Commit 1d2f1f90 authored by David S. Miller's avatar David S. Miller

[SPARC64]: Sun4v cross-call sending support.

Technically the hypervisor call supports sending in a list
of all cpus to get the cross-call, but I only pass in one
cpu at a time for now.

The multi-cpu support is there, just ifdef'd out so it's easy to
enable or delete it later.
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent 5b0c0572
......@@ -900,6 +900,24 @@ static void __cpuinit init_one_kbuf(unsigned long *pa_ptr)
*pa_ptr = __pa(page);
}
static void __cpuinit init_cpu_send_mondo_info(struct trap_per_cpu *tb)
{
#ifdef CONFIG_SMP
unsigned long page;
BUILD_BUG_ON((NR_CPUS * sizeof(u16)) > (PAGE_SIZE - 64));
page = get_zeroed_page(GFP_ATOMIC);
if (!page) {
prom_printf("SUN4V: Error, cannot allocate cpu mondo page.\n");
prom_halt();
}
tb->cpu_mondo_block_pa = __pa(page);
tb->cpu_list_pa = __pa(page + 64);
#endif
}
/* Allocate and init the mondo and error queues for this cpu. */
void __cpuinit sun4v_init_mondo_queues(void)
{
......@@ -908,10 +926,14 @@ void __cpuinit sun4v_init_mondo_queues(void)
init_one_mondo(&tb->cpu_mondo_pa, HV_CPU_QUEUE_CPU_MONDO);
init_one_mondo(&tb->dev_mondo_pa, HV_CPU_QUEUE_DEVICE_MONDO);
init_one_mondo(&tb->resum_mondo_pa, HV_CPU_QUEUE_RES_ERROR);
init_one_kbuf(&tb->resum_kernel_buf_pa);
init_one_mondo(&tb->nonresum_mondo_pa, HV_CPU_QUEUE_NONRES_ERROR);
init_one_kbuf(&tb->nonresum_kernel_buf_pa);
init_cpu_send_mondo_info(tb);
}
/* Only invoked on boot processor. */
......
......@@ -531,10 +531,133 @@ retry:
}
}
#if 0
/* Multi-cpu list version. */
static int init_cpu_list(u16 *list, cpumask_t mask)
{
int i, cnt;
cnt = 0;
for_each_cpu_mask(i, mask)
list[cnt++] = i;
return cnt;
}
static int update_cpu_list(u16 *list, int orig_cnt, cpumask_t mask)
{
int i;
for (i = 0; i < orig_cnt; i++) {
if (list[i] == 0xffff)
cpu_clear(i, mask);
}
return init_cpu_list(list, mask);
}
static void hypervisor_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask)
{
int this_cpu = get_cpu();
struct trap_per_cpu *tb = &trap_block[this_cpu];
u64 *mondo = __va(tb->cpu_mondo_block_pa);
u16 *cpu_list = __va(tb->cpu_list_pa);
int cnt, retries;
mondo[0] = data0;
mondo[1] = data1;
mondo[2] = data2;
wmb();
retries = 0;
cnt = init_cpu_list(cpu_list, mask);
do {
register unsigned long func __asm__("%o0");
register unsigned long arg0 __asm__("%o1");
register unsigned long arg1 __asm__("%o2");
register unsigned long arg2 __asm__("%o3");
func = HV_FAST_CPU_MONDO_SEND;
arg0 = cnt;
arg1 = tb->cpu_list_pa;
arg2 = tb->cpu_mondo_block_pa;
__asm__ __volatile__("ta %8"
: "=&r" (func), "=&r" (arg0),
"=&r" (arg1), "=&r" (arg2)
: "0" (func), "1" (arg0),
"2" (arg1), "3" (arg2),
"i" (HV_FAST_TRAP)
: "memory");
if (likely(func == HV_EOK))
break;
if (unlikely(++retries > 100)) {
printk("CPU[%d]: sun4v mondo error %lu\n",
this_cpu, func);
break;
}
cnt = update_cpu_list(cpu_list, cnt, mask);
udelay(2 * cnt);
} while (1);
put_cpu();
}
#else
/* Single-cpu list version. */
static void hypervisor_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask)
{
/* XXX implement me */
int this_cpu = get_cpu();
struct trap_per_cpu *tb = &trap_block[this_cpu];
u64 *mondo = __va(tb->cpu_mondo_block_pa);
u16 *cpu_list = __va(tb->cpu_list_pa);
int i;
mondo[0] = data0;
mondo[1] = data1;
mondo[2] = data2;
wmb();
for_each_cpu_mask(i, mask) {
int retries = 0;
do {
register unsigned long func __asm__("%o0");
register unsigned long arg0 __asm__("%o1");
register unsigned long arg1 __asm__("%o2");
register unsigned long arg2 __asm__("%o3");
cpu_list[0] = i;
func = HV_FAST_CPU_MONDO_SEND;
arg0 = 1;
arg1 = tb->cpu_list_pa;
arg2 = tb->cpu_mondo_block_pa;
__asm__ __volatile__("ta %8"
: "=&r" (func), "=&r" (arg0),
"=&r" (arg1), "=&r" (arg2)
: "0" (func), "1" (arg0),
"2" (arg1), "3" (arg2),
"i" (HV_FAST_TRAP)
: "memory");
if (likely(func == HV_EOK))
break;
if (unlikely(++retries > 100)) {
printk("CPU[%d]: sun4v mondo error %lu\n",
this_cpu, func);
break;
}
udelay(2 * i);
} while (1);
}
put_cpu();
}
#endif
/* Send cross call to all processors mentioned in MASK
* except self.
......
......@@ -2377,7 +2377,11 @@ void __init trap_init(void)
(TRAP_PER_CPU_NONRESUM_KBUF_PA !=
offsetof(struct trap_per_cpu, nonresum_kernel_buf_pa)) ||
(TRAP_PER_CPU_FAULT_INFO !=
offsetof(struct trap_per_cpu, fault_info)))
offsetof(struct trap_per_cpu, fault_info)) ||
(TRAP_PER_CPU_CPU_MONDO_BLOCK_PA !=
offsetof(struct trap_per_cpu, cpu_mondo_block_pa)) ||
(TRAP_PER_CPU_CPU_LIST_PA !=
offsetof(struct trap_per_cpu, cpu_list_pa)))
trap_per_cpu_offsets_are_bolixed_dave();
/* Attach to the address space of init_task. On SMP we
......
......@@ -65,8 +65,16 @@ struct trap_per_cpu {
unsigned long nonresum_mondo_pa;
unsigned long nonresum_kernel_buf_pa;
/* Dcache lines 3 and 4: Hypervisor Fault Status */
/* Dcache lines 3, 4, 5, and 6: Hypervisor Fault Status */
struct hv_fault_status fault_info;
/* Dcache line 7: Physical addresses of CPU send mondo block and CPU list. */
unsigned long cpu_mondo_block_pa;
unsigned long cpu_list_pa;
unsigned long __pad1[2];
/* Dcache line 8: Unused, needed to keep trap_block a power-of-2 in size. */
unsigned long __pad2[4];
} __attribute__((aligned(64)));
extern struct trap_per_cpu trap_block[NR_CPUS];
extern void init_cur_cpu_trap(void);
......@@ -108,8 +116,10 @@ extern struct sun4v_2insn_patch_entry __sun4v_2insn_patch,
#define TRAP_PER_CPU_NONRESUM_MONDO_PA 0x30
#define TRAP_PER_CPU_NONRESUM_KBUF_PA 0x38
#define TRAP_PER_CPU_FAULT_INFO 0x40
#define TRAP_PER_CPU_CPU_MONDO_BLOCK_PA 0xc0
#define TRAP_PER_CPU_CPU_LIST_PA 0xc8
#define TRAP_BLOCK_SZ_SHIFT 7
#define TRAP_BLOCK_SZ_SHIFT 8
#include <asm/scratchpad.h>
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment