count, gts->ts_dsr_au_count);
@@ -773,8 +787,8 @@ int gru_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
return VM_FAULT_SIGBUS;
again:
- preempt_disable();
mutex_lock(&gts->ts_ctxlock);
+ preempt_disable();
if (gts->ts_gru) {
if (gts->ts_gru->gs_blade_id != uv_numa_blade_id()) {
STAT(migrated_nopfn_unload);
diff --git a/drivers/misc/sgi-gru/gruprocfs.c b/drivers/misc/sgi-gru/gruprocfs.c
index 73b0ca061bb..ee74821b171 100644
--- a/drivers/misc/sgi-gru/gruprocfs.c
+++ b/drivers/misc/sgi-gru/gruprocfs.c
@@ -62,7 +62,9 @@ static int statistics_show(struct seq_file *s, void *p)
printstat(s, asid_wrap);
printstat(s, asid_reuse);
printstat(s, intr);
+ printstat(s, intr_mm_lock_failed);
printstat(s, call_os);
+ printstat(s, call_os_offnode_reference);
printstat(s, call_os_check_for_bug);
printstat(s, call_os_wait_queue);
printstat(s, user_flush_tlb);
@@ -120,6 +122,30 @@ static ssize_t statistics_write(struct file *file, const char __user *userbuf,
return count;
}
+static int mcs_statistics_show(struct seq_file *s, void *p)
+{
+ int op;
+ unsigned long total, count, max;
+ static char *id[] = {"cch_allocate", "cch_start", "cch_interrupt",
+ "cch_interrupt_sync", "cch_deallocate", "tgh_invalidate"};
+
+ for (op = 0; op < mcsop_last; op++) {
+ count = atomic_long_read(&mcs_op_statistics[op].count);
+ total = atomic_long_read(&mcs_op_statistics[op].total);
+ max = mcs_op_statistics[op].max;
+ seq_printf(s, "%-20s%12ld%12ld%12ld\n", id[op], count,
+ count ? total / count : 0, max);
+ }
+ return 0;
+}
+
+static ssize_t mcs_statistics_write(struct file *file,
+ const char __user *userbuf, size_t count, loff_t *data)
+{
+ memset(mcs_op_statistics, 0, sizeof(mcs_op_statistics));
+ return count;
+}
+
static int options_show(struct seq_file *s, void *p)
{
seq_printf(s, "0x%lx\n", gru_options);
@@ -135,6 +161,7 @@ static ssize_t options_write(struct file *file, const char __user *userbuf,
if (copy_from_user
(buf, userbuf, count < sizeof(buf) ? count : sizeof(buf)))
return -EFAULT;
+ buf[count - 1] = '\0';
if (!strict_strtoul(buf, 10, &val))
gru_options = val;
@@ -199,7 +226,7 @@ static void seq_stop(struct seq_file *file, void *data)
static void *seq_start(struct seq_file *file, loff_t *gid)
{
- if (*gid < GRU_MAX_GRUS)
+ if (*gid < gru_max_gids)
return gid;
return NULL;
}
@@ -207,7 +234,7 @@ static void *seq_start(struct seq_file *file, loff_t *gid)
static void *seq_next(struct seq_file *file, void *data, loff_t *gid)
{
(*gid)++;
- if (*gid < GRU_MAX_GRUS)
+ if (*gid < gru_max_gids)
return gid;
return NULL;
}
@@ -231,6 +258,11 @@ static int statistics_open(struct inode *inode, struct file *file)
return single_open(file, statistics_show, NULL);
}
+static int mcs_statistics_open(struct inode *inode, struct file *file)
+{
+ return single_open(file, mcs_statistics_show, NULL);
+}
+
static int options_open(struct inode *inode, struct file *file)
{
return single_open(file, options_show, NULL);
@@ -255,6 +287,14 @@ static const struct file_operations statistics_fops = {
.release = single_release,
};
+static const struct file_operations mcs_statistics_fops = {
+ .open = mcs_statistics_open,
+ .read = seq_read,
+ .write = mcs_statistics_write,
+ .llseek = seq_lseek,
+ .release = single_release,
+};
+
static const struct file_operations options_fops = {
.open = options_open,
.read = seq_read,
@@ -283,6 +323,7 @@ static struct proc_entry {
struct proc_dir_entry *entry;
} proc_files[] = {
{"statistics", 0644, &statistics_fops},
+ {"mcs_statistics", 0644, &mcs_statistics_fops},
{"debug_options", 0644, &options_fops},
{"cch_status", 0444, &cch_fops},
{"gru_status", 0444, &gru_fops},
diff --git a/drivers/misc/sgi-gru/grutables.h b/drivers/misc/sgi-gru/grutables.h
index a78f70deeb5..bf1eeb7553e 100644
--- a/drivers/misc/sgi-gru/grutables.h
+++ b/drivers/misc/sgi-gru/grutables.h
@@ -153,6 +153,7 @@
extern struct gru_stats_s gru_stats;
extern struct gru_blade_state *gru_base[];
extern unsigned long gru_start_paddr, gru_end_paddr;
+extern unsigned int gru_max_gids;
#define GRU_MAX_BLADES MAX_NUMNODES
#define GRU_MAX_GRUS (GRU_MAX_BLADES * GRU_CHIPLETS_PER_BLADE)
@@ -184,7 +185,9 @@ struct gru_stats_s {
atomic_long_t asid_wrap;
atomic_long_t asid_reuse;
atomic_long_t intr;
+ atomic_long_t intr_mm_lock_failed;
atomic_long_t call_os;
+ atomic_long_t call_os_offnode_reference;
atomic_long_t call_os_check_for_bug;
atomic_long_t call_os_wait_queue;
atomic_long_t user_flush_tlb;
@@ -237,6 +240,17 @@ struct gru_stats_s {
};
+enum mcs_op {cchop_allocate, cchop_start, cchop_interrupt, cchop_interrupt_sync,
+ cchop_deallocate, tghop_invalidate, mcsop_last};
+
+struct mcs_op_statistic {
+ atomic_long_t count;
+ atomic_long_t total;
+ unsigned long max;
+};
+
+extern struct mcs_op_statistic mcs_op_statistics[mcsop_last];
+
#define OPT_DPRINT 1
#define OPT_STATS 2
#define GRU_QUICKLOOK 4
@@ -278,13 +292,12 @@ struct gru_stats_s {
/* Generate a GRU asid value from a GRU base asid & a virtual address. */
#if defined CONFIG_IA64
#define VADDR_HI_BIT 64
-#define GRUREGION(addr) ((addr) >> (VADDR_HI_BIT - 3) & 3)
#elif defined CONFIG_X86_64
#define VADDR_HI_BIT 48
-#define GRUREGION(addr) (0) /* ZZZ could do better */
#else
#error "Unsupported architecture"
#endif
+#define GRUREGION(addr) ((addr) >> (VADDR_HI_BIT - 3) & 3)
#define GRUASID(asid, addr) ((asid) + GRUREGION(addr))
/*------------------------------------------------------------------------------
@@ -297,12 +310,12 @@ struct gru_state;
* This structure is pointed to from the mmstruct via the notifier pointer.
* There is one of these per address space.
*/
-struct gru_mm_tracker {
- unsigned int mt_asid_gen; /* ASID wrap count */
- int mt_asid; /* current base ASID for gru */
- unsigned short mt_ctxbitmap; /* bitmap of contexts using
+struct gru_mm_tracker { /* pack to reduce size */
+ unsigned int mt_asid_gen:24; /* ASID wrap count */
+ unsigned int mt_asid:24; /* current base ASID for gru */
+ unsigned short mt_ctxbitmap:16;/* bitmap of contexts using
asid */
-};
+} __attribute__ ((packed));
struct gru_mm_struct {
struct mmu_notifier ms_notifier;
@@ -348,6 +361,7 @@ struct gru_thread_state {
long ts_user_options;/* misc user option flags */
pid_t ts_tgid_owner; /* task that is using the
context - for migration */
+ unsigned short ts_sizeavail; /* Pagesizes in use */
int ts_tsid; /* thread that owns the
structure */
int ts_tlb_int_select;/* target cpu if interrupts
@@ -359,6 +373,9 @@ struct gru_thread_state {
required for contest */
unsigned char ts_cbr_au_count;/* Number of CBR resources
required for contest */
+ char ts_blade; /* If >= 0, migrate context if
+ ref from diferent blade */
+ char ts_force_cch_reload;
char ts_force_unload;/* force context to be unloaded
after migration */
char ts_cbr_idx[GRU_CBR_AU];/* CBR numbers of each
@@ -392,12 +409,12 @@ struct gru_state {
gru segments (64) */
void *gs_gru_base_vaddr; /* Virtual address of
gru segments (64) */
- unsigned char gs_gid; /* unique GRU number */
+ unsigned short gs_gid; /* unique GRU number */
+ unsigned short gs_blade_id; /* blade of GRU */
unsigned char gs_tgh_local_shift; /* used to pick TGH for
local flush */
unsigned char gs_tgh_first_remote; /* starting TGH# for
remote flush */
- unsigned short gs_blade_id; /* blade of GRU */
spinlock_t gs_asid_lock; /* lock used for
assigning asids */
spinlock_t gs_lock; /* lock used for
@@ -492,6 +509,10 @@ struct gru_blade_state {
(i) < GRU_CHIPLETS_PER_BLADE; \
(i)++, (gru)++)
+/* Scan all GRUs */
+#define foreach_gid(gid) \
+ for ((gid) = 0; (gid) < gru_max_gids; (gid)++)
+
/* Scan all active GTSs on a gru. Note: must hold ss_lock to use this macro. */
#define for_each_gts_on_gru(gts, gru, ctxnum) \
for ((ctxnum) = 0; (ctxnum) < GRU_NUM_CCH; (ctxnum)++) \
@@ -578,9 +599,11 @@ extern struct gru_thread_state *gru_find_thread_state(struct vm_area_struct
extern struct gru_thread_state *gru_alloc_thread_state(struct vm_area_struct
*vma, int tsid);
extern void gru_unload_context(struct gru_thread_state *gts, int savestate);
+extern int gru_update_cch(struct gru_thread_state *gts, int force_unload);
extern void gts_drop(struct gru_thread_state *gts);
extern void gru_tgh_flush_init(struct gru_state *gru);
extern int gru_kservices_init(struct gru_state *gru);
+extern void gru_kservices_exit(struct gru_state *gru);
extern irqreturn_t gru_intr(int irq, void *dev_id);
extern int gru_handle_user_call_os(unsigned long address);
extern int gru_user_flush_tlb(unsigned long arg);
diff --git a/drivers/misc/sgi-gru/grutlbpurge.c b/drivers/misc/sgi-gru/grutlbpurge.c
index c84496a7769..1d125091f5e 100644
--- a/drivers/misc/sgi-gru/grutlbpurge.c
+++ b/drivers/misc/sgi-gru/grutlbpurge.c
@@ -187,7 +187,7 @@ void gru_flush_tlb_range(struct gru_mm_struct *gms, unsigned long start,
" FLUSH gruid %d, asid 0x%x, num %ld, cbmap 0x%x\n",
gid, asid, num, asids->mt_ctxbitmap);
tgh = get_lock_tgh_handle(gru);
- tgh_invalidate(tgh, start, 0, asid, grupagesize, 0,
+ tgh_invalidate(tgh, start, ~0, asid, grupagesize, 0,
num - 1, asids->mt_ctxbitmap);
get_unlock_tgh_handle(tgh);
} else {
@@ -210,11 +210,10 @@ void gru_flush_all_tlb(struct gru_state *gru)
{
struct gru_tlb_global_handle *tgh;
- gru_dbg(grudev, "gru %p, gid %d\n", gru, gru->gs_gid);
+ gru_dbg(grudev, "gid %d\n", gru->gs_gid);
tgh = get_lock_tgh_handle(gru);
- tgh_invalidate(tgh, 0, ~0, 0, 1, 1, GRUMAXINVAL - 1, 0);
+ tgh_invalidate(tgh, 0, ~0, 0, 1, 1, GRUMAXINVAL - 1, 0xffff);
get_unlock_tgh_handle(tgh);
- preempt_enable();
}
/*
diff --git a/drivers/misc/sgi-xp/xpc.h b/drivers/misc/sgi-xp/xpc.h
index 275b78896a7..114444cfd49 100644
--- a/drivers/misc/sgi-xp/xpc.h
+++ b/drivers/misc/sgi-xp/xpc.h
@@ -92,7 +92,9 @@ struct xpc_rsvd_page {
u8 pad1[3]; /* align to next u64 in 1st 64-byte cacheline */
union {
unsigned long vars_pa; /* phys address of struct xpc_vars */
- unsigned long activate_mq_gpa; /* gru phy addr of activate_mq */
+ unsigned long activate_gru_mq_desc_gpa; /* phys addr of */
+ /* activate mq's */
+ /* gru mq descriptor */
} sn;
unsigned long ts_jiffies; /* timestamp when rsvd pg was setup by XPC */
u64 pad2[10]; /* align to last u64 in 2nd 64-byte cacheline */
@@ -189,7 +191,9 @@ struct xpc_gru_mq_uv {
int irq; /* irq raised when message is received in mq */
int mmr_blade; /* blade where watchlist was allocated from */
unsigned long mmr_offset; /* offset of irq mmr located on mmr_blade */
+ unsigned long mmr_value; /* value of irq mmr located on mmr_blade */
int watchlist_num; /* number of watchlist allocatd by BIOS */
+ void *gru_mq_desc; /* opaque structure used by the GRU driver */
};
/*
@@ -197,6 +201,7 @@ struct xpc_gru_mq_uv {
* heartbeat, partition active state, and channel state. This is UV only.
*/
struct xpc_activate_mq_msghdr_uv {
+ unsigned int gru_msg_hdr; /* FOR GRU INTERNAL USE ONLY */
short partid; /* sender's partid */
u8 act_state; /* sender's act_state at time msg sent */
u8 type; /* message's type */
@@ -232,7 +237,7 @@ struct xpc_activate_mq_msg_heartbeat_req_uv {
struct xpc_activate_mq_msg_activate_req_uv {
struct xpc_activate_mq_msghdr_uv hdr;
unsigned long rp_gpa;
- unsigned long activate_mq_gpa;
+ unsigned long activate_gru_mq_desc_gpa;
};
struct xpc_activate_mq_msg_deactivate_req_uv {
@@ -263,7 +268,7 @@ struct xpc_activate_mq_msg_chctl_openreply_uv {
short ch_number;
short remote_nentries; /* ??? Is this needed? What is? */
short local_nentries; /* ??? Is this needed? What is? */
- unsigned long local_notify_mq_gpa;
+ unsigned long notify_gru_mq_desc_gpa;
};
/*
@@ -510,8 +515,8 @@ struct xpc_channel_sn2 {
};
struct xpc_channel_uv {
- unsigned long remote_notify_mq_gpa; /* gru phys address of remote */
- /* partition's notify mq */
+ void *cached_notify_gru_mq_desc; /* remote partition's notify mq's */
+ /* gru mq descriptor */
struct xpc_send_msg_slot_uv *send_msg_slots;
void *recv_msg_slots; /* each slot will hold a xpc_notify_mq_msg_uv */
@@ -682,8 +687,12 @@ struct xpc_partition_sn2 {
};
struct xpc_partition_uv {
- unsigned long remote_activate_mq_gpa; /* gru phys address of remote */
- /* partition's activate mq */
+ unsigned long activate_gru_mq_desc_gpa; /* phys addr of parititon's */
+ /* activate mq's gru mq */
+ /* descriptor */
+ void *cached_activate_gru_mq_desc; /* cached copy of partition's */
+ /* activate mq's gru mq descriptor */
+ struct mutex cached_activate_gru_mq_desc_mutex;
spinlock_t flags_lock; /* protect updating of flags */
unsigned int flags; /* general flags */
u8 remote_act_state; /* remote partition's act_state */
@@ -694,8 +703,9 @@ struct xpc_partition_uv {
/* struct xpc_partition_uv flags */
-#define XPC_P_HEARTBEAT_OFFLINE_UV 0x00000001
-#define XPC_P_ENGAGED_UV 0x00000002
+#define XPC_P_HEARTBEAT_OFFLINE_UV 0x00000001
+#define XPC_P_ENGAGED_UV 0x00000002
+#define XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV 0x00000004
/* struct xpc_partition_uv act_state change requests */
@@ -804,6 +814,7 @@ extern void xpc_activate_kthreads(struct xpc_channel *, int);
extern void xpc_create_kthreads(struct xpc_channel *, int, int);
extern void xpc_disconnect_wait(int);
extern int (*xpc_setup_partitions_sn) (void);
+extern void (*xpc_teardown_partitions_sn) (void);
extern enum xp_retval (*xpc_get_partition_rsvd_page_pa) (void *, u64 *,
unsigned long *,
size_t *);
@@ -846,8 +857,8 @@ extern void (*xpc_send_chctl_openrequest) (struct xpc_channel *,
unsigned long *);
extern void (*xpc_send_chctl_openreply) (struct xpc_channel *, unsigned long *);
-extern void (*xpc_save_remote_msgqueue_pa) (struct xpc_channel *,
- unsigned long);
+extern enum xp_retval (*xpc_save_remote_msgqueue_pa) (struct xpc_channel *,
+ unsigned long);
extern enum xp_retval (*xpc_send_payload) (struct xpc_channel *, u32, void *,
u16, u8, xpc_notify_func, void *);
diff --git a/drivers/misc/sgi-xp/xpc_channel.c b/drivers/misc/sgi-xp/xpc_channel.c
index 45fd653dbe3..99a2534c38a 100644
--- a/drivers/misc/sgi-xp/xpc_channel.c
+++ b/drivers/misc/sgi-xp/xpc_channel.c
@@ -183,6 +183,7 @@ xpc_process_openclose_chctl_flags(struct xpc_partition *part, int ch_number,
&part->remote_openclose_args[ch_number];
struct xpc_channel *ch = &part->channels[ch_number];
enum xp_retval reason;
+ enum xp_retval ret;
spin_lock_irqsave(&ch->lock, irq_flags);
@@ -399,8 +400,13 @@ again:
DBUG_ON(args->local_nentries == 0);
DBUG_ON(args->remote_nentries == 0);
+ ret = xpc_save_remote_msgqueue_pa(ch, args->local_msgqueue_pa);
+ if (ret != xpSuccess) {
+ XPC_DISCONNECT_CHANNEL(ch, ret, &irq_flags);
+ spin_unlock_irqrestore(&ch->lock, irq_flags);
+ return;
+ }
ch->flags |= XPC_C_ROPENREPLY;
- xpc_save_remote_msgqueue_pa(ch, args->local_msgqueue_pa);
if (args->local_nentries < ch->remote_nentries) {
dev_dbg(xpc_chan, "XPC_CHCTL_OPENREPLY: new "
diff --git a/drivers/misc/sgi-xp/xpc_main.c b/drivers/misc/sgi-xp/xpc_main.c
index 6576170de96..1ab9fda87fa 100644
--- a/drivers/misc/sgi-xp/xpc_main.c
+++ b/drivers/misc/sgi-xp/xpc_main.c
@@ -171,6 +171,7 @@ static struct notifier_block xpc_die_notifier = {
};
int (*xpc_setup_partitions_sn) (void);
+void (*xpc_teardown_partitions_sn) (void);
enum xp_retval (*xpc_get_partition_rsvd_page_pa) (void *buf, u64 *cookie,
unsigned long *rp_pa,
size_t *len);
@@ -217,8 +218,8 @@ void (*xpc_send_chctl_openrequest) (struct xpc_channel *ch,
void (*xpc_send_chctl_openreply) (struct xpc_channel *ch,
unsigned long *irq_flags);
-void (*xpc_save_remote_msgqueue_pa) (struct xpc_channel *ch,
- unsigned long msgqueue_pa);
+enum xp_retval (*xpc_save_remote_msgqueue_pa) (struct xpc_channel *ch,
+ unsigned long msgqueue_pa);
enum xp_retval (*xpc_send_payload) (struct xpc_channel *ch, u32 flags,
void *payload, u16 payload_size,
@@ -998,6 +999,7 @@ xpc_setup_partitions(void)
static void
xpc_teardown_partitions(void)
{
+ xpc_teardown_partitions_sn();
kfree(xpc_partitions);
}
diff --git a/drivers/misc/sgi-xp/xpc_sn2.c b/drivers/misc/sgi-xp/xpc_sn2.c
index 2e975762c32..eaaa964942d 100644
--- a/drivers/misc/sgi-xp/xpc_sn2.c
+++ b/drivers/misc/sgi-xp/xpc_sn2.c
@@ -66,6 +66,12 @@ xpc_setup_partitions_sn_sn2(void)
return 0;
}
+static void
+xpc_teardown_partitions_sn_sn2(void)
+{
+ /* nothing needs to be done */
+}
+
/* SH_IPI_ACCESS shub register value on startup */
static u64 xpc_sh1_IPI_access_sn2;
static u64 xpc_sh2_IPI_access0_sn2;
@@ -436,11 +442,12 @@ xpc_send_chctl_local_msgrequest_sn2(struct xpc_channel *ch)
XPC_SEND_LOCAL_NOTIFY_IRQ_SN2(ch, XPC_CHCTL_MSGREQUEST);
}
-static void
+static enum xp_retval
xpc_save_remote_msgqueue_pa_sn2(struct xpc_channel *ch,
unsigned long msgqueue_pa)
{
ch->sn.sn2.remote_msgqueue_pa = msgqueue_pa;
+ return xpSuccess;
}
/*
@@ -1737,20 +1744,20 @@ xpc_clear_remote_msgqueue_flags_sn2(struct xpc_channel *ch)
{
struct xpc_channel_sn2 *ch_sn2 = &ch->sn.sn2;
struct xpc_msg_sn2 *msg;
- s64 put;
+ s64 put, remote_nentries = ch->remote_nentries;
/* flags are zeroed when the buffer is allocated */
- if (ch_sn2->remote_GP.put < ch->remote_nentries)
+ if (ch_sn2->remote_GP.put < remote_nentries)
return;
- put = max(ch_sn2->w_remote_GP.put, ch->remote_nentries);
+ put = max(ch_sn2->w_remote_GP.put, remote_nentries);
do {
msg = (struct xpc_msg_sn2 *)((u64)ch_sn2->remote_msgqueue +
- (put % ch->remote_nentries) *
+ (put % remote_nentries) *
ch->entry_size);
DBUG_ON(!(msg->flags & XPC_M_SN2_READY));
DBUG_ON(!(msg->flags & XPC_M_SN2_DONE));
- DBUG_ON(msg->number != put - ch->remote_nentries);
+ DBUG_ON(msg->number != put - remote_nentries);
msg->flags = 0;
} while (++put < ch_sn2->remote_GP.put);
}
@@ -2315,6 +2322,7 @@ xpc_init_sn2(void)
size_t buf_size;
xpc_setup_partitions_sn = xpc_setup_partitions_sn_sn2;
+ xpc_teardown_partitions_sn = xpc_teardown_partitions_sn_sn2;
xpc_get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_sn2;
xpc_setup_rsvd_page_sn = xpc_setup_rsvd_page_sn_sn2;
xpc_increment_heartbeat = xpc_increment_heartbeat_sn2;
diff --git a/drivers/misc/sgi-xp/xpc_uv.c b/drivers/misc/sgi-xp/xpc_uv.c
index 29c0502a96b..f7fff4727ed 100644
--- a/drivers/misc/sgi-xp/xpc_uv.c
+++ b/drivers/misc/sgi-xp/xpc_uv.c
@@ -31,6 +31,21 @@
#include "../sgi-gru/grukservices.h"
#include "xpc.h"
+#if defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
+struct uv_IO_APIC_route_entry {
+ __u64 vector : 8,
+ delivery_mode : 3,
+ dest_mode : 1,
+ delivery_status : 1,
+ polarity : 1,
+ __reserved_1 : 1,
+ trigger : 1,
+ mask : 1,
+ __reserved_2 : 15,
+ dest : 32;
+};
+#endif
+
static atomic64_t xpc_heartbeat_uv;
static DECLARE_BITMAP(xpc_heartbeating_to_mask_uv, XP_MAX_NPARTITIONS_UV);
@@ -56,26 +71,52 @@ xpc_setup_partitions_sn_uv(void)
for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
part_uv = &xpc_partitions[partid].sn.uv;
+ mutex_init(&part_uv->cached_activate_gru_mq_desc_mutex);
spin_lock_init(&part_uv->flags_lock);
part_uv->remote_act_state = XPC_P_AS_INACTIVE;
}
return 0;
}
+static void
+xpc_teardown_partitions_sn_uv(void)
+{
+ short partid;
+ struct xpc_partition_uv *part_uv;
+ unsigned long irq_flags;
+
+ for (partid = 0; partid < XP_MAX_NPARTITIONS_UV; partid++) {
+ part_uv = &xpc_partitions[partid].sn.uv;
+
+ if (part_uv->cached_activate_gru_mq_desc != NULL) {
+ mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
+ spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
+ part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
+ spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
+ kfree(part_uv->cached_activate_gru_mq_desc);
+ part_uv->cached_activate_gru_mq_desc = NULL;
+ mutex_unlock(&part_uv->
+ cached_activate_gru_mq_desc_mutex);
+ }
+ }
+}
+
static int
xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name)
{
+ int mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
+
#if defined CONFIG_X86_64
mq->irq = uv_setup_irq(irq_name, cpu, mq->mmr_blade, mq->mmr_offset);
if (mq->irq < 0) {
dev_err(xpc_part, "uv_setup_irq() returned error=%d\n",
- mq->irq);
+ -mq->irq);
+ return mq->irq;
}
-#elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
- int mmr_pnode;
- unsigned long mmr_value;
+ mq->mmr_value = uv_read_global_mmr64(mmr_pnode, mq->mmr_offset);
+#elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
if (strcmp(irq_name, XPC_ACTIVATE_IRQ_NAME) == 0)
mq->irq = SGI_XPC_ACTIVATE;
else if (strcmp(irq_name, XPC_NOTIFY_IRQ_NAME) == 0)
@@ -83,10 +124,8 @@ xpc_get_gru_mq_irq_uv(struct xpc_gru_mq_uv *mq, int cpu, char *irq_name)
else
return -EINVAL;
- mmr_pnode = uv_blade_to_pnode(mq->mmr_blade);
- mmr_value = (unsigned long)cpu_physical_id(cpu) << 32 | mq->irq;
-
- uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mmr_value);
+ mq->mmr_value = (unsigned long)cpu_physical_id(cpu) << 32 | mq->irq;
+ uv_write_global_mmr64(mmr_pnode, mq->mmr_offset, mq->mmr_value);
#else
#error not a supported configuration
#endif
@@ -127,7 +166,7 @@ xpc_gru_mq_watchlist_alloc_uv(struct xpc_gru_mq_uv *mq)
return ret;
}
#elif defined CONFIG_IA64_GENERIC || defined CONFIG_IA64_SGI_UV
- ret = sn_mq_watchlist_alloc(mq->mmr_blade, uv_gpa(mq->address),
+ ret = sn_mq_watchlist_alloc(mq->mmr_blade, (void *)uv_gpa(mq->address),
mq->order, &mq->mmr_offset);
if (ret < 0) {
dev_err(xpc_part, "sn_mq_watchlist_alloc() failed, ret=%d\n",
@@ -168,12 +207,22 @@ xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
int pg_order;
struct page *page;
struct xpc_gru_mq_uv *mq;
+ struct uv_IO_APIC_route_entry *mmr_value;
mq = kmalloc(sizeof(struct xpc_gru_mq_uv), GFP_KERNEL);
if (mq == NULL) {
dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
"a xpc_gru_mq_uv structure\n");
ret = -ENOMEM;
+ goto out_0;
+ }
+
+ mq->gru_mq_desc = kzalloc(sizeof(struct gru_message_queue_desc),
+ GFP_KERNEL);
+ if (mq->gru_mq_desc == NULL) {
+ dev_err(xpc_part, "xpc_create_gru_mq_uv() failed to kmalloc() "
+ "a gru_message_queue_desc structure\n");
+ ret = -ENOMEM;
goto out_1;
}
@@ -194,14 +243,6 @@ xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
}
mq->address = page_address(page);
- ret = gru_create_message_queue(mq->address, mq_size);
- if (ret != 0) {
- dev_err(xpc_part, "gru_create_message_queue() returned "
- "error=%d\n", ret);
- ret = -EINVAL;
- goto out_3;
- }
-
/* enable generation of irq when GRU mq operation occurs to this mq */
ret = xpc_gru_mq_watchlist_alloc_uv(mq);
if (ret != 0)
@@ -214,10 +255,20 @@ xpc_create_gru_mq_uv(unsigned int mq_size, int cpu, char *irq_name,
ret = request_irq(mq->irq, irq_handler, 0, irq_name, NULL);
if (ret != 0) {
dev_err(xpc_part, "request_irq(irq=%d) returned error=%d\n",
- mq->irq, ret);
+ mq->irq, -ret);
goto out_5;
}
+ mmr_value = (struct uv_IO_APIC_route_entry *)&mq->mmr_value;
+ ret = gru_create_message_queue(mq->gru_mq_desc, mq->address, mq_size,
+ nid, mmr_value->vector, mmr_value->dest);
+ if (ret != 0) {
+ dev_err(xpc_part, "gru_create_message_queue() returned "
+ "error=%d\n", ret);
+ ret = -EINVAL;
+ goto out_6;
+ }
+
/* allow other partitions to access this GRU mq */
xp_ret = xp_expand_memprotect(xp_pa(mq->address), mq_size);
if (xp_ret != xpSuccess) {
@@ -237,8 +288,10 @@ out_4:
out_3:
free_pages((unsigned long)mq->address, pg_order);
out_2:
- kfree(mq);
+ kfree(mq->gru_mq_desc);
out_1:
+ kfree(mq);
+out_0:
return ERR_PTR(ret);
}
@@ -268,13 +321,14 @@ xpc_destroy_gru_mq_uv(struct xpc_gru_mq_uv *mq)
}
static enum xp_retval
-xpc_send_gru_msg(unsigned long mq_gpa, void *msg, size_t msg_size)
+xpc_send_gru_msg(struct gru_message_queue_desc *gru_mq_desc, void *msg,
+ size_t msg_size)
{
enum xp_retval xp_ret;
int ret;
while (1) {
- ret = gru_send_message_gpa(mq_gpa, msg, msg_size);
+ ret = gru_send_message_gpa(gru_mq_desc, msg, msg_size);
if (ret == MQE_OK) {
xp_ret = xpSuccess;
break;
@@ -421,7 +475,15 @@ xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
part_uv->act_state_req = XPC_P_ASR_ACTIVATE_UV;
part->remote_rp_pa = msg->rp_gpa; /* !!! _pa is _gpa */
part->remote_rp_ts_jiffies = msg_hdr->rp_ts_jiffies;
- part_uv->remote_activate_mq_gpa = msg->activate_mq_gpa;
+
+ if (msg->activate_gru_mq_desc_gpa !=
+ part_uv->activate_gru_mq_desc_gpa) {
+ spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
+ part_uv->flags &= ~XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
+ spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
+ part_uv->activate_gru_mq_desc_gpa =
+ msg->activate_gru_mq_desc_gpa;
+ }
spin_unlock_irqrestore(&xpc_activate_IRQ_rcvd_lock, irq_flags);
(*wakeup_hb_checker)++;
@@ -498,7 +560,7 @@ xpc_handle_activate_mq_msg_uv(struct xpc_partition *part,
args = &part->remote_openclose_args[msg->ch_number];
args->remote_nentries = msg->remote_nentries;
args->local_nentries = msg->local_nentries;
- args->local_msgqueue_pa = msg->local_notify_mq_gpa;
+ args->local_msgqueue_pa = msg->notify_gru_mq_desc_gpa;
spin_lock_irqsave(&part->chctl_lock, irq_flags);
part->chctl.flags[msg->ch_number] |= XPC_CHCTL_OPENREPLY;
@@ -558,9 +620,10 @@ xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
short partid;
struct xpc_partition *part;
int wakeup_hb_checker = 0;
+ int part_referenced;
while (1) {
- msg_hdr = gru_get_next_message(xpc_activate_mq_uv->address);
+ msg_hdr = gru_get_next_message(xpc_activate_mq_uv->gru_mq_desc);
if (msg_hdr == NULL)
break;
@@ -571,14 +634,15 @@ xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
partid);
} else {
part = &xpc_partitions[partid];
- if (xpc_part_ref(part)) {
- xpc_handle_activate_mq_msg_uv(part, msg_hdr,
- &wakeup_hb_checker);
+
+ part_referenced = xpc_part_ref(part);
+ xpc_handle_activate_mq_msg_uv(part, msg_hdr,
+ &wakeup_hb_checker);
+ if (part_referenced)
xpc_part_deref(part);
- }
}
- gru_free_message(xpc_activate_mq_uv->address, msg_hdr);
+ gru_free_message(xpc_activate_mq_uv->gru_mq_desc, msg_hdr);
}
if (wakeup_hb_checker)
@@ -588,21 +652,73 @@ xpc_handle_activate_IRQ_uv(int irq, void *dev_id)
}
static enum xp_retval
+xpc_cache_remote_gru_mq_desc_uv(struct gru_message_queue_desc *gru_mq_desc,
+ unsigned long gru_mq_desc_gpa)
+{
+ enum xp_retval ret;
+
+ ret = xp_remote_memcpy(uv_gpa(gru_mq_desc), gru_mq_desc_gpa,
+ sizeof(struct gru_message_queue_desc));
+ if (ret == xpSuccess)
+ gru_mq_desc->mq = NULL;
+
+ return ret;
+}
+
+static enum xp_retval
xpc_send_activate_IRQ_uv(struct xpc_partition *part, void *msg, size_t msg_size,
int msg_type)
{
struct xpc_activate_mq_msghdr_uv *msg_hdr = msg;
+ struct xpc_partition_uv *part_uv = &part->sn.uv;
+ struct gru_message_queue_desc *gru_mq_desc;
+ unsigned long irq_flags;
+ enum xp_retval ret;
DBUG_ON(msg_size > XPC_ACTIVATE_MSG_SIZE_UV);
msg_hdr->type = msg_type;
- msg_hdr->partid = XPC_PARTID(part);
+ msg_hdr->partid = xp_partition_id;
msg_hdr->act_state = part->act_state;
msg_hdr->rp_ts_jiffies = xpc_rsvd_page->ts_jiffies;
+ mutex_lock(&part_uv->cached_activate_gru_mq_desc_mutex);
+again:
+ if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV)) {
+ gru_mq_desc = part_uv->cached_activate_gru_mq_desc;
+ if (gru_mq_desc == NULL) {
+ gru_mq_desc = kmalloc(sizeof(struct
+ gru_message_queue_desc),
+ GFP_KERNEL);
+ if (gru_mq_desc == NULL) {
+ ret = xpNoMemory;
+ goto done;
+ }
+ part_uv->cached_activate_gru_mq_desc = gru_mq_desc;
+ }
+
+ ret = xpc_cache_remote_gru_mq_desc_uv(gru_mq_desc,
+ part_uv->
+ activate_gru_mq_desc_gpa);
+ if (ret != xpSuccess)
+ goto done;
+
+ spin_lock_irqsave(&part_uv->flags_lock, irq_flags);
+ part_uv->flags |= XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV;
+ spin_unlock_irqrestore(&part_uv->flags_lock, irq_flags);
+ }
+
/* ??? Is holding a spin_lock (ch->lock) during this call a bad idea? */
- return xpc_send_gru_msg(part->sn.uv.remote_activate_mq_gpa, msg,
- msg_size);
+ ret = xpc_send_gru_msg(part_uv->cached_activate_gru_mq_desc, msg,
+ msg_size);
+ if (ret != xpSuccess) {
+ smp_rmb(); /* ensure a fresh copy of part_uv->flags */
+ if (!(part_uv->flags & XPC_P_CACHED_ACTIVATE_GRU_MQ_DESC_UV))
+ goto again;
+ }
+done:
+ mutex_unlock(&part_uv->cached_activate_gru_mq_desc_mutex);
+ return ret;
}
static void
@@ -620,7 +736,7 @@ static void
xpc_send_activate_IRQ_ch_uv(struct xpc_channel *ch, unsigned long *irq_flags,
void *msg, size_t msg_size, int msg_type)
{
- struct xpc_partition *part = &xpc_partitions[ch->number];
+ struct xpc_partition *part = &xpc_partitions[ch->partid];
enum xp_retval ret;
ret = xpc_send_activate_IRQ_uv(part, msg, msg_size, msg_type);
@@ -692,7 +808,8 @@ xpc_get_partition_rsvd_page_pa_uv(void *buf, u64 *cookie, unsigned long *rp_pa,
static int
xpc_setup_rsvd_page_sn_uv(struct xpc_rsvd_page *rp)
{
- rp->sn.activate_mq_gpa = uv_gpa(xpc_activate_mq_uv->address);
+ rp->sn.activate_gru_mq_desc_gpa =
+ uv_gpa(xpc_activate_mq_uv->gru_mq_desc);
return 0;
}
@@ -787,7 +904,8 @@ xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
part->remote_rp_pa = remote_rp_gpa; /* !!! _pa here is really _gpa */
part->remote_rp_ts_jiffies = remote_rp->ts_jiffies;
- part->sn.uv.remote_activate_mq_gpa = remote_rp->sn.activate_mq_gpa;
+ part->sn.uv.activate_gru_mq_desc_gpa =
+ remote_rp->sn.activate_gru_mq_desc_gpa;
/*
* ??? Is it a good idea to make this conditional on what is
@@ -795,7 +913,8 @@ xpc_request_partition_activation_uv(struct xpc_rsvd_page *remote_rp,
*/
if (part->sn.uv.remote_act_state == XPC_P_AS_INACTIVE) {
msg.rp_gpa = uv_gpa(xpc_rsvd_page);
- msg.activate_mq_gpa = xpc_rsvd_page->sn.activate_mq_gpa;
+ msg.activate_gru_mq_desc_gpa =
+ xpc_rsvd_page->sn.activate_gru_mq_desc_gpa;
xpc_send_activate_IRQ_part_uv(part, &msg, sizeof(msg),
XPC_ACTIVATE_MQ_MSG_ACTIVATE_REQ_UV);
}
@@ -857,7 +976,8 @@ xpc_get_fifo_entry_uv(struct xpc_fifo_head_uv *head)
if (head->first == NULL)
head->last = NULL;
}
- head->n_entries++;
+ head->n_entries--;
+ BUG_ON(head->n_entries < 0);
spin_unlock_irqrestore(&head->lock, irq_flags);
first->next = NULL;
return first;
@@ -876,8 +996,7 @@ xpc_put_fifo_entry_uv(struct xpc_fifo_head_uv *head,
else
head->first = last;
head->last = last;
- head->n_entries--;
- BUG_ON(head->n_entries < 0);
+ head->n_entries++;
spin_unlock_irqrestore(&head->lock, irq_flags);
}
@@ -1037,6 +1156,12 @@ xpc_setup_msg_structures_uv(struct xpc_channel *ch)
DBUG_ON(ch->flags & XPC_C_SETUP);
+ ch_uv->cached_notify_gru_mq_desc = kmalloc(sizeof(struct
+ gru_message_queue_desc),
+ GFP_KERNEL);
+ if (ch_uv->cached_notify_gru_mq_desc == NULL)
+ return xpNoMemory;
+
ret = xpc_allocate_send_msg_slot_uv(ch);
if (ret == xpSuccess) {
@@ -1060,7 +1185,8 @@ xpc_teardown_msg_structures_uv(struct xpc_channel *ch)
DBUG_ON(!spin_is_locked(&ch->lock));
- ch_uv->remote_notify_mq_gpa = 0;
+ kfree(ch_uv->cached_notify_gru_mq_desc);
+ ch_uv->cached_notify_gru_mq_desc = NULL;
if (ch->flags & XPC_C_SETUP) {
xpc_init_fifo_uv(&ch_uv->msg_slot_free_list);
@@ -1111,7 +1237,7 @@ xpc_send_chctl_openreply_uv(struct xpc_channel *ch, unsigned long *irq_flags)
msg.ch_number = ch->number;
msg.local_nentries = ch->local_nentries;
msg.remote_nentries = ch->remote_nentries;
- msg.local_notify_mq_gpa = uv_gpa(xpc_notify_mq_uv);
+ msg.notify_gru_mq_desc_gpa = uv_gpa(xpc_notify_mq_uv->gru_mq_desc);
xpc_send_activate_IRQ_ch_uv(ch, irq_flags, &msg, sizeof(msg),
XPC_ACTIVATE_MQ_MSG_CHCTL_OPENREPLY_UV);
}
@@ -1128,11 +1254,15 @@ xpc_send_chctl_local_msgrequest_uv(struct xpc_partition *part, int ch_number)
xpc_wakeup_channel_mgr(part);
}
-static void
+static enum xp_retval
xpc_save_remote_msgqueue_pa_uv(struct xpc_channel *ch,
- unsigned long msgqueue_pa)
+ unsigned long gru_mq_desc_gpa)
{
- ch->sn.uv.remote_notify_mq_gpa = msgqueue_pa;
+ struct xpc_channel_uv *ch_uv = &ch->sn.uv;
+
+ DBUG_ON(ch_uv->cached_notify_gru_mq_desc == NULL);
+ return xpc_cache_remote_gru_mq_desc_uv(ch_uv->cached_notify_gru_mq_desc,
+ gru_mq_desc_gpa);
}
static void
@@ -1339,7 +1469,8 @@ xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
short partid;
struct xpc_partition *part;
- while ((msg = gru_get_next_message(xpc_notify_mq_uv)) != NULL) {
+ while ((msg = gru_get_next_message(xpc_notify_mq_uv->gru_mq_desc)) !=
+ NULL) {
partid = msg->hdr.partid;
if (partid < 0 || partid >= XP_MAX_NPARTITIONS_UV) {
@@ -1354,7 +1485,7 @@ xpc_handle_notify_IRQ_uv(int irq, void *dev_id)
}
}
- gru_free_message(xpc_notify_mq_uv, msg);
+ gru_free_message(xpc_notify_mq_uv->gru_mq_desc, msg);
}
return IRQ_HANDLED;
@@ -1438,7 +1569,8 @@ xpc_send_payload_uv(struct xpc_channel *ch, u32 flags, void *payload,
msg->hdr.msg_slot_number = msg_slot->msg_slot_number;
memcpy(&msg->payload, payload, payload_size);
- ret = xpc_send_gru_msg(ch->sn.uv.remote_notify_mq_gpa, msg, msg_size);
+ ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
+ msg_size);
if (ret == xpSuccess)
goto out_1;
@@ -1529,7 +1661,7 @@ xpc_received_payload_uv(struct xpc_channel *ch, void *payload)
msg->hdr.partid = xp_partition_id;
msg->hdr.size = 0; /* size of zero indicates this is an ACK */
- ret = xpc_send_gru_msg(ch->sn.uv.remote_notify_mq_gpa, msg,
+ ret = xpc_send_gru_msg(ch->sn.uv.cached_notify_gru_mq_desc, msg,
sizeof(struct xpc_notify_mq_msghdr_uv));
if (ret != xpSuccess)
XPC_DEACTIVATE_PARTITION(&xpc_partitions[ch->partid], ret);
@@ -1541,6 +1673,7 @@ int
xpc_init_uv(void)
{
xpc_setup_partitions_sn = xpc_setup_partitions_sn_uv;
+ xpc_teardown_partitions_sn = xpc_teardown_partitions_sn_uv;
xpc_process_activate_IRQ_rcvd = xpc_process_activate_IRQ_rcvd_uv;
xpc_get_partition_rsvd_page_pa = xpc_get_partition_rsvd_page_pa_uv;
xpc_setup_rsvd_page_sn = xpc_setup_rsvd_page_sn_uv;