lib: Factor-out TLB management from IPI management

This patch factor-out TLB management from IPI management to separate
sources sbi_tlb.c and sbi_tlb.h.

Signed-off-by: Anup Patel <anup.patel@wdc.com>
Reviewed-by: Atish Patra <atish.patra@wdc.com>
This commit is contained in:
Anup Patel 2019-05-24 07:58:05 +05:30 committed by Anup Patel
parent 2dfed32c46
commit 95b7480ab4
7 changed files with 292 additions and 207 deletions

View file

@ -32,9 +32,6 @@
#define PAGE_SIZE (_AC(1, UL) << PAGE_SHIFT)
#define PAGE_MASK (~(PAGE_SIZE - 1))
#define SBI_TLB_FLUSH_ALL ((unsigned long)-1)
#define SBI_TLB_FLUSH_MAX_SIZE (1UL << 30)
#define REG_L __REG_SEL(ld, lw)
#define REG_S __REG_SEL(sd, sw)
#define SZREG __REG_SEL(8, 4)

View file

@ -22,28 +22,12 @@
/* clang-format on */
#define SBI_TLB_FIFO_NUM_ENTRIES 4
enum sbi_tlb_info_types {
SBI_TLB_FLUSH_VMA,
SBI_TLB_FLUSH_VMA_ASID,
SBI_TLB_FLUSH_VMA_VMID
};
struct sbi_scratch;
struct sbi_ipi_data {
unsigned long ipi_type;
};
struct sbi_tlb_info {
unsigned long start;
unsigned long size;
unsigned long asid;
unsigned long type;
};
#define SBI_TLB_INFO_SIZE sizeof(struct sbi_tlb_info)
int sbi_ipi_send_many(struct sbi_scratch *scratch, ulong *pmask, u32 event,
void *data);

48
include/sbi/sbi_tlb.h Normal file
View file

@ -0,0 +1,48 @@
/*
* SPDX-License-Identifier: BSD-2-Clause
*
* Copyright (c) 2019 Western Digital Corporation or its affiliates.
*
* Authors:
* Atish Patra <atish.patra@wdc.com>
* Anup Patel <anup.patel@wdc.com>
*/
#ifndef __SBI_TLB_H__
#define __SBI_TLB_H__
#include <sbi/sbi_types.h>
/* clang-format off */
#define SBI_TLB_FLUSH_ALL ((unsigned long)-1)
#define SBI_TLB_FLUSH_MAX_SIZE (1UL << 30)
/* clang-format on */
#define SBI_TLB_FIFO_NUM_ENTRIES 4
enum sbi_tlb_info_types {
SBI_TLB_FLUSH_VMA,
SBI_TLB_FLUSH_VMA_ASID,
SBI_TLB_FLUSH_VMA_VMID
};
struct sbi_scratch;
struct sbi_tlb_info {
unsigned long start;
unsigned long size;
unsigned long asid;
unsigned long type;
};
#define SBI_TLB_INFO_SIZE sizeof(struct sbi_tlb_info)
int sbi_tlb_fifo_update(struct sbi_scratch *scratch, u32 event, void *data);
void sbi_tlb_fifo_process(struct sbi_scratch *scratch, u32 event);
int sbi_tlb_fifo_init(struct sbi_scratch *scratch, bool cold_boot);
#endif

View file

@ -24,6 +24,7 @@ lib-objs-y += sbi_misaligned_ldst.o
lib-objs-y += sbi_scratch.o
lib-objs-y += sbi_system.o
lib-objs-y += sbi_timer.o
lib-objs-y += sbi_tlb.o
lib-objs-y += sbi_trap.o
# External Libraries to include

View file

@ -14,6 +14,7 @@
#include <sbi/sbi_ipi.h>
#include <sbi/sbi_system.h>
#include <sbi/sbi_timer.h>
#include <sbi/sbi_tlb.h>
#include <sbi/sbi_trap.h>
#define SBI_ECALL_VERSION_MAJOR 0

View file

@ -13,92 +13,23 @@
#include <sbi/riscv_atomic.h>
#include <sbi/riscv_unpriv.h>
#include <sbi/sbi_error.h>
#include <sbi/sbi_fifo.h>
#include <sbi/sbi_hart.h>
#include <sbi/sbi_bitops.h>
#include <sbi/sbi_hart.h>
#include <sbi/sbi_ipi.h>
#include <sbi/sbi_platform.h>
#include <sbi/sbi_timer.h>
#include <sbi/sbi_tlb.h>
#include <plat/string.h>
static unsigned long ipi_data_off;
static unsigned long ipi_tlb_fifo_off;
static unsigned long ipi_tlb_fifo_mem_off;
static inline int __sbi_tlb_fifo_range_check(struct sbi_tlb_info *curr,
struct sbi_tlb_info *next)
{
unsigned long curr_end;
unsigned long next_end;
int ret = SBI_FIFO_UNCHANGED;
if (!curr || !next)
return ret;
next_end = next->start + next->size;
curr_end = curr->start + curr->size;
if (next->start <= curr->start && next_end > curr_end) {
curr->start = next->start;
curr->size = next->size;
ret = SBI_FIFO_UPDATED;
} else if (next->start >= curr->start && next_end <= curr_end) {
ret = SBI_FIFO_SKIP;
}
return ret;
}
/**
* Call back to decide if an inplace fifo update is required or next entry can
* can be skipped. Here are the different cases that are being handled.
*
* Case1:
* if next flush request range lies within one of the existing entry, skip
* the next entry.
* Case2:
* if flush request range in current fifo entry lies within next flush
* request, update the current entry.
* Case3:
if a complete vma flush is requested, then all entries can be deleted
and new request can be enqueued. This will not be done for ASID case
as that means we have to iterate again in the fifo to figure out which
entries belong to that ASID.
*/
int sbi_tlb_fifo_update_cb(void *in, void *data)
{
struct sbi_tlb_info *curr;
struct sbi_tlb_info *next;
int ret = SBI_FIFO_UNCHANGED;
if (!in && !!data)
return ret;
curr = (struct sbi_tlb_info *)data;
next = (struct sbi_tlb_info *)in;
if (next->type == SBI_TLB_FLUSH_VMA_ASID &&
curr->type == SBI_TLB_FLUSH_VMA_ASID) {
if (next->asid == curr->asid)
ret = __sbi_tlb_fifo_range_check(curr, next);
} else if (next->type == SBI_TLB_FLUSH_VMA &&
curr->type == SBI_TLB_FLUSH_VMA) {
if (next->size == SBI_TLB_FLUSH_ALL)
ret = SBI_FIFO_RESET;
else
ret = __sbi_tlb_fifo_range_check(curr, next);
}
return ret;
}
static int sbi_ipi_send(struct sbi_scratch *scratch, u32 hartid, u32 event,
void *data)
{
int ret;
struct sbi_scratch *remote_scratch = NULL;
const struct sbi_platform *plat = sbi_platform_ptr(scratch);
struct sbi_ipi_data *ipi_data;
struct sbi_fifo *ipi_tlb_fifo;
struct sbi_tlb_info *tinfo = data;
int ret = SBI_FIFO_UNCHANGED;
if (sbi_platform_hart_disabled(plat, hartid))
return -1;
@ -109,45 +40,21 @@ static int sbi_ipi_send(struct sbi_scratch *scratch, u32 hartid, u32 event,
*/
remote_scratch = sbi_hart_id_to_scratch(scratch, hartid);
ipi_data = sbi_scratch_offset_ptr(remote_scratch, ipi_data_off);
ipi_tlb_fifo = sbi_scratch_offset_ptr(remote_scratch,
ipi_tlb_fifo_off);
if (event == SBI_IPI_EVENT_SFENCE_VMA ||
event == SBI_IPI_EVENT_SFENCE_VMA_ASID) {
/*
* If address range to flush is too big then simply
* upgrade it to flush all because we can only flush
* 4KB at a time.
*/
if (tinfo->size >= SBI_TLB_FLUSH_MAX_SIZE) {
tinfo->start = 0;
tinfo->size = SBI_TLB_FLUSH_ALL;
}
ret = sbi_fifo_inplace_update(ipi_tlb_fifo, data,
sbi_tlb_fifo_update_cb);
if (ret == SBI_FIFO_SKIP || ret == SBI_FIFO_UPDATED) {
ret = sbi_tlb_fifo_update(remote_scratch, event, data);
if (ret > 0)
goto done;
}
while (sbi_fifo_enqueue(ipi_tlb_fifo, data) < 0) {
/**
* For now, Busy loop until there is space in the fifo.
* There may be case where target hart is also
* enqueue in source hart's fifo. Both hart may busy
* loop leading to a deadlock.
* TODO: Introduce a wait/wakeup event mechansim to handle
* this properly.
*/
__asm__ __volatile("nop");
__asm__ __volatile("nop");
}
else if (ret < 0)
return ret;
}
atomic_raw_set_bit(event, &ipi_data->ipi_type);
mb();
sbi_platform_ipi_send(plat, hartid);
if (event != SBI_IPI_EVENT_SOFT)
sbi_platform_ipi_sync(plat, hartid);
done:
done:
return 0;
}
@ -180,69 +87,13 @@ void sbi_ipi_clear_smode(struct sbi_scratch *scratch)
csr_clear(CSR_MIP, MIP_SSIP);
}
static void sbi_ipi_tlb_flush_all()
{
__asm__ __volatile("sfence.vma");
}
static void sbi_ipi_sfence_vma(struct sbi_tlb_info *tinfo)
{
unsigned long start = tinfo->start;
unsigned long size = tinfo->size;
unsigned long i;
if ((start == 0 && size == 0) || (size == SBI_TLB_FLUSH_ALL)) {
sbi_ipi_tlb_flush_all();
return;
}
for (i = 0; i < size; i += PAGE_SIZE) {
__asm__ __volatile__("sfence.vma %0"
:
: "r"(start + i)
: "memory");
}
}
static void sbi_ipi_sfence_vma_asid(struct sbi_tlb_info *tinfo)
{
unsigned long start = tinfo->start;
unsigned long size = tinfo->size;
unsigned long asid = tinfo->asid;
unsigned long i;
if (start == 0 && size == 0) {
sbi_ipi_tlb_flush_all();
return;
}
/* Flush entire MM context for a given ASID */
if (size == SBI_TLB_FLUSH_ALL) {
__asm__ __volatile__("sfence.vma x0, %0"
:
: "r"(asid)
: "memory");
return;
}
for (i = 0; i < size; i += PAGE_SIZE) {
__asm__ __volatile__("sfence.vma %0, %1"
:
: "r"(start + i), "r"(asid)
: "memory");
}
}
void sbi_ipi_process(struct sbi_scratch *scratch)
{
volatile unsigned long ipi_type;
struct sbi_tlb_info tinfo;
unsigned int ipi_event;
const struct sbi_platform *plat = sbi_platform_ptr(scratch);
struct sbi_ipi_data *ipi_data =
sbi_scratch_offset_ptr(scratch, ipi_data_off);
struct sbi_fifo *ipi_tlb_fifo =
sbi_scratch_offset_ptr(scratch, ipi_tlb_fifo_off);
u32 hartid = sbi_current_hartid();
sbi_platform_ipi_clear(plat, hartid);
@ -260,13 +111,7 @@ void sbi_ipi_process(struct sbi_scratch *scratch)
break;
case SBI_IPI_EVENT_SFENCE_VMA:
case SBI_IPI_EVENT_SFENCE_VMA_ASID:
while (!sbi_fifo_dequeue(ipi_tlb_fifo, &tinfo)) {
if (tinfo.type == SBI_TLB_FLUSH_VMA)
sbi_ipi_sfence_vma(&tinfo);
else if (tinfo.type == SBI_TLB_FLUSH_VMA_ASID)
sbi_ipi_sfence_vma_asid(&tinfo);
memset(&tinfo, 0, SBI_TLB_INFO_SIZE);
}
sbi_tlb_fifo_process(scratch, ipi_event);
break;
case SBI_IPI_EVENT_HALT:
sbi_hart_hang();
@ -278,8 +123,7 @@ void sbi_ipi_process(struct sbi_scratch *scratch)
int sbi_ipi_init(struct sbi_scratch *scratch, bool cold_boot)
{
void *ipi_tlb_mem;
struct sbi_fifo *ipi_tlb_q;
int ret;
struct sbi_ipi_data *ipi_data;
if (cold_boot) {
@ -287,34 +131,17 @@ int sbi_ipi_init(struct sbi_scratch *scratch, bool cold_boot)
"IPI_DATA");
if (!ipi_data_off)
return SBI_ENOMEM;
ipi_tlb_fifo_off = sbi_scratch_alloc_offset(sizeof(*ipi_tlb_q),
"IPI_TLB_FIFO");
if (!ipi_tlb_fifo_off) {
sbi_scratch_free_offset(ipi_data_off);
return SBI_ENOMEM;
}
ipi_tlb_fifo_mem_off = sbi_scratch_alloc_offset(
SBI_TLB_FIFO_NUM_ENTRIES * SBI_TLB_INFO_SIZE,
"IPI_TLB_FIFO_MEM");
if (!ipi_tlb_fifo_mem_off) {
sbi_scratch_free_offset(ipi_tlb_fifo_off);
sbi_scratch_free_offset(ipi_data_off);
return SBI_ENOMEM;
}
} else {
if (!ipi_data_off ||
!ipi_tlb_fifo_off ||
!ipi_tlb_fifo_mem_off)
if (!ipi_data_off)
return SBI_ENOMEM;
}
ipi_data = sbi_scratch_offset_ptr(scratch, ipi_data_off);
ipi_tlb_q = sbi_scratch_offset_ptr(scratch, ipi_tlb_fifo_off);
ipi_tlb_mem = sbi_scratch_offset_ptr(scratch, ipi_tlb_fifo_mem_off);
ipi_data->ipi_type = 0x00;
sbi_fifo_init(ipi_tlb_q, ipi_tlb_mem,
SBI_TLB_FIFO_NUM_ENTRIES, SBI_TLB_INFO_SIZE);
ret = sbi_tlb_fifo_init(scratch, cold_boot);
if (ret)
return ret;
/* Enable software interrupts */
csr_set(CSR_MIE, MIP_MSIP);

227
lib/sbi_tlb.c Normal file
View file

@ -0,0 +1,227 @@
/*
* SPDX-License-Identifier: BSD-2-Clause
*
* Copyright (c) 2019 Western Digital Corporation or its affiliates.
*
* Authors:
* Atish Patra <atish.patra@wdc.com>
* Anup Patel <anup.patel@wdc.com>
*/
#include <sbi/riscv_asm.h>
#include <sbi/riscv_barrier.h>
#include <sbi/sbi_error.h>
#include <sbi/sbi_fifo.h>
#include <sbi/sbi_hart.h>
#include <sbi/sbi_bitops.h>
#include <sbi/sbi_scratch.h>
#include <sbi/sbi_tlb.h>
#include <plat/string.h>
static unsigned long ipi_tlb_fifo_off;
static unsigned long ipi_tlb_fifo_mem_off;
static inline int __sbi_tlb_fifo_range_check(struct sbi_tlb_info *curr,
struct sbi_tlb_info *next)
{
unsigned long curr_end;
unsigned long next_end;
int ret = SBI_FIFO_UNCHANGED;
if (!curr || !next)
return ret;
next_end = next->start + next->size;
curr_end = curr->start + curr->size;
if (next->start <= curr->start && next_end > curr_end) {
curr->start = next->start;
curr->size = next->size;
ret = SBI_FIFO_UPDATED;
} else if (next->start >= curr->start && next_end <= curr_end) {
ret = SBI_FIFO_SKIP;
}
return ret;
}
/**
* Call back to decide if an inplace fifo update is required or next entry can
* can be skipped. Here are the different cases that are being handled.
*
* Case1:
* if next flush request range lies within one of the existing entry, skip
* the next entry.
* Case2:
* if flush request range in current fifo entry lies within next flush
* request, update the current entry.
* Case3:
if a complete vma flush is requested, then all entries can be deleted
and new request can be enqueued. This will not be done for ASID case
as that means we have to iterate again in the fifo to figure out which
entries belong to that ASID.
*/
static int sbi_tlb_fifo_update_cb(void *in, void *data)
{
struct sbi_tlb_info *curr;
struct sbi_tlb_info *next;
int ret = SBI_FIFO_UNCHANGED;
if (!in && !!data)
return ret;
curr = (struct sbi_tlb_info *)data;
next = (struct sbi_tlb_info *)in;
if (next->type == SBI_TLB_FLUSH_VMA_ASID &&
curr->type == SBI_TLB_FLUSH_VMA_ASID) {
if (next->asid == curr->asid)
ret = __sbi_tlb_fifo_range_check(curr, next);
} else if (next->type == SBI_TLB_FLUSH_VMA &&
curr->type == SBI_TLB_FLUSH_VMA) {
if (next->size == SBI_TLB_FLUSH_ALL)
ret = SBI_FIFO_RESET;
else
ret = __sbi_tlb_fifo_range_check(curr, next);
}
return ret;
}
int sbi_tlb_fifo_update(struct sbi_scratch *scratch, u32 event, void *data)
{
int ret;
struct sbi_fifo *ipi_tlb_fifo;
struct sbi_tlb_info *tinfo = data;
ipi_tlb_fifo = sbi_scratch_offset_ptr(scratch,
ipi_tlb_fifo_off);
/*
* If address range to flush is too big then simply
* upgrade it to flush all because we can only flush
* 4KB at a time.
*/
if (tinfo->size >= SBI_TLB_FLUSH_MAX_SIZE) {
tinfo->start = 0;
tinfo->size = SBI_TLB_FLUSH_ALL;
}
ret = sbi_fifo_inplace_update(ipi_tlb_fifo, data,
sbi_tlb_fifo_update_cb);
if (ret == SBI_FIFO_SKIP || ret == SBI_FIFO_UPDATED) {
return 1;
}
while (sbi_fifo_enqueue(ipi_tlb_fifo, data) < 0) {
/**
* For now, Busy loop until there is space in the fifo.
* There may be case where target hart is also
* enqueue in source hart's fifo. Both hart may busy
* loop leading to a deadlock.
* TODO: Introduce a wait/wakeup event mechansim to handle
* this properly.
*/
__asm__ __volatile("nop");
__asm__ __volatile("nop");
}
return 0;
}
static void sbi_tlb_flush_all(void)
{
__asm__ __volatile("sfence.vma");
}
static void sbi_tlb_fifo_sfence_vma(struct sbi_tlb_info *tinfo)
{
unsigned long start = tinfo->start;
unsigned long size = tinfo->size;
unsigned long i;
if ((start == 0 && size == 0) || (size == SBI_TLB_FLUSH_ALL)) {
sbi_tlb_flush_all();
return;
}
for (i = 0; i < size; i += PAGE_SIZE) {
__asm__ __volatile__("sfence.vma %0"
:
: "r"(start + i)
: "memory");
}
}
static void sbi_tlb_fifo_sfence_vma_asid(struct sbi_tlb_info *tinfo)
{
unsigned long start = tinfo->start;
unsigned long size = tinfo->size;
unsigned long asid = tinfo->asid;
unsigned long i;
if (start == 0 && size == 0) {
sbi_tlb_flush_all();
return;
}
/* Flush entire MM context for a given ASID */
if (size == SBI_TLB_FLUSH_ALL) {
__asm__ __volatile__("sfence.vma x0, %0"
:
: "r"(asid)
: "memory");
return;
}
for (i = 0; i < size; i += PAGE_SIZE) {
__asm__ __volatile__("sfence.vma %0, %1"
:
: "r"(start + i), "r"(asid)
: "memory");
}
}
void sbi_tlb_fifo_process(struct sbi_scratch *scratch, u32 event)
{
struct sbi_tlb_info tinfo;
struct sbi_fifo *ipi_tlb_fifo =
sbi_scratch_offset_ptr(scratch, ipi_tlb_fifo_off);
while (!sbi_fifo_dequeue(ipi_tlb_fifo, &tinfo)) {
if (tinfo.type == SBI_TLB_FLUSH_VMA)
sbi_tlb_fifo_sfence_vma(&tinfo);
else if (tinfo.type == SBI_TLB_FLUSH_VMA_ASID)
sbi_tlb_fifo_sfence_vma_asid(&tinfo);
memset(&tinfo, 0, SBI_TLB_INFO_SIZE);
}
}
int sbi_tlb_fifo_init(struct sbi_scratch *scratch, bool cold_boot)
{
void *ipi_tlb_mem;
struct sbi_fifo *ipi_tlb_q;
if (cold_boot) {
ipi_tlb_fifo_off = sbi_scratch_alloc_offset(sizeof(*ipi_tlb_q),
"IPI_TLB_FIFO");
if (!ipi_tlb_fifo_off)
return SBI_ENOMEM;
ipi_tlb_fifo_mem_off = sbi_scratch_alloc_offset(
SBI_TLB_FIFO_NUM_ENTRIES * SBI_TLB_INFO_SIZE,
"IPI_TLB_FIFO_MEM");
if (!ipi_tlb_fifo_mem_off) {
sbi_scratch_free_offset(ipi_tlb_fifo_off);
return SBI_ENOMEM;
}
} else {
if (!ipi_tlb_fifo_off ||
!ipi_tlb_fifo_mem_off)
return SBI_ENOMEM;
}
ipi_tlb_q = sbi_scratch_offset_ptr(scratch, ipi_tlb_fifo_off);
ipi_tlb_mem = sbi_scratch_offset_ptr(scratch, ipi_tlb_fifo_mem_off);
sbi_fifo_init(ipi_tlb_q, ipi_tlb_mem,
SBI_TLB_FIFO_NUM_ENTRIES, SBI_TLB_INFO_SIZE);
return 0;
}