xref: /illumos-gate/usr/src/uts/common/io/qede/579xx/drivers/ecore/documentation/osal.txt (revision 14b24e2b)

Ecore Operating System Abstraction Layer (osal) documentation
=============================================================

Introduction
============
This document contains the osal information - functions utilized by the ecore
which the various parties incorporating the ecore into their own sourcecode
need to implement in order for the ecore to function.

Everything osal should be prefixed with osal*. Structures should be prefixed by
lowercase `osal_' and be lowercase themselves, while functions should be
prefixed with upper case letters `OSAL_' and be uppercased.

Notice that some are very trivial, and can be easily replaced by a single
function call or preprocessor macro. Also notice some are weakly-typed, or
include values [e.g., struct types] that should probably be implemented
as preprocessor macros.

* - as always, there are a couple of exceptions.

Time related osals
==================
- OSAL_UDELAY(int)
	The function should delay for said amount of micro-seconds.

- OSAL_MSLEEP(int)
	The function should sleep for said amount of mili-seconds, releasing the CPU.
	(The existing of such a call in a flow demands the flow to be run from a
	context supporting sleep).

Memory related osals
====================
- void* OSAL_ALLOC(struct *ecore_dev, u32 mode, u32 size)
	Allocate `size' memory for said device. The mode comes from linux's GFP_*
	defines [Notice this means you need to define those values, at least
	KERNEL & ATOMIC even if you don't need them on your OS].
	Return value should be NULL if allocation fails, and otherwise return the
	pointer to allocated memory.

- void* OSAL_ZALLOC(struct *ecore_dev, u32 mode, u32 size)
	Like OSAL_ALLOC, only the memory [if succesfully allocated] should be set
	to 0s.

- void* OSAL_CALLOC(struct *ecore_dev, u32 mode, u32 num, u32 elem_size)
	Like OSAL_ALLOC, only this allocates memory sufficient for `num' elements,
	each of size `elem_size'.

- void* OSAL_VZALLOC(struct *ecore_dev, u32 size)
	Allocate `size' memory for said device, but as opposed to OSAL_ALLOC,
	the allocated memory is only virtually contiguous and not necessarily
	physically contiguous. Content of memory should be zeroed.
	Return value should be NULL if allocation fails, and otherwise return the
	pointer to allocated memory.

- void OSAL_FREE(struct *ecore_dev, void* memory)
	frees a memory previously dynamically allocated by OSAL_([Z|C]?)ALLOC.
	Notice this should succeed even if the pointer is NULL.

- void OSAL_VFREE(struct *ecore_dev, void* memory)
	frees a memory previously dynamically allocated by OSAL_VZALLOC.
	Notice this should succeed even if the pointer is NULL.

- void OSAL_MEM_ZERO(void* memory, u32 size)
	Set `size' bytes starting at the address pointed by `memory' with 0s.

- void OSAL_MEMCPY(void* dst, void* src, u32 size)
	Copy `size' bytes from address pointed by `src' to address pointed by `dst'.

- void OSAL_MEMSET(void *s, int c, u32  size)
	Set `size' bytes starting at the address pointed to by s with c.

- int OSAL_MEMCMP(void *s1, void *s2, u32 n)
	Comapre the first n bytes of memory areas s1 and s2.
	The function returns an integer less than, equal to, or
	greater than zero if the first n bytes of s1 is found, respectively,
	to be less than, to match, or be greater than the first n bytes of
	s2.

dma_addr_t - this variable type should be defined as a variable that can contain
	physical addresses. It's utilized in OSAL_DMA_* functions.

- void* OSAL_DMA_ALLOC_COHERENT(struct *ecore_dev, dma_addr_t *phys, u32 size)
	Allocate `size' bytes of DMA-able memory [which will later be accessed by
	our HW]. The physical address will by filled in `phys' [NULL if allocation
	fails], and the virtual address will return [NULL if allocation fails].

- void OSAL_DMA_FREE_COHERENT(struct *ecore_dev, void* virt, dma_addr_t *phys, u32 size)
	Frees previously allocated memory [via OSAL_DMA_ALLOC_COHERENT].

Memory Access related osals
===========================
Notice - these do not begin with the OSAL convention.

- void REG_WR(struct *ecore_hwfn, u32 addr, u32 value)
- void REG_WR16(struct *ecore_hwfn, u32 addr, u16 value)
	These should perform the memory write toward chip. Address is an offset
	inside the Bar.

- void DOORBELL(struct *ecore_hwfn, u32 addr, u32 value)
	This should perform memory write toward chip. Addrees is an offset
	inside the doorbell bar.

- void DIRECT_REG_WR(struct *ecore_hwfn, u32 addr, u32 value)
	Perform a memory write toward chip. Address is absolute [i.e., includes
	bar offset in memory].
	Notice most ecore clients don't need the ecore_hwfn; Notice that unless
	ECORE_CONFIG_DIRECT_HWFN is set during compilation, this function will
	be called sometimes when the ecore_hwfn is set to be OSAL_NULL.

- void DIRECT_REG_RD(struct *ecore_hwfn, u32 addr)
	Perform a memory read from chip. Address is absolute.
	[Same comment as for DIRECT_REG_WR]

- u32 REG_RD(struct *ecore_hwfn, u32 addr)
	This should perform the memory read from chip. Address is an offset
	inside the Bar.

Synchronization related osals
=============================
For most upper-layer driver there's no need to allocate memory for lock primitives -
for such upper-layer drivers, it should suffice to give an empty implementations
to - MUTEX_ALLOC(), MUTEX_DEALLOC(), SPIN_LOCK_ALLOC(), SPIN_LOCK_DEALLOC().
In case this is actually needed, the ecore should also be compile with the
CONFIG_ECORE_LOCK_ALLOC flag [since sometimes flows diverge a bit].

osal_spinlock_t - a type of variable [struct] which is used as a spinlock in the
		  various OSAL_SPIN_* calls. Spinlock assumes locking is blocking
		  and interrupt disabling on all CPUs, protecting an exclusive
		  critical section.

osal_mutex_t - a type of variable [struct] which is used as a mutex in the
	       various OSAL_MUTEX_* calls. Mutices assume locking is sleeping,
	       protecting an exclusive critical section.

- void OSAL_MUTEX_ALLOC(struct ecore_hwfn*, osal_mutex*)
- void OSAL_SPIN_LOCK_ALLOC(struct ecore_hwfn*, osal_spinlock_t*)
	Used to allocate [if upper-layer needs to] the spinlock/mutex.
	NOTICE - counter-intuitively ecore assuumes this always succeeds.
	But upper-layer can assume this will only be called by rescource alloc &
	init schemes [and not dynamically], and thus prepare the memory before-hand.

- void OSAL_MUTEX_DEALLOC(osal_mutex*)
- void OSAL_SPIN_LOCK_DEALLOC(osal_spinlock_t*)
	Used to free any previously allocated memory of the spinlock/mutex.

- void OSAL_MUTEX_INIT(osal_mutex_t*)
- void OSAL_SPIN_LOCK_INIT(osal_spinlock_t*)
	Initializes the given spinlock/mutex.

- void OSAL_MUTEX_ACQUIRE(osal_mutex_t*)
- void OSAL_SPIN_LOCK(osal_spinlock_t*)
	Wait until spinlock/mutex is exclusively held - blocking for spinlocks,
	sleeping for mutices.
- void OSAL_SPIN_LOCK_IRQSAVE(osal_spinlock_t*, flags)
	Wait until spinlock/mutex is exclusively held - blocking for spinlocks,
	sleeping for mutices. Use the type of spinlock that also syncronizes
	with irqs. Relevant only for some OSs, in most cases an implementation
	that is identical to OSAL_SPIN_LOCK will suffice.

- OSAL_MUTEX_RELEASE(osal_mutex_t*)
- void OSAL_SPIN_UNLOCK(osal_spinlock_t*)
	Release a held spinlock/mutex [no logic necessary for checking whether
	it's already locked; that's the caller's responsibility].
- void OSAL_SPIN_UNLOCK_IRQSAVE(osal_spinlock_t*, flags)
	Release a held spinlock/mutex [no logic necessary for checking whether
	it's already locked; that's the caller's responsibility]. Should be
	called on a lock that was taken using OSAL_SPIN_LOCK_IRQSAVE.
	Relevant only for some OSs, in most cases an implementation
	that is identical to OSAL_SPIN_LOCK will suffice.

osal_dpc_t - a variable representing a non-sleepable context, used by
	     bcm_osal_dpc_* calls.

- osal_dpc_t OSAL_DPC_ALLOC(struct ecore_hwfn)
	Allocates the dpc.

- void OSAL_DPC_INIT(osal_dpc_t dpc, struct ecore_hwfn)
	Initializes the dpc.

- void OSAL_DPC_SYNC(struct ecore_hwfn *p_hwfn)
	Make sure all slowpath OS dpc's are synced/flushed.

- void OSAL_POLL_MODE_DPC(struct ecore_hnfn*)
	Polls ecore_int_sp_dpc during blocking ramrods if needed.  Used when
	no other thread will service the sp dpc, such as with single threaded
	systems.

Notice that the DPC interface from ecore is probably the worst [or at least
one-of the worst] implementation in the API - this API does not determine
when the DPC will run and what will run in its context.
At the moment, the only DPC used by the ecore is sp_dpc in ecore_int.c, and
the missing logic here is that it should somehow be connected by the upper
layer to the slowpath interrupt ISR [either directly or indirectly], and that
it should run [or call in some manner] `ecore_int_sp_dpc'.

Linked List osals
=================
osal_list_t - this struct is the linked-list root, i.e., from it the linked
	      list can be travelled; Most OSAL_LIST_* functions will require
	      such an anchor to be passed.
osal_list_entry_t - each struct which will be added to a list should contain
		    such a field.

- void OSAL_LIST_INIT(osal_list_t*)
	Initializes the list passed to the function [so it can be utilized
	later on by other OSAL_LIST_* calls].

- void OSAL_LIST_PUSH_HEAD(osal_list_entry_t *entry, osal_list_t *list)
- void OSAL_LIST_PUSH_TAIL(osal_list_entry_t *entry, osal_list_t *list)
	Add `entry' to the beginning/end of a list anchored by `list'.

- type* OSAL_LIST_FIRST_ENTRY(osal_list_t*, type, name_of_list_field)
	[Notice - this is weakly typed and should probably only be
	implemented as a macro]
	Assume the first entry in the list is of type `type', and that
	the osal_list_t entry inside it is named `name_of_list_field'.
	Return a pointer to that entry.

	E.g., a short example of list functions uses:

	struct foo {
		osal_struct_t list_head;
		int x;
	};
	osal_struct_list_t root;
	struct foo foo1, *tmp_foo;

	OSAL_LIST_INIT(&root)
	foo1.x = 10;
	OSAL_LIST_PUSH_HEAD(&foo1.head, &root);
	tmp_foo = OSAL_LIST_FIRST_ENTRY(&root, struct foo, list_head)
	/* tmp_foo now points to foo1 */

- void OSAL_LIST_REMOVE_ENTRY(osal_list_entry_t *entry, osal_list_t *list)
	Removes `entry' from `list' [Notice no need to check that entry is
	part of list; That's the caller's responsiblity]

- bool OSAL_LIST_IS_EMPTY(osal_list_t*)
	return true iff the list passed to the function contains at least one
	element.

- void OSAL_LIST_FOR_EACH_ENTRY(type *entry, osal_list_t* list,
				type, name_of_list_field)
- void OSAL_LIST_FOR_EACH_ENTRY_SAFE(type *entry, type *tmp_entry,
				     osal_list_t* list,
				     type, name_of_list_field)
	[This is a loop macro, which opens a new statement block and should
	 probably be implemented as a preprocessor macro]
	In every iteration of this for-loop `entry' will be filled with the
	subsequent entry from the list anchored at `list'. The type of the
	elements of the list and the name of the osal_list_entry_t fields
	they contain should be explicitly stated [as is the case in
	OSAL_LIST_FIRST_ENTRY].
	The 'SAFE' variant requires an additional temporary variable which
	points to the same type as `entry' (`tmp_entry'), and is safe in the
	sense that `entry' can be removed from the list and the iteration
	logic should not break.
	If breaking from the loop one can assume `tmp_entry' would point to
	the element it pointed to while breaking. But no assumption can be
	made on its value assuming the iteration runs on all list elements.

 - void OSAL_LIST_INSERT_ENTRY_BEFORE(type *new_entry, type *entry,
                                                            osal_list_t* list)
   Insert a new entry before the specified entry in the list,
   or insert at head of list if specified entry is 0.

- void OSAL_LIST_INSERT_ENTRY_AFTER(type *new_entry, type *entry,
                                                            osal_list_t* list)
   Insert a new entry after the specified entry in the list,
   or insert at tail of list if specified entry is 0.

- void OSAL_LIST_SPLICE_INIT(osal_list_t* new_list, osal_list_t* list)
   Update new_list by splicing list to the head of new_list.

- void OSAL_LIST_SPLICE_TAIL_INIT(osal_list_t* new_list, osal_list_t* list)
   Update new_list by splicing list to the tail of new_list.

PCI access osals
================
- void OSAL_PCI_READ_CONFIG_BYTE(struct *ecore_dev, u32 address, u8 *dst)
- void OSAL_PCI_READ_CONFIG_WORD(struct *ecore_dev, u32 address, u16 *dst)
- void OSAL_PCI_READ_CONFIG_DWORD(struct *ecore_dev, u32 address, u32 *dst)
	Read the appropraite data size and fill in `dst' from offset
	`address' in the device's configuration space.
- void OSAL_PCI_WRITE_CONFIG_WORD(struct *ecore_dev, u32 address, u16 val)
	Write the given value at the offset `address' in the device's
	configuration space.
- int OSAL_PCI_FIND_CAPABILITY(struct *ecore_dev, u16 pcie_id)
	Returns the offset in the PCI configuration space of the PCIe capability
	`pcie_id', or 0 if no such if capability is found.
- int OSAL_PCI_FIND_EXT_CAPABILITY(struct *ecore_dev, u16 pcie_id)
	Returns the offset in the PCI configuration space of the PCIe extended
	capability `pcie_id', or 0 if no such if capability is found.
- u32 OSAL_BAR_SIZE(struct *ecore_dev, u8 bar_id)
	Returns the PCI bar size of the bar specified by `bar_id' - id 0 refers
	to the regview bar and id 1 to the doorbell bar.

Memory Barriers
===============
- void OSAL_MMIOWB(strcut *ecore_dev)
- void OSAL_BARRIER(struct *ecore_dev)
- void OSAL_SMP_RMB(struct *ecore_dev)
- void OSAL_SMP_WMB(struct *ecore_dev)
- void OSAL_RMB(struct *ecore_dev)
- void OSAL_WMB(struct *ecore_dev)

Memory barriers mostly follow the linux's definition of memory barriers,
as can be found in the linux kernel under `Documentation/memory-barriers.txt'.
It's possible that certain Operating systems will not require all of said
memory barriers, e.g., if there's no need to support weak-memory ordered
system with the OS; In such a case, they should implement the barriers as
empty functions.
Also notice that on most implementations the ecore_dev pointer isn't needed for
the barrier implementation.

- void OSAL_DMA_SYNC(struct *ecore_dev, void* addr, u32 size, bool is_post)
	Used in upper-layer drivers that need to sync memory and caches
	surrounding a dma transaction; This will be called with
	'is_post == false' prior to the DMA transaction, and 'is_post == true'
	following it.
	`addr' will be the physical address of the source buffer.

Bit operations
==============
- void OSAL_SET_BIT(u8 bit, unsigned long *bitmap)
	Set a bit in a bitmap; Logically *bitmap |= (1ULL << bit)

- void OSAL_CLEAR_BIT(u8 bit, unsigned long *bitmap)
	Clears a bit in a bitmap; Logically *bitmap &= ~(1ULL << bit)

- bool OSAL_TEST_BIT(u8 bit, unsigned long *bitmap)
	Tests whether a bit is set in a bitmap.
	Logically !!(*bitmap & (1ULL << bit))

- bool OSAL_TEST_AND_CLEAR_BIT(u8 bit, unsigned long *bitmap)
	Tests whether a bit is set in a bitmap and clears it.
	Logically returns !!(*bitmap & (1ULL << bit)) and then performs
	*bitmap &= ~(1ULL << bit)

- bool OSAL_TEST_AND_FLIP_BIT(u8 bit, unsigned long *bitmap)
	Tests whether a bit is set in a bitmap and flips it.
	Logically returns !!(*bitmap & (1ULL << bit)) and then performs
	*bitmap ^= (1ULL << bit)

- u8 OSAL_FIND_FIRST_ZERO_BIT(unsigned long *bitmap, u8 length)
	Returns the bit-index of the first non-set bit in a bitmap of
	length `length'. Logically, returns min bit s.t.
	(bit < length) && !(*bitmap & (1ULL << bit))

- u8 OSAL_FIND_FIRST_BIT(unsigned long *bitmap, u8 length)
        Returns the bit-index of the first non-set bit in a bitmap of
        length `length'. Logically, returns min bit s.t.
        (bit < length) && (*bitmap & (1ULL << bit)).
	In case no bit is set, should return length.

- int OSAL_BITMAP_WEIGHT(u32 *bitmap, u32 nbits)
	Returns the Hamming weight (number of set bits) in the bitmap.

Endianess
=========
OSAL_BE32 - a variable representing a 32-bit data in BE format.
- __be32 OSAL_CPU_TO_BE32(u32 val)
	convert a u32 val into big-endian format.

- __be64 OSAL_CPU_TO_BE64(u64 val)
	convert a u64 val into big-endian format

- u32 OSAL_BE32_TO_CPU(u32 val)
	convert a 32-bit value in big-endian format into current architecture.

- __be16 OSAL_CPU_TO_BE16(u16 val)
	convert a u16 val into big-endian format.

- u16 OSAL_BE16_TO_CPU(u16 val)
	convert a 16-bit value in big-endian format into current architecture.

- u32 OSAL_CPU_TO_LE32(u32 val)
	convert a 32-bit value from native byte ordering to little endian format.

- u16 OSAL_CPU_TO_LE16(u16 val)
	convert a 16-bit value from native byte ordering to little endian format.

- u32 OSAL_LE32_TO_CPU(u32 val)
	convert a 32-bit little endian value to native byte ordering.

- u16 OSAL_LE16_TO_CPU(u32 val)
	convert a 16-bit little endian value to native byte ordering.

Physical link
=============
- void OSAL_LINK_UPDATE(struct ecore_hwfn*)
	ecore uses this callback to inform upper-layer driver that a link
	change has been informed by the MFW, and the link state is parsed
	inside the hwfn->mcp_info->link_output.
	Notice this is called from non-sleepable context, so it's possible
	upper layer driver will have to schedule in its implementation to
	process the change in some other context.

Single Root IOV
===============
- ? OSAL_VF_SEND_MSG2PF(struct *ecore_dev, u8* done, union vfpf_tlvs *msg,
			union pfvf_tlvs *reply_addr, u32 msg_size,
			u32 reply_size)
	This should be implemented by drivers utilizing a SW channel
	[as opposed to the ecore hw_channel] for sending messages from VF
	to PF. This should send a buffer pointed by `msg' of length `msg_size'
	to the PF, where the PF's answer should be written to `reply_addr'
	of maximum size `reply_size'. Writing to `done' should signal the
	transaction is complete.

- int OSAL_PF_VF_MSG(struct ecore_hwfn*, u8 relative_vfid)
	This should be implemented by drivers utilizing the HW channel
	for sending messages from VF to PF. This is called on the PF by
	ecore on the EQ-handling context to signal to upper-layer that
	the vf with the relative vfid has sent a message, and allow the
	upper-layer to handle the request. Should return 0 on success,
	non-zero otherwise.

- void OSAL_PF_VF_MALICIOUS(struct ecore_hwfn*, u8 relative_vfid)
	This indicates to ecore-client that the specific VF is now considered
	malicious by FW. Exact implication may vary [I.e., it's possible that
	the hw-channel is still operational or it might be disabled, depending
	on the exact nature of the vf maliciousness].

- enum _ecore_status OSAL_IOV_CHK_UCAST(struct *ecore_hwfn,
					int vfid,
					struct ecore_filter_ucast *params)
	Called on the PF whenever the VF requests to configure a unicast
	filter, to allow the upper layer to decide whether to allow the
	unicast configuration or not. Should return ECORE_SUCCESS if
	configuration is allowed, ECORE_EXISTS if it's already configured
	and ECORE_INVAL if its forbidden.

- enum _ecore_status
  OSAL_IOV_PRE_START_VPORT(struct *ecore_hwfn,
			   int relative_vf_id,
			   struct ecore_sp_vport_start_params *params)
	Called on the PF before starting a VF's vport, to check the
	OS-specific setting for that vport.

- void OSAL_IOV_POST_START_VPORT(struct *ecore_hwfn,
				 int relative_vf_id,
				 u8 vport_id,
				 u16 opaque_fid)
	Called on the PF after succesfully starting a VF's vport,
	to allow the upper layer the chance of doing some additional
	OS-specific work.

- enum _ecore_status_t
  OSAL_IOV_VF_VPORT_UPDATE(struct *ecore_hwfn, u8 relative_vfid,
			   struct *ecore_sp_vport_update_params,
			   u16 *p_tlvs_mask)
 	Called on PF side to allow upper-client to manipulate the configuration
        of a vport update request by VF.
        `p_tlv_mask' is a bit mask of ECORE_IOV_VP_UPDATE_* values, and it's
        the ecore-client responsibility to clear bits which are NOT to be
	configured.
	If the ecore-client wants to fail the configuaration altogether, it
	should return a value other than ECORE_SUCCESS.

- void OSAL_IOV_VF_VPORT_STOP(struct ecore_hwfn *p_hwfn,
			       struct ecore_vf_info *vf)
 	Called on PF side to allow upper-client to to execute required operations
        before closing vport.

- void OSAL_VF_FLR_UPDATE(struct *ecore_hwfn)
	Called on PF side to indicate MFW signaled some of the PF's
	vfs were Function Level Resetted. When this is called,
	pending_flr field in the PF's iov information should be set
	to be a bitmask of all relative VFIDs that need FLR.

- void OSAL_IOV_VF_CLEANUP(struct *ecore_hwfn, u8 relative_vfid)
	Called on PF side to indicate that VF has been cleaned, to allow
	ecore-client the change of cleaning any database it had for that vf.

- void OSAL_VF_FILL_ACQUIRE_RESC_REQ(struct ecore_hwfn *p_hwfn,
                                     struct vf_pf_resc_request *p_resc_req,
				     struct ecore_vf_acquire_sw_info *p_sw_info)
        Called on the VF before sending acquire messgae to the PF,
        to allow the upper layer to fill the num of requested resources.

- enum _ecore_status_t
  OSAL_VF_UPDATE_ACQUIRE_RESC_RESP(struct ecore_hwfn *p_hwfn,
                                   struct pf_vf_resc *p_resc_resp)
        Called on the VF after acquire response has recieved to update
        upper layer with the acquired resources. The upper layer should
        return error in case there is a problem with the amount of acquired
        resources, otherwise success.

- enum _ecore_status_t
  OSAL_IOV_VF_ACQUIRE(struct ecore_hwfn *p_hwfn, u8 relative_vfid)
	Called on PF side while processing a VF acquisition message, to allow
        the ecore-client to prevent the acquisition if required.

- enum _ecore_status_t
  OSAL_VF_CQE_COMPLETION(struct ecore_hwfn *p_hwfn,
                         struct eth_slow_path_rx_cqe *cqe,
                         enum protocol_type protocol)
        Called on the VF to let the upper layer chance to handle the cqe.

- u8 OSAL_IOV_GET_OS_TYPE()
	Called on PF side - should return a VFPF_ACQUIRE_OS_* value matching
	the OS hypervisor type.

- void OSAL_IOV_VF_MSG_TYPE(struct ecore_hwfn *p_hwfn,
                            u8 vf_id,
                            int vf_msg_type)
	Called on PF side to indicate  the VF to PF TLV Request type send by VF in
	the MailBox request through HW Channel.

- void OSAL_IOV_PF_RESP_TYPE(struct ecore_hwfn *p_hwfn,
                             u8 vf_id,
                             enum ecore_iov_pf_to_vf_status pf_resp_type)
	Called on PF side to indicate the PF to VF Response type after the processing
        of VF Mailbox request through HW Channel.

- int OSAL_PF_VALIDATE_MODIFY_TUNN_CONFIG(struct ecore_hwfn *p_hwfn,
					  u16 *feature_mask, bool *update,
					  struct ecore_tunnel_info *p_tun)
	Called on PF side to validate and modify VF's requested tunnel mode,
	classes and udp ports based on update masks/flags. In case PF modifies
	VF requested data then it should still return an error
	to indicate to VF flow. Finally, If any tunnel configuration update
	is required then they must set "true" in "update" address.
	PF should also set the features in "feature_mask" which are enabled
	or requested from VF to be enabled and can be kept intact
	[i.e they can no longer be modified further by PF in terms of tunn
	mode or tun classes].

HW errors & Recovery
====================
- void OSAL_SCHEDULE_RECOVERY_HANDLER(struct ecore_hwfn* p_hwfn)
	ecore uses this callback to inform the upper-layer driver that a process
	kill indication has been received from the MFW, and that a recovery
	handler should be scheduled to handle the recovery flow.

- void OSAL_HW_ERROR_OCCURRED(struct ecore_hwfn *p_hwfn,
			      enum ecore_hw_err_type err_type)
	ecore uses this callback to inform the upper-layer driver about an error
	in the HW/FW.

Unzipping functionality related osal
====================================
- u32 OSAL_UNZIP_DATA(struct ecore_hwfn *p_hwfn, u32 input_len, u8 *input_buf,
		      u32 max_size, u8 *unzip_buf)
	Unzip zipped data of length "input_len" present in "input_buf" and
	write unzipped data into "unzip_buf". "max_size" is max length of unzipped
	data which can be written into "unzip_buf".
	returns the length of unzipped data in dwords, in case of failure returns 0.

Note -	All drivers has to define/set CONFIG_ECORE_ZIPPED_FW to take zipped
	firmware file and implement their own functionality to unzip data.

String handling related osals
=============================
- int OSAL_SPRINTF(char *str, const char *format, ...)
	Write a formatted output to a string pointed to by str.

- int OSAL_SNPRINTF(char *str, osal_size_t n, const char *format, ...)
	Write a formatted output to a string pointed to by str, taking n as the maximum number of characters to write.
	The remaining characters are discarded and not stored, but counted for the value returned by the function.

- u32 OSAL_STRLEN(const char *str)
	Return the length of the string pointed to by str, excluding the terminating null byte ('\0').

- char *OSAL_STRCPY(char *dest, const char *src)
	Copy the string pointed to by src, including the terminating null byte ('\0'), to the buffer  pointed to by dest.

- char *OSAL_STRNCPY(char *dest, const char *src, osal_size_t n)
	Copies up to n characters from the string pointed to by src to dest.
	In the case where the length of src is less than that of n, the remainder of dest will be padded with null bytes.

- int OSAL_STRCMP(const char *str1, const char *str2)
	Compares the string pointed by str1 to the string pointed by str2.

- int OSAL_STRTOUL(const char *str, unsigned int base, unsigned long *res)
	Converts a string pointed by str to an unsigned long.
	The string base can be given explicitly, or "0" can be provided and then it will be automatically detected with the conventional semantics
	(begins with 0x - the number will be parsed as a hexadecimal, if it otherwise begins with 0 - octal, otherwise - decimal).
	The the result of the conversion is written to the variable pointed by res.

Miscellaneous
=============
osal_size_t  - The type is used to store the result of sizeof operator.
		On certain platfroms it is typedefed to __SIZE_TYPE__, a
		compiler predefined macro.

osal_int_ptr_t - Integer type large enough to hold a pointer. The data type is
	          useful for casting pointers when we want to do address
	          arithmetic.

OSAL_NULL -  Null value defintion.

OSAL_INLINE - The specifier instructs the compiler to insert a copy of the
	       function body into each place the function is called.

OSAL_BUILD_BUG_ON(condition) - Break compilation if the "condition" is true.

OSAL_PAGE_SIZE - Macro for the page size value i.e., number of bytes in a
	          memory page.

OSAL_CACHE_LINE_SIZE - Macro for the cache line size value in bytes.

OSAL_IOMEM - Annotation used to mark pointers to I/O memory. It is used by
	      Sparse, a tool used to find possible coding faults in the
	      kernel. The annotation is ignored in the normal compilation.
	      When checking the code with sparse, however, developers will see
	      a whole new set of warnings caused by code which mixes normal
	      pointers with OSAL_IOMEM pointers, or which dereferences those
	      pointers.

OSAL_UNLIKELY(condition) -  An instruction to the compiler to emit instructions
	     in the favor of "condition" value likely to be false (zero).

- type OSAL_MIN_T(type, val1, val2)
	Returns the minimum of (val1, val2) which are of data type "type".

- type OSAL_MAX_T(type, val1, val2)
	Returns the maximum of (val1, val2) which are of data type "type".

- u32* OSAL_REG_ADDR(struct *ecore_hwfn, u32 hw_offset)
	Returns the memory address value of the device at the offset hw_offset.

- u32 OSAL_NUM_ACTIVE_CPU()
	Returns the number of active CPUs on the machine

- u32 DIV_ROUND_UP(u32 size, u32 divisor_size)
	Returns number of elements required of `divisor_size' each,
        required to hold at least `size' data.

- u32 ROUNDUP(u32 size, u32 divisor_size)
	Returns DIV_ROUND_UP * divisor_size [how much memory will be
	required for holding `size' data, if we can only allocate
	in granularity of `divisor_size'].

- u32 OSAL_ROUNDUP_POW_OF_TWO(u32)
	Return a round-up to closest value which is a power of 2.

- u32 OSAL_LOG2(u32)
	Returns a log on a 2-basis of the value.

- void OSAL_ASSERT(bool cond)
	Should perform some sort of panic in case `cond' is false.

- u32 OFFSETOF(struct, field)
	returns the offset in bytes of field inside struct.
	TBD - isn't it ANSI-C? If so, can be removed from ecore.

- void PRINT(void *dp_ctx, char *format_string, ...)
- void PRINT_ERR(void *dp_ctx, char *format_string, ...)
	Macros used by the ecore debug facilities to print verbose & error logs.

- void OSAL_WARN(bool condition, const char *format, ...)
	Should cause a warning [at least on debug mode] if hit.

- void OSAL_BEFORE_PF_START(struct file, engine id)
	Perform some action just before first interaction with FW.

- void OSAL_AFTER_PF_STOP(struct file, engine id)
	Perform some action just aftter last interaction with FW.

- u32 ARRAY_SIZE(array[])
	Returns the number of elements in an array.

- osal_uintptr_t - Used for casting pointer to integer value.
	Some 32 bit OS complains about "cast from pointer to integer of different size". As on those platform
	integers and pointers are of different sizes. Due to this fact we couldn't use any generic data type in
	ecore to cast pointer to integer data type which can be appropriate for both 32 bit and 64 bit platform
	for all OS distributions, as a data type size on each OS distribution may vary. So this osal serves the
	purpose of to define OS specific data type cast which can be used on both 32 bit and 64 bit platform.

- void OSAL_GET_PROTOCOL_STATS(struct *ecore_dev, enum ecore_mcp_protocol_type type,
			       union ecore_mcp_protocol_stats *stats);
	Call from the ecore to get the statististics of a protocol driver. Ecore client
	need to populate the requested statistics. If the PF has more than one function,
	driver should return the statistics sum of all the interfaces under the PF.

- int OSAL_SLOWPATH_IRQ_REQ(struct ecore_hwfn *p_hwfn)
	Call from ecore to the upper layer driver to request IRQs for the slowpath
	interrupts handling.

- u32 OSAL_GET_RDMA_SB_ID(struct ecore_hwfn *p_hwfn, u32 relative_sb_id)
	Call from ecore to obtain real SB ID from upper layer.

- void OSAL_DCBX_AEN(struct ecore_hwfn *p_hwfn, enum ecore_mib_read_type mib_type)
	Call from ecore to notify the dcbx related asynchronous events from the MFW.

- u32 OSAL_CRC32(u32 crc, u8 *buf, osal_size_t length)
	Compute a CRC-32 value using a Castagnoli polynomial.
	`crc' is the previous value for the checksum, and `buf' shall point to
	an array of `length' bytes of data to be added to this checksum.

- void OSAL_CRC8_POPULATE(u8 table[CRC8_TABLE_SIZE], u8 polynomial)
	Fill the provided crc table for given polynomial in reverse bit order (msb first).
	The table size is 256 ('CRC8_TABLE_SIZE').

- u8 OSAL_CRC8(const u8 table[CRC8_TABLE_SIZE], u8 *pdata, osal_size_t nbytes, u8 crc)
	Calculate a crc8 over the given input data.
	`table' is the crc table used for calculation, `pdata' is a pointer to data buffer,
	`nbytes' is the size of the data buffer, and `crc' is the previous returned crc8 value.

- s64 OSAL_DIV_S64(s64 value, s64 base)
	Some distros [32-bit] are having problems with direct division of 64-bit variables.
	This should logically return (value / base), and for distros that don't care about this
	sort of issues it can be defined to be exactly that.

NVRAM related osals
===================
- bool OSAL_NVM_IS_ACCESS_ENABLED(struct ecore_hwfn *p_hwfn)
        This osal allows to upper layer decide if it is allowed to access
        the NVRAM.

Management changes related osals
================================
- int OSAL_MFW_TLV_REQ(struct ecore_hwfn *p_hwfn)
	Called to inform that MFW has requested for TLVs. We are in the interrupt
	context here, ecore client need to schedule a thread/bottom-half context
	to handle this task, and return the control immediately.
	The bottom-half thread will need to invoke ecore_mfw_process_tlv_req()
	for further processing of the TLV request.

- int OSAL_MFW_FILL_TLV_DATA(struct ecore_hwfn *p_hwfn, enum ecore_mfw_tlv_type type,
                             union ecore_mfw_tlv_data *data)
	Called from ecore to get the TLV values of a given type. Ecore client
	need to fill in the values for all the fields that it's aware of, and
	also need to set the flags associated with the respective fields. For instance,
	if client sets value for 'npi_enabled' field, it needs to set the flag
	'npiv_enabled_set' to true.

- void OSAL_HW_INFO_CHANGE(struct ecore_hwfn *p_hwfn, enum ecore_hw_info_change)
	Called after management has changed some property of a hw_info field.
	The enum value indicates which field has changed. This is a FYI kind of
	notification - direct HW/FW changes [if applicable] were already done.