Merge branch 'linus' into release
[firefly-linux-kernel-4.4.55.git] / drivers / firewire / fw-iso.c
index e14c03dc006588d1b98d0b10ded8bd210d8fdb8d..2baf1007253e66a972b357d37b62b1e245d26bcd 100644 (file)
@@ -1,5 +1,7 @@
 /*
- * Isochronous IO functionality
+ * Isochronous I/O functionality:
+ *   - Isochronous DMA context management
+ *   - Isochronous bus resource management (channels, bandwidth), client side
  *
  * Copyright (C) 2006 Kristian Hoegsberg <krh@bitplanet.net>
  *
  * Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
  */
 
-#include <linux/kernel.h>
-#include <linux/module.h>
 #include <linux/dma-mapping.h>
-#include <linux/vmalloc.h>
+#include <linux/errno.h>
+#include <linux/firewire-constants.h>
+#include <linux/kernel.h>
 #include <linux/mm.h>
+#include <linux/spinlock.h>
+#include <linux/vmalloc.h>
 
-#include "fw-transaction.h"
 #include "fw-topology.h"
-#include "fw-device.h"
+#include "fw-transaction.h"
 
-int
-fw_iso_buffer_init(struct fw_iso_buffer *buffer, struct fw_card *card,
-                  int page_count, enum dma_data_direction direction)
+/*
+ * Isochronous DMA context management
+ */
+
+int fw_iso_buffer_init(struct fw_iso_buffer *buffer, struct fw_card *card,
+                      int page_count, enum dma_data_direction direction)
 {
-       int i, j, retval = -ENOMEM;
+       int i, j;
        dma_addr_t address;
 
        buffer->page_count = page_count;
@@ -69,19 +75,21 @@ fw_iso_buffer_init(struct fw_iso_buffer *buffer, struct fw_card *card,
        kfree(buffer->pages);
  out:
        buffer->pages = NULL;
-       return retval;
+
+       return -ENOMEM;
 }
 
 int fw_iso_buffer_map(struct fw_iso_buffer *buffer, struct vm_area_struct *vma)
 {
        unsigned long uaddr;
-       int i, retval;
+       int i, err;
 
        uaddr = vma->vm_start;
        for (i = 0; i < buffer->page_count; i++) {
-               retval = vm_insert_page(vma, uaddr, buffer->pages[i]);
-               if (retval)
-                       return retval;
+               err = vm_insert_page(vma, uaddr, buffer->pages[i]);
+               if (err)
+                       return err;
+
                uaddr += PAGE_SIZE;
        }
 
@@ -105,14 +113,14 @@ void fw_iso_buffer_destroy(struct fw_iso_buffer *buffer,
        buffer->pages = NULL;
 }
 
-struct fw_iso_context *
-fw_iso_context_create(struct fw_card *card, int type,
-                     int channel, int speed, size_t header_size,
-                     fw_iso_callback_t callback, void *callback_data)
+struct fw_iso_context *fw_iso_context_create(struct fw_card *card,
+               int type, int channel, int speed, size_t header_size,
+               fw_iso_callback_t callback, void *callback_data)
 {
        struct fw_iso_context *ctx;
 
-       ctx = card->driver->allocate_iso_context(card, type, header_size);
+       ctx = card->driver->allocate_iso_context(card,
+                                                type, channel, header_size);
        if (IS_ERR(ctx))
                return ctx;
 
@@ -134,25 +142,186 @@ void fw_iso_context_destroy(struct fw_iso_context *ctx)
        card->driver->free_iso_context(ctx);
 }
 
-int
-fw_iso_context_start(struct fw_iso_context *ctx, int cycle, int sync, int tags)
+int fw_iso_context_start(struct fw_iso_context *ctx,
+                        int cycle, int sync, int tags)
 {
        return ctx->card->driver->start_iso(ctx, cycle, sync, tags);
 }
 
-int
-fw_iso_context_queue(struct fw_iso_context *ctx,
-                    struct fw_iso_packet *packet,
-                    struct fw_iso_buffer *buffer,
-                    unsigned long payload)
+int fw_iso_context_queue(struct fw_iso_context *ctx,
+                        struct fw_iso_packet *packet,
+                        struct fw_iso_buffer *buffer,
+                        unsigned long payload)
 {
        struct fw_card *card = ctx->card;
 
        return card->driver->queue_iso(ctx, packet, buffer, payload);
 }
 
-int
-fw_iso_context_stop(struct fw_iso_context *ctx)
+int fw_iso_context_stop(struct fw_iso_context *ctx)
 {
        return ctx->card->driver->stop_iso(ctx);
 }
+
+/*
+ * Isochronous bus resource management (channels, bandwidth), client side
+ */
+
+static int manage_bandwidth(struct fw_card *card, int irm_id, int generation,
+                           int bandwidth, bool allocate)
+{
+       __be32 data[2];
+       int try, new, old = allocate ? BANDWIDTH_AVAILABLE_INITIAL : 0;
+
+       /*
+        * On a 1394a IRM with low contention, try < 1 is enough.
+        * On a 1394-1995 IRM, we need at least try < 2.
+        * Let's just do try < 5.
+        */
+       for (try = 0; try < 5; try++) {
+               new = allocate ? old - bandwidth : old + bandwidth;
+               if (new < 0 || new > BANDWIDTH_AVAILABLE_INITIAL)
+                       break;
+
+               data[0] = cpu_to_be32(old);
+               data[1] = cpu_to_be32(new);
+               switch (fw_run_transaction(card, TCODE_LOCK_COMPARE_SWAP,
+                               irm_id, generation, SCODE_100,
+                               CSR_REGISTER_BASE + CSR_BANDWIDTH_AVAILABLE,
+                               data, sizeof(data))) {
+               case RCODE_GENERATION:
+                       /* A generation change frees all bandwidth. */
+                       return allocate ? -EAGAIN : bandwidth;
+
+               case RCODE_COMPLETE:
+                       if (be32_to_cpup(data) == old)
+                               return bandwidth;
+
+                       old = be32_to_cpup(data);
+                       /* Fall through. */
+               }
+       }
+
+       return -EIO;
+}
+
+static int manage_channel(struct fw_card *card, int irm_id, int generation,
+                         u32 channels_mask, u64 offset, bool allocate)
+{
+       __be32 data[2], c, all, old;
+       int i, retry = 5;
+
+       old = all = allocate ? cpu_to_be32(~0) : 0;
+
+       for (i = 0; i < 32; i++) {
+               if (!(channels_mask & 1 << i))
+                       continue;
+
+               c = cpu_to_be32(1 << (31 - i));
+               if ((old & c) != (all & c))
+                       continue;
+
+               data[0] = old;
+               data[1] = old ^ c;
+               switch (fw_run_transaction(card, TCODE_LOCK_COMPARE_SWAP,
+                                          irm_id, generation, SCODE_100,
+                                          offset, data, sizeof(data))) {
+               case RCODE_GENERATION:
+                       /* A generation change frees all channels. */
+                       return allocate ? -EAGAIN : i;
+
+               case RCODE_COMPLETE:
+                       if (data[0] == old)
+                               return i;
+
+                       old = data[0];
+
+                       /* Is the IRM 1394a-2000 compliant? */
+                       if ((data[0] & c) == (data[1] & c))
+                               continue;
+
+                       /* 1394-1995 IRM, fall through to retry. */
+               default:
+                       if (retry--)
+                               i--;
+               }
+       }
+
+       return -EIO;
+}
+
+static void deallocate_channel(struct fw_card *card, int irm_id,
+                              int generation, int channel)
+{
+       u32 mask;
+       u64 offset;
+
+       mask = channel < 32 ? 1 << channel : 1 << (channel - 32);
+       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);
+}
+
+/**
+ * fw_iso_resource_manage - Allocate or deallocate a channel and/or bandwidth
+ *
+ * In parameters: card, generation, channels_mask, bandwidth, allocate
+ * Out parameters: channel, bandwidth
+ * This function blocks (sleeps) during communication with the IRM.
+ *
+ * Allocates or deallocates at most one channel out of channels_mask.
+ * channels_mask is a bitfield with MSB for channel 63 and LSB for channel 0.
+ * (Note, the IRM's CHANNELS_AVAILABLE is a big-endian bitfield with MSB for
+ * channel 0 and LSB for channel 63.)
+ * Allocates or deallocates as many bandwidth allocation units as specified.
+ *
+ * Returns channel < 0 if no channel was allocated or deallocated.
+ * Returns bandwidth = 0 if no bandwidth was allocated or deallocated.
+ *
+ * If generation is stale, deallocations succeed but allocations fail with
+ * channel = -EAGAIN.
+ *
+ * If channel allocation fails, no bandwidth will be allocated either.
+ * If bandwidth allocation fails, no channel will be allocated either.
+ * But deallocations of channel and bandwidth are tried independently
+ * of each other's success.
+ */
+void fw_iso_resource_manage(struct fw_card *card, int generation,
+                           u64 channels_mask, int *channel, int *bandwidth,
+                           bool allocate)
+{
+       u32 channels_hi = channels_mask;        /* channels 31...0 */
+       u32 channels_lo = channels_mask >> 32;  /* channels 63...32 */
+       int irm_id, ret, c = -EINVAL;
+
+       spin_lock_irq(&card->lock);
+       irm_id = card->irm_node->node_id;
+       spin_unlock_irq(&card->lock);
+
+       if (channels_hi)
+               c = manage_channel(card, irm_id, generation, channels_hi,
+                   CSR_REGISTER_BASE + CSR_CHANNELS_AVAILABLE_HI, allocate);
+       if (channels_lo && c < 0) {
+               c = manage_channel(card, irm_id, generation, channels_lo,
+                   CSR_REGISTER_BASE + CSR_CHANNELS_AVAILABLE_LO, allocate);
+               if (c >= 0)
+                       c += 32;
+       }
+       *channel = c;
+
+       if (allocate && channels_mask != 0 && c < 0)
+               *bandwidth = 0;
+
+       if (*bandwidth == 0)
+               return;
+
+       ret = manage_bandwidth(card, irm_id, generation, *bandwidth, allocate);
+       if (ret < 0)
+               *bandwidth = 0;
+
+       if (allocate && ret < 0 && c >= 0) {
+               deallocate_channel(card, irm_id, generation, c);
+               *channel = ret;
+       }
+}