1
linux/drivers/net/ipa/ipa_table.c

776 lines
24 KiB
C
Raw Permalink Normal View History

// SPDX-License-Identifier: GPL-2.0
/* Copyright (c) 2012-2018, The Linux Foundation. All rights reserved.
* Copyright (C) 2018-2024 Linaro Ltd.
*/
#include <linux/bitops.h>
#include <linux/build_bug.h>
#include <linux/device.h>
#include <linux/dma-mapping.h>
#include <linux/io.h>
#include <linux/types.h>
#include "gsi.h"
#include "gsi_trans.h"
#include "ipa.h"
#include "ipa_cmd.h"
#include "ipa_endpoint.h"
#include "ipa_mem.h"
#include "ipa_reg.h"
#include "ipa_table.h"
#include "ipa_version.h"
/**
* DOC: IPA Filter and Route Tables
*
* The IPA has tables defined in its local (IPA-resident) memory that define
* filter and routing rules. An entry in either of these tables is a little
* endian 64-bit "slot" that holds the address of a rule definition. (The
* size of these slots is 64 bits regardless of the host DMA address size.)
*
* Separate tables (both filter and route) are used for IPv4 and IPv6. There
* is normally another set of "hashed" filter and route tables, which are
* used with a hash of message metadata. Hashed operation is not supported
* by all IPA hardware (IPA v4.2 doesn't support hashed tables).
*
* Rules can be in local memory or in DRAM (system memory). The offset of
* an object (such as a route or filter table) in IPA-resident memory must
* 128-byte aligned. An object in system memory (such as a route or filter
* rule) must be at an 8-byte aligned address. We currently only place
* route or filter rules in system memory.
*
* A rule consists of a contiguous block of 32-bit values terminated with
* 32 zero bits. A special "zero entry" rule consisting of 64 zero bits
* represents "no filtering" or "no routing," and is the reset value for
* filter or route table rules.
*
* Each filter rule is associated with an AP or modem TX endpoint, though
* not all TX endpoints support filtering. The first 64-bit slot in a
* filter table is a bitmap indicating which endpoints have entries in
* the table. Each set bit in this bitmap indicates the presence of the
* address of a filter rule in the memory following the bitmap. Until IPA
* v5.0, the low-order bit (bit 0) in this bitmap represents a special
* global filter, which applies to all traffic. Otherwise the position of
* each set bit represents an endpoint for which a filter rule is defined.
*
* The global rule is not used in current code, and support for it is
* removed starting at IPA v5.0. For IPA v5.0+, the endpoint bitmap
* position defines the endpoint ID--i.e. if bit 1 is set in the endpoint
* bitmap, endpoint 1 has a filter rule. Older versions of IPA represent
* the presence of a filter rule for endpoint X by bit (X + 1) being set.
* I.e., bit 1 set indicates the presence of a filter rule for endpoint 0,
* and bit 3 set means there is a filter rule present for endpoint 2.
*
* Each filter table entry has the address of a set of equations that
* implement a filter rule. So following the endpoint bitmap there
* will be such an address/entry for each endpoint with a set bit in
* the bitmap.
*
* The AP initializes all entries in a filter table to refer to a "zero"
* rule. Once initialized, the modem and AP update the entries for
* endpoints they "own" directly. Currently the AP does not use the IPA
* filtering functionality.
*
* This diagram shows an example of a filter table with an endpoint
* bitmap as defined prior to IPA v5.0.
*
* IPA Filter Table
* ----------------------
* endpoint bitmap | 0x0000000000000048 | Bits 3 and 6 set (endpoints 2 and 5)
* |--------------------|
* 1st endpoint | 0x000123456789abc0 | DMA address for modem endpoint 2 rule
* |--------------------|
* 2nd endpoint | 0x000123456789abf0 | DMA address for AP endpoint 5 rule
* |--------------------|
* (unused) | | (Unused space in filter table)
* |--------------------|
* . . .
* |--------------------|
* (unused) | | (Unused space in filter table)
* ----------------------
*
* The set of available route rules is divided about equally between the AP
* and modem. The AP initializes all entries in a route table to refer to
* a "zero entry". Once initialized, the modem and AP are responsible for
* updating their own entries. All entries in a route table are usable,
* though the AP currently does not use the IPA routing functionality.
*
* IPA Route Table
* ----------------------
* 1st modem route | 0x0001234500001100 | DMA address for first route rule
* |--------------------|
* 2nd modem route | 0x0001234500001140 | DMA address for second route rule
* |--------------------|
* . . .
* |--------------------|
* Last modem route| 0x0001234500002280 | DMA address for Nth route rule
* |--------------------|
* 1st AP route | 0x0001234500001100 | DMA address for route rule (N+1)
* |--------------------|
* 2nd AP route | 0x0001234500001140 | DMA address for next route rule
* |--------------------|
* . . .
* |--------------------|
* Last AP route | 0x0001234500002280 | DMA address for last route rule
* ----------------------
*/
/* Filter or route rules consist of a set of 32-bit values followed by a
* 32-bit all-zero rule list terminator. The "zero rule" is simply an
* all-zero rule followed by the list terminator.
*/
#define IPA_ZERO_RULE_SIZE (2 * sizeof(__le32))
/* Check things that can be validated at build time. */
static void ipa_table_validate_build(void)
{
/* Filter and route tables contain DMA addresses that refer
* to filter or route rules. But the size of a table entry
* is 64 bits regardless of what the size of an AP DMA address
* is. A fixed constant defines the size of an entry, and
* code in ipa_table_init() uses a pointer to __le64 to
* initialize tables.
*/
BUILD_BUG_ON(sizeof(dma_addr_t) > sizeof(__le64));
/* A "zero rule" is used to represent no filtering or no routing.
* It is a 64-bit block of zeroed memory. Code in ipa_table_init()
* assumes that it can be written using a pointer to __le64.
*/
BUILD_BUG_ON(IPA_ZERO_RULE_SIZE != sizeof(__le64));
}
static const struct ipa_mem *
ipa_table_mem(struct ipa *ipa, bool filter, bool hashed, bool ipv6)
{
enum ipa_mem_id mem_id;
mem_id = filter ? hashed ? ipv6 ? IPA_MEM_V6_FILTER_HASHED
: IPA_MEM_V4_FILTER_HASHED
: ipv6 ? IPA_MEM_V6_FILTER
: IPA_MEM_V4_FILTER
: hashed ? ipv6 ? IPA_MEM_V6_ROUTE_HASHED
: IPA_MEM_V4_ROUTE_HASHED
: ipv6 ? IPA_MEM_V6_ROUTE
: IPA_MEM_V4_ROUTE;
return ipa_mem_find(ipa, mem_id);
}
/* Return true if hashed tables are supported */
bool ipa_table_hash_support(struct ipa *ipa)
{
return ipa->version != IPA_VERSION_4_2;
}
bool ipa_filtered_valid(struct ipa *ipa, u64 filtered)
{
struct device *dev = ipa->dev;
u32 count;
if (!filtered) {
dev_err(dev, "at least one filtering endpoint is required\n");
return false;
}
count = hweight64(filtered);
if (count > ipa->filter_count) {
dev_err(dev, "too many filtering endpoints (%u > %u)\n",
count, ipa->filter_count);
return false;
}
return true;
}
/* Zero entry count means no table, so just return a 0 address */
static dma_addr_t ipa_table_addr(struct ipa *ipa, bool filter_mask, u16 count)
{
u32 skip;
if (!count)
return 0;
WARN_ON(count > max_t(u32, ipa->filter_count, ipa->route_count));
/* Skip over the zero rule and possibly the filter mask */
skip = filter_mask ? 1 : 2;
return ipa->table_addr + skip * sizeof(*ipa->table_virt);
}
static void ipa_table_reset_add(struct gsi_trans *trans, bool filter,
bool hashed, bool ipv6, u16 first, u16 count)
{
struct ipa *ipa = container_of(trans->gsi, struct ipa, gsi);
const struct ipa_mem *mem;
dma_addr_t addr;
u32 offset;
u16 size;
/* Nothing to do if the memory region is doesn't exist or is empty */
mem = ipa_table_mem(ipa, filter, hashed, ipv6);
if (!mem || !mem->size)
return;
if (filter)
first++; /* skip over bitmap */
offset = mem->offset + first * sizeof(__le64);
size = count * sizeof(__le64);
addr = ipa_table_addr(ipa, false, count);
ipa_cmd_dma_shared_mem_add(trans, offset, size, addr, true);
}
/* Reset entries in a single filter table belonging to either the AP or
* modem to refer to the zero entry. The memory region supplied will be
* for the IPv4 and IPv6 non-hashed and hashed filter tables.
*/
static int
ipa_filter_reset_table(struct ipa *ipa, bool hashed, bool ipv6, bool modem)
{
u64 ep_mask = ipa->filtered;
struct gsi_trans *trans;
enum gsi_ee_id ee_id;
trans = ipa_cmd_trans_alloc(ipa, hweight64(ep_mask));
if (!trans) {
dev_err(ipa->dev, "no transaction for %s filter reset\n",
modem ? "modem" : "AP");
return -EBUSY;
}
ee_id = modem ? GSI_EE_MODEM : GSI_EE_AP;
while (ep_mask) {
u32 endpoint_id = __ffs(ep_mask);
struct ipa_endpoint *endpoint;
ep_mask ^= BIT(endpoint_id);
endpoint = &ipa->endpoint[endpoint_id];
if (endpoint->ee_id != ee_id)
continue;
ipa_table_reset_add(trans, true, hashed, ipv6, endpoint_id, 1);
}
gsi_trans_commit_wait(trans);
return 0;
}
/* Theoretically, each filter table could have more filter slots to
* update than the maximum number of commands in a transaction. So
* we do each table separately.
*/
static int ipa_filter_reset(struct ipa *ipa, bool modem)
{
int ret;
ret = ipa_filter_reset_table(ipa, false, false, modem);
if (ret)
return ret;
net: ipa: only reset hashed tables when supported Last year, the code that manages GSI channel transactions switched from using spinlock-protected linked lists to using indexes into the ring buffer used for a channel. Recently, Google reported seeing transaction reference count underflows occasionally during shutdown. Doug Anderson found a way to reproduce the issue reliably, and bisected the issue to the commit that eliminated the linked lists and the lock. The root cause was ultimately determined to be related to unused transactions being committed as part of the modem shutdown cleanup activity. Unused transactions are not normally expected (except in error cases). The modem uses some ranges of IPA-resident memory, and whenever it shuts down we zero those ranges. In ipa_filter_reset_table() a transaction is allocated to zero modem filter table entries. If hashing is not supported, hashed table memory should not be zeroed. But currently nothing prevents that, and the result is an unused transaction. Something similar occurs when we zero routing table entries for the modem. By preventing any attempt to clear hashed tables when hashing is not supported, the reference count underflow is avoided in this case. Note that there likely remains an issue with properly freeing unused transactions (if they occur due to errors). This patch addresses only the underflows that Google originally reported. Cc: <stable@vger.kernel.org> # 6.1.x Fixes: d338ae28d8a8 ("net: ipa: kill all other transaction lists") Tested-by: Douglas Anderson <dianders@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20230724224055.1688854-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-07-24 15:40:55 -07:00
ret = ipa_filter_reset_table(ipa, false, true, modem);
if (ret || !ipa_table_hash_support(ipa))
return ret;
net: ipa: only reset hashed tables when supported Last year, the code that manages GSI channel transactions switched from using spinlock-protected linked lists to using indexes into the ring buffer used for a channel. Recently, Google reported seeing transaction reference count underflows occasionally during shutdown. Doug Anderson found a way to reproduce the issue reliably, and bisected the issue to the commit that eliminated the linked lists and the lock. The root cause was ultimately determined to be related to unused transactions being committed as part of the modem shutdown cleanup activity. Unused transactions are not normally expected (except in error cases). The modem uses some ranges of IPA-resident memory, and whenever it shuts down we zero those ranges. In ipa_filter_reset_table() a transaction is allocated to zero modem filter table entries. If hashing is not supported, hashed table memory should not be zeroed. But currently nothing prevents that, and the result is an unused transaction. Something similar occurs when we zero routing table entries for the modem. By preventing any attempt to clear hashed tables when hashing is not supported, the reference count underflow is avoided in this case. Note that there likely remains an issue with properly freeing unused transactions (if they occur due to errors). This patch addresses only the underflows that Google originally reported. Cc: <stable@vger.kernel.org> # 6.1.x Fixes: d338ae28d8a8 ("net: ipa: kill all other transaction lists") Tested-by: Douglas Anderson <dianders@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20230724224055.1688854-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-07-24 15:40:55 -07:00
ret = ipa_filter_reset_table(ipa, true, false, modem);
if (ret)
return ret;
net: ipa: only reset hashed tables when supported Last year, the code that manages GSI channel transactions switched from using spinlock-protected linked lists to using indexes into the ring buffer used for a channel. Recently, Google reported seeing transaction reference count underflows occasionally during shutdown. Doug Anderson found a way to reproduce the issue reliably, and bisected the issue to the commit that eliminated the linked lists and the lock. The root cause was ultimately determined to be related to unused transactions being committed as part of the modem shutdown cleanup activity. Unused transactions are not normally expected (except in error cases). The modem uses some ranges of IPA-resident memory, and whenever it shuts down we zero those ranges. In ipa_filter_reset_table() a transaction is allocated to zero modem filter table entries. If hashing is not supported, hashed table memory should not be zeroed. But currently nothing prevents that, and the result is an unused transaction. Something similar occurs when we zero routing table entries for the modem. By preventing any attempt to clear hashed tables when hashing is not supported, the reference count underflow is avoided in this case. Note that there likely remains an issue with properly freeing unused transactions (if they occur due to errors). This patch addresses only the underflows that Google originally reported. Cc: <stable@vger.kernel.org> # 6.1.x Fixes: d338ae28d8a8 ("net: ipa: kill all other transaction lists") Tested-by: Douglas Anderson <dianders@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20230724224055.1688854-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-07-24 15:40:55 -07:00
return ipa_filter_reset_table(ipa, true, true, modem);
}
/* The AP routes and modem routes are each contiguous within the
* table. We can update each table with a single command, and we
* won't exceed the per-transaction command limit.
* */
static int ipa_route_reset(struct ipa *ipa, bool modem)
{
net: ipa: only reset hashed tables when supported Last year, the code that manages GSI channel transactions switched from using spinlock-protected linked lists to using indexes into the ring buffer used for a channel. Recently, Google reported seeing transaction reference count underflows occasionally during shutdown. Doug Anderson found a way to reproduce the issue reliably, and bisected the issue to the commit that eliminated the linked lists and the lock. The root cause was ultimately determined to be related to unused transactions being committed as part of the modem shutdown cleanup activity. Unused transactions are not normally expected (except in error cases). The modem uses some ranges of IPA-resident memory, and whenever it shuts down we zero those ranges. In ipa_filter_reset_table() a transaction is allocated to zero modem filter table entries. If hashing is not supported, hashed table memory should not be zeroed. But currently nothing prevents that, and the result is an unused transaction. Something similar occurs when we zero routing table entries for the modem. By preventing any attempt to clear hashed tables when hashing is not supported, the reference count underflow is avoided in this case. Note that there likely remains an issue with properly freeing unused transactions (if they occur due to errors). This patch addresses only the underflows that Google originally reported. Cc: <stable@vger.kernel.org> # 6.1.x Fixes: d338ae28d8a8 ("net: ipa: kill all other transaction lists") Tested-by: Douglas Anderson <dianders@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20230724224055.1688854-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-07-24 15:40:55 -07:00
bool hash_support = ipa_table_hash_support(ipa);
u32 modem_route_count = ipa->modem_route_count;
struct gsi_trans *trans;
u16 first;
u16 count;
net: ipa: only reset hashed tables when supported Last year, the code that manages GSI channel transactions switched from using spinlock-protected linked lists to using indexes into the ring buffer used for a channel. Recently, Google reported seeing transaction reference count underflows occasionally during shutdown. Doug Anderson found a way to reproduce the issue reliably, and bisected the issue to the commit that eliminated the linked lists and the lock. The root cause was ultimately determined to be related to unused transactions being committed as part of the modem shutdown cleanup activity. Unused transactions are not normally expected (except in error cases). The modem uses some ranges of IPA-resident memory, and whenever it shuts down we zero those ranges. In ipa_filter_reset_table() a transaction is allocated to zero modem filter table entries. If hashing is not supported, hashed table memory should not be zeroed. But currently nothing prevents that, and the result is an unused transaction. Something similar occurs when we zero routing table entries for the modem. By preventing any attempt to clear hashed tables when hashing is not supported, the reference count underflow is avoided in this case. Note that there likely remains an issue with properly freeing unused transactions (if they occur due to errors). This patch addresses only the underflows that Google originally reported. Cc: <stable@vger.kernel.org> # 6.1.x Fixes: d338ae28d8a8 ("net: ipa: kill all other transaction lists") Tested-by: Douglas Anderson <dianders@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20230724224055.1688854-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-07-24 15:40:55 -07:00
trans = ipa_cmd_trans_alloc(ipa, hash_support ? 4 : 2);
if (!trans) {
dev_err(ipa->dev, "no transaction for %s route reset\n",
modem ? "modem" : "AP");
return -EBUSY;
}
if (modem) {
first = 0;
count = modem_route_count;
} else {
first = modem_route_count;
count = ipa->route_count - modem_route_count;
}
ipa_table_reset_add(trans, false, false, false, first, count);
ipa_table_reset_add(trans, false, false, true, first, count);
net: ipa: only reset hashed tables when supported Last year, the code that manages GSI channel transactions switched from using spinlock-protected linked lists to using indexes into the ring buffer used for a channel. Recently, Google reported seeing transaction reference count underflows occasionally during shutdown. Doug Anderson found a way to reproduce the issue reliably, and bisected the issue to the commit that eliminated the linked lists and the lock. The root cause was ultimately determined to be related to unused transactions being committed as part of the modem shutdown cleanup activity. Unused transactions are not normally expected (except in error cases). The modem uses some ranges of IPA-resident memory, and whenever it shuts down we zero those ranges. In ipa_filter_reset_table() a transaction is allocated to zero modem filter table entries. If hashing is not supported, hashed table memory should not be zeroed. But currently nothing prevents that, and the result is an unused transaction. Something similar occurs when we zero routing table entries for the modem. By preventing any attempt to clear hashed tables when hashing is not supported, the reference count underflow is avoided in this case. Note that there likely remains an issue with properly freeing unused transactions (if they occur due to errors). This patch addresses only the underflows that Google originally reported. Cc: <stable@vger.kernel.org> # 6.1.x Fixes: d338ae28d8a8 ("net: ipa: kill all other transaction lists") Tested-by: Douglas Anderson <dianders@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20230724224055.1688854-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2023-07-24 15:40:55 -07:00
if (hash_support) {
ipa_table_reset_add(trans, false, true, false, first, count);
ipa_table_reset_add(trans, false, true, true, first, count);
}
gsi_trans_commit_wait(trans);
return 0;
}
void ipa_table_reset(struct ipa *ipa, bool modem)
{
struct device *dev = ipa->dev;
const char *ee_name;
int ret;
ee_name = modem ? "modem" : "AP";
/* Report errors, but reset filter and route tables */
ret = ipa_filter_reset(ipa, modem);
if (ret)
dev_err(dev, "error %d resetting filter table for %s\n",
ret, ee_name);
ret = ipa_route_reset(ipa, modem);
if (ret)
dev_err(dev, "error %d resetting route table for %s\n",
ret, ee_name);
}
int ipa_table_hash_flush(struct ipa *ipa)
{
struct gsi_trans *trans;
const struct reg *reg;
u32 val;
if (!ipa_table_hash_support(ipa))
return 0;
trans = ipa_cmd_trans_alloc(ipa, 1);
if (!trans) {
dev_err(ipa->dev, "no transaction for hash flush\n");
return -EBUSY;
}
if (ipa->version < IPA_VERSION_5_0) {
reg = ipa_reg(ipa, FILT_ROUT_HASH_FLUSH);
val = reg_bit(reg, IPV6_ROUTER_HASH);
val |= reg_bit(reg, IPV6_FILTER_HASH);
val |= reg_bit(reg, IPV4_ROUTER_HASH);
val |= reg_bit(reg, IPV4_FILTER_HASH);
} else {
reg = ipa_reg(ipa, FILT_ROUT_CACHE_FLUSH);
/* IPA v5.0+ uses a unified cache (both IPv4 and IPv6) */
val = reg_bit(reg, ROUTER_CACHE);
val |= reg_bit(reg, FILTER_CACHE);
}
ipa_cmd_register_write_add(trans, reg_offset(reg), val, val, false);
gsi_trans_commit_wait(trans);
return 0;
}
static void ipa_table_init_add(struct gsi_trans *trans, bool filter, bool ipv6)
{
struct ipa *ipa = container_of(trans->gsi, struct ipa, gsi);
const struct ipa_mem *hash_mem;
enum ipa_cmd_opcode opcode;
const struct ipa_mem *mem;
dma_addr_t hash_addr;
dma_addr_t addr;
u32 hash_offset;
u32 zero_offset;
u16 hash_count;
u32 zero_size;
u16 hash_size;
u16 count;
u16 size;
opcode = filter ? ipv6 ? IPA_CMD_IP_V6_FILTER_INIT
: IPA_CMD_IP_V4_FILTER_INIT
: ipv6 ? IPA_CMD_IP_V6_ROUTING_INIT
: IPA_CMD_IP_V4_ROUTING_INIT;
/* The non-hashed region will exist (see ipa_table_mem_valid()) */
mem = ipa_table_mem(ipa, filter, false, ipv6);
hash_mem = ipa_table_mem(ipa, filter, true, ipv6);
hash_offset = hash_mem ? hash_mem->offset : 0;
/* Compute the number of table entries to initialize */
if (filter) {
/* The number of filtering endpoints determines number of
* entries in the filter table; we also add one more "slot"
* to hold the bitmap itself. The size of the hashed filter
* table is either the same as the non-hashed one, or zero.
*/
count = 1 + hweight64(ipa->filtered);
hash_count = hash_mem && hash_mem->size ? count : 0;
} else {
/* The size of a route table region determines the number
* of entries it has.
*/
count = mem->size / sizeof(__le64);
hash_count = hash_mem ? hash_mem->size / sizeof(__le64) : 0;
}
size = count * sizeof(__le64);
hash_size = hash_count * sizeof(__le64);
addr = ipa_table_addr(ipa, filter, count);
hash_addr = ipa_table_addr(ipa, filter, hash_count);
ipa_cmd_table_init_add(trans, opcode, size, mem->offset, addr,
hash_size, hash_offset, hash_addr);
if (!filter)
return;
/* Zero the unused space in the filter table */
zero_offset = mem->offset + size;
zero_size = mem->size - size;
ipa_cmd_dma_shared_mem_add(trans, zero_offset, zero_size,
ipa->zero_addr, true);
if (!hash_size)
return;
/* Zero the unused space in the hashed filter table */
zero_offset = hash_offset + hash_size;
zero_size = hash_mem->size - hash_size;
ipa_cmd_dma_shared_mem_add(trans, zero_offset, zero_size,
ipa->zero_addr, true);
}
int ipa_table_setup(struct ipa *ipa)
{
struct gsi_trans *trans;
/* We will need at most 8 TREs:
* - IPv4:
* - One for route table initialization (non-hashed and hashed)
* - One for filter table initialization (non-hashed and hashed)
* - One to zero unused entries in the non-hashed filter table
* - One to zero unused entries in the hashed filter table
* - IPv6:
* - One for route table initialization (non-hashed and hashed)
* - One for filter table initialization (non-hashed and hashed)
* - One to zero unused entries in the non-hashed filter table
* - One to zero unused entries in the hashed filter table
* All platforms support at least 8 TREs in a transaction.
*/
trans = ipa_cmd_trans_alloc(ipa, 8);
if (!trans) {
dev_err(ipa->dev, "no transaction for table setup\n");
return -EBUSY;
}
ipa_table_init_add(trans, false, false);
ipa_table_init_add(trans, false, true);
ipa_table_init_add(trans, true, false);
ipa_table_init_add(trans, true, true);
gsi_trans_commit_wait(trans);
return 0;
}
/**
* ipa_filter_tuple_zero() - Zero an endpoint's hashed filter tuple
* @endpoint: Endpoint whose filter hash tuple should be zeroed
*
* Endpoint must be for the AP (not modem) and support filtering. Updates
* the filter hash values without changing route ones.
*/
static void ipa_filter_tuple_zero(struct ipa_endpoint *endpoint)
{
u32 endpoint_id = endpoint->endpoint_id;
struct ipa *ipa = endpoint->ipa;
const struct reg *reg;
u32 offset;
u32 val;
if (ipa->version < IPA_VERSION_5_0) {
reg = ipa_reg(ipa, ENDP_FILTER_ROUTER_HSH_CFG);
offset = reg_n_offset(reg, endpoint_id);
val = ioread32(endpoint->ipa->reg_virt + offset);
/* Zero all filter-related fields, preserving the rest */
val &= ~reg_fmask(reg, FILTER_HASH_MSK_ALL);
} else {
/* IPA v5.0 separates filter and router cache configuration */
reg = ipa_reg(ipa, ENDP_FILTER_CACHE_CFG);
offset = reg_n_offset(reg, endpoint_id);
/* Zero all filter-related fields */
val = 0;
}
iowrite32(val, endpoint->ipa->reg_virt + offset);
}
/* Configure a hashed filter table; there is no ipa_filter_deconfig() */
static void ipa_filter_config(struct ipa *ipa, bool modem)
{
enum gsi_ee_id ee_id = modem ? GSI_EE_MODEM : GSI_EE_AP;
u64 ep_mask = ipa->filtered;
if (!ipa_table_hash_support(ipa))
return;
while (ep_mask) {
u32 endpoint_id = __ffs(ep_mask);
struct ipa_endpoint *endpoint;
ep_mask ^= BIT(endpoint_id);
endpoint = &ipa->endpoint[endpoint_id];
if (endpoint->ee_id == ee_id)
ipa_filter_tuple_zero(endpoint);
}
}
static bool ipa_route_id_modem(struct ipa *ipa, u32 route_id)
{
return route_id < ipa->modem_route_count;
}
/**
* ipa_route_tuple_zero() - Zero a hashed route table entry tuple
* @ipa: IPA pointer
* @route_id: Route table entry whose hash tuple should be zeroed
*
* Updates the route hash values without changing filter ones.
*/
static void ipa_route_tuple_zero(struct ipa *ipa, u32 route_id)
{
const struct reg *reg;
u32 offset;
u32 val;
if (ipa->version < IPA_VERSION_5_0) {
reg = ipa_reg(ipa, ENDP_FILTER_ROUTER_HSH_CFG);
offset = reg_n_offset(reg, route_id);
val = ioread32(ipa->reg_virt + offset);
/* Zero all route-related fields, preserving the rest */
val &= ~reg_fmask(reg, ROUTER_HASH_MSK_ALL);
} else {
/* IPA v5.0 separates filter and router cache configuration */
reg = ipa_reg(ipa, ENDP_ROUTER_CACHE_CFG);
offset = reg_n_offset(reg, route_id);
/* Zero all route-related fields */
val = 0;
}
iowrite32(val, ipa->reg_virt + offset);
}
/* Configure a hashed route table; there is no ipa_route_deconfig() */
static void ipa_route_config(struct ipa *ipa, bool modem)
{
u32 route_id;
if (!ipa_table_hash_support(ipa))
return;
for (route_id = 0; route_id < ipa->route_count; route_id++)
if (ipa_route_id_modem(ipa, route_id) == modem)
ipa_route_tuple_zero(ipa, route_id);
}
/* Configure a filter and route tables; there is no ipa_table_deconfig() */
void ipa_table_config(struct ipa *ipa)
{
ipa_filter_config(ipa, false);
ipa_filter_config(ipa, true);
ipa_route_config(ipa, false);
ipa_route_config(ipa, true);
}
/* Verify the sizes of all IPA table filter or routing table memory regions
* are valid. If valid, this records the size of the routing table.
*/
bool ipa_table_mem_valid(struct ipa *ipa, bool filter)
{
bool hash_support = ipa_table_hash_support(ipa);
const struct ipa_mem *mem_hashed;
const struct ipa_mem *mem_ipv4;
const struct ipa_mem *mem_ipv6;
u32 count;
/* IPv4 and IPv6 non-hashed tables are expected to be defined and
* have the same size. Both must have at least two entries (and
* would normally have more than that).
*/
mem_ipv4 = ipa_table_mem(ipa, filter, false, false);
if (!mem_ipv4)
return false;
mem_ipv6 = ipa_table_mem(ipa, filter, false, true);
if (!mem_ipv6)
return false;
if (mem_ipv4->size != mem_ipv6->size)
return false;
/* Compute and record the number of entries for each table type */
count = mem_ipv4->size / sizeof(__le64);
if (count < 2)
return false;
if (filter)
ipa->filter_count = count - 1; /* Filter map in first entry */
else
ipa->route_count = count;
/* Table offset and size must fit in TABLE_INIT command fields */
if (!ipa_cmd_table_init_valid(ipa, mem_ipv4, !filter))
return false;
/* Make sure the regions are big enough */
if (filter) {
/* Filter tables must able to hold the endpoint bitmap plus
* an entry for each endpoint that supports filtering
*/
if (count < 1 + hweight64(ipa->filtered))
return false;
} else {
/* Routing tables must be able to hold all modem entries,
* plus at least one entry for the AP.
*/
if (count < ipa->modem_route_count + 1)
return false;
}
/* If hashing is supported, hashed tables are expected to be defined,
* and have the same size as non-hashed tables. If hashing is not
* supported, hashed tables are expected to have zero size (or not
* be defined).
*/
mem_hashed = ipa_table_mem(ipa, filter, true, false);
if (hash_support) {
if (!mem_hashed || mem_hashed->size != mem_ipv4->size)
return false;
} else {
if (mem_hashed && mem_hashed->size)
return false;
}
/* Same check for IPv6 tables */
mem_hashed = ipa_table_mem(ipa, filter, true, true);
if (hash_support) {
if (!mem_hashed || mem_hashed->size != mem_ipv6->size)
return false;
} else {
if (mem_hashed && mem_hashed->size)
return false;
}
return true;
}
/* Initialize a coherent DMA allocation containing initialized filter and
* route table data. This is used when initializing or resetting the IPA
* filter or route table.
*
* The first entry in a filter table contains a bitmap indicating which
* endpoints contain entries in the table. In addition to that first entry,
* there is a fixed maximum number of entries that follow. Filter table
* entries are 64 bits wide, and (other than the bitmap) contain the DMA
* address of a filter rule. A "zero rule" indicates no filtering, and
* consists of 64 bits of zeroes. When a filter table is initialized (or
* reset) its entries are made to refer to the zero rule.
*
* Each entry in a route table is the DMA address of a routing rule. For
* routing there is also a 64-bit "zero rule" that means no routing, and
* when a route table is initialized or reset, its entries are made to refer
* to the zero rule. The zero rule is shared for route and filter tables.
*
* +-------------------+
* --> | zero rule |
* / |-------------------|
* | | filter mask |
* |\ |-------------------|
* | ---- zero rule address | \
* |\ |-------------------| |
* | ---- zero rule address | | Max IPA filter count
* | |-------------------| > or IPA route count,
* | ... | whichever is greater
* \ |-------------------| |
* ---- zero rule address | /
* +-------------------+
*/
int ipa_table_init(struct ipa *ipa)
{
struct device *dev = ipa->dev;
dma_addr_t addr;
__le64 le_addr;
__le64 *virt;
size_t size;
u32 count;
ipa_table_validate_build();
count = max_t(u32, ipa->filter_count, ipa->route_count);
/* The IPA hardware requires route and filter table rules to be
* aligned on a 128-byte boundary. We put the "zero rule" at the
* base of the table area allocated here. The DMA address returned
* by dma_alloc_coherent() is guaranteed to be a power-of-2 number
* of pages, which satisfies the rule alignment requirement.
*/
size = IPA_ZERO_RULE_SIZE + (1 + count) * sizeof(__le64);
virt = dma_alloc_coherent(dev, size, &addr, GFP_KERNEL);
if (!virt)
return -ENOMEM;
ipa->table_virt = virt;
ipa->table_addr = addr;
/* First slot is the zero rule */
*virt++ = 0;
/* Next is the filter table bitmap. The "soft" bitmap value might
* need to be converted to the hardware representation by shifting
* it left one position. Prior to IPA v5.0, bit 0 repesents global
* filtering, which is possible but not used. IPA v5.0+ eliminated
* that option, so there's no shifting required.
*/
if (ipa->version < IPA_VERSION_5_0)
*virt++ = cpu_to_le64(ipa->filtered << 1);
else
*virt++ = cpu_to_le64(ipa->filtered);
/* All the rest contain the DMA address of the zero rule */
le_addr = cpu_to_le64(addr);
while (count--)
*virt++ = le_addr;
return 0;
}
void ipa_table_exit(struct ipa *ipa)
{
u32 count = max_t(u32, 1 + ipa->filter_count, ipa->route_count);
struct device *dev = ipa->dev;
size_t size;
size = IPA_ZERO_RULE_SIZE + (1 + count) * sizeof(__le64);
dma_free_coherent(dev, size, ipa->table_virt, ipa->table_addr);
ipa->table_addr = 0;
ipa->table_virt = NULL;
}