firewire: octlet AT payloads can be stack-allocated

We do not need slab allocations anymore in order to satisfy
streaming DMA mapping constraints, thanks to commit da28947e7e36
"firewire: ohci: avoid separate DMA mapping for small AT payloads".

(Besides, the slab-allocated buffers that firewire-core, firewire-sbp2,
and firedtv used to provide for 8-byte write and lock requests were
still not fully portable since they crossed cacheline boundaries or
shared a cacheline with unrelated CPU-accessed data.  snd-firewire-lib
got this aspect right by using an extra kmalloc/ kfree just for the
8-byte transaction buffer.)

This change replaces kmalloc'ed lock transaction scratch buffers in
firewire-core, firedtv, and snd-firewire-lib by local stack allocations.
Perhaps the most notable result of the change is simpler locking because
there is no need to serialize usages of preallocated per-device buffers
anymore.  Also, allocations and deallocations are simpler.

Signed-off-by: Stefan Richter <stefanr@s5r6.in-berlin.de>
Acked-by: Clemens Ladisch <clemens@ladisch.de>

4 files changed, 24 insertions, 24 deletions

diff --git a/drivers/firewire/core-card.c b/drivers/firewire/core-card.c
index 3c44fbc81ac..e119f1e6ba4 100644
--- a/drivers/firewire/core-card.c
+++ b/drivers/firewire/core-card.c
@@ -258,8 +258,7 @@ static void allocate_broadcast_channel(struct fw_card *card, int generation)
 
         if (!card->broadcast_channel_allocated) {
                 fw_iso_resource_manage(card, generation, 1ULL << 31,
-                                       &channel, &bandwidth, true,
-                                       card->bm_transaction_data);
+                                       &channel, &bandwidth, true);
                 if (channel != 31) {
                         fw_notify("failed to allocate broadcast channel\n");
                         return;
@@ -294,6 +293,7 @@ static void bm_work(struct work_struct *work)
         bool root_device_is_cmc;
         bool irm_is_1394_1995_only;
         bool keep_this_irm;
+        __be32 transaction_data[2];
 
         spin_lock_irq(&card->lock);
 
@@ -355,21 +355,21 @@ static void bm_work(struct work_struct *work)
                         goto pick_me;
                 }
 
-                card->bm_transaction_data[0] = cpu_to_be32(0x3f);
-                card->bm_transaction_data[1] = cpu_to_be32(local_id);
+                transaction_data[0] = cpu_to_be32(0x3f);
+                transaction_data[1] = cpu_to_be32(local_id);
 
                 spin_unlock_irq(&card->lock);
 
                 rcode = fw_run_transaction(card, TCODE_LOCK_COMPARE_SWAP,
                                 irm_id, generation, SCODE_100,
                                 CSR_REGISTER_BASE + CSR_BUS_MANAGER_ID,
-                                card->bm_transaction_data, 8);
+                                transaction_data, 8);
 
                 if (rcode == RCODE_GENERATION)
                         /* Another bus reset, BM work has been rescheduled. */
                         goto out;
 
-                bm_id = be32_to_cpu(card->bm_transaction_data[0]);
+                bm_id = be32_to_cpu(transaction_data[0]);
 
                 spin_lock_irq(&card->lock);
                 if (rcode == RCODE_COMPLETE && generation == card->generation)
@@ -490,11 +490,11 @@ static void bm_work(struct work_struct *work)
                 /*
                  * Make sure that the cycle master sends cycle start packets.
                  */
-                card->bm_transaction_data[0] = cpu_to_be32(CSR_STATE_BIT_CMSTR);
+                transaction_data[0] = cpu_to_be32(CSR_STATE_BIT_CMSTR);
                 rcode = fw_run_transaction(card, TCODE_WRITE_QUADLET_REQUEST,
                                 root_id, generation, SCODE_100,
                                 CSR_REGISTER_BASE + CSR_STATE_SET,
-                                card->bm_transaction_data, 4);
+                                transaction_data, 4);
                 if (rcode == RCODE_GENERATION)
                         goto out;
         }
diff --git a/drivers/firewire/core-cdev.c b/drivers/firewire/core-cdev.c
index 62ac111af24..2a3f1c4d690 100644
--- a/drivers/firewire/core-cdev.c
+++ b/drivers/firewire/core-cdev.c
@@ -141,7 +141,6 @@ struct iso_resource {
         int generation;
         u64 channels;
         s32 bandwidth;
-        __be32 transaction_data[2];
         struct iso_resource_event *e_alloc, *e_dealloc;
 };
 
@@ -1229,8 +1228,7 @@ static void iso_resource_work(struct work_struct *work)
                         r->channels, &channel, &bandwidth,
                         todo == ISO_RES_ALLOC ||
                         todo == ISO_RES_REALLOC ||
-                        todo == ISO_RES_ALLOC_ONCE,
-                        r->transaction_data);
+                        todo == ISO_RES_ALLOC_ONCE);
         /*
          * Is this generation outdated already?  As long as this resource sticks
          * in the idr, it will be scheduled again for a newer generation or at
diff --git a/drivers/firewire/core-iso.c b/drivers/firewire/core-iso.c
index 481056df926..f872ede5af3 100644
--- a/drivers/firewire/core-iso.c
+++ b/drivers/firewire/core-iso.c
@@ -196,9 +196,10 @@ EXPORT_SYMBOL(fw_iso_context_stop);
  */
 
 static int manage_bandwidth(struct fw_card *card, int irm_id, int generation,
-                            int bandwidth, bool allocate, __be32 data[2])
+                            int bandwidth, bool allocate)
 {
         int try, new, old = allocate ? BANDWIDTH_AVAILABLE_INITIAL : 0;
+        __be32 data[2];
 
         /*
          * On a 1394a IRM with low contention, try < 1 is enough.
@@ -233,9 +234,10 @@ static int manage_bandwidth(struct fw_card *card, int irm_id, int generation,
 }
 
 static int manage_channel(struct fw_card *card, int irm_id, int generation,
-                u32 channels_mask, u64 offset, bool allocate, __be32 data[2])
+                u32 channels_mask, u64 offset, bool allocate)
 {
         __be32 bit, all, old;
+        __be32 data[2];
         int channel, ret = -EIO, retry = 5;
 
         old = all = allocate ? cpu_to_be32(~0) : 0;
@@ -284,7 +286,7 @@ static int manage_channel(struct fw_card *card, int irm_id, int generation,
 }
 
 static void deallocate_channel(struct fw_card *card, int irm_id,
-                               int generation, int channel, __be32 buffer[2])
+                               int generation, int channel)
 {
         u32 mask;
         u64 offset;
@@ -293,7 +295,7 @@ static void deallocate_channel(struct fw_card *card, int irm_id,
         offset = channel < 32 ? CSR_REGISTER_BASE + CSR_CHANNELS_AVAILABLE_HI :
                                 CSR_REGISTER_BASE + CSR_CHANNELS_AVAILABLE_LO;
 
-        manage_channel(card, irm_id, generation, mask, offset, false, buffer);
+        manage_channel(card, irm_id, generation, mask, offset, false);
 }
 
 /**
@@ -322,7 +324,7 @@ static void deallocate_channel(struct fw_card *card, int irm_id,
  */
 void fw_iso_resource_manage(struct fw_card *card, int generation,
                             u64 channels_mask, int *channel, int *bandwidth,
-                            bool allocate, __be32 buffer[2])
+                            bool allocate)
 {
         u32 channels_hi = channels_mask;        /* channels 31...0 */
         u32 channels_lo = channels_mask >> 32;  /* channels 63...32 */
@@ -335,11 +337,11 @@ void fw_iso_resource_manage(struct fw_card *card, int generation,
         if (channels_hi)
                 c = manage_channel(card, irm_id, generation, channels_hi,
                                 CSR_REGISTER_BASE + CSR_CHANNELS_AVAILABLE_HI,
-                                allocate, buffer);
+                                allocate);
         if (channels_lo && c < 0) {
                 c = manage_channel(card, irm_id, generation, channels_lo,
                                 CSR_REGISTER_BASE + CSR_CHANNELS_AVAILABLE_LO,
-                                allocate, buffer);
+                                allocate);
                 if (c >= 0)
                         c += 32;
         }
@@ -351,14 +353,13 @@ void fw_iso_resource_manage(struct fw_card *card, int generation,
         if (*bandwidth == 0)
                 return;
 
-        ret = manage_bandwidth(card, irm_id, generation, *bandwidth,
-                               allocate, buffer);
+        ret = manage_bandwidth(card, irm_id, generation, *bandwidth, allocate);
         if (ret < 0)
                 *bandwidth = 0;
 
         if (allocate && ret < 0) {
                 if (c >= 0)
-                        deallocate_channel(card, irm_id, generation, c, buffer);
+                        deallocate_channel(card, irm_id, generation, c);
                 *channel = ret;
         }
 }
diff --git a/drivers/firewire/core-transaction.c b/drivers/firewire/core-transaction.c
index d00f8ce902c..77275fdf6c1 100644
--- a/drivers/firewire/core-transaction.c
+++ b/drivers/firewire/core-transaction.c
@@ -326,8 +326,8 @@ static int allocate_tlabel(struct fw_card *card)
  * It will contain tag, channel, and sy data instead of a node ID then.
  *
  * The payload buffer at @data is going to be DMA-mapped except in case of
- * quadlet-sized payload or of local (loopback) requests.  Hence make sure that
- * the buffer complies with the restrictions for DMA-mapped memory.  The
+ * @length <= 8 or of local (loopback) requests.  Hence make sure that the
+ * buffer complies with the restrictions of the streaming DMA mapping API.
  * @payload must not be freed before the @callback is called.
  *
  * In case of request types without payload, @data is NULL and @length is 0.
@@ -411,7 +411,8 @@ static void transaction_callback(struct fw_card *card, int rcode,
  *
  * Returns the RCODE.  See fw_send_request() for parameter documentation.
  * Unlike fw_send_request(), @data points to the payload of the request or/and
- * to the payload of the response.
+ * to the payload of the response.  DMA mapping restrictions apply to outbound
+ * request payloads of >= 8 bytes but not to inbound response payloads.
  */
 int fw_run_transaction(struct fw_card *card, int tcode, int destination_id,
                        int generation, int speed, unsigned long long offset,