linux/drivers/net/ipa/gsi_trans.c

805 lines
24 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0
/* Copyright (c) 2012-2018, The Linux Foundation. All rights reserved.
* Copyright (C) 2019-2020 Linaro Ltd.
*/
#include <linux/types.h>
#include <linux/bits.h>
#include <linux/bitfield.h>
#include <linux/refcount.h>
#include <linux/scatterlist.h>
#include <linux/dma-direction.h>
#include "gsi.h"
#include "gsi_private.h"
#include "gsi_trans.h"
#include "ipa_gsi.h"
#include "ipa_data.h"
#include "ipa_cmd.h"
/**
* DOC: GSI Transactions
*
* A GSI transaction abstracts the behavior of a GSI channel by representing
* everything about a related group of IPA commands in a single structure.
* (A "command" in this sense is either a data transfer or an IPA immediate
* command.) Most details of interaction with the GSI hardware are managed
* by the GSI transaction core, allowing users to simply describe commands
* to be performed. When a transaction has completed a callback function
* (dependent on the type of endpoint associated with the channel) allows
* cleanup of resources associated with the transaction.
*
* To perform a command (or set of them), a user of the GSI transaction
* interface allocates a transaction, indicating the number of TREs required
* (one per command). If sufficient TREs are available, they are reserved
* for use in the transaction and the allocation succeeds. This way
* exhaustion of the available TREs in a channel ring is detected
* as early as possible. All resources required to complete a transaction
* are allocated at transaction allocation time.
*
* Commands performed as part of a transaction are represented in an array
* of Linux scatterlist structures. This array is allocated with the
* transaction, and its entries are initialized using standard scatterlist
* functions (such as sg_set_buf() or skb_to_sgvec()).
*
* Once a transaction's scatterlist structures have been initialized, the
* transaction is committed. The caller is responsible for mapping buffers
* for DMA if necessary, and this should be done *before* allocating
* the transaction. Between a successful allocation and commit of a
* transaction no errors should occur.
*
* Committing transfers ownership of the entire transaction to the GSI
* transaction core. The GSI transaction code formats the content of
* the scatterlist array into the channel ring buffer and informs the
* hardware that new TREs are available to process.
*
* The last TRE in each transaction is marked to interrupt the AP when the
* GSI hardware has completed it. Because transfers described by TREs are
* performed strictly in order, signaling the completion of just the last
* TRE in the transaction is sufficient to indicate the full transaction
* is complete.
*
* When a transaction is complete, ipa_gsi_trans_complete() is called by the
* GSI code into the IPA layer, allowing it to perform any final cleanup
* required before the transaction is freed.
*/
/* Hardware values representing a transfer element type */
enum gsi_tre_type {
GSI_RE_XFER = 0x2,
GSI_RE_IMMD_CMD = 0x3,
};
/* An entry in a channel ring */
struct gsi_tre {
__le64 addr; /* DMA address */
__le16 len_opcode; /* length in bytes or enum IPA_CMD_* */
__le16 reserved;
__le32 flags; /* TRE_FLAGS_* */
};
/* gsi_tre->flags mask values (in CPU byte order) */
#define TRE_FLAGS_CHAIN_FMASK GENMASK(0, 0)
#define TRE_FLAGS_IEOT_FMASK GENMASK(9, 9)
#define TRE_FLAGS_BEI_FMASK GENMASK(10, 10)
#define TRE_FLAGS_TYPE_FMASK GENMASK(23, 16)
int gsi_trans_pool_init(struct gsi_trans_pool *pool, size_t size, u32 count,
u32 max_alloc)
{
void *virt;
#ifdef IPA_VALIDATE
if (!size || size % 8)
return -EINVAL;
if (count < max_alloc)
return -EINVAL;
if (!max_alloc)
return -EINVAL;
#endif /* IPA_VALIDATE */
/* By allocating a few extra entries in our pool (one less
* than the maximum number that will be requested in a
* single allocation), we can always satisfy requests without
* ever worrying about straddling the end of the pool array.
* If there aren't enough entries starting at the free index,
* we just allocate free entries from the beginning of the pool.
*/
virt = kcalloc(count + max_alloc - 1, size, GFP_KERNEL);
if (!virt)
return -ENOMEM;
pool->base = virt;
/* If the allocator gave us any extra memory, use it */
pool->count = ksize(pool->base) / size;
pool->free = 0;
pool->max_alloc = max_alloc;
pool->size = size;
pool->addr = 0; /* Only used for DMA pools */
return 0;
}
void gsi_trans_pool_exit(struct gsi_trans_pool *pool)
{
kfree(pool->base);
memset(pool, 0, sizeof(*pool));
}
/* Allocate the requested number of (zeroed) entries from the pool */
/* Home-grown DMA pool. This way we can preallocate and use the tre_count
* to guarantee allocations will succeed. Even though we specify max_alloc
* (and it can be more than one), we only allow allocation of a single
* element from a DMA pool.
*/
int gsi_trans_pool_init_dma(struct device *dev, struct gsi_trans_pool *pool,
size_t size, u32 count, u32 max_alloc)
{
size_t total_size;
dma_addr_t addr;
void *virt;
#ifdef IPA_VALIDATE
if (!size || size % 8)
return -EINVAL;
if (count < max_alloc)
return -EINVAL;
if (!max_alloc)
return -EINVAL;
#endif /* IPA_VALIDATE */
/* Don't let allocations cross a power-of-two boundary */
size = __roundup_pow_of_two(size);
total_size = (count + max_alloc - 1) * size;
/* The allocator will give us a power-of-2 number of pages. But we
* can't guarantee that, so request it. That way we won't waste any
* memory that would be available beyond the required space.
*/
total_size = get_order(total_size) << PAGE_SHIFT;
virt = dma_alloc_coherent(dev, total_size, &addr, GFP_KERNEL);
if (!virt)
return -ENOMEM;
pool->base = virt;
pool->count = total_size / size;
pool->free = 0;
pool->size = size;
pool->max_alloc = max_alloc;
pool->addr = addr;
return 0;
}
void gsi_trans_pool_exit_dma(struct device *dev, struct gsi_trans_pool *pool)
{
dma_free_coherent(dev, pool->size, pool->base, pool->addr);
memset(pool, 0, sizeof(*pool));
}
/* Return the byte offset of the next free entry in the pool */
static u32 gsi_trans_pool_alloc_common(struct gsi_trans_pool *pool, u32 count)
{
u32 offset;
/* assert(count > 0); */
/* assert(count <= pool->max_alloc); */
/* Allocate from beginning if wrap would occur */
if (count > pool->count - pool->free)
pool->free = 0;
offset = pool->free * pool->size;
pool->free += count;
memset(pool->base + offset, 0, count * pool->size);
return offset;
}
/* Allocate a contiguous block of zeroed entries from a pool */
void *gsi_trans_pool_alloc(struct gsi_trans_pool *pool, u32 count)
{
return pool->base + gsi_trans_pool_alloc_common(pool, count);
}
/* Allocate a single zeroed entry from a DMA pool */
void *gsi_trans_pool_alloc_dma(struct gsi_trans_pool *pool, dma_addr_t *addr)
{
u32 offset = gsi_trans_pool_alloc_common(pool, 1);
*addr = pool->addr + offset;
return pool->base + offset;
}
/* Return the pool element that immediately follows the one given.
* This only works done if elements are allocated one at a time.
*/
void *gsi_trans_pool_next(struct gsi_trans_pool *pool, void *element)
{
void *end = pool->base + pool->count * pool->size;
/* assert(element >= pool->base); */
/* assert(element < end); */
/* assert(pool->max_alloc == 1); */
element += pool->size;
return element < end ? element : pool->base;
}
/* Map a given ring entry index to the transaction associated with it */
static void gsi_channel_trans_map(struct gsi_channel *channel, u32 index,
struct gsi_trans *trans)
{
/* Note: index *must* be used modulo the ring count here */
channel->trans_info.map[index % channel->tre_ring.count] = trans;
}
/* Return the transaction mapped to a given ring entry */
struct gsi_trans *
gsi_channel_trans_mapped(struct gsi_channel *channel, u32 index)
{
/* Note: index *must* be used modulo the ring count here */
return channel->trans_info.map[index % channel->tre_ring.count];
}
/* Return the oldest completed transaction for a channel (or null) */
struct gsi_trans *gsi_channel_trans_complete(struct gsi_channel *channel)
{
return list_first_entry_or_null(&channel->trans_info.complete,
struct gsi_trans, links);
}
/* Move a transaction from the allocated list to the pending list */
static void gsi_trans_move_pending(struct gsi_trans *trans)
{
struct gsi_channel *channel = &trans->gsi->channel[trans->channel_id];
struct gsi_trans_info *trans_info = &channel->trans_info;
spin_lock_bh(&trans_info->spinlock);
list_move_tail(&trans->links, &trans_info->pending);
spin_unlock_bh(&trans_info->spinlock);
}
/* Move a transaction and all of its predecessors from the pending list
* to the completed list.
*/
void gsi_trans_move_complete(struct gsi_trans *trans)
{
struct gsi_channel *channel = &trans->gsi->channel[trans->channel_id];
struct gsi_trans_info *trans_info = &channel->trans_info;
struct list_head list;
spin_lock_bh(&trans_info->spinlock);
/* Move this transaction and all predecessors to completed list */
list_cut_position(&list, &trans_info->pending, &trans->links);
list_splice_tail(&list, &trans_info->complete);
spin_unlock_bh(&trans_info->spinlock);
}
/* Move a transaction from the completed list to the polled list */
void gsi_trans_move_polled(struct gsi_trans *trans)
{
struct gsi_channel *channel = &trans->gsi->channel[trans->channel_id];
struct gsi_trans_info *trans_info = &channel->trans_info;
spin_lock_bh(&trans_info->spinlock);
list_move_tail(&trans->links, &trans_info->polled);
spin_unlock_bh(&trans_info->spinlock);
}
/* Reserve some number of TREs on a channel. Returns true if successful */
static bool
gsi_trans_tre_reserve(struct gsi_trans_info *trans_info, u32 tre_count)
{
int avail = atomic_read(&trans_info->tre_avail);
int new;
do {
new = avail - (int)tre_count;
if (unlikely(new < 0))
return false;
} while (!atomic_try_cmpxchg(&trans_info->tre_avail, &avail, new));
return true;
}
/* Release previously-reserved TRE entries to a channel */
static void
gsi_trans_tre_release(struct gsi_trans_info *trans_info, u32 tre_count)
{
atomic_add(tre_count, &trans_info->tre_avail);
}
/* Allocate a GSI transaction on a channel */
struct gsi_trans *gsi_channel_trans_alloc(struct gsi *gsi, u32 channel_id,
u32 tre_count,
enum dma_data_direction direction)
{
struct gsi_channel *channel = &gsi->channel[channel_id];
struct gsi_trans_info *trans_info;
struct gsi_trans *trans;
/* assert(tre_count <= gsi_channel_trans_tre_max(gsi, channel_id)); */
trans_info = &channel->trans_info;
/* We reserve the TREs now, but consume them at commit time.
* If there aren't enough available, we're done.
*/
if (!gsi_trans_tre_reserve(trans_info, tre_count))
return NULL;
/* Allocate and initialize non-zero fields in the the transaction */
trans = gsi_trans_pool_alloc(&trans_info->pool, 1);
trans->gsi = gsi;
trans->channel_id = channel_id;
trans->tre_count = tre_count;
init_completion(&trans->completion);
/* Allocate the scatterlist and (if requested) info entries. */
trans->sgl = gsi_trans_pool_alloc(&trans_info->sg_pool, tre_count);
sg_init_marker(trans->sgl, tre_count);
trans->direction = direction;
spin_lock_bh(&trans_info->spinlock);
list_add_tail(&trans->links, &trans_info->alloc);
spin_unlock_bh(&trans_info->spinlock);
refcount_set(&trans->refcount, 1);
return trans;
}
net: ipa: lock when freeing transaction Transactions sit on one of several lists, depending on their state (allocated, pending, complete, or polled). A spinlock protects against concurrent access when transactions are moved between these lists. Transactions are also reference counted. A newly-allocated transaction has an initial count of 1; a transaction is released in gsi_trans_free() only if its decremented reference count reaches 0. Releasing a transaction includes removing it from the polled (or if unused, allocated) list, so the spinlock is acquired when we release a transaction. The reference count is used to allow a caller to synchronously wait for a committed transaction to complete. In this case, the waiter takes an extra reference to the transaction *before* committing it (so it won't be freed), and releases its reference (calls gsi_trans_free()) when it is done with it. Similarly, gsi_channel_update() takes an extra reference to ensure a transaction isn't released before the function is done operating on it. Until the transaction is moved to the completed list (by this function) it won't be freed, so this reference is taken "safely." But in the quiesce path, we want to wait for the "last" transaction, which we find in the completed or polled list. Transactions on these lists can be freed at any time, so we (try to) prevent that by taking the reference while holding the spinlock. Currently gsi_trans_free() decrements a transaction's reference count unconditionally, acquiring the lock to remove the transaction from its list *only* when the count reaches 0. This does not protect the quiesce path, which depends on the lock to ensure its extra reference prevents release of the transaction. Fix this by only dropping the last reference to a transaction in gsi_trans_free() while holding the spinlock. Fixes: 9dd441e4ed575 ("soc: qcom: ipa: GSI transactions") Reported-by: Stephen Boyd <swboyd@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20201114182017.28270-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2020-11-15 02:20:17 +08:00
/* Free a previously-allocated transaction */
void gsi_trans_free(struct gsi_trans *trans)
{
net: ipa: lock when freeing transaction Transactions sit on one of several lists, depending on their state (allocated, pending, complete, or polled). A spinlock protects against concurrent access when transactions are moved between these lists. Transactions are also reference counted. A newly-allocated transaction has an initial count of 1; a transaction is released in gsi_trans_free() only if its decremented reference count reaches 0. Releasing a transaction includes removing it from the polled (or if unused, allocated) list, so the spinlock is acquired when we release a transaction. The reference count is used to allow a caller to synchronously wait for a committed transaction to complete. In this case, the waiter takes an extra reference to the transaction *before* committing it (so it won't be freed), and releases its reference (calls gsi_trans_free()) when it is done with it. Similarly, gsi_channel_update() takes an extra reference to ensure a transaction isn't released before the function is done operating on it. Until the transaction is moved to the completed list (by this function) it won't be freed, so this reference is taken "safely." But in the quiesce path, we want to wait for the "last" transaction, which we find in the completed or polled list. Transactions on these lists can be freed at any time, so we (try to) prevent that by taking the reference while holding the spinlock. Currently gsi_trans_free() decrements a transaction's reference count unconditionally, acquiring the lock to remove the transaction from its list *only* when the count reaches 0. This does not protect the quiesce path, which depends on the lock to ensure its extra reference prevents release of the transaction. Fix this by only dropping the last reference to a transaction in gsi_trans_free() while holding the spinlock. Fixes: 9dd441e4ed575 ("soc: qcom: ipa: GSI transactions") Reported-by: Stephen Boyd <swboyd@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20201114182017.28270-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2020-11-15 02:20:17 +08:00
refcount_t *refcount = &trans->refcount;
struct gsi_trans_info *trans_info;
net: ipa: lock when freeing transaction Transactions sit on one of several lists, depending on their state (allocated, pending, complete, or polled). A spinlock protects against concurrent access when transactions are moved between these lists. Transactions are also reference counted. A newly-allocated transaction has an initial count of 1; a transaction is released in gsi_trans_free() only if its decremented reference count reaches 0. Releasing a transaction includes removing it from the polled (or if unused, allocated) list, so the spinlock is acquired when we release a transaction. The reference count is used to allow a caller to synchronously wait for a committed transaction to complete. In this case, the waiter takes an extra reference to the transaction *before* committing it (so it won't be freed), and releases its reference (calls gsi_trans_free()) when it is done with it. Similarly, gsi_channel_update() takes an extra reference to ensure a transaction isn't released before the function is done operating on it. Until the transaction is moved to the completed list (by this function) it won't be freed, so this reference is taken "safely." But in the quiesce path, we want to wait for the "last" transaction, which we find in the completed or polled list. Transactions on these lists can be freed at any time, so we (try to) prevent that by taking the reference while holding the spinlock. Currently gsi_trans_free() decrements a transaction's reference count unconditionally, acquiring the lock to remove the transaction from its list *only* when the count reaches 0. This does not protect the quiesce path, which depends on the lock to ensure its extra reference prevents release of the transaction. Fix this by only dropping the last reference to a transaction in gsi_trans_free() while holding the spinlock. Fixes: 9dd441e4ed575 ("soc: qcom: ipa: GSI transactions") Reported-by: Stephen Boyd <swboyd@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20201114182017.28270-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2020-11-15 02:20:17 +08:00
bool last;
net: ipa: lock when freeing transaction Transactions sit on one of several lists, depending on their state (allocated, pending, complete, or polled). A spinlock protects against concurrent access when transactions are moved between these lists. Transactions are also reference counted. A newly-allocated transaction has an initial count of 1; a transaction is released in gsi_trans_free() only if its decremented reference count reaches 0. Releasing a transaction includes removing it from the polled (or if unused, allocated) list, so the spinlock is acquired when we release a transaction. The reference count is used to allow a caller to synchronously wait for a committed transaction to complete. In this case, the waiter takes an extra reference to the transaction *before* committing it (so it won't be freed), and releases its reference (calls gsi_trans_free()) when it is done with it. Similarly, gsi_channel_update() takes an extra reference to ensure a transaction isn't released before the function is done operating on it. Until the transaction is moved to the completed list (by this function) it won't be freed, so this reference is taken "safely." But in the quiesce path, we want to wait for the "last" transaction, which we find in the completed or polled list. Transactions on these lists can be freed at any time, so we (try to) prevent that by taking the reference while holding the spinlock. Currently gsi_trans_free() decrements a transaction's reference count unconditionally, acquiring the lock to remove the transaction from its list *only* when the count reaches 0. This does not protect the quiesce path, which depends on the lock to ensure its extra reference prevents release of the transaction. Fix this by only dropping the last reference to a transaction in gsi_trans_free() while holding the spinlock. Fixes: 9dd441e4ed575 ("soc: qcom: ipa: GSI transactions") Reported-by: Stephen Boyd <swboyd@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20201114182017.28270-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2020-11-15 02:20:17 +08:00
/* We must hold the lock to release the last reference */
if (refcount_dec_not_one(refcount))
return;
trans_info = &trans->gsi->channel[trans->channel_id].trans_info;
spin_lock_bh(&trans_info->spinlock);
net: ipa: lock when freeing transaction Transactions sit on one of several lists, depending on their state (allocated, pending, complete, or polled). A spinlock protects against concurrent access when transactions are moved between these lists. Transactions are also reference counted. A newly-allocated transaction has an initial count of 1; a transaction is released in gsi_trans_free() only if its decremented reference count reaches 0. Releasing a transaction includes removing it from the polled (or if unused, allocated) list, so the spinlock is acquired when we release a transaction. The reference count is used to allow a caller to synchronously wait for a committed transaction to complete. In this case, the waiter takes an extra reference to the transaction *before* committing it (so it won't be freed), and releases its reference (calls gsi_trans_free()) when it is done with it. Similarly, gsi_channel_update() takes an extra reference to ensure a transaction isn't released before the function is done operating on it. Until the transaction is moved to the completed list (by this function) it won't be freed, so this reference is taken "safely." But in the quiesce path, we want to wait for the "last" transaction, which we find in the completed or polled list. Transactions on these lists can be freed at any time, so we (try to) prevent that by taking the reference while holding the spinlock. Currently gsi_trans_free() decrements a transaction's reference count unconditionally, acquiring the lock to remove the transaction from its list *only* when the count reaches 0. This does not protect the quiesce path, which depends on the lock to ensure its extra reference prevents release of the transaction. Fix this by only dropping the last reference to a transaction in gsi_trans_free() while holding the spinlock. Fixes: 9dd441e4ed575 ("soc: qcom: ipa: GSI transactions") Reported-by: Stephen Boyd <swboyd@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20201114182017.28270-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2020-11-15 02:20:17 +08:00
/* Reference might have been added before we got the lock */
last = refcount_dec_and_test(refcount);
if (last)
list_del(&trans->links);
spin_unlock_bh(&trans_info->spinlock);
net: ipa: lock when freeing transaction Transactions sit on one of several lists, depending on their state (allocated, pending, complete, or polled). A spinlock protects against concurrent access when transactions are moved between these lists. Transactions are also reference counted. A newly-allocated transaction has an initial count of 1; a transaction is released in gsi_trans_free() only if its decremented reference count reaches 0. Releasing a transaction includes removing it from the polled (or if unused, allocated) list, so the spinlock is acquired when we release a transaction. The reference count is used to allow a caller to synchronously wait for a committed transaction to complete. In this case, the waiter takes an extra reference to the transaction *before* committing it (so it won't be freed), and releases its reference (calls gsi_trans_free()) when it is done with it. Similarly, gsi_channel_update() takes an extra reference to ensure a transaction isn't released before the function is done operating on it. Until the transaction is moved to the completed list (by this function) it won't be freed, so this reference is taken "safely." But in the quiesce path, we want to wait for the "last" transaction, which we find in the completed or polled list. Transactions on these lists can be freed at any time, so we (try to) prevent that by taking the reference while holding the spinlock. Currently gsi_trans_free() decrements a transaction's reference count unconditionally, acquiring the lock to remove the transaction from its list *only* when the count reaches 0. This does not protect the quiesce path, which depends on the lock to ensure its extra reference prevents release of the transaction. Fix this by only dropping the last reference to a transaction in gsi_trans_free() while holding the spinlock. Fixes: 9dd441e4ed575 ("soc: qcom: ipa: GSI transactions") Reported-by: Stephen Boyd <swboyd@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20201114182017.28270-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2020-11-15 02:20:17 +08:00
if (!last)
return;
ipa_gsi_trans_release(trans);
/* Releasing the reserved TREs implicitly frees the sgl[] and
* (if present) info[] arrays, plus the transaction itself.
*/
gsi_trans_tre_release(trans_info, trans->tre_count);
}
/* Add an immediate command to a transaction */
void gsi_trans_cmd_add(struct gsi_trans *trans, void *buf, u32 size,
dma_addr_t addr, enum dma_data_direction direction,
enum ipa_cmd_opcode opcode)
{
struct ipa_cmd_info *info;
u32 which = trans->used++;
struct scatterlist *sg;
/* assert(which < trans->tre_count); */
net: ipa: command payloads already mapped IPA transactions describe actions to be performed by the IPA hardware. Three cases use IPA transactions: transmitting a socket buffer; providing a page to receive packet data; and issuing an IPA immediate command. An IPA transaction contains a scatter/gather list (SGL) to hold the set of actions to be performed. We map buffers in the SGL for DMA at the time they are added to the transaction. For skb TX transactions, we fill the SGL with a call to skb_to_sgvec(). Page RX transactions involve a single page pointer, and that is recorded in the SGL with sg_set_page(). In both of these cases we then map the SGL for DMA with a call to dma_map_sg(). Immediate commands are different. The payload for an immediate command comes from a region of coherent DMA memory, which must *not* be mapped for DMA. For that reason, gsi_trans_cmd_add() sort of hand-crafts each SGL entry added to a command transaction. This patch fixes a problem with the code that crafts the SGL entry for an immediate command. Previously a portion of the SGL entry was updated using sg_set_buf(). However this is not valid because it includes a call to virt_to_page() on the buffer, but the command buffer pointer is not a linear address. Since we never actually map the SGL for command transactions, there are very few fields in the SGL we need to fill. Specifically, we only need to record the DMA address and the length, so they can be used by __gsi_trans_commit() to fill a TRE. We additionally need to preserve the SGL flags so for_each_sg() still works. For that we can simply assign a null page pointer for command SGL entries. Fixes: 9dd441e4ed575 ("soc: qcom: ipa: GSI transactions") Reported-by: Stephen Boyd <swboyd@chromium.org> Tested-by: Stephen Boyd <swboyd@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20201022010029.11877-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2020-10-22 09:00:29 +08:00
/* Commands are quite different from data transfer requests.
* Their payloads come from a pool whose memory is allocated
* using dma_alloc_coherent(). We therefore do *not* map them
* for DMA (unlike what we do for pages and skbs).
*
* When a transaction completes, the SGL is normally unmapped.
* A command transaction has direction DMA_NONE, which tells
* gsi_trans_complete() to skip the unmapping step.
*
* The only things we use directly in a command scatter/gather
* entry are the DMA address and length. We still need the SG
* table flags to be maintained though, so assign a NULL page
* pointer for that purpose.
*/
sg = &trans->sgl[which];
net: ipa: command payloads already mapped IPA transactions describe actions to be performed by the IPA hardware. Three cases use IPA transactions: transmitting a socket buffer; providing a page to receive packet data; and issuing an IPA immediate command. An IPA transaction contains a scatter/gather list (SGL) to hold the set of actions to be performed. We map buffers in the SGL for DMA at the time they are added to the transaction. For skb TX transactions, we fill the SGL with a call to skb_to_sgvec(). Page RX transactions involve a single page pointer, and that is recorded in the SGL with sg_set_page(). In both of these cases we then map the SGL for DMA with a call to dma_map_sg(). Immediate commands are different. The payload for an immediate command comes from a region of coherent DMA memory, which must *not* be mapped for DMA. For that reason, gsi_trans_cmd_add() sort of hand-crafts each SGL entry added to a command transaction. This patch fixes a problem with the code that crafts the SGL entry for an immediate command. Previously a portion of the SGL entry was updated using sg_set_buf(). However this is not valid because it includes a call to virt_to_page() on the buffer, but the command buffer pointer is not a linear address. Since we never actually map the SGL for command transactions, there are very few fields in the SGL we need to fill. Specifically, we only need to record the DMA address and the length, so they can be used by __gsi_trans_commit() to fill a TRE. We additionally need to preserve the SGL flags so for_each_sg() still works. For that we can simply assign a null page pointer for command SGL entries. Fixes: 9dd441e4ed575 ("soc: qcom: ipa: GSI transactions") Reported-by: Stephen Boyd <swboyd@chromium.org> Tested-by: Stephen Boyd <swboyd@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20201022010029.11877-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2020-10-22 09:00:29 +08:00
sg_assign_page(sg, NULL);
sg_dma_address(sg) = addr;
net: ipa: command payloads already mapped IPA transactions describe actions to be performed by the IPA hardware. Three cases use IPA transactions: transmitting a socket buffer; providing a page to receive packet data; and issuing an IPA immediate command. An IPA transaction contains a scatter/gather list (SGL) to hold the set of actions to be performed. We map buffers in the SGL for DMA at the time they are added to the transaction. For skb TX transactions, we fill the SGL with a call to skb_to_sgvec(). Page RX transactions involve a single page pointer, and that is recorded in the SGL with sg_set_page(). In both of these cases we then map the SGL for DMA with a call to dma_map_sg(). Immediate commands are different. The payload for an immediate command comes from a region of coherent DMA memory, which must *not* be mapped for DMA. For that reason, gsi_trans_cmd_add() sort of hand-crafts each SGL entry added to a command transaction. This patch fixes a problem with the code that crafts the SGL entry for an immediate command. Previously a portion of the SGL entry was updated using sg_set_buf(). However this is not valid because it includes a call to virt_to_page() on the buffer, but the command buffer pointer is not a linear address. Since we never actually map the SGL for command transactions, there are very few fields in the SGL we need to fill. Specifically, we only need to record the DMA address and the length, so they can be used by __gsi_trans_commit() to fill a TRE. We additionally need to preserve the SGL flags so for_each_sg() still works. For that we can simply assign a null page pointer for command SGL entries. Fixes: 9dd441e4ed575 ("soc: qcom: ipa: GSI transactions") Reported-by: Stephen Boyd <swboyd@chromium.org> Tested-by: Stephen Boyd <swboyd@chromium.org> Signed-off-by: Alex Elder <elder@linaro.org> Link: https://lore.kernel.org/r/20201022010029.11877-1-elder@linaro.org Signed-off-by: Jakub Kicinski <kuba@kernel.org>
2020-10-22 09:00:29 +08:00
sg_dma_len(sg) = size;
info = &trans->info[which];
info->opcode = opcode;
info->direction = direction;
}
/* Add a page transfer to a transaction. It will fill the only TRE. */
int gsi_trans_page_add(struct gsi_trans *trans, struct page *page, u32 size,
u32 offset)
{
struct scatterlist *sg = &trans->sgl[0];
int ret;
/* assert(trans->tre_count == 1); */
/* assert(!trans->used); */
sg_set_page(sg, page, size, offset);
ret = dma_map_sg(trans->gsi->dev, sg, 1, trans->direction);
if (!ret)
return -ENOMEM;
trans->used++; /* Transaction now owns the (DMA mapped) page */
return 0;
}
/* Add an SKB transfer to a transaction. No other TREs will be used. */
int gsi_trans_skb_add(struct gsi_trans *trans, struct sk_buff *skb)
{
struct scatterlist *sg = &trans->sgl[0];
u32 used;
int ret;
/* assert(trans->tre_count == 1); */
/* assert(!trans->used); */
/* skb->len will not be 0 (checked early) */
ret = skb_to_sgvec(skb, sg, 0, skb->len);
if (ret < 0)
return ret;
used = ret;
ret = dma_map_sg(trans->gsi->dev, sg, used, trans->direction);
if (!ret)
return -ENOMEM;
trans->used += used; /* Transaction now owns the (DMA mapped) skb */
return 0;
}
/* Compute the length/opcode value to use for a TRE */
static __le16 gsi_tre_len_opcode(enum ipa_cmd_opcode opcode, u32 len)
{
return opcode == IPA_CMD_NONE ? cpu_to_le16((u16)len)
: cpu_to_le16((u16)opcode);
}
/* Compute the flags value to use for a given TRE */
static __le32 gsi_tre_flags(bool last_tre, bool bei, enum ipa_cmd_opcode opcode)
{
enum gsi_tre_type tre_type;
u32 tre_flags;
tre_type = opcode == IPA_CMD_NONE ? GSI_RE_XFER : GSI_RE_IMMD_CMD;
tre_flags = u32_encode_bits(tre_type, TRE_FLAGS_TYPE_FMASK);
/* Last TRE contains interrupt flags */
if (last_tre) {
/* All transactions end in a transfer completion interrupt */
tre_flags |= TRE_FLAGS_IEOT_FMASK;
/* Don't interrupt when outbound commands are acknowledged */
if (bei)
tre_flags |= TRE_FLAGS_BEI_FMASK;
} else { /* All others indicate there's more to come */
tre_flags |= TRE_FLAGS_CHAIN_FMASK;
}
return cpu_to_le32(tre_flags);
}
static void gsi_trans_tre_fill(struct gsi_tre *dest_tre, dma_addr_t addr,
u32 len, bool last_tre, bool bei,
enum ipa_cmd_opcode opcode)
{
struct gsi_tre tre;
tre.addr = cpu_to_le64(addr);
tre.len_opcode = gsi_tre_len_opcode(opcode, len);
tre.reserved = 0;
tre.flags = gsi_tre_flags(last_tre, bei, opcode);
/* ARM64 can write 16 bytes as a unit with a single instruction.
* Doing the assignment this way is an attempt to make that happen.
*/
*dest_tre = tre;
}
/**
* __gsi_trans_commit() - Common GSI transaction commit code
* @trans: Transaction to commit
* @ring_db: Whether to tell the hardware about these queued transfers
*
* Formats channel ring TRE entries based on the content of the scatterlist.
* Maps a transaction pointer to the last ring entry used for the transaction,
* so it can be recovered when it completes. Moves the transaction to the
* pending list. Finally, updates the channel ring pointer and optionally
* rings the doorbell.
*/
static void __gsi_trans_commit(struct gsi_trans *trans, bool ring_db)
{
struct gsi_channel *channel = &trans->gsi->channel[trans->channel_id];
struct gsi_ring *ring = &channel->tre_ring;
enum ipa_cmd_opcode opcode = IPA_CMD_NONE;
bool bei = channel->toward_ipa;
struct ipa_cmd_info *info;
struct gsi_tre *dest_tre;
struct scatterlist *sg;
u32 byte_count = 0;
u32 avail;
u32 i;
/* assert(trans->used > 0); */
/* Consume the entries. If we cross the end of the ring while
* filling them we'll switch to the beginning to finish.
* If there is no info array we're doing a simple data
* transfer request, whose opcode is IPA_CMD_NONE.
*/
info = trans->info ? &trans->info[0] : NULL;
avail = ring->count - ring->index % ring->count;
dest_tre = gsi_ring_virt(ring, ring->index);
for_each_sg(trans->sgl, sg, trans->used, i) {
bool last_tre = i == trans->used - 1;
dma_addr_t addr = sg_dma_address(sg);
u32 len = sg_dma_len(sg);
byte_count += len;
if (!avail--)
dest_tre = gsi_ring_virt(ring, 0);
if (info)
opcode = info++->opcode;
gsi_trans_tre_fill(dest_tre, addr, len, last_tre, bei, opcode);
dest_tre++;
}
ring->index += trans->used;
if (channel->toward_ipa) {
/* We record TX bytes when they are sent */
trans->len = byte_count;
trans->trans_count = channel->trans_count;
trans->byte_count = channel->byte_count;
channel->trans_count++;
channel->byte_count += byte_count;
}
/* Associate the last TRE with the transaction */
gsi_channel_trans_map(channel, ring->index - 1, trans);
gsi_trans_move_pending(trans);
/* Ring doorbell if requested, or if all TREs are allocated */
if (ring_db || !atomic_read(&channel->trans_info.tre_avail)) {
/* Report what we're handing off to hardware for TX channels */
if (channel->toward_ipa)
gsi_channel_tx_queued(channel);
gsi_channel_doorbell(channel);
}
}
/* Commit a GSI transaction */
void gsi_trans_commit(struct gsi_trans *trans, bool ring_db)
{
if (trans->used)
__gsi_trans_commit(trans, ring_db);
else
gsi_trans_free(trans);
}
/* Commit a GSI transaction and wait for it to complete */
void gsi_trans_commit_wait(struct gsi_trans *trans)
{
if (!trans->used)
goto out_trans_free;
refcount_inc(&trans->refcount);
__gsi_trans_commit(trans, true);
wait_for_completion(&trans->completion);
out_trans_free:
gsi_trans_free(trans);
}
/* Commit a GSI transaction and wait for it to complete, with timeout */
int gsi_trans_commit_wait_timeout(struct gsi_trans *trans,
unsigned long timeout)
{
unsigned long timeout_jiffies = msecs_to_jiffies(timeout);
unsigned long remaining = 1; /* In case of empty transaction */
if (!trans->used)
goto out_trans_free;
refcount_inc(&trans->refcount);
__gsi_trans_commit(trans, true);
remaining = wait_for_completion_timeout(&trans->completion,
timeout_jiffies);
out_trans_free:
gsi_trans_free(trans);
return remaining ? 0 : -ETIMEDOUT;
}
/* Process the completion of a transaction; called while polling */
void gsi_trans_complete(struct gsi_trans *trans)
{
/* If the entire SGL was mapped when added, unmap it now */
if (trans->direction != DMA_NONE)
dma_unmap_sg(trans->gsi->dev, trans->sgl, trans->used,
trans->direction);
ipa_gsi_trans_complete(trans);
complete(&trans->completion);
gsi_trans_free(trans);
}
/* Cancel a channel's pending transactions */
void gsi_channel_trans_cancel_pending(struct gsi_channel *channel)
{
struct gsi_trans_info *trans_info = &channel->trans_info;
struct gsi_trans *trans;
bool cancelled;
/* channel->gsi->mutex is held by caller */
spin_lock_bh(&trans_info->spinlock);
cancelled = !list_empty(&trans_info->pending);
list_for_each_entry(trans, &trans_info->pending, links)
trans->cancelled = true;
list_splice_tail_init(&trans_info->pending, &trans_info->complete);
spin_unlock_bh(&trans_info->spinlock);
/* Schedule NAPI polling to complete the cancelled transactions */
if (cancelled)
napi_schedule(&channel->napi);
}
/* Issue a command to read a single byte from a channel */
int gsi_trans_read_byte(struct gsi *gsi, u32 channel_id, dma_addr_t addr)
{
struct gsi_channel *channel = &gsi->channel[channel_id];
struct gsi_ring *ring = &channel->tre_ring;
struct gsi_trans_info *trans_info;
struct gsi_tre *dest_tre;
trans_info = &channel->trans_info;
/* First reserve the TRE, if possible */
if (!gsi_trans_tre_reserve(trans_info, 1))
return -EBUSY;
/* Now fill the the reserved TRE and tell the hardware */
dest_tre = gsi_ring_virt(ring, ring->index);
gsi_trans_tre_fill(dest_tre, addr, 1, true, false, IPA_CMD_NONE);
ring->index++;
gsi_channel_doorbell(channel);
return 0;
}
/* Mark a gsi_trans_read_byte() request done */
void gsi_trans_read_byte_done(struct gsi *gsi, u32 channel_id)
{
struct gsi_channel *channel = &gsi->channel[channel_id];
gsi_trans_tre_release(&channel->trans_info, 1);
}
/* Initialize a channel's GSI transaction info */
int gsi_channel_trans_init(struct gsi *gsi, u32 channel_id)
{
struct gsi_channel *channel = &gsi->channel[channel_id];
struct gsi_trans_info *trans_info;
u32 tre_max;
int ret;
/* Ensure the size of a channel element is what's expected */
BUILD_BUG_ON(sizeof(struct gsi_tre) != GSI_RING_ELEMENT_SIZE);
/* The map array is used to determine what transaction is associated
* with a TRE that the hardware reports has completed. We need one
* map entry per TRE.
*/
trans_info = &channel->trans_info;
trans_info->map = kcalloc(channel->tre_count, sizeof(*trans_info->map),
GFP_KERNEL);
if (!trans_info->map)
return -ENOMEM;
/* We can't use more TREs than there are available in the ring.
* This limits the number of transactions that can be oustanding.
* Worst case is one TRE per transaction (but we actually limit
* it to something a little less than that). We allocate resources
* for transactions (including transaction structures) based on
* this maximum number.
*/
tre_max = gsi_channel_tre_max(channel->gsi, channel_id);
/* Transactions are allocated one at a time. */
ret = gsi_trans_pool_init(&trans_info->pool, sizeof(struct gsi_trans),
tre_max, 1);
if (ret)
goto err_kfree;
/* A transaction uses a scatterlist array to represent the data
* transfers implemented by the transaction. Each scatterlist
* element is used to fill a single TRE when the transaction is
* committed. So we need as many scatterlist elements as the
* maximum number of TREs that can be outstanding.
*
* All TREs in a transaction must fit within the channel's TLV FIFO.
* A transaction on a channel can allocate as many TREs as that but
* no more.
*/
ret = gsi_trans_pool_init(&trans_info->sg_pool,
sizeof(struct scatterlist),
tre_max, channel->tlv_count);
if (ret)
goto err_trans_pool_exit;
/* Finally, the tre_avail field is what ultimately limits the number
* of outstanding transactions and their resources. A transaction
* allocation succeeds only if the TREs available are sufficient for
* what the transaction might need. Transaction resource pools are
* sized based on the maximum number of outstanding TREs, so there
* will always be resources available if there are TREs available.
*/
atomic_set(&trans_info->tre_avail, tre_max);
spin_lock_init(&trans_info->spinlock);
INIT_LIST_HEAD(&trans_info->alloc);
INIT_LIST_HEAD(&trans_info->pending);
INIT_LIST_HEAD(&trans_info->complete);
INIT_LIST_HEAD(&trans_info->polled);
return 0;
err_trans_pool_exit:
gsi_trans_pool_exit(&trans_info->pool);
err_kfree:
kfree(trans_info->map);
dev_err(gsi->dev, "error %d initializing channel %u transactions\n",
ret, channel_id);
return ret;
}
/* Inverse of gsi_channel_trans_init() */
void gsi_channel_trans_exit(struct gsi_channel *channel)
{
struct gsi_trans_info *trans_info = &channel->trans_info;
gsi_trans_pool_exit(&trans_info->sg_pool);
gsi_trans_pool_exit(&trans_info->pool);
kfree(trans_info->map);
}