[XEN PATCH v9 21/24] xen/arm: ffa: improve lock granularity

[Jens Wiklander <jens.wiklander@xxxxxxxxxx>]

The single lock in struct ffa_ctx is complemented with rx_lock and tx_lock.

The old lock is used for small critical sections, like increasing
shm_count or adding another shm to shm_list.

rx_lock and tx_lock are only acquired using spin_trylock() which for
well-behaving guests should always succeed. Guests using the RX and TX
buffers are expected to serialize accesses before doing the FF-A
request.

Signed-off-by: Jens Wiklander <jens.wiklander@xxxxxxxxxx>
---
 xen/arch/arm/tee/ffa.c | 121 ++++++++++++++++++++++++++++++-----------
 1 file changed, 89 insertions(+), 32 deletions(-)

diff --git a/xen/arch/arm/tee/ffa.c b/xen/arch/arm/tee/ffa.c
index ac23b9edc74c..59832f73c2be 100644
--- a/xen/arch/arm/tee/ffa.c
+++ b/xen/arch/arm/tee/ffa.c
@@ -365,6 +365,13 @@ struct ffa_ctx {
     struct list_head shm_list;
     /* Number of allocated shared memory object */
     unsigned int shm_count;
+    /*
+     * tx_lock is used to serialize access to tx
+     * rx_lock is used to serialize access to rx
+     * lock is used for the rest in this struct
+     */
+    spinlock_t tx_lock;
+    spinlock_t rx_lock;
     spinlock_t lock;
 };
 
@@ -815,7 +822,9 @@ static int32_t handle_partition_info_get(uint32_t w1, 
uint32_t w2, uint32_t w3,
     if ( !ffa_rx )
         return FFA_RET_DENIED;
 
-    spin_lock(&ctx->lock);
+    if ( !spin_trylock(&ctx->rx_lock) )
+        return FFA_RET_BUSY;
+
     if ( !ctx->page_count || !ctx->rx_is_free )
         goto out;
     spin_lock(&ffa_rx_buffer_lock);
@@ -866,7 +875,7 @@ out_rx_release:
 out_rx_buf_unlock:
     spin_unlock(&ffa_rx_buffer_lock);
 out:
-    spin_unlock(&ctx->lock);
+    spin_unlock(&ctx->rx_lock);
 
     return ret;
 }
@@ -877,13 +886,15 @@ static int32_t handle_rx_release(void)
     struct domain *d = current->domain;
     struct ffa_ctx *ctx = d->arch.tee;
 
-    spin_lock(&ctx->lock);
+    if ( !spin_trylock(&ctx->rx_lock) )
+        return FFA_RET_BUSY;
+
     if ( !ctx->page_count || ctx->rx_is_free )
         goto out;
     ret = FFA_RET_OK;
     ctx->rx_is_free = true;
 out:
-    spin_unlock(&ctx->lock);
+    spin_unlock(&ctx->rx_lock);
 
     return ret;
 }
@@ -994,21 +1005,43 @@ static void put_shm_pages(struct ffa_shm_mem *shm)
     }
 }
 
+static bool inc_ctx_shm_count(struct ffa_ctx *ctx)
+{
+    bool ret = true;
+
+    spin_lock(&ctx->lock);
+    if (ctx->shm_count >= FFA_MAX_SHM_COUNT)
+        ret = false;
+    else
+        ctx->shm_count++;
+    spin_unlock(&ctx->lock);
+
+    return ret;
+}
+
+static void dec_ctx_shm_count(struct ffa_ctx *ctx)
+{
+    spin_lock(&ctx->lock);
+    ASSERT(ctx->shm_count > 0);
+    ctx->shm_count--;
+    spin_unlock(&ctx->lock);
+}
+
 static struct ffa_shm_mem *alloc_ffa_shm_mem(struct ffa_ctx *ctx,
                                              unsigned int page_count)
 {
     struct ffa_shm_mem *shm;
 
-    if ( page_count >= FFA_MAX_SHM_PAGE_COUNT ||
-         ctx->shm_count >= FFA_MAX_SHM_COUNT )
+    if ( page_count >= FFA_MAX_SHM_PAGE_COUNT )
+        return NULL;
+    if ( !inc_ctx_shm_count(ctx) )
         return NULL;
 
     shm = xzalloc_flex_struct(struct ffa_shm_mem, pages, page_count);
     if ( shm )
-    {
-        ctx->shm_count++;
         shm->page_count = page_count;
-    }
+    else
+        dec_ctx_shm_count(ctx);
 
     return shm;
 }
@@ -1018,8 +1051,7 @@ static void free_ffa_shm_mem(struct ffa_ctx *ctx, struct 
ffa_shm_mem *shm)
     if ( !shm )
         return;
 
-    ASSERT(ctx->shm_count > 0);
-    ctx->shm_count--;
+    dec_ctx_shm_count(ctx);
     put_shm_pages(shm);
     xfree(shm);
 }
@@ -1299,7 +1331,11 @@ static void handle_mem_share(struct cpu_user_regs *regs)
         goto out_set_ret;
     }
 
-    spin_lock(&ctx->lock);
+    if ( !spin_trylock(&ctx->tx_lock) )
+    {
+        ret = FFA_RET_BUSY;
+        goto out_set_ret;
+    }
 
     if ( frag_len > ctx->page_count * FFA_PAGE_SIZE )
         goto out_unlock;
@@ -1421,7 +1457,9 @@ static void handle_mem_share(struct cpu_user_regs *regs)
     if ( ret )
         goto out;
 
+    spin_lock(&ctx->lock);
     list_add_tail(&shm->list, &ctx->shm_list);
+    spin_unlock(&ctx->lock);
 
     uint64_to_regpair(&handle_hi, &handle_lo, shm->handle);
 
@@ -1429,7 +1467,7 @@ out:
     if ( ret )
         free_ffa_shm_mem(ctx, shm);
 out_unlock:
-    spin_unlock(&ctx->lock);
+    spin_unlock(&ctx->tx_lock);
 
 out_set_ret:
     if ( ret > 0 )
@@ -1464,7 +1502,12 @@ static void handle_mem_frag_tx(struct cpu_user_regs 
*regs)
     uint16_t sender_id = 0;
     int ret;
 
-    spin_lock(&ctx->lock);
+    if ( !spin_trylock(&ctx->tx_lock) )
+    {
+        ret = FFA_RET_BUSY;
+        goto out_set_ret;
+    }
+
     s = find_frag_state(ctx, handle);
     if ( !s )
     {
@@ -1489,15 +1532,20 @@ static void handle_mem_frag_tx(struct cpu_user_regs 
*regs)
     spin_unlock(&ffa_tx_buffer_lock);
     if ( ret < 0 )
         goto out_free_s;
+
+    spin_lock(&ctx->lock);
     list_add_tail(&s->shm->list, &ctx->shm_list);
+    spin_unlock(&ctx->lock);
+
 out_free_s:
     if ( ret < 0 )
         free_ffa_shm_mem(ctx, s->shm);
     list_del(&s->list);
     xfree(s);
 out:
-    spin_unlock(&ctx->lock);
+    spin_unlock(&ctx->tx_lock);
 
+out_set_ret:
     if ( ret > 0 )
             set_regs_frag_rx(regs, handle_lo, handle_hi, ret, sender_id);
     else if ( ret == 0)
@@ -1506,6 +1554,18 @@ out:
             set_regs_error(regs, ret);
 }
 
+/* Must only be called with ctx->lock held */
+static struct ffa_shm_mem *find_shm_mem(struct ffa_ctx *ctx, uint64_t handle)
+{
+    struct ffa_shm_mem *shm;
+
+    list_for_each_entry(shm, &ctx->shm_list, list)
+        if ( shm->handle == handle )
+            return shm;
+
+    return NULL;
+}
+
 static int handle_mem_reclaim(uint64_t handle, uint32_t flags)
 {
     struct domain *d = current->domain;
@@ -1516,29 +1576,26 @@ static int handle_mem_reclaim(uint64_t handle, uint32_t 
flags)
     int ret;
 
     spin_lock(&ctx->lock);
-    list_for_each_entry(shm, &ctx->shm_list, list)
-    {
-        if ( shm->handle == handle )
-            goto found_it;
-    }
-    shm = NULL;
-    ret = FFA_RET_INVALID_PARAMETERS;
-    goto out;
-found_it:
+    shm = find_shm_mem(ctx, handle);
+    if ( shm )
+        list_del(&shm->list);
+    spin_unlock(&ctx->lock);
+    if ( !shm )
+        return FFA_RET_INVALID_PARAMETERS;
 
     uint64_to_regpair(&handle_hi, &handle_lo, handle);
     ret = ffa_mem_reclaim(handle_lo, handle_hi, flags);
+
     if ( ret )
     {
-        shm = NULL;
-        goto out;
+        spin_lock(&ctx->lock);
+        list_add_tail(&shm->list, &ctx->shm_list);
+        spin_unlock(&ctx->lock);
+    }
+    else
+    {
+        free_ffa_shm_mem(ctx, shm);
     }
-
-    list_del(&shm->list);
-
-out:
-    free_ffa_shm_mem(ctx, shm);
-    spin_unlock(&ctx->lock);
 
     return ret;
 }
-- 
2.34.1