net/cxgbe: fix slot allocation for IPv6 flows
authorRahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Fri, 27 Sep 2019 20:30:03 +0000 (02:00 +0530)
committerFerruh Yigit <ferruh.yigit@intel.com>
Mon, 7 Oct 2019 13:00:57 +0000 (15:00 +0200)
IPv6 flows occupy only 2 slots on Chelsio T6 NICs. Fix the slot
calculation logic to return correct number of slots.

Fixes: ee61f5113b17 ("net/cxgbe: parse and validate flows")
Fixes: 9eb2c9a48072 ("net/cxgbe: implement flow create operation")
Fixes: 3f2c1e209cfc ("net/cxgbe: add Compressed Local IP region")
Cc: stable@dpdk.org
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
drivers/net/cxgbe/cxgbe_filter.c
drivers/net/cxgbe/cxgbe_filter.h
drivers/net/cxgbe/cxgbe_flow.c

index 3b7966d..33b95a6 100644 (file)
@@ -213,20 +213,32 @@ static inline void mk_set_tcb_field_ulp(struct filter_entry *f,
 }
 
 /**
- * Check if entry already filled.
+ * IPv6 requires 2 slots on T6 and 4 slots for cards below T6.
+ * IPv4 requires only 1 slot on all cards.
  */
-bool cxgbe_is_filter_set(struct tid_info *t, int fidx, int family)
+u8 cxgbe_filter_slots(struct adapter *adap, u8 family)
 {
-       bool result = FALSE;
-       int i, max;
+       if (family == FILTER_TYPE_IPV6) {
+               if (CHELSIO_CHIP_VERSION(adap->params.chip) < CHELSIO_T6)
+                       return 4;
 
-       /* IPv6 requires four slots and IPv4 requires only 1 slot.
-        * Ensure, there's enough slots available.
-        */
-       max = family == FILTER_TYPE_IPV6 ? fidx + 3 : fidx;
+               return 2;
+       }
+
+       return 1;
+}
+
+/**
+ * Check if entries are already filled.
+ */
+bool cxgbe_is_filter_set(struct tid_info *t, u32 fidx, u8 nentries)
+{
+       bool result = FALSE;
+       u32 i;
 
+       /* Ensure there's enough slots available. */
        t4_os_lock(&t->ftid_lock);
-       for (i = fidx; i <= max; i++) {
+       for (i = fidx; i < fidx + nentries; i++) {
                if (rte_bitmap_get(t->ftid_bmap, i)) {
                        result = TRUE;
                        break;
@@ -237,17 +249,18 @@ bool cxgbe_is_filter_set(struct tid_info *t, int fidx, int family)
 }
 
 /**
- * Allocate a available free entry
+ * Allocate available free entries.
  */
-int cxgbe_alloc_ftid(struct adapter *adap, unsigned int family)
+int cxgbe_alloc_ftid(struct adapter *adap, u8 nentries)
 {
        struct tid_info *t = &adap->tids;
        int pos;
        int size = t->nftids;
 
        t4_os_lock(&t->ftid_lock);
-       if (family == FILTER_TYPE_IPV6)
-               pos = cxgbe_bitmap_find_free_region(t->ftid_bmap, size, 4);
+       if (nentries > 1)
+               pos = cxgbe_bitmap_find_free_region(t->ftid_bmap, size,
+                                                   nentries);
        else
                pos = cxgbe_find_first_zero_bit(t->ftid_bmap, size);
        t4_os_unlock(&t->ftid_lock);
@@ -565,7 +578,7 @@ static int cxgbe_set_hash_filter(struct rte_eth_dev *dev,
        if (atid < 0)
                goto out_err;
 
-       if (f->fs.type) {
+       if (f->fs.type == FILTER_TYPE_IPV6) {
                /* IPv6 hash filter */
                f->clipt = cxgbe_clip_alloc(f->dev, (u32 *)&f->fs.val.lip);
                if (!f->clipt)
@@ -804,44 +817,34 @@ out:
 }
 
 /**
- * Set the corresponding entry in the bitmap. 4 slots are
- * marked for IPv6, whereas only 1 slot is marked for IPv4.
+ * Set the corresponding entries in the bitmap.
  */
-static int cxgbe_set_ftid(struct tid_info *t, int fidx, int family)
+static int cxgbe_set_ftid(struct tid_info *t, u32 fidx, u8 nentries)
 {
+       u32 i;
+
        t4_os_lock(&t->ftid_lock);
        if (rte_bitmap_get(t->ftid_bmap, fidx)) {
                t4_os_unlock(&t->ftid_lock);
                return -EBUSY;
        }
 
-       if (family == FILTER_TYPE_IPV4) {
-               rte_bitmap_set(t->ftid_bmap, fidx);
-       } else {
-               rte_bitmap_set(t->ftid_bmap, fidx);
-               rte_bitmap_set(t->ftid_bmap, fidx + 1);
-               rte_bitmap_set(t->ftid_bmap, fidx + 2);
-               rte_bitmap_set(t->ftid_bmap, fidx + 3);
-       }
+       for (i = fidx; i < fidx + nentries; i++)
+               rte_bitmap_set(t->ftid_bmap, i);
        t4_os_unlock(&t->ftid_lock);
        return 0;
 }
 
 /**
- * Clear the corresponding entry in the bitmap. 4 slots are
- * cleared for IPv6, whereas only 1 slot is cleared for IPv4.
+ * Clear the corresponding entries in the bitmap.
  */
-static void cxgbe_clear_ftid(struct tid_info *t, int fidx, int family)
+static void cxgbe_clear_ftid(struct tid_info *t, u32 fidx, u8 nentries)
 {
+       u32 i;
+
        t4_os_lock(&t->ftid_lock);
-       if (family == FILTER_TYPE_IPV4) {
-               rte_bitmap_clear(t->ftid_bmap, fidx);
-       } else {
-               rte_bitmap_clear(t->ftid_bmap, fidx);
-               rte_bitmap_clear(t->ftid_bmap, fidx + 1);
-               rte_bitmap_clear(t->ftid_bmap, fidx + 2);
-               rte_bitmap_clear(t->ftid_bmap, fidx + 3);
-       }
+       for (i = fidx; i < fidx + nentries; i++)
+               rte_bitmap_clear(t->ftid_bmap, i);
        t4_os_unlock(&t->ftid_lock);
 }
 
@@ -859,6 +862,7 @@ int cxgbe_del_filter(struct rte_eth_dev *dev, unsigned int filter_id,
        struct adapter *adapter = pi->adapter;
        struct filter_entry *f;
        unsigned int chip_ver;
+       u8 nentries;
        int ret;
 
        if (is_hashfilter(adapter) && fs->cap)
@@ -869,24 +873,25 @@ int cxgbe_del_filter(struct rte_eth_dev *dev, unsigned int filter_id,
 
        chip_ver = CHELSIO_CHIP_VERSION(adapter->params.chip);
 
-       ret = cxgbe_is_filter_set(&adapter->tids, filter_id, fs->type);
-       if (!ret) {
-               dev_warn(adap, "%s: could not find filter entry: %u\n",
-                        __func__, filter_id);
-               return -EINVAL;
-       }
-
        /*
-        * Ensure filter id is aligned on the 2 slot boundary for T6,
+        * Ensure IPv6 filter id is aligned on the 2 slot boundary for T6,
         * and 4 slot boundary for cards below T6.
         */
-       if (fs->type) {
+       if (fs->type == FILTER_TYPE_IPV6) {
                if (chip_ver < CHELSIO_T6)
                        filter_id &= ~(0x3);
                else
                        filter_id &= ~(0x1);
        }
 
+       nentries = cxgbe_filter_slots(adapter, fs->type);
+       ret = cxgbe_is_filter_set(&adapter->tids, filter_id, nentries);
+       if (!ret) {
+               dev_warn(adap, "%s: could not find filter entry: %u\n",
+                        __func__, filter_id);
+               return -EINVAL;
+       }
+
        f = &adapter->tids.ftid_tab[filter_id];
        ret = writable_filter(f);
        if (ret)
@@ -896,8 +901,7 @@ int cxgbe_del_filter(struct rte_eth_dev *dev, unsigned int filter_id,
                f->ctx = ctx;
                cxgbe_clear_ftid(&adapter->tids,
                                 f->tid - adapter->tids.ftid_base,
-                                f->fs.type ? FILTER_TYPE_IPV6 :
-                                             FILTER_TYPE_IPV4);
+                                nentries);
                return del_filter_wr(dev, filter_id);
        }
 
@@ -927,10 +931,10 @@ int cxgbe_set_filter(struct rte_eth_dev *dev, unsigned int filter_id,
 {
        struct port_info *pi = ethdev2pinfo(dev);
        struct adapter *adapter = pi->adapter;
-       unsigned int fidx, iq, fid_bit = 0;
+       unsigned int fidx, iq;
        struct filter_entry *f;
        unsigned int chip_ver;
-       uint8_t bitoff[16] = {0};
+       u8 nentries, bitoff[16] = {0};
        int ret;
 
        if (is_hashfilter(adapter) && fs->cap)
@@ -945,80 +949,31 @@ int cxgbe_set_filter(struct rte_eth_dev *dev, unsigned int filter_id,
        if (ret)
                return ret;
 
-       /*
-        * Ensure filter id is aligned on the 4 slot boundary for IPv6
-        * maskfull filters.
-        */
-       if (fs->type)
-               filter_id &= ~(0x3);
-
-       ret = cxgbe_is_filter_set(&adapter->tids, filter_id, fs->type);
-       if (ret)
-               return -EBUSY;
-
-       iq = get_filter_steerq(dev, fs);
-
        /*
         * IPv6 filters occupy four slots and must be aligned on four-slot
         * boundaries for T5. On T6, IPv6 filters occupy two-slots and
         * must be aligned on two-slot boundaries.
         *
         * IPv4 filters only occupy a single slot and have no alignment
-        * requirements but writing a new IPv4 filter into the middle
-        * of an existing IPv6 filter requires clearing the old IPv6
-        * filter.
+        * requirements.
         */
-       if (fs->type == FILTER_TYPE_IPV4) { /* IPv4 */
-               /*
-                * For T6, If our IPv4 filter isn't being written to a
-                * multiple of two filter index and there's an IPv6
-                * filter at the multiple of 2 base slot, then we need
-                * to delete that IPv6 filter ...
-                * For adapters below T6, IPv6 filter occupies 4 entries.
-                */
+       fidx = filter_id;
+       if (fs->type == FILTER_TYPE_IPV6) {
                if (chip_ver < CHELSIO_T6)
-                       fidx = filter_id & ~0x3;
+                       fidx &= ~(0x3);
                else
-                       fidx = filter_id & ~0x1;
-
-               if (fidx != filter_id && adapter->tids.ftid_tab[fidx].fs.type) {
-                       f = &adapter->tids.ftid_tab[fidx];
-                       if (f->valid)
-                               return -EBUSY;
-               }
-       } else { /* IPv6 */
-               unsigned int max_filter_id;
-
-               if (chip_ver < CHELSIO_T6) {
-                       /*
-                        * Ensure that the IPv6 filter is aligned on a
-                        * multiple of 4 boundary.
-                        */
-                       if (filter_id & 0x3)
-                               return -EINVAL;
+                       fidx &= ~(0x1);
+       }
 
-                       max_filter_id = filter_id + 4;
-               } else {
-                       /*
-                        * For T6, CLIP being enabled, IPv6 filter would occupy
-                        * 2 entries.
-                        */
-                       if (filter_id & 0x1)
-                               return -EINVAL;
+       if (fidx != filter_id)
+               return -EINVAL;
 
-                       max_filter_id = filter_id + 2;
-               }
+       nentries = cxgbe_filter_slots(adapter, fs->type);
+       ret = cxgbe_is_filter_set(&adapter->tids, filter_id, nentries);
+       if (ret)
+               return -EBUSY;
 
-               /*
-                * Check all except the base overlapping IPv4 filter
-                * slots.
-                */
-               for (fidx = filter_id + 1; fidx < max_filter_id; fidx++) {
-                       f = &adapter->tids.ftid_tab[fidx];
-                       if (f->valid)
-                               return -EBUSY;
-               }
-       }
+       iq = get_filter_steerq(dev, fs);
 
        /*
         * Check to make sure that provided filter index is not
@@ -1029,9 +984,7 @@ int cxgbe_set_filter(struct rte_eth_dev *dev, unsigned int filter_id,
                return -EBUSY;
 
        fidx = adapter->tids.ftid_base + filter_id;
-       fid_bit = filter_id;
-       ret = cxgbe_set_ftid(&adapter->tids, fid_bit,
-                            fs->type ? FILTER_TYPE_IPV6 : FILTER_TYPE_IPV4);
+       ret = cxgbe_set_ftid(&adapter->tids, filter_id, nentries);
        if (ret)
                return ret;
 
@@ -1041,9 +994,7 @@ int cxgbe_set_filter(struct rte_eth_dev *dev, unsigned int filter_id,
        ret = writable_filter(f);
        if (ret) {
                /* Clear the bits we have set above */
-               cxgbe_clear_ftid(&adapter->tids, fid_bit,
-                                fs->type ? FILTER_TYPE_IPV6 :
-                                           FILTER_TYPE_IPV4);
+               cxgbe_clear_ftid(&adapter->tids, filter_id, nentries);
                return ret;
        }
 
@@ -1074,17 +1025,13 @@ int cxgbe_set_filter(struct rte_eth_dev *dev, unsigned int filter_id,
        f->ctx = ctx;
        f->tid = fidx; /* Save the actual tid */
        ret = set_filter_wr(dev, filter_id);
-       if (ret) {
-               fid_bit = f->tid - adapter->tids.ftid_base;
+       if (ret)
                goto free_tid;
-       }
 
        return ret;
 
 free_tid:
-       cxgbe_clear_ftid(&adapter->tids, fid_bit,
-                        fs->type ? FILTER_TYPE_IPV6 :
-                                   FILTER_TYPE_IPV4);
+       cxgbe_clear_ftid(&adapter->tids, filter_id, nentries);
        clear_filter(f);
        return ret;
 }
index 1964730..06021c8 100644 (file)
@@ -248,7 +248,8 @@ cxgbe_bitmap_find_free_region(struct rte_bitmap *bmap, unsigned int size,
        return idx;
 }
 
-bool cxgbe_is_filter_set(struct tid_info *, int fidx, int family);
+u8 cxgbe_filter_slots(struct adapter *adap, u8 family);
+bool cxgbe_is_filter_set(struct tid_info *t, u32 fidx, u8 nentries);
 void cxgbe_filter_rpl(struct adapter *adap, const struct cpl_set_tcb_rpl *rpl);
 int cxgbe_set_filter(struct rte_eth_dev *dev, unsigned int filter_id,
                     struct ch_filter_specification *fs,
@@ -256,7 +257,7 @@ int cxgbe_set_filter(struct rte_eth_dev *dev, unsigned int filter_id,
 int cxgbe_del_filter(struct rte_eth_dev *dev, unsigned int filter_id,
                     struct ch_filter_specification *fs,
                     struct filter_ctx *ctx);
-int cxgbe_alloc_ftid(struct adapter *adap, unsigned int family);
+int cxgbe_alloc_ftid(struct adapter *adap, u8 nentries);
 int cxgbe_init_hash_filter(struct adapter *adap);
 void cxgbe_hash_filter_rpl(struct adapter *adap,
                           const struct cpl_act_open_rpl *rpl);
index 848c61f..8a5d06f 100644 (file)
@@ -304,12 +304,15 @@ static int cxgbe_validate_fidxondel(struct filter_entry *f, unsigned int fidx)
 {
        struct adapter *adap = ethdev2adap(f->dev);
        struct ch_filter_specification fs = f->fs;
+       u8 nentries;
 
        if (fidx >= adap->tids.nftids) {
                dev_err(adap, "invalid flow index %d.\n", fidx);
                return -EINVAL;
        }
-       if (!cxgbe_is_filter_set(&adap->tids, fidx, fs.type)) {
+
+       nentries = cxgbe_filter_slots(adap, fs.type);
+       if (!cxgbe_is_filter_set(&adap->tids, fidx, nentries)) {
                dev_err(adap, "Already free fidx:%d f:%p\n", fidx, f);
                return -EINVAL;
        }
@@ -321,10 +324,14 @@ static int
 cxgbe_validate_fidxonadd(struct ch_filter_specification *fs,
                         struct adapter *adap, unsigned int fidx)
 {
-       if (cxgbe_is_filter_set(&adap->tids, fidx, fs->type)) {
+       u8 nentries;
+
+       nentries = cxgbe_filter_slots(adap, fs->type);
+       if (cxgbe_is_filter_set(&adap->tids, fidx, nentries)) {
                dev_err(adap, "filter index: %d is busy.\n", fidx);
                return -EBUSY;
        }
+
        if (fidx >= adap->tids.nftids) {
                dev_err(adap, "filter index (%u) >= max(%u)\n",
                        fidx, adap->tids.nftids);
@@ -351,9 +358,11 @@ static int cxgbe_get_fidx(struct rte_flow *flow, unsigned int *fidx)
 
        /* For tcam get the next available slot, if default value specified */
        if (flow->fidx == FILTER_ID_MAX) {
+               u8 nentries;
                int idx;
 
-               idx = cxgbe_alloc_ftid(adap, fs->type);
+               nentries = cxgbe_filter_slots(adap, fs->type);
+               idx = cxgbe_alloc_ftid(adap, nentries);
                if (idx < 0) {
                        dev_err(adap, "unable to get a filter index in tcam\n");
                        return -ENOMEM;