qcacmn: Add a set of nbuf fragment allocation APIs

It adds a set of frag way allocation APIs explicitly.
IPA Tx nbufs change to frag way allocation API, which saves memory.

Current __qdf_nbuf_alloc changes to non-frag way allocation by
default if SKB RECYLCER not defined.

Change-Id: I5f87ffac54c49f9af920775c13b6dfdd147476dd
CRs-Fixed: 3534452

dp/wifi3.0/dp_ipa.c

@@ -644,7 +644,8 @@ static int dp_ipa_tx_alt_pool_attach(struct dp_soc *soc)
 	 */
 	for (tx_buffer_count = 0;
 		tx_buffer_count < max_alloc_count - 1; tx_buffer_count++) {
-		nbuf = qdf_nbuf_alloc(soc->osdev, alloc_size, 0, 256, FALSE);
+		nbuf = qdf_nbuf_frag_alloc(soc->osdev, alloc_size, 0,
+					   256, FALSE);
 		if (!nbuf)
 			break;
 
@@ -1494,7 +1495,8 @@ static int dp_tx_ipa_uc_attach(struct dp_soc *soc, struct dp_pdev *pdev)
 	 */
 	for (tx_buffer_count = 0;
 		tx_buffer_count < max_alloc_count - 1; tx_buffer_count++) {
-		nbuf = qdf_nbuf_alloc(soc->osdev, alloc_size, 0, 256, FALSE);
+		nbuf = qdf_nbuf_frag_alloc(soc->osdev, alloc_size, 0,
+					   256, FALSE);
 		if (!nbuf)
 			break;
 

qdf/inc/qdf_nbuf.h

@@ -2286,6 +2286,13 @@ qdf_nbuf_t qdf_nbuf_alloc_debug(qdf_device_t osdev, qdf_size_t size,
 				int reserve, int align, int prio,
 				const char *func, uint32_t line);
 
+#define qdf_nbuf_frag_alloc(d, s, r, a, p) \
+	qdf_nbuf_frag_alloc_debug(d, s, r, a, p, __func__, __LINE__)
+
+qdf_nbuf_t qdf_nbuf_frag_alloc_debug(qdf_device_t osdev, qdf_size_t size,
+				     int reserve, int align, int prio,
+				     const char *func, uint32_t line);
+
 /**
  * qdf_nbuf_alloc_no_recycler() - Allocates skb
  * @size: Size to be allocated for skb
@@ -2461,6 +2468,18 @@ qdf_nbuf_alloc_fl(qdf_device_t osdev, qdf_size_t size, int reserve, int align,
 	return __qdf_nbuf_alloc(osdev, size, reserve, align, prio, func, line);
 }
 
+#define qdf_nbuf_frag_alloc(osdev, size, reserve, align, prio) \
+	qdf_nbuf_frag_alloc_fl(osdev, size, reserve, align, prio, \
+			  __func__, __LINE__)
+
+static inline qdf_nbuf_t
+qdf_nbuf_frag_alloc_fl(qdf_device_t osdev, qdf_size_t size, int reserve,
+		       int align, int prio, const char *func, uint32_t line)
+{
+	return __qdf_nbuf_frag_alloc(osdev, size, reserve, align, prio,
+				     func, line);
+}
+
 /**
  * qdf_nbuf_alloc_no_recycler_fl() - Allocate SKB
  * @size: Size to be allocated for skb

qdf/linux/src/i_qdf_mem.h

@@ -49,6 +49,8 @@
  */
 #define GFP_KERNEL 0
 #define GFP_ATOMIC 0
+#define __GFP_KSWAPD_RECLAIM 0
+#define __GFP_DIRECT_RECLAIM 0
 #define kzalloc(size, flags) NULL
 #define vmalloc(size)        NULL
 #define kfree(buf)

qdf/linux/src/i_qdf_nbuf.h

@@ -310,7 +310,7 @@ typedef void (*qdf_nbuf_free_t)(__qdf_nbuf_t);
  * @func: Function name of the call site
  * @line: line number of the call site
  *
- * This allocates an nbuf aligns if needed and reserves some space in the front,
+ * This allocates a nbuf aligns if needed and reserves some space in the front,
  * since the reserve is done after alignment the reserve value if being
  * unaligned will result in an unaligned address.
  *
@@ -345,6 +345,28 @@ __qdf_nbuf_t __qdf_nbuf_alloc_ppe_ds(__qdf_device_t osdev, size_t size,
 				     const char *func, uint32_t line);
 #endif /* QCA_DP_NBUF_FAST_PPEDS */
 
+/**
+ * __qdf_nbuf_frag_alloc() - Allocate nbuf in page fragment way.
+ * @osdev: Device handle
+ * @size: Netbuf requested size
+ * @reserve: headroom to start with
+ * @align: Align
+ * @prio: Priority
+ * @func: Function name of the call site
+ * @line: line number of the call site
+ *
+ * This allocates a nbuf aligns if needed and reserves some space in the front,
+ * since the reserve is done after alignment the reserve value if being
+ * unaligned will result in an unaligned address.
+ * It will call into kernel page fragment APIs, long time keeping for scattered
+ * allocations should be considered for avoidance.
+ *
+ * Return: nbuf or %NULL if no memory
+ */
+__qdf_nbuf_t
+__qdf_nbuf_frag_alloc(__qdf_device_t osdev, size_t size, int reserve, int align,
+		      int prio, const char *func, uint32_t line);
+
 /**
  * __qdf_nbuf_alloc_no_recycler() - Allocates skb
  * @size: Size to be allocated for skb

qdf/linux/src/qdf_nbuf.c

@@ -555,6 +555,16 @@ skb_alloc:
 }
 #else
 
+#ifdef QCA_DP_NBUF_FAST_RECYCLE_CHECK
+struct sk_buff *__qdf_nbuf_alloc(qdf_device_t osdev, size_t size, int reserve,
+				 int align, int prio, const char *func,
+				 uint32_t line)
+{
+	return __qdf_nbuf_frag_alloc(osdev, size, reserve, align, prio, func,
+				     line);
+}
+
+#else
 struct sk_buff *__qdf_nbuf_alloc(qdf_device_t osdev, size_t size, int reserve,
 				 int align, int prio, const char *func,
 				 uint32_t line)
@@ -579,6 +589,81 @@ struct sk_buff *__qdf_nbuf_alloc(qdf_device_t osdev, size_t size, int reserve,
 #endif
 	}
 
+	skb =  alloc_skb(size, flags);
+
+	if (skb)
+		goto skb_alloc;
+
+	skb = pld_nbuf_pre_alloc(size);
+
+	if (!skb) {
+		qdf_rl_nofl_err("NBUF alloc failed %zuB @ %s:%d",
+				size, func, line);
+		__qdf_nbuf_start_replenish_timer();
+		return NULL;
+	}
+
+	__qdf_nbuf_stop_replenish_timer();
+
+skb_alloc:
+	memset(skb->cb, 0x0, sizeof(skb->cb));
+	skb->dev = NULL;
+	/*
+	 * The default is for netbuf fragments to be interpreted
+	 * as wordstreams rather than bytestreams.
+	 */
+	QDF_NBUF_CB_TX_EXTRA_FRAG_WORDSTR_EFRAG(skb) = 1;
+	QDF_NBUF_CB_TX_EXTRA_FRAG_WORDSTR_NBUF(skb) = 1;
+
+	/*
+	 * XXX:how about we reserve first then align
+	 * Align & make sure that the tail & data are adjusted properly
+	 */
+
+	if (align) {
+		offset = ((unsigned long)skb->data) % align;
+		if (offset)
+			skb_reserve(skb, align - offset);
+	}
+
+	/*
+	 * NOTE:alloc doesn't take responsibility if reserve unaligns the data
+	 * pointer
+	 */
+	skb_reserve(skb, reserve);
+	qdf_nbuf_count_inc(skb);
+
+	return skb;
+}
+#endif
+
+#endif
+qdf_export_symbol(__qdf_nbuf_alloc);
+
+struct sk_buff *__qdf_nbuf_frag_alloc(qdf_device_t osdev, size_t size,
+				      int reserve, int align, int prio,
+				      const char *func, uint32_t line)
+{
+	struct sk_buff *skb;
+	unsigned long offset;
+	int flags = GFP_KERNEL & ~__GFP_DIRECT_RECLAIM;
+
+	if (align)
+		size += (align - 1);
+
+	if (in_interrupt() || irqs_disabled() || in_atomic()) {
+		flags = GFP_ATOMIC;
+#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 4, 0)
+		/*
+		 * Observed that kcompactd burns out CPU to make order-3 page.
+		 *__netdev_alloc_skb has 4k page fallback option just in case of
+		 * failing high order page allocation so we don't need to be
+		 * hard. Make kcompactd rest in piece.
+		 */
+		flags = flags & ~__GFP_KSWAPD_RECLAIM;
+#endif
+	}
+
 	skb = __netdev_alloc_skb(NULL, size, flags);
 
 	if (skb)
@@ -591,13 +676,12 @@ struct sk_buff *__qdf_nbuf_alloc(qdf_device_t osdev, size_t size, int reserve,
 				size, func, line);
 		__qdf_nbuf_start_replenish_timer();
 		return NULL;
-	} else {
-		__qdf_nbuf_stop_replenish_timer();
 	}
 
+	__qdf_nbuf_stop_replenish_timer();
+
 skb_alloc:
 	memset(skb->cb, 0x0, sizeof(skb->cb));
-
 	/*
 	 * The default is for netbuf fragments to be interpreted
 	 * as wordstreams rather than bytestreams.
@@ -625,8 +709,8 @@ skb_alloc:
 
 	return skb;
 }
-#endif
-qdf_export_symbol(__qdf_nbuf_alloc);
+
+qdf_export_symbol(__qdf_nbuf_frag_alloc);
 
 __qdf_nbuf_t __qdf_nbuf_alloc_no_recycler(size_t size, int reserve, int align,
 					  const char *func, uint32_t line)
@@ -3377,6 +3461,33 @@ qdf_nbuf_t qdf_nbuf_alloc_debug(qdf_device_t osdev, qdf_size_t size,
 }
 qdf_export_symbol(qdf_nbuf_alloc_debug);
 
+qdf_nbuf_t qdf_nbuf_frag_alloc_debug(qdf_device_t osdev, qdf_size_t size,
+				     int reserve, int align, int prio,
+				     const char *func, uint32_t line)
+{
+	qdf_nbuf_t nbuf;
+
+	if (is_initial_mem_debug_disabled)
+		return __qdf_nbuf_frag_alloc(osdev, size,
+					reserve, align,
+					prio, func, line);
+
+	nbuf = __qdf_nbuf_frag_alloc(osdev, size, reserve, align, prio,
+				     func, line);
+
+	/* Store SKB in internal QDF tracking table */
+	if (qdf_likely(nbuf)) {
+		qdf_net_buf_debug_add_node(nbuf, size, func, line);
+		qdf_nbuf_history_add(nbuf, func, line, QDF_NBUF_ALLOC);
+	} else {
+		qdf_nbuf_history_add(nbuf, func, line, QDF_NBUF_ALLOC_FAILURE);
+	}
+
+	return nbuf;
+}
+
+qdf_export_symbol(qdf_nbuf_frag_alloc_debug);
+
 qdf_nbuf_t qdf_nbuf_alloc_no_recycler_debug(size_t size, int reserve, int align,
 					    const char *func, uint32_t line)
 {