[PATCH net-next 00/15] nfp: allow concurrency in core and minor fixes

[PATCH net-next 00/15] nfp: allow concurrency in core and minor fixes

[Jakub Kicinski]

Hi!

The first 10 patches of this series prepare nfpcore for concurrent 
accesses.  This will be needed by upcoming hwmon and devlink patches.
Most locking is already in place, the patches in this series iron out
a few bugs.

Last 5 patches are fixes and cleanups to the netdev code, including
removal of doorbell pointers used only on old versions of the chip,
removal of unnecessarily defensive code and flushing xmit_more more
carefully on error paths.

Jakub Kicinski (15):
  nfp: disallow sharing mutexes on the same machine
  nfp: fail graciously when someone tries to grab global lock
  nfp: remove cpp mutex cache
  nfp: move mutex code out of nfp_cppcore.c
  nfp: document expected locking in the core
  nfp: lock area cache earlier
  nfp: correct return codes when msleep gets interrupted
  nfp: don't ignore return value of wait_event_interruptible
  nfp: fix invalid area detection
  nfp: fix nfp_cpp_read()/nfp_cpp_write() error paths
  nfp: don't use netdev_warn() before netdev is registered
  nfp: remove RX queue pointers
  nfp: flush xmit_more on error paths
  nfp: remove defensive checks around ndo_open()/ndo_close()
  nfp: disable FW on reconfiguration errors

 drivers/net/ethernet/netronome/nfp/Makefile        |   1 +
 drivers/net/ethernet/netronome/nfp/nfp_net.h       |   8 +-
 .../net/ethernet/netronome/nfp/nfp_net_common.c    |  50 +--
 .../net/ethernet/netronome/nfp/nfp_net_debugfs.c   |  15 +-
 drivers/net/ethernet/netronome/nfp/nfpcore/nfp.h   |   9 +-
 .../ethernet/netronome/nfp/nfpcore/nfp6000_pcie.c  |  29 +-
 .../ethernet/netronome/nfp/nfpcore/nfp_cppcore.c   | 467 ++++-----------------
 .../net/ethernet/netronome/nfp/nfpcore/nfp_mutex.c | 345 +++++++++++++++
 .../net/ethernet/netronome/nfp/nfpcore/nfp_nsp.c   |  12 +-
 .../ethernet/netronome/nfp/nfpcore/nfp_resource.c  |  15 +-
 10 files changed, 484 insertions(+), 467 deletions(-)
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfpcore/nfp_mutex.c

-- 
2.11.0

[PATCH net-next 01/15] nfp: disallow sharing mutexes on the same machine

[Jakub Kicinski]

NFP can be connected to multiple machines via PCI or other buses.
Access to hardware resources is arbitrated using locks residing
in device memory.  Currently nfpcore only respects the mutexes
when it comes to inter-host locking, but if we try to acquire
the same lock again, on one host - it will simply return success
because owner of the lock is already set to that host.

This makes the locks useless for arbitration within one host
and unfair because whichever host grabbed the lock will have
a chance to reacquire it without others getting a shot.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c | 6 ------
 1 file changed, 6 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
index 40108e66c654..e2267b421af0 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
@@ -1736,11 +1736,5 @@ int nfp_cpp_mutex_trylock(struct nfp_cpp_mutex *mutex)
 		return 0;
 	}
 
-	/* Already locked by us? Success! */
-	if (tmp == value) {
-		mutex->depth = 1;
-		return 0;
-	}
-
 	return nfp_mutex_is_locked(tmp) ? -EBUSY : -EINVAL;
 }
-- 
2.11.0

[PATCH net-next 02/15] nfp: fail graciously when someone tries to grab global lock

[Jakub Kicinski]

The global device lock is acquired to search the resource table.
The lock is actually itself part of the table (entry 0).
Therefore if someone asks for resource 0 we would deadlock since
double locking is no longer allowed.

Currently the driver doesn't try to lock that resource so let's
simply make sure we fail graciously and not add special handling
of this case until really need.  Hide the relevant defines in
the source file.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfpcore/nfp.h          |  9 +--------
 drivers/net/ethernet/netronome/nfp/nfpcore/nfp_resource.c | 15 ++++++++++++---
 2 files changed, 13 insertions(+), 11 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp.h b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp.h
index 42cb720b696d..f7ca8e374923 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp.h
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp.h
@@ -66,14 +66,7 @@ int nfp_nsp_write_eth_table(struct nfp_nsp *state,
 
 /* Implemented in nfp_resource.c */
 
-#define NFP_RESOURCE_TBL_TARGET		NFP_CPP_TARGET_MU
-#define NFP_RESOURCE_TBL_BASE		0x8100000000ULL
-
-/* NFP Resource Table self-identifier */
-#define NFP_RESOURCE_TBL_NAME		"nfp.res"
-#define NFP_RESOURCE_TBL_KEY		0x00000000 /* Special key for entry 0 */
-
-/* All other keys are CRC32-POSIX of the 8-byte identification string */
+/* All keys are CRC32-POSIX of the 8-byte identification string */
 
 /* ARM/PCI vNIC Interfaces 0..3 */
 #define NFP_RESOURCE_VNIC_PCI_0		"vnic.p0"
diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_resource.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_resource.c
index a2850344f8b4..2d15a7c9d0de 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_resource.c
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_resource.c
@@ -45,6 +45,13 @@
 #include "nfp_cpp.h"
 #include "nfp6000/nfp6000.h"
 
+#define NFP_RESOURCE_TBL_TARGET		NFP_CPP_TARGET_MU
+#define NFP_RESOURCE_TBL_BASE		0x8100000000ULL
+
+/* NFP Resource Table self-identifier */
+#define NFP_RESOURCE_TBL_NAME		"nfp.res"
+#define NFP_RESOURCE_TBL_KEY		0x00000000 /* Special key for entry 0 */
+
 #define NFP_RESOURCE_ENTRY_NAME_SZ	8
 
 /**
@@ -100,9 +107,11 @@ static int nfp_cpp_resource_find(struct nfp_cpp *cpp, struct nfp_resource *res)
 	strncpy(name_pad, res->name, sizeof(name_pad));
 
 	/* Search for a matching entry */
-	key = NFP_RESOURCE_TBL_KEY;
-	if (memcmp(name_pad, NFP_RESOURCE_TBL_NAME "\0\0\0\0\0\0\0\0", 8))
-		key = crc32_posix(name_pad, sizeof(name_pad));
+	if (!memcmp(name_pad, NFP_RESOURCE_TBL_NAME "\0\0\0\0\0\0\0\0", 8)) {
+		nfp_err(cpp, "Grabbing device lock not supported\n");
+		return -EOPNOTSUPP;
+	}
+	key = crc32_posix(name_pad, sizeof(name_pad));
 
 	for (i = 0; i < NFP_RESOURCE_TBL_ENTRIES; i++) {
 		u64 addr = NFP_RESOURCE_TBL_BASE +
-- 
2.11.0

[PATCH net-next 03/15] nfp: remove cpp mutex cache

[Jakub Kicinski]

CPP mutex cache was introduced to work around the fact that the
same host could successfully acquire a lock multiple times.  It
used to collapse multiple users to the same struct nfp_cpp_mutex
and track use count.  Unfortunately it's racy.  Since we now force
all nfp_mutex_lock() callers within the host to actually succeed
at acquiring the lock we no longer need the cache, let's remove it.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 .../ethernet/netronome/nfp/nfpcore/nfp_cppcore.c   | 43 +---------------------
 1 file changed, 2 insertions(+), 41 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
index e2267b421af0..6337342c5b62 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
@@ -66,10 +66,8 @@ struct nfp_cpp_resource {
 };
 
 struct nfp_cpp_mutex {
-	struct list_head list;
 	struct nfp_cpp *cpp;
 	int target;
-	u16 usage;
 	u16 depth;
 	unsigned long long address;
 	u32 key;
@@ -86,7 +84,6 @@ struct nfp_cpp {
 
 	const struct nfp_cpp_operations *op;
 	struct list_head resource_list;	/* NFP CPP resource list */
-	struct list_head mutex_cache;	/* Mutex cache */
 	rwlock_t resource_lock;
 	wait_queue_head_t waitq;
 
@@ -187,24 +184,6 @@ void nfp_cpp_free(struct nfp_cpp *cpp)
 {
 	struct nfp_cpp_area_cache *cache, *ctmp;
 	struct nfp_cpp_resource *res, *rtmp;
-	struct nfp_cpp_mutex *mutex, *mtmp;
-
-	/* There should be no mutexes in the cache at this point. */
-	WARN_ON(!list_empty(&cpp->mutex_cache));
-	/* .. but if there are, unlock them and complain. */
-	list_for_each_entry_safe(mutex, mtmp, &cpp->mutex_cache, list) {
-		dev_err(cpp->dev.parent, "Dangling mutex: @%d::0x%llx, %d locks held by %d owners\n",
-			mutex->target, (unsigned long long)mutex->address,
-			mutex->depth, mutex->usage);
-
-		/* Forcing an unlock */
-		mutex->depth = 1;
-		nfp_cpp_mutex_unlock(mutex);
-
-		/* Forcing a free */
-		mutex->usage = 1;
-		nfp_cpp_mutex_free(mutex);
-	}
 
 	/* Remove all caches */
 	list_for_each_entry_safe(cache, ctmp, &cpp->area_cache_list, entry) {
@@ -1127,7 +1106,6 @@ nfp_cpp_from_operations(const struct nfp_cpp_operations *ops,
 	rwlock_init(&cpp->resource_lock);
 	init_waitqueue_head(&cpp->waitq);
 	lockdep_set_class(&cpp->resource_lock, &nfp_cpp_resource_lock_key);
-	INIT_LIST_HEAD(&cpp->mutex_cache);
 	INIT_LIST_HEAD(&cpp->resource_list);
 	INIT_LIST_HEAD(&cpp->area_cache_list);
 	mutex_init(&cpp->area_cache_mutex);
@@ -1536,14 +1514,6 @@ struct nfp_cpp_mutex *nfp_cpp_mutex_alloc(struct nfp_cpp *cpp, int target,
 	if (err)
 		return NULL;
 
-	/* Look for mutex on cache list */
-	list_for_each_entry(mutex, &cpp->mutex_cache, list) {
-		if (mutex->target == target && mutex->address == address) {
-			mutex->usage++;
-			return mutex;
-		}
-	}
-
 	err = nfp_cpp_readl(cpp, mur, address + 4, &tmp);
 	if (err < 0)
 		return NULL;
@@ -1560,10 +1530,6 @@ struct nfp_cpp_mutex *nfp_cpp_mutex_alloc(struct nfp_cpp *cpp, int target,
 	mutex->address = address;
 	mutex->key = key;
 	mutex->depth = 0;
-	mutex->usage = 1;
-
-	/* Add mutex to cache list */
-	list_add(&mutex->list, &cpp->mutex_cache);
 
 	return mutex;
 }
@@ -1574,11 +1540,6 @@ struct nfp_cpp_mutex *nfp_cpp_mutex_alloc(struct nfp_cpp *cpp, int target,
  */
 void nfp_cpp_mutex_free(struct nfp_cpp_mutex *mutex)
 {
-	if (--mutex->usage)
-		return;
-
-	/* Remove mutex from cache */
-	list_del(&mutex->list);
 	kfree(mutex);
 }
 
@@ -1611,8 +1572,8 @@ int nfp_cpp_mutex_lock(struct nfp_cpp_mutex *mutex)
 		if (time_is_before_eq_jiffies(warn_at)) {
 			warn_at = jiffies + 60 * HZ;
 			dev_warn(mutex->cpp->dev.parent,
-				 "Warning: waiting for NFP mutex [usage:%hd depth:%hd target:%d addr:%llx key:%08x]\n",
-				 mutex->usage, mutex->depth,
+				 "Warning: waiting for NFP mutex [depth:%hd target:%d addr:%llx key:%08x]\n",
+				 mutex->depth,
 				 mutex->target, mutex->address, mutex->key);
 		}
 	}
-- 
2.11.0

[PATCH net-next 04/15] nfp: move mutex code out of nfp_cppcore.c

[Jakub Kicinski]

After mutex cache removal we can put the mutex code in a separate
source file.  This makes it clear it doesn't play with internals
of struct nfp_cpp any more.

No functional changes.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/Makefile        |   1 +
 .../ethernet/netronome/nfp/nfpcore/nfp_cppcore.c   | 304 ------------------
 .../net/ethernet/netronome/nfp/nfpcore/nfp_mutex.c | 345 +++++++++++++++++++++
 3 files changed, 346 insertions(+), 304 deletions(-)
 create mode 100644 drivers/net/ethernet/netronome/nfp/nfpcore/nfp_mutex.c

diff --git a/drivers/net/ethernet/netronome/nfp/Makefile b/drivers/net/ethernet/netronome/nfp/Makefile
index 6933afa69df2..4a5d13ef92a4 100644
--- a/drivers/net/ethernet/netronome/nfp/Makefile
+++ b/drivers/net/ethernet/netronome/nfp/Makefile
@@ -6,6 +6,7 @@ nfp-objs := \
 	    nfpcore/nfp_cpplib.o \
 	    nfpcore/nfp_hwinfo.o \
 	    nfpcore/nfp_mip.o \
+	    nfpcore/nfp_mutex.o \
 	    nfpcore/nfp_nffw.o \
 	    nfpcore/nfp_nsp.o \
 	    nfpcore/nfp_nsp_eth.o \
diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
index 6337342c5b62..62aa7bcee93d 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
@@ -65,14 +65,6 @@ struct nfp_cpp_resource {
 	u64 end;
 };
 
-struct nfp_cpp_mutex {
-	struct nfp_cpp *cpp;
-	int target;
-	u16 depth;
-	unsigned long long address;
-	u32 key;
-};
-
 struct nfp_cpp {
 	struct device dev;
 
@@ -1403,299 +1395,3 @@ void *nfp_cpp_explicit_priv(struct nfp_cpp_explicit *cpp_explicit)
 {
 	return &cpp_explicit[1];
 }
-
-/* THIS FUNCTION IS NOT EXPORTED */
-static u32 nfp_mutex_locked(u16 interface)
-{
-	return (u32)interface << 16 | 0x000f;
-}
-
-static u32 nfp_mutex_unlocked(u16 interface)
-{
-	return (u32)interface << 16 | 0x0000;
-}
-
-static bool nfp_mutex_is_locked(u32 val)
-{
-	return (val & 0xffff) == 0x000f;
-}
-
-static bool nfp_mutex_is_unlocked(u32 val)
-{
-	return (val & 0xffff) == 0000;
-}
-
-/* If you need more than 65536 recursive locks, please rethink your code. */
-#define MUTEX_DEPTH_MAX         0xffff
-
-static int
-nfp_cpp_mutex_validate(u16 interface, int *target, unsigned long long address)
-{
-	/* Not permitted on invalid interfaces */
-	if (NFP_CPP_INTERFACE_TYPE_of(interface) ==
-	    NFP_CPP_INTERFACE_TYPE_INVALID)
-		return -EINVAL;
-
-	/* Address must be 64-bit aligned */
-	if (address & 7)
-		return -EINVAL;
-
-	if (*target != NFP_CPP_TARGET_MU)
-		return -EINVAL;
-
-	return 0;
-}
-
-/**
- * nfp_cpp_mutex_init() - Initialize a mutex location
- * @cpp:	NFP CPP handle
- * @target:	NFP CPP target ID (ie NFP_CPP_TARGET_CLS or NFP_CPP_TARGET_MU)
- * @address:	Offset into the address space of the NFP CPP target ID
- * @key:	Unique 32-bit value for this mutex
- *
- * The CPP target:address must point to a 64-bit aligned location, and
- * will initialize 64 bits of data at the location.
- *
- * This creates the initial mutex state, as locked by this
- * nfp_cpp_interface().
- *
- * This function should only be called when setting up
- * the initial lock state upon boot-up of the system.
- *
- * Return: 0 on success, or -errno on failure
- */
-int nfp_cpp_mutex_init(struct nfp_cpp *cpp,
-		       int target, unsigned long long address, u32 key)
-{
-	const u32 muw = NFP_CPP_ID(target, 4, 0);    /* atomic_write */
-	u16 interface = nfp_cpp_interface(cpp);
-	int err;
-
-	err = nfp_cpp_mutex_validate(interface, &target, address);
-	if (err)
-		return err;
-
-	err = nfp_cpp_writel(cpp, muw, address + 4, key);
-	if (err)
-		return err;
-
-	err = nfp_cpp_writel(cpp, muw, address, nfp_mutex_locked(interface));
-	if (err)
-		return err;
-
-	return 0;
-}
-
-/**
- * nfp_cpp_mutex_alloc() - Create a mutex handle
- * @cpp:	NFP CPP handle
- * @target:	NFP CPP target ID (ie NFP_CPP_TARGET_CLS or NFP_CPP_TARGET_MU)
- * @address:	Offset into the address space of the NFP CPP target ID
- * @key:	32-bit unique key (must match the key at this location)
- *
- * The CPP target:address must point to a 64-bit aligned location, and
- * reserve 64 bits of data at the location for use by the handle.
- *
- * Only target/address pairs that point to entities that support the
- * MU Atomic Engine's CmpAndSwap32 command are supported.
- *
- * Return:	A non-NULL struct nfp_cpp_mutex * on success, NULL on failure.
- */
-struct nfp_cpp_mutex *nfp_cpp_mutex_alloc(struct nfp_cpp *cpp, int target,
-					  unsigned long long address, u32 key)
-{
-	const u32 mur = NFP_CPP_ID(target, 3, 0);    /* atomic_read */
-	u16 interface = nfp_cpp_interface(cpp);
-	struct nfp_cpp_mutex *mutex;
-	int err;
-	u32 tmp;
-
-	err = nfp_cpp_mutex_validate(interface, &target, address);
-	if (err)
-		return NULL;
-
-	err = nfp_cpp_readl(cpp, mur, address + 4, &tmp);
-	if (err < 0)
-		return NULL;
-
-	if (tmp != key)
-		return NULL;
-
-	mutex = kzalloc(sizeof(*mutex), GFP_KERNEL);
-	if (!mutex)
-		return NULL;
-
-	mutex->cpp = cpp;
-	mutex->target = target;
-	mutex->address = address;
-	mutex->key = key;
-	mutex->depth = 0;
-
-	return mutex;
-}
-
-/**
- * nfp_cpp_mutex_free() - Free a mutex handle - does not alter the lock state
- * @mutex:	NFP CPP Mutex handle
- */
-void nfp_cpp_mutex_free(struct nfp_cpp_mutex *mutex)
-{
-	kfree(mutex);
-}
-
-/**
- * nfp_cpp_mutex_lock() - Lock a mutex handle, using the NFP MU Atomic Engine
- * @mutex:	NFP CPP Mutex handle
- *
- * Return: 0 on success, or -errno on failure
- */
-int nfp_cpp_mutex_lock(struct nfp_cpp_mutex *mutex)
-{
-	unsigned long warn_at = jiffies + 15 * HZ;
-	unsigned int timeout_ms = 1;
-	int err;
-
-	/* We can't use a waitqueue here, because the unlocker
-	 * might be on a separate CPU.
-	 *
-	 * So just wait for now.
-	 */
-	for (;;) {
-		err = nfp_cpp_mutex_trylock(mutex);
-		if (err != -EBUSY)
-			break;
-
-		err = msleep_interruptible(timeout_ms);
-		if (err != 0)
-			return -ERESTARTSYS;
-
-		if (time_is_before_eq_jiffies(warn_at)) {
-			warn_at = jiffies + 60 * HZ;
-			dev_warn(mutex->cpp->dev.parent,
-				 "Warning: waiting for NFP mutex [depth:%hd target:%d addr:%llx key:%08x]\n",
-				 mutex->depth,
-				 mutex->target, mutex->address, mutex->key);
-		}
-	}
-
-	return err;
-}
-
-/**
- * nfp_cpp_mutex_unlock() - Unlock a mutex handle, using the MU Atomic Engine
- * @mutex:	NFP CPP Mutex handle
- *
- * Return: 0 on success, or -errno on failure
- */
-int nfp_cpp_mutex_unlock(struct nfp_cpp_mutex *mutex)
-{
-	const u32 muw = NFP_CPP_ID(mutex->target, 4, 0);    /* atomic_write */
-	const u32 mur = NFP_CPP_ID(mutex->target, 3, 0);    /* atomic_read */
-	struct nfp_cpp *cpp = mutex->cpp;
-	u32 key, value;
-	u16 interface;
-	int err;
-
-	interface = nfp_cpp_interface(cpp);
-
-	if (mutex->depth > 1) {
-		mutex->depth--;
-		return 0;
-	}
-
-	err = nfp_cpp_readl(mutex->cpp, mur, mutex->address + 4, &key);
-	if (err < 0)
-		return err;
-
-	if (key != mutex->key)
-		return -EPERM;
-
-	err = nfp_cpp_readl(mutex->cpp, mur, mutex->address, &value);
-	if (err < 0)
-		return err;
-
-	if (value != nfp_mutex_locked(interface))
-		return -EACCES;
-
-	err = nfp_cpp_writel(cpp, muw, mutex->address,
-			     nfp_mutex_unlocked(interface));
-	if (err < 0)
-		return err;
-
-	mutex->depth = 0;
-	return 0;
-}
-
-/**
- * nfp_cpp_mutex_trylock() - Attempt to lock a mutex handle
- * @mutex:	NFP CPP Mutex handle
- *
- * Return:      0 if the lock succeeded, -errno on failure
- */
-int nfp_cpp_mutex_trylock(struct nfp_cpp_mutex *mutex)
-{
-	const u32 muw = NFP_CPP_ID(mutex->target, 4, 0);    /* atomic_write */
-	const u32 mus = NFP_CPP_ID(mutex->target, 5, 3);    /* test_set_imm */
-	const u32 mur = NFP_CPP_ID(mutex->target, 3, 0);    /* atomic_read */
-	struct nfp_cpp *cpp = mutex->cpp;
-	u32 key, value, tmp;
-	int err;
-
-	if (mutex->depth > 0) {
-		if (mutex->depth == MUTEX_DEPTH_MAX)
-			return -E2BIG;
-		mutex->depth++;
-		return 0;
-	}
-
-	/* Verify that the lock marker is not damaged */
-	err = nfp_cpp_readl(cpp, mur, mutex->address + 4, &key);
-	if (err < 0)
-		return err;
-
-	if (key != mutex->key)
-		return -EPERM;
-
-	/* Compare against the unlocked state, and if true,
-	 * write the interface id into the top 16 bits, and
-	 * mark as locked.
-	 */
-	value = nfp_mutex_locked(nfp_cpp_interface(cpp));
-
-	/* We use test_set_imm here, as it implies a read
-	 * of the current state, and sets the bits in the
-	 * bytemask of the command to 1s. Since the mutex
-	 * is guaranteed to be 64-bit aligned, the bytemask
-	 * of this 32-bit command is ensured to be 8'b00001111,
-	 * which implies that the lower 4 bits will be set to
-	 * ones regardless of the initial state.
-	 *
-	 * Since this is a 'Readback' operation, with no Pull
-	 * data, we can treat this as a normal Push (read)
-	 * atomic, which returns the original value.
-	 */
-	err = nfp_cpp_readl(cpp, mus, mutex->address, &tmp);
-	if (err < 0)
-		return err;
-
-	/* Was it unlocked? */
-	if (nfp_mutex_is_unlocked(tmp)) {
-		/* The read value can only be 0x....0000 in the unlocked state.
-		 * If there was another contending for this lock, then
-		 * the lock state would be 0x....000f
-		 */
-
-		/* Write our owner ID into the lock
-		 * While not strictly necessary, this helps with
-		 * debug and bookkeeping.
-		 */
-		err = nfp_cpp_writel(cpp, muw, mutex->address, value);
-		if (err < 0)
-			return err;
-
-		mutex->depth = 1;
-		return 0;
-	}
-
-	return nfp_mutex_is_locked(tmp) ? -EBUSY : -EINVAL;
-}
diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_mutex.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_mutex.c
new file mode 100644
index 000000000000..8a99c189efa8
--- /dev/null
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_mutex.c
@@ -0,0 +1,345 @@
+/*
+ * Copyright (C) 2015-2017 Netronome Systems, Inc.
+ *
+ * This software is dual licensed under the GNU General License Version 2,
+ * June 1991 as shown in the file COPYING in the top-level directory of this
+ * source tree or the BSD 2-Clause License provided below.  You have the
+ * option to license this software under the complete terms of either license.
+ *
+ * The BSD 2-Clause License:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      1. Redistributions of source code must retain the above
+ *         copyright notice, this list of conditions and the following
+ *         disclaimer.
+ *
+ *      2. Redistributions in binary form must reproduce the above
+ *         copyright notice, this list of conditions and the following
+ *         disclaimer in the documentation and/or other materials
+ *         provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/delay.h>
+#include <linux/device.h>
+#include <linux/jiffies.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/wait.h>
+
+#include "nfp_cpp.h"
+#include "nfp6000/nfp6000.h"
+
+struct nfp_cpp_mutex {
+	struct nfp_cpp *cpp;
+	int target;
+	u16 depth;
+	unsigned long long address;
+	u32 key;
+};
+
+static u32 nfp_mutex_locked(u16 interface)
+{
+	return (u32)interface << 16 | 0x000f;
+}
+
+static u32 nfp_mutex_unlocked(u16 interface)
+{
+	return (u32)interface << 16 | 0x0000;
+}
+
+static bool nfp_mutex_is_locked(u32 val)
+{
+	return (val & 0xffff) == 0x000f;
+}
+
+static bool nfp_mutex_is_unlocked(u32 val)
+{
+	return (val & 0xffff) == 0000;
+}
+
+/* If you need more than 65536 recursive locks, please rethink your code. */
+#define NFP_MUTEX_DEPTH_MAX         0xffff
+
+static int
+nfp_cpp_mutex_validate(u16 interface, int *target, unsigned long long address)
+{
+	/* Not permitted on invalid interfaces */
+	if (NFP_CPP_INTERFACE_TYPE_of(interface) ==
+	    NFP_CPP_INTERFACE_TYPE_INVALID)
+		return -EINVAL;
+
+	/* Address must be 64-bit aligned */
+	if (address & 7)
+		return -EINVAL;
+
+	if (*target != NFP_CPP_TARGET_MU)
+		return -EINVAL;
+
+	return 0;
+}
+
+/**
+ * nfp_cpp_mutex_init() - Initialize a mutex location
+ * @cpp:	NFP CPP handle
+ * @target:	NFP CPP target ID (ie NFP_CPP_TARGET_CLS or NFP_CPP_TARGET_MU)
+ * @address:	Offset into the address space of the NFP CPP target ID
+ * @key:	Unique 32-bit value for this mutex
+ *
+ * The CPP target:address must point to a 64-bit aligned location, and
+ * will initialize 64 bits of data at the location.
+ *
+ * This creates the initial mutex state, as locked by this
+ * nfp_cpp_interface().
+ *
+ * This function should only be called when setting up
+ * the initial lock state upon boot-up of the system.
+ *
+ * Return: 0 on success, or -errno on failure
+ */
+int nfp_cpp_mutex_init(struct nfp_cpp *cpp,
+		       int target, unsigned long long address, u32 key)
+{
+	const u32 muw = NFP_CPP_ID(target, 4, 0);    /* atomic_write */
+	u16 interface = nfp_cpp_interface(cpp);
+	int err;
+
+	err = nfp_cpp_mutex_validate(interface, &target, address);
+	if (err)
+		return err;
+
+	err = nfp_cpp_writel(cpp, muw, address + 4, key);
+	if (err)
+		return err;
+
+	err = nfp_cpp_writel(cpp, muw, address, nfp_mutex_locked(interface));
+	if (err)
+		return err;
+
+	return 0;
+}
+
+/**
+ * nfp_cpp_mutex_alloc() - Create a mutex handle
+ * @cpp:	NFP CPP handle
+ * @target:	NFP CPP target ID (ie NFP_CPP_TARGET_CLS or NFP_CPP_TARGET_MU)
+ * @address:	Offset into the address space of the NFP CPP target ID
+ * @key:	32-bit unique key (must match the key at this location)
+ *
+ * The CPP target:address must point to a 64-bit aligned location, and
+ * reserve 64 bits of data at the location for use by the handle.
+ *
+ * Only target/address pairs that point to entities that support the
+ * MU Atomic Engine's CmpAndSwap32 command are supported.
+ *
+ * Return:	A non-NULL struct nfp_cpp_mutex * on success, NULL on failure.
+ */
+struct nfp_cpp_mutex *nfp_cpp_mutex_alloc(struct nfp_cpp *cpp, int target,
+					  unsigned long long address, u32 key)
+{
+	const u32 mur = NFP_CPP_ID(target, 3, 0);    /* atomic_read */
+	u16 interface = nfp_cpp_interface(cpp);
+	struct nfp_cpp_mutex *mutex;
+	int err;
+	u32 tmp;
+
+	err = nfp_cpp_mutex_validate(interface, &target, address);
+	if (err)
+		return NULL;
+
+	err = nfp_cpp_readl(cpp, mur, address + 4, &tmp);
+	if (err < 0)
+		return NULL;
+
+	if (tmp != key)
+		return NULL;
+
+	mutex = kzalloc(sizeof(*mutex), GFP_KERNEL);
+	if (!mutex)
+		return NULL;
+
+	mutex->cpp = cpp;
+	mutex->target = target;
+	mutex->address = address;
+	mutex->key = key;
+	mutex->depth = 0;
+
+	return mutex;
+}
+
+/**
+ * nfp_cpp_mutex_free() - Free a mutex handle - does not alter the lock state
+ * @mutex:	NFP CPP Mutex handle
+ */
+void nfp_cpp_mutex_free(struct nfp_cpp_mutex *mutex)
+{
+	kfree(mutex);
+}
+
+/**
+ * nfp_cpp_mutex_lock() - Lock a mutex handle, using the NFP MU Atomic Engine
+ * @mutex:	NFP CPP Mutex handle
+ *
+ * Return: 0 on success, or -errno on failure
+ */
+int nfp_cpp_mutex_lock(struct nfp_cpp_mutex *mutex)
+{
+	unsigned long warn_at = jiffies + 15 * HZ;
+	unsigned int timeout_ms = 1;
+	int err;
+
+	/* We can't use a waitqueue here, because the unlocker
+	 * might be on a separate CPU.
+	 *
+	 * So just wait for now.
+	 */
+	for (;;) {
+		err = nfp_cpp_mutex_trylock(mutex);
+		if (err != -EBUSY)
+			break;
+
+		err = msleep_interruptible(timeout_ms);
+		if (err != 0)
+			return -ERESTARTSYS;
+
+		if (time_is_before_eq_jiffies(warn_at)) {
+			warn_at = jiffies + 60 * HZ;
+			nfp_warn(mutex->cpp,
+				 "Warning: waiting for NFP mutex [depth:%hd target:%d addr:%llx key:%08x]\n",
+				 mutex->depth,
+				 mutex->target, mutex->address, mutex->key);
+		}
+	}
+
+	return err;
+}
+
+/**
+ * nfp_cpp_mutex_unlock() - Unlock a mutex handle, using the MU Atomic Engine
+ * @mutex:	NFP CPP Mutex handle
+ *
+ * Return: 0 on success, or -errno on failure
+ */
+int nfp_cpp_mutex_unlock(struct nfp_cpp_mutex *mutex)
+{
+	const u32 muw = NFP_CPP_ID(mutex->target, 4, 0);    /* atomic_write */
+	const u32 mur = NFP_CPP_ID(mutex->target, 3, 0);    /* atomic_read */
+	struct nfp_cpp *cpp = mutex->cpp;
+	u32 key, value;
+	u16 interface;
+	int err;
+
+	interface = nfp_cpp_interface(cpp);
+
+	if (mutex->depth > 1) {
+		mutex->depth--;
+		return 0;
+	}
+
+	err = nfp_cpp_readl(mutex->cpp, mur, mutex->address + 4, &key);
+	if (err < 0)
+		return err;
+
+	if (key != mutex->key)
+		return -EPERM;
+
+	err = nfp_cpp_readl(mutex->cpp, mur, mutex->address, &value);
+	if (err < 0)
+		return err;
+
+	if (value != nfp_mutex_locked(interface))
+		return -EACCES;
+
+	err = nfp_cpp_writel(cpp, muw, mutex->address,
+			     nfp_mutex_unlocked(interface));
+	if (err < 0)
+		return err;
+
+	mutex->depth = 0;
+	return 0;
+}
+
+/**
+ * nfp_cpp_mutex_trylock() - Attempt to lock a mutex handle
+ * @mutex:	NFP CPP Mutex handle
+ *
+ * Return:      0 if the lock succeeded, -errno on failure
+ */
+int nfp_cpp_mutex_trylock(struct nfp_cpp_mutex *mutex)
+{
+	const u32 muw = NFP_CPP_ID(mutex->target, 4, 0);    /* atomic_write */
+	const u32 mus = NFP_CPP_ID(mutex->target, 5, 3);    /* test_set_imm */
+	const u32 mur = NFP_CPP_ID(mutex->target, 3, 0);    /* atomic_read */
+	struct nfp_cpp *cpp = mutex->cpp;
+	u32 key, value, tmp;
+	int err;
+
+	if (mutex->depth > 0) {
+		if (mutex->depth == NFP_MUTEX_DEPTH_MAX)
+			return -E2BIG;
+		mutex->depth++;
+		return 0;
+	}
+
+	/* Verify that the lock marker is not damaged */
+	err = nfp_cpp_readl(cpp, mur, mutex->address + 4, &key);
+	if (err < 0)
+		return err;
+
+	if (key != mutex->key)
+		return -EPERM;
+
+	/* Compare against the unlocked state, and if true,
+	 * write the interface id into the top 16 bits, and
+	 * mark as locked.
+	 */
+	value = nfp_mutex_locked(nfp_cpp_interface(cpp));
+
+	/* We use test_set_imm here, as it implies a read
+	 * of the current state, and sets the bits in the
+	 * bytemask of the command to 1s. Since the mutex
+	 * is guaranteed to be 64-bit aligned, the bytemask
+	 * of this 32-bit command is ensured to be 8'b00001111,
+	 * which implies that the lower 4 bits will be set to
+	 * ones regardless of the initial state.
+	 *
+	 * Since this is a 'Readback' operation, with no Pull
+	 * data, we can treat this as a normal Push (read)
+	 * atomic, which returns the original value.
+	 */
+	err = nfp_cpp_readl(cpp, mus, mutex->address, &tmp);
+	if (err < 0)
+		return err;
+
+	/* Was it unlocked? */
+	if (nfp_mutex_is_unlocked(tmp)) {
+		/* The read value can only be 0x....0000 in the unlocked state.
+		 * If there was another contending for this lock, then
+		 * the lock state would be 0x....000f
+		 */
+
+		/* Write our owner ID into the lock
+		 * While not strictly necessary, this helps with
+		 * debug and bookkeeping.
+		 */
+		err = nfp_cpp_writel(cpp, muw, mutex->address, value);
+		if (err < 0)
+			return err;
+
+		mutex->depth = 1;
+		return 0;
+	}
+
+	return nfp_mutex_is_locked(tmp) ? -EBUSY : -EINVAL;
+}
-- 
2.11.0

[PATCH net-next 05/15] nfp: document expected locking in the core

[Jakub Kicinski]

Document which fields of nfp_cpp are protected by which locks.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 .../ethernet/netronome/nfp/nfpcore/nfp_cppcore.c   | 33 ++++++++++++++++++----
 1 file changed, 27 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
index 62aa7bcee93d..4e08362d8c97 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
@@ -65,28 +65,49 @@ struct nfp_cpp_resource {
 	u64 end;
 };
 
+/**
+ * struct nfp_cpp - main nfpcore device structure
+ * Following fields are read-only after probe() exits or netdevs are spawned.
+ * @dev:		embedded device structure
+ * @op:			low-level implementation ops
+ * @priv:		private data of the low-level implementation
+ * @model:		chip model
+ * @interface:		chip interface id we are using to reach it
+ * @serial:		chip serial number
+ * @imb_cat_table:	CPP Mapping Table
+ *
+ * Following fields can be used only in probe() or with rtnl held:
+ * @hwinfo:		HWInfo database fetched from the device
+ * @rtsym:		firmware run time symbols
+ *
+ * Following fields use explicit locking:
+ * @resource_list:	NFP CPP resource list
+ * @resource_lock:	protects @resource_list
+ *
+ * @area_cache_list:	cached areas for cpp/xpb read/write speed up
+ * @area_cache_mutex:	protects @area_cache_list
+ *
+ * @waitq:		area wait queue
+ */
 struct nfp_cpp {
 	struct device dev;
 
-	void *priv; /* Private data of the low-level implementation */
+	void *priv;
 
 	u32 model;
 	u16 interface;
 	u8 serial[NFP_SERIAL_LEN];
 
 	const struct nfp_cpp_operations *op;
-	struct list_head resource_list;	/* NFP CPP resource list */
+	struct list_head resource_list;
 	rwlock_t resource_lock;
 	wait_queue_head_t waitq;
 
-	/* NFP6000 CPP Mapping Table */
 	u32 imb_cat_table[16];
 
-	/* Cached areas for cpp/xpb readl/writel speedups */
-	struct mutex area_cache_mutex;  /* Lock for the area cache */
+	struct mutex area_cache_mutex;
 	struct list_head area_cache_list;
 
-	/* Cached information */
 	void *hwinfo;
 	void *rtsym;
 };
-- 
2.11.0

[PATCH net-next 06/15] nfp: lock area cache earlier

[Jakub Kicinski]

We shouldn't access area_cache_list without its lock even
to check if it's empty.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c | 14 ++++++++------
 1 file changed, 8 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
index 4e08362d8c97..5189fedb0f4f 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
@@ -821,10 +821,7 @@ area_cache_get(struct nfp_cpp *cpp, u32 id,
 	 * the need for special case code below when
 	 * checking against available cache size.
 	 */
-	if (length == 0)
-		return NULL;
-
-	if (list_empty(&cpp->area_cache_list) || id == 0)
+	if (length == 0 || id == 0)
 		return NULL;
 
 	/* Remap from cpp_island to cpp_target */
@@ -832,10 +829,15 @@ area_cache_get(struct nfp_cpp *cpp, u32 id,
 	if (err < 0)
 		return NULL;
 
-	addr += *offset;
-
 	mutex_lock(&cpp->area_cache_mutex);
 
+	if (list_empty(&cpp->area_cache_list)) {
+		mutex_unlock(&cpp->area_cache_mutex);
+		return NULL;
+	}
+
+	addr += *offset;
+
 	/* See if we have a match */
 	list_for_each_entry(cache, &cpp->area_cache_list, entry) {
 		if (id == cache->id &&
-- 
2.11.0

[PATCH net-next 07/15] nfp: correct return codes when msleep gets interrupted

[Jakub Kicinski]

msleep_interruptible() returns time left to wait, not error
code.  Return ERESTARTSYS when interrupted.

While at it correct a comment and make the polling a bit
more aggressive.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp.c
index 34c50987c377..17822ae4a17f 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp.c
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_nsp.c
@@ -209,9 +209,8 @@ nfp_nsp_wait_reg(struct nfp_cpp *cpp, u64 *reg,
 		if ((*reg & mask) == val)
 			return 0;
 
-		err = msleep_interruptible(100);
-		if (err)
-			return err;
+		if (msleep_interruptible(25))
+			return -ERESTARTSYS;
 
 		if (time_after(start_time, wait_until))
 			return -ETIMEDOUT;
@@ -228,7 +227,7 @@ nfp_nsp_wait_reg(struct nfp_cpp *cpp, u64 *reg,
  *
  * Return: 0 for success with no result
  *
- *	 1..255 for NSP completion with a result code
+ *	 positive value for NSP completion with a result code
  *
  *	-EAGAIN if the NSP is not yet present
  *	-ENODEV if the NSP is not a supported model
@@ -380,9 +379,10 @@ int nfp_nsp_wait(struct nfp_nsp *state)
 		if (err != -EAGAIN)
 			break;
 
-		err = msleep_interruptible(100);
-		if (err)
+		if (msleep_interruptible(25)) {
+			err = -ERESTARTSYS;
 			break;
+		}
 
 		if (time_after(start_time, wait_until)) {
 			err = -ETIMEDOUT;
-- 
2.11.0

[PATCH net-next 08/15] nfp: don't ignore return value of wait_event_interruptible

[Jakub Kicinski]

When signal interrupts waiting for an area to become available
we assume success.  Pay attention to the return code.  Unpack
the code a little bit to make it more readable.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 .../ethernet/netronome/nfp/nfpcore/nfp_cppcore.c   | 56 +++++++++++++++-------
 1 file changed, 38 insertions(+), 18 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
index 5189fedb0f4f..2e4796b52b84 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
@@ -411,9 +411,43 @@ nfp_cpp_area_alloc(struct nfp_cpp *cpp, u32 dest,
  */
 void nfp_cpp_area_free(struct nfp_cpp_area *area)
 {
+	if (atomic_read(&area->refcount))
+		nfp_warn(area->cpp, "Warning: freeing busy area\n");
 	nfp_cpp_area_put(area);
 }
 
+static bool nfp_cpp_area_acquire_try(struct nfp_cpp_area *area, int *status)
+{
+	*status = area->cpp->op->area_acquire(area);
+
+	return *status != -EAGAIN;
+}
+
+static int __nfp_cpp_area_acquire(struct nfp_cpp_area *area)
+{
+	int err, status;
+
+	if (atomic_inc_return(&area->refcount) > 1)
+		return 0;
+
+	if (!area->cpp->op->area_acquire)
+		return 0;
+
+	err = wait_event_interruptible(area->cpp->waitq,
+				       nfp_cpp_area_acquire_try(area, &status));
+	if (!err)
+		err = status;
+	if (err) {
+		nfp_warn(area->cpp, "Warning: area wait failed: %d\n", err);
+		atomic_dec(&area->refcount);
+		return err;
+	}
+
+	nfp_cpp_area_get(area);
+
+	return 0;
+}
+
 /**
  * nfp_cpp_area_acquire() - lock down a CPP area for access
  * @area:	CPP area handle
@@ -425,27 +459,13 @@ void nfp_cpp_area_free(struct nfp_cpp_area *area)
  */
 int nfp_cpp_area_acquire(struct nfp_cpp_area *area)
 {
-	mutex_lock(&area->mutex);
-	if (atomic_inc_return(&area->refcount) == 1) {
-		int (*a_a)(struct nfp_cpp_area *);
-
-		a_a = area->cpp->op->area_acquire;
-		if (a_a) {
-			int err;
+	int ret;
 
-			wait_event_interruptible(area->cpp->waitq,
-						 (err = a_a(area)) != -EAGAIN);
-			if (err < 0) {
-				atomic_dec(&area->refcount);
-				mutex_unlock(&area->mutex);
-				return err;
-			}
-		}
-	}
+	mutex_lock(&area->mutex);
+	ret = __nfp_cpp_area_acquire(area);
 	mutex_unlock(&area->mutex);
 
-	nfp_cpp_area_get(area);
-	return 0;
+	return ret;
 }
 
 /**
-- 
2.11.0

[PATCH net-next 09/15] nfp: fix invalid area detection

[Jakub Kicinski]

Core should detect when someone is trying to request an access
window which is too large for a given type of access.  Otherwise
the requester will be put on a wait queue for ever without any
error message.

Add const qualifiers to clarify that we are only looking at read-
-only members in relevant functions.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 .../ethernet/netronome/nfp/nfpcore/nfp6000_pcie.c  | 29 +++++++++++-----------
 1 file changed, 15 insertions(+), 14 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp6000_pcie.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp6000_pcie.c
index 15cc3e77cf6a..43dc68e01274 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp6000_pcie.c
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp6000_pcie.c
@@ -217,7 +217,7 @@ static resource_size_t nfp_bar_resource_start(struct nfp_bar *bar)
 #define TARGET_WIDTH_64    8
 
 static int
-compute_bar(struct nfp6000_pcie *nfp, struct nfp_bar *bar,
+compute_bar(const struct nfp6000_pcie *nfp, const struct nfp_bar *bar,
 	    u32 *bar_config, u64 *bar_base,
 	    int tgt, int act, int tok, u64 offset, size_t size, int width)
 {
@@ -410,35 +410,36 @@ find_matching_bar(struct nfp6000_pcie *nfp,
 
 /* Return EAGAIN if no resource is available */
 static int
-find_unused_bar_noblock(struct nfp6000_pcie *nfp,
+find_unused_bar_noblock(const struct nfp6000_pcie *nfp,
 			int tgt, int act, int tok,
 			u64 offset, size_t size, int width)
 {
-	int n, invalid = 0;
+	int n, busy = 0;
 
 	for (n = 0; n < nfp->bars; n++) {
-		struct nfp_bar *bar = &nfp->bar[n];
+		const struct nfp_bar *bar = &nfp->bar[n];
 		int err;
 
-		if (bar->bitsize == 0) {
-			invalid++;
-			continue;
-		}
-
-		if (atomic_read(&bar->refcnt) != 0)
+		if (!bar->bitsize)
 			continue;
 
 		/* Just check to see if we can make it fit... */
 		err = compute_bar(nfp, bar, NULL, NULL,
 				  tgt, act, tok, offset, size, width);
+		if (err)
+			continue;
 
-		if (err < 0)
-			invalid++;
-		else
+		if (!atomic_read(&bar->refcnt))
 			return n;
+
+		busy++;
 	}
 
-	return (n == invalid) ? -EINVAL : -EAGAIN;
+	if (WARN(!busy, "No suitable BAR found for request tgt:0x%x act:0x%x tok:0x%x off:0x%llx size:%zd width:%d\n",
+		 tgt, act, tok, offset, size, width))
+		return -EINVAL;
+
+	return -EAGAIN;
 }
 
 static int
-- 
2.11.0

[PATCH net-next 10/15] nfp: fix nfp_cpp_read()/nfp_cpp_write() error paths

[Jakub Kicinski]

When acquiring an area fails we can't call function doing both
release and free.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c | 15 +++++++++------
 1 file changed, 9 insertions(+), 6 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
index 2e4796b52b84..e2abba4c3a3f 100644
--- a/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
+++ b/drivers/net/ethernet/netronome/nfp/nfpcore/nfp_cppcore.c
@@ -951,12 +951,14 @@ int nfp_cpp_read(struct nfp_cpp *cpp, u32 destination,
 			return -ENOMEM;
 
 		err = nfp_cpp_area_acquire(area);
-		if (err)
-			goto out;
+		if (err) {
+			nfp_cpp_area_free(area);
+			return err;
+		}
 	}
 
 	err = nfp_cpp_area_read(area, offset, kernel_vaddr, length);
-out:
+
 	if (cache)
 		area_cache_put(cpp, cache);
 	else
@@ -993,13 +995,14 @@ int nfp_cpp_write(struct nfp_cpp *cpp, u32 destination,
 			return -ENOMEM;
 
 		err = nfp_cpp_area_acquire(area);
-		if (err)
-			goto out;
+		if (err) {
+			nfp_cpp_area_free(area);
+			return err;
+		}
 	}
 
 	err = nfp_cpp_area_write(area, offset, kernel_vaddr, length);
 
-out:
 	if (cache)
 		area_cache_put(cpp, cache);
 	else
-- 
2.11.0

[PATCH net-next 11/15] nfp: don't use netdev_warn() before netdev is registered

[Jakub Kicinski]

Fix warning which was using netdev_warn() instead of dev_warn()
to early.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_net_common.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index f134f1808b9a..19f9d95faea4 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -336,9 +336,9 @@ nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
 
 	if (dp->num_rx_rings > dp->num_r_vecs ||
 	    dp->num_tx_rings > dp->num_r_vecs)
-		nn_warn(nn, "More rings (%d,%d) than vectors (%d).\n",
-			dp->num_rx_rings, dp->num_tx_rings,
-			dp->num_r_vecs);
+		dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n",
+			 dp->num_rx_rings, dp->num_tx_rings,
+			 dp->num_r_vecs);
 
 	dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings);
 	dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings);
-- 
2.11.0

[PATCH net-next 12/15] nfp: remove RX queue pointers

[Jakub Kicinski]

NFP6000 doesn't use queue pointers/doorbells for RX, it uses
'done' bit in descriptors.  Remove the pointers from data structures.
Since we are saving space in rx_ring structure make fields we
previously compressed to 16bits word size again.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_net.h         |  8 ++------
 drivers/net/ethernet/netronome/nfp/nfp_net_common.c  |  3 ---
 drivers/net/ethernet/netronome/nfp/nfp_net_debugfs.c | 15 ++++-----------
 3 files changed, 6 insertions(+), 20 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net.h b/drivers/net/ethernet/netronome/nfp/nfp_net.h
index 4d45f4573b57..8e04aa0e6e87 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net.h
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net.h
@@ -307,9 +307,7 @@ struct nfp_net_rx_buf {
  * @rd_p:       FL/RX ring read pointer (free running)
  * @idx:        Ring index from Linux's perspective
  * @fl_qcidx:   Queue Controller Peripheral (QCP) queue index for the freelist
- * @rx_qcidx:   Queue Controller Peripheral (QCP) queue index for the RX queue
  * @qcp_fl:     Pointer to base of the QCP freelist queue
- * @qcp_rx:     Pointer to base of the QCP RX queue
  * @wr_ptr_add: Accumulated number of buffers to add to QCP write pointer
  *              (used for free list batching)
  * @rxbufs:     Array of transmitted FL/RX buffers
@@ -324,13 +322,11 @@ struct nfp_net_rx_ring {
 	u32 wr_p;
 	u32 rd_p;
 
-	u16 idx;
-	u16 wr_ptr_add;
+	u32 idx;
+	u32 wr_ptr_add;
 
 	int fl_qcidx;
-	int rx_qcidx;
 	u8 __iomem *qcp_fl;
-	u8 __iomem *qcp_rx;
 
 	struct nfp_net_rx_buf *rxbufs;
 	struct nfp_net_rx_desc *rxds;
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index 19f9d95faea4..255294b8bc5f 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -479,10 +479,7 @@ nfp_net_rx_ring_init(struct nfp_net_rx_ring *rx_ring,
 	rx_ring->r_vec = r_vec;
 
 	rx_ring->fl_qcidx = rx_ring->idx * nn->stride_rx;
-	rx_ring->rx_qcidx = rx_ring->fl_qcidx + (nn->stride_rx - 1);
-
 	rx_ring->qcp_fl = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->fl_qcidx);
-	rx_ring->qcp_rx = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->rx_qcidx);
 }
 
 /**
diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_debugfs.c b/drivers/net/ethernet/netronome/nfp/nfp_net_debugfs.c
index 74125584260b..4077c59bf782 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_debugfs.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_debugfs.c
@@ -40,9 +40,9 @@ static struct dentry *nfp_dir;
 
 static int nfp_net_debugfs_rx_q_read(struct seq_file *file, void *data)
 {
-	int fl_rd_p, fl_wr_p, rx_rd_p, rx_wr_p, rxd_cnt;
 	struct nfp_net_r_vector *r_vec = file->private;
 	struct nfp_net_rx_ring *rx_ring;
+	int fl_rd_p, fl_wr_p, rxd_cnt;
 	struct nfp_net_rx_desc *rxd;
 	struct nfp_net *nn;
 	void *frag;
@@ -61,14 +61,11 @@ static int nfp_net_debugfs_rx_q_read(struct seq_file *file, void *data)
 
 	fl_rd_p = nfp_qcp_rd_ptr_read(rx_ring->qcp_fl);
 	fl_wr_p = nfp_qcp_wr_ptr_read(rx_ring->qcp_fl);
-	rx_rd_p = nfp_qcp_rd_ptr_read(rx_ring->qcp_rx);
-	rx_wr_p = nfp_qcp_wr_ptr_read(rx_ring->qcp_rx);
 
-	seq_printf(file, "RX[%02d,%02d,%02d]: cnt=%d dma=%pad host=%p   H_RD=%d H_WR=%d FL_RD=%d FL_WR=%d RX_RD=%d RX_WR=%d\n",
-		   rx_ring->idx, rx_ring->fl_qcidx, rx_ring->rx_qcidx,
+	seq_printf(file, "RX[%02d,%02d]: cnt=%d dma=%pad host=%p   H_RD=%d H_WR=%d FL_RD=%d FL_WR=%d\n",
+		   rx_ring->idx, rx_ring->fl_qcidx,
 		   rx_ring->cnt, &rx_ring->dma, rx_ring->rxds,
-		   rx_ring->rd_p, rx_ring->wr_p,
-		   fl_rd_p, fl_wr_p, rx_rd_p, rx_wr_p);
+		   rx_ring->rd_p, rx_ring->wr_p, fl_rd_p, fl_wr_p);
 
 	for (i = 0; i < rxd_cnt; i++) {
 		rxd = &rx_ring->rxds[i];
@@ -91,10 +88,6 @@ static int nfp_net_debugfs_rx_q_read(struct seq_file *file, void *data)
 			seq_puts(file, " FL_RD");
 		if (i == fl_wr_p % rxd_cnt)
 			seq_puts(file, " FL_WR");
-		if (i == rx_rd_p % rxd_cnt)
-			seq_puts(file, " RX_RD");
-		if (i == rx_wr_p % rxd_cnt)
-			seq_puts(file, " RX_WR");
 
 		seq_putc(file, '\n');
 	}
-- 
2.11.0

[PATCH net-next 13/15] nfp: flush xmit_more on error paths

[Jakub Kicinski]

In case of ring full or DMA mapping error remember to flush xmit_more
delayed kicks.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_net_common.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index 255294b8bc5f..d35eeba86bac 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -759,6 +759,7 @@ static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
 		nn_dp_warn(dp, "TX ring %d busy. wrp=%u rdp=%u\n",
 			   qidx, tx_ring->wr_p, tx_ring->rd_p);
 		netif_tx_stop_queue(nd_q);
+		nfp_net_tx_xmit_more_flush(tx_ring);
 		u64_stats_update_begin(&r_vec->tx_sync);
 		r_vec->tx_busy++;
 		u64_stats_update_end(&r_vec->tx_sync);
@@ -867,6 +868,7 @@ static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
 	tx_ring->txbufs[wr_idx].fidx = -2;
 err_free:
 	nn_dp_warn(dp, "Failed to map DMA TX buffer\n");
+	nfp_net_tx_xmit_more_flush(tx_ring);
 	u64_stats_update_begin(&r_vec->tx_sync);
 	r_vec->tx_errors++;
 	u64_stats_update_end(&r_vec->tx_sync);
-- 
2.11.0

[PATCH net-next 14/15] nfp: remove defensive checks around ndo_open()/ndo_close()

[Jakub Kicinski]

Device open and close handlers check if the device is already
in the desired state.  Thanks to our reconfig infrastructure
this should not be necessary, there doesn't seem to be any
code in the driver which depends on it.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 drivers/net/ethernet/netronome/nfp/nfp_net_common.c | 10 ----------
 1 file changed, 10 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index d35eeba86bac..e12353a7c83c 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -2233,11 +2233,6 @@ static int nfp_net_netdev_open(struct net_device *netdev)
 	struct nfp_net *nn = netdev_priv(netdev);
 	int err, r;
 
-	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_ENABLE) {
-		nn_err(nn, "Dev is already enabled: 0x%08x\n", nn->dp.ctrl);
-		return -EBUSY;
-	}
-
 	/* Step 1: Allocate resources for rings and the like
 	 * - Request interrupts
 	 * - Allocate RX and TX ring resources
@@ -2368,11 +2363,6 @@ static int nfp_net_netdev_close(struct net_device *netdev)
 {
 	struct nfp_net *nn = netdev_priv(netdev);
 
-	if (!(nn->dp.ctrl & NFP_NET_CFG_CTRL_ENABLE)) {
-		nn_err(nn, "Dev is not up: 0x%08x\n", nn->dp.ctrl);
-		return 0;
-	}
-
 	/* Step 1: Disable RX and TX rings from the Linux kernel perspective
 	 */
 	nfp_net_close_stack(nn);
-- 
2.11.0

[PATCH net-next 15/15] nfp: disable FW on reconfiguration errors

[Jakub Kicinski]

Since we no longer need to keep the FW enabled for .ndo_close()
to work we can always stop FW after reconfiguration failure.
This seems to make most FWs more resilient to faults (at least
in error injection scenarios).

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
---
 .../net/ethernet/netronome/nfp/nfp_net_common.c    | 29 ++++++++--------------
 1 file changed, 11 insertions(+), 18 deletions(-)

diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
index e12353a7c83c..8f2da128ce0f 100644
--- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
+++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
@@ -2127,7 +2127,11 @@ nfp_net_tx_ring_hw_cfg_write(struct nfp_net *nn,
 	nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), tx_ring->r_vec->irq_entry);
 }
 
-static int __nfp_net_set_config_and_enable(struct nfp_net *nn)
+/**
+ * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
+ * @nn:      NFP Net device to reconfigure
+ */
+static int nfp_net_set_config_and_enable(struct nfp_net *nn)
 {
 	u32 new_ctrl, update = 0;
 	unsigned int r;
@@ -2176,6 +2180,10 @@ static int __nfp_net_set_config_and_enable(struct nfp_net *nn)
 
 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
 	err = nfp_net_reconfig(nn, update);
+	if (err) {
+		nfp_net_clear_config_and_disable(nn);
+		return err;
+	}
 
 	nn->dp.ctrl = new_ctrl;
 
@@ -2191,22 +2199,7 @@ static int __nfp_net_set_config_and_enable(struct nfp_net *nn)
 		udp_tunnel_get_rx_info(nn->dp.netdev);
 	}
 
-	return err;
-}
-
-/**
- * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
- * @nn:      NFP Net device to reconfigure
- */
-static int nfp_net_set_config_and_enable(struct nfp_net *nn)
-{
-	int err;
-
-	err = __nfp_net_set_config_and_enable(nn);
-	if (err)
-		nfp_net_clear_config_and_disable(nn);
-
-	return err;
+	return 0;
 }
 
 /**
@@ -2447,7 +2440,7 @@ static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp)
 			return err;
 	}
 
-	return __nfp_net_set_config_and_enable(nn);
+	return nfp_net_set_config_and_enable(nn);
 }
 
 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn)
-- 
2.11.0

Re: [PATCH net-next 11/15] nfp: don't use netdev_warn() before netdev is registered

[Sergei Shtylyov]

Hello!

On 3/22/2017 3:59 AM, Jakub Kicinski wrote:

> Fix warning which was using netdev_warn() instead of dev_warn()

    I'm seeing nn_warn()...

> to early.

    Too. :-)

> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> ---
>  drivers/net/ethernet/netronome/nfp/nfp_net_common.c | 6 +++---
>  1 file changed, 3 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
> index f134f1808b9a..19f9d95faea4 100644
> --- a/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
> +++ b/drivers/net/ethernet/netronome/nfp/nfp_net_common.c
> @@ -336,9 +336,9 @@ nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
>
>  	if (dp->num_rx_rings > dp->num_r_vecs ||
>  	    dp->num_tx_rings > dp->num_r_vecs)
> -		nn_warn(nn, "More rings (%d,%d) than vectors (%d).\n",
> -			dp->num_rx_rings, dp->num_tx_rings,
> -			dp->num_r_vecs);
> +		dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n",
> +			 dp->num_rx_rings, dp->num_tx_rings,
> +			 dp->num_r_vecs);
>
>  	dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings);
>  	dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings);

MBR, Sergei

Re: [PATCH net-next 00/15] nfp: allow concurrency in core and minor fixes

[David Miller]

From: Jakub Kicinski <jakub.kicinski@netronome.com>
Date: Tue, 21 Mar 2017 17:59:06 -0700

> The first 10 patches of this series prepare nfpcore for concurrent 
> accesses.  This will be needed by upcoming hwmon and devlink patches.
> Most locking is already in place, the patches in this series iron out
> a few bugs.
> 
> Last 5 patches are fixes and cleanups to the netdev code, including
> removal of doorbell pointers used only on old versions of the chip,
> removal of unnecessarily defensive code and flushing xmit_more more
> carefully on error paths.

Series applied, thank you.