NVMe: Simplify completion handling
authorMatthew Wilcox <matthew.r.wilcox@intel.com>
Sat, 15 Oct 2011 11:33:46 +0000 (07:33 -0400)
committerMatthew Wilcox <matthew.r.wilcox@intel.com>
Tue, 10 Jan 2012 19:47:46 +0000 (14:47 -0500)
Instead of encoding the handler type in the bottom two bits of the
per-completion context pointer, store the handler function as well
as the context pointer.  This gives us more flexibility and the code
is clearer.  It comes at the cost of an extra 8k of memory per queue,
but this feels like a reasonable price to pay.

Signed-off-by: Matthew Wilcox <matthew.r.wilcox@intel.com>
drivers/block/nvme.c

index a17f80f..4724655 100644 (file)
@@ -135,8 +135,12 @@ static inline void _nvme_check_size(void)
        BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64);
 }
 
+typedef void (*nvme_completion_fn)(struct nvme_queue *, void *,
+                                               struct nvme_completion *);
+
 struct nvme_cmd_info {
-       unsigned long ctx;
+       nvme_completion_fn fn;
+       void *ctx;
        unsigned long timeout;
 };
 
@@ -149,7 +153,7 @@ static struct nvme_cmd_info *nvme_cmd_info(struct nvme_queue *nvmeq)
  * alloc_cmdid() - Allocate a Command ID
  * @nvmeq: The queue that will be used for this command
  * @ctx: A pointer that will be passed to the handler
- * @handler: The ID of the handler to call
+ * @handler: The function to call on completion
  *
  * Allocate a Command ID for a queue.  The data passed in will
  * be passed to the completion handler.  This is implemented by using
@@ -160,28 +164,27 @@ static struct nvme_cmd_info *nvme_cmd_info(struct nvme_queue *nvmeq)
  * May be called with local interrupts disabled and the q_lock held,
  * or with interrupts enabled and no locks held.
  */
-static int alloc_cmdid(struct nvme_queue *nvmeq, void *ctx, int handler,
-                                                       unsigned timeout)
+static int alloc_cmdid(struct nvme_queue *nvmeq, void *ctx,
+                               nvme_completion_fn handler, unsigned timeout)
 {
        int depth = nvmeq->q_depth - 1;
        struct nvme_cmd_info *info = nvme_cmd_info(nvmeq);
        int cmdid;
 
-       BUG_ON((unsigned long)ctx & 3);
-
        do {
                cmdid = find_first_zero_bit(nvmeq->cmdid_data, depth);
                if (cmdid >= depth)
                        return -EBUSY;
        } while (test_and_set_bit(cmdid, nvmeq->cmdid_data));
 
-       info[cmdid].ctx = (unsigned long)ctx | handler;
+       info[cmdid].fn = handler;
+       info[cmdid].ctx = ctx;
        info[cmdid].timeout = jiffies + timeout;
        return cmdid;
 }
 
 static int alloc_cmdid_killable(struct nvme_queue *nvmeq, void *ctx,
-                                               int handler, unsigned timeout)
+                               nvme_completion_fn handler, unsigned timeout)
 {
        int cmdid;
        wait_event_killable(nvmeq->sq_full,
@@ -189,47 +192,69 @@ static int alloc_cmdid_killable(struct nvme_queue *nvmeq, void *ctx,
        return (cmdid < 0) ? -EINTR : cmdid;
 }
 
-/*
- * If you need more than four handlers, you'll need to change how
- * alloc_cmdid and nvme_process_cq work.  Consider using a special
- * CMD_CTX value instead, if that works for your situation.
- */
-enum {
-       sync_completion_id = 0,
-       bio_completion_id,
-};
-
-/* Special values must be a multiple of 4, and less than 0x1000 */
-#define CMD_CTX_BASE           (POISON_POINTER_DELTA + sync_completion_id)
+/* Special values must be less than 0x1000 */
+#define CMD_CTX_BASE           ((void *)POISON_POINTER_DELTA)
 #define CMD_CTX_CANCELLED      (0x30C + CMD_CTX_BASE)
 #define CMD_CTX_COMPLETED      (0x310 + CMD_CTX_BASE)
 #define CMD_CTX_INVALID                (0x314 + CMD_CTX_BASE)
 #define CMD_CTX_FLUSH          (0x318 + CMD_CTX_BASE)
 
+static void special_completion(struct nvme_queue *nvmeq, void *ctx,
+                                               struct nvme_completion *cqe)
+{
+       if (ctx == CMD_CTX_CANCELLED)
+               return;
+       if (ctx == CMD_CTX_FLUSH)
+               return;
+       if (ctx == CMD_CTX_COMPLETED) {
+               dev_warn(nvmeq->q_dmadev,
+                               "completed id %d twice on queue %d\n",
+                               cqe->command_id, le16_to_cpup(&cqe->sq_id));
+               return;
+       }
+       if (ctx == CMD_CTX_INVALID) {
+               dev_warn(nvmeq->q_dmadev,
+                               "invalid id %d completed on queue %d\n",
+                               cqe->command_id, le16_to_cpup(&cqe->sq_id));
+               return;
+       }
+
+       dev_warn(nvmeq->q_dmadev, "Unknown special completion %p\n", ctx);
+}
+
 /*
  * Called with local interrupts disabled and the q_lock held.  May not sleep.
  */
-static unsigned long free_cmdid(struct nvme_queue *nvmeq, int cmdid)
+static void *free_cmdid(struct nvme_queue *nvmeq, int cmdid,
+                                               nvme_completion_fn *fn)
 {
-       unsigned long data;
+       void *ctx;
        struct nvme_cmd_info *info = nvme_cmd_info(nvmeq);
 
-       if (cmdid >= nvmeq->q_depth)
+       if (cmdid >= nvmeq->q_depth) {
+               *fn = special_completion;
                return CMD_CTX_INVALID;
-       data = info[cmdid].ctx;
+       }
+       *fn = info[cmdid].fn;
+       ctx = info[cmdid].ctx;
+       info[cmdid].fn = special_completion;
        info[cmdid].ctx = CMD_CTX_COMPLETED;
        clear_bit(cmdid, nvmeq->cmdid_data);
        wake_up(&nvmeq->sq_full);
-       return data;
+       return ctx;
 }
 
-static unsigned long cancel_cmdid(struct nvme_queue *nvmeq, int cmdid)
+static void *cancel_cmdid(struct nvme_queue *nvmeq, int cmdid,
+                                               nvme_completion_fn *fn)
 {
-       unsigned long data;
+       void *ctx;
        struct nvme_cmd_info *info = nvme_cmd_info(nvmeq);
-       data = info[cmdid].ctx;
+       if (fn)
+               *fn = info[cmdid].fn;
+       ctx = info[cmdid].ctx;
+       info[cmdid].fn = special_completion;
        info[cmdid].ctx = CMD_CTX_CANCELLED;
-       return data;
+       return ctx;
 }
 
 static struct nvme_queue *get_nvmeq(struct nvme_ns *ns)
@@ -485,7 +510,7 @@ static int nvme_submit_flush(struct nvme_queue *nvmeq, struct nvme_ns *ns,
 static int nvme_submit_flush_data(struct nvme_queue *nvmeq, struct nvme_ns *ns)
 {
        int cmdid = alloc_cmdid(nvmeq, (void *)CMD_CTX_FLUSH,
-                                               sync_completion_id, IO_TIMEOUT);
+                                               special_completion, IO_TIMEOUT);
        if (unlikely(cmdid < 0))
                return cmdid;
 
@@ -518,7 +543,7 @@ static int nvme_submit_bio_queue(struct nvme_queue *nvmeq, struct nvme_ns *ns,
        nbio->bio = bio;
 
        result = -EBUSY;
-       cmdid = alloc_cmdid(nvmeq, nbio, bio_completion_id, IO_TIMEOUT);
+       cmdid = alloc_cmdid(nvmeq, nbio, bio_completion, IO_TIMEOUT);
        if (unlikely(cmdid < 0))
                goto free_nbio;
 
@@ -599,45 +624,6 @@ static int nvme_make_request(struct request_queue *q, struct bio *bio)
        return 0;
 }
 
-struct sync_cmd_info {
-       struct task_struct *task;
-       u32 result;
-       int status;
-};
-
-static void sync_completion(struct nvme_queue *nvmeq, void *ctx,
-                                               struct nvme_completion *cqe)
-{
-       struct sync_cmd_info *cmdinfo = ctx;
-       if (unlikely((unsigned long)cmdinfo == CMD_CTX_CANCELLED))
-               return;
-       if ((unsigned long)cmdinfo == CMD_CTX_FLUSH)
-               return;
-       if (unlikely((unsigned long)cmdinfo == CMD_CTX_COMPLETED)) {
-               dev_warn(nvmeq->q_dmadev,
-                               "completed id %d twice on queue %d\n",
-                               cqe->command_id, le16_to_cpup(&cqe->sq_id));
-               return;
-       }
-       if (unlikely((unsigned long)cmdinfo == CMD_CTX_INVALID)) {
-               dev_warn(nvmeq->q_dmadev,
-                               "invalid id %d completed on queue %d\n",
-                               cqe->command_id, le16_to_cpup(&cqe->sq_id));
-               return;
-       }
-       cmdinfo->result = le32_to_cpup(&cqe->result);
-       cmdinfo->status = le16_to_cpup(&cqe->status) >> 1;
-       wake_up_process(cmdinfo->task);
-}
-
-typedef void (*completion_fn)(struct nvme_queue *, void *,
-                                               struct nvme_completion *);
-
-static const completion_fn nvme_completions[4] = {
-       [sync_completion_id] = sync_completion,
-       [bio_completion_id]  = bio_completion,
-};
-
 static irqreturn_t nvme_process_cq(struct nvme_queue *nvmeq)
 {
        u16 head, phase;
@@ -646,9 +632,8 @@ static irqreturn_t nvme_process_cq(struct nvme_queue *nvmeq)
        phase = nvmeq->cq_phase;
 
        for (;;) {
-               unsigned long data;
-               void *ptr;
-               unsigned char handler;
+               void *ctx;
+               nvme_completion_fn fn;
                struct nvme_completion cqe = nvmeq->cqes[head];
                if ((le16_to_cpu(cqe.status) & 1) != phase)
                        break;
@@ -658,10 +643,8 @@ static irqreturn_t nvme_process_cq(struct nvme_queue *nvmeq)
                        phase = !phase;
                }
 
-               data = free_cmdid(nvmeq, cqe.command_id);
-               handler = data & 3;
-               ptr = (void *)(data & ~3UL);
-               nvme_completions[handler](nvmeq, ptr, &cqe);
+               ctx = free_cmdid(nvmeq, cqe.command_id, &fn);
+               fn(nvmeq, ctx, &cqe);
        }
 
        /* If the controller ignores the cq head doorbell and continuously
@@ -702,10 +685,25 @@ static irqreturn_t nvme_irq_check(int irq, void *data)
 static void nvme_abort_command(struct nvme_queue *nvmeq, int cmdid)
 {
        spin_lock_irq(&nvmeq->q_lock);
-       cancel_cmdid(nvmeq, cmdid);
+       cancel_cmdid(nvmeq, cmdid, NULL);
        spin_unlock_irq(&nvmeq->q_lock);
 }
 
+struct sync_cmd_info {
+       struct task_struct *task;
+       u32 result;
+       int status;
+};
+
+static void sync_completion(struct nvme_queue *nvmeq, void *ctx,
+                                               struct nvme_completion *cqe)
+{
+       struct sync_cmd_info *cmdinfo = ctx;
+       cmdinfo->result = le32_to_cpup(&cqe->result);
+       cmdinfo->status = le16_to_cpup(&cqe->status) >> 1;
+       wake_up_process(cmdinfo->task);
+}
+
 /*
  * Returns 0 on success.  If the result is negative, it's a Linux error code;
  * if the result is positive, it's an NVM Express status code
@@ -719,7 +717,7 @@ static int nvme_submit_sync_cmd(struct nvme_queue *nvmeq,
        cmdinfo.task = current;
        cmdinfo.status = -EINTR;
 
-       cmdid = alloc_cmdid_killable(nvmeq, &cmdinfo, sync_completion_id,
+       cmdid = alloc_cmdid_killable(nvmeq, &cmdinfo, sync_completion,
                                                                timeout);
        if (cmdid < 0)
                return cmdid;
@@ -1201,18 +1199,15 @@ static void nvme_timeout_ios(struct nvme_queue *nvmeq)
        int cmdid;
 
        for_each_set_bit(cmdid, nvmeq->cmdid_data, depth) {
-               unsigned long data;
-               void *ptr;
-               unsigned char handler;
+               void *ctx;
+               nvme_completion_fn fn;
                static struct nvme_completion cqe = { .status = cpu_to_le16(NVME_SC_ABORT_REQ) << 1, };
 
                if (!time_after(now, info[cmdid].timeout))
                        continue;
                dev_warn(nvmeq->q_dmadev, "Timing out I/O %d\n", cmdid);
-               data = cancel_cmdid(nvmeq, cmdid);
-               handler = data & 3;
-               ptr = (void *)(data & ~3UL);
-               nvme_completions[handler](nvmeq, ptr, &cqe);
+               ctx = cancel_cmdid(nvmeq, cmdid, &fn);
+               fn(nvmeq, ctx, &cqe);
        }
 }