/* * Copyright (C) 2014 Freescale Semiconductor * * SPDX-License-Identifier: GPL-2.0+ */ #ifndef _FSL_QBMAN_BASE_H #define _FSL_QBMAN_BASE_H /* Descriptor for a QBMan instance on the SoC. On partitions/targets that do not * control this QBMan instance, these values may simply be place-holders. The * idea is simply that we be able to distinguish between them, eg. so that SWP * descriptors can identify which QBMan instance they belong to. */ struct qbman_block_desc { void *ccsr_reg_bar; /* CCSR register map */ int irq_rerr; /* Recoverable error interrupt line */ int irq_nrerr; /* Non-recoverable error interrupt line */ }; /* Descriptor for a QBMan software portal, expressed in terms that make sense to * the user context. Ie. on MC, this information is likely to be true-physical, * and instantiated statically at compile-time. On GPP, this information is * likely to be obtained via "discovery" over a partition's "layerscape bus" * (ie. in response to a MC portal command), and would take into account any * virtualisation of the GPP user's address space and/or interrupt numbering. */ struct qbman_swp_desc { const struct qbman_block_desc *block; /* The QBMan instance */ void *cena_bar; /* Cache-enabled portal register map */ void *cinh_bar; /* Cache-inhibited portal register map */ }; /* Driver object for managing a QBMan portal */ struct qbman_swp; /* Place-holder for FDs, we represent it via the simplest form that we need for * now. Different overlays may be needed to support different options, etc. (It * is impractical to define One True Struct, because the resulting encoding * routines (lots of read-modify-writes) would be worst-case performance whether * or not circumstances required them.) * * Note, as with all data-structures exchanged between software and hardware (be * they located in the portal register map or DMA'd to and from main-memory), * the driver ensures that the caller of the driver API sees the data-structures * in host-endianness. "struct qbman_fd" is no exception. The 32-bit words * contained within this structure are represented in host-endianness, even if * hardware always treats them as little-endian. As such, if any of these fields * are interpreted in a binary (rather than numerical) fashion by hardware * blocks (eg. accelerators), then the user should be careful. We illustrate * with an example; * * Suppose the desired behaviour of an accelerator is controlled by the "frc" * field of the FDs that are sent to it. Suppose also that the behaviour desired * by the user corresponds to an "frc" value which is expressed as the literal * sequence of bytes 0xfe, 0xed, 0xab, and 0xba. So "frc" should be the 32-bit * value in which 0xfe is the first byte and 0xba is the last byte, and as * hardware is little-endian, this amounts to a 32-bit "value" of 0xbaabedfe. If * the software is little-endian also, this can simply be achieved by setting * frc=0xbaabedfe. On the other hand, if software is big-endian, it should set * frc=0xfeedabba! The best away of avoiding trouble with this sort of thing is * to treat the 32-bit words as numerical values, in which the offset of a field * from the beginning of the first byte (as required or generated by hardware) * is numerically encoded by a left-shift (ie. by raising the field to a * corresponding power of 2). Ie. in the current example, software could set * "frc" in the following way, and it would work correctly on both little-endian * and big-endian operation; * fd.frc = (0xfe << 0) | (0xed << 8) | (0xab << 16) | (0xba << 24); */ struct qbman_fd { union { uint32_t words[8]; struct qbman_fd_simple { uint32_t addr_lo; uint32_t addr_hi; uint32_t len; /* offset in the MS 16 bits, BPID in the LS 16 bits */ uint32_t bpid_offset; uint32_t frc; /* frame context */ /* "err", "va", "cbmt", "asal", [...] */ uint32_t ctrl; /* flow context */ uint32_t flc_lo; uint32_t flc_hi; } simple; }; }; #endif /* !_FSL_QBMAN_BASE_H */