freebsd-dev/sys/dev/sfxge/common/efx_rx.c

1720 lines
40 KiB
C
Raw Normal View History

/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 2007-2016 Solarflare Communications Inc.
* All rights reserved.
*
* 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.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
* THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
* OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
* WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
* OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE,
* EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* The views and conclusions contained in the software and documentation are
* those of the authors and should not be interpreted as representing official
* policies, either expressed or implied, of the FreeBSD Project.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "efx.h"
#include "efx_impl.h"
#if EFSYS_OPT_SIENA
static __checkReturn efx_rc_t
siena_rx_init(
__in efx_nic_t *enp);
static void
siena_rx_fini(
__in efx_nic_t *enp);
#if EFSYS_OPT_RX_SCATTER
static __checkReturn efx_rc_t
siena_rx_scatter_enable(
__in efx_nic_t *enp,
__in unsigned int buf_size);
#endif /* EFSYS_OPT_RX_SCATTER */
#if EFSYS_OPT_RX_SCALE
static __checkReturn efx_rc_t
siena_rx_scale_mode_set(
__in efx_nic_t *enp,
__in uint32_t rss_context,
__in efx_rx_hash_alg_t alg,
__in efx_rx_hash_type_t type,
__in boolean_t insert);
static __checkReturn efx_rc_t
siena_rx_scale_key_set(
__in efx_nic_t *enp,
__in uint32_t rss_context,
__in_ecount(n) uint8_t *key,
__in size_t n);
static __checkReturn efx_rc_t
siena_rx_scale_tbl_set(
__in efx_nic_t *enp,
__in uint32_t rss_context,
__in_ecount(n) unsigned int *table,
__in size_t n);
static __checkReturn uint32_t
siena_rx_prefix_hash(
__in efx_nic_t *enp,
__in efx_rx_hash_alg_t func,
__in uint8_t *buffer);
#endif /* EFSYS_OPT_RX_SCALE */
static __checkReturn efx_rc_t
siena_rx_prefix_pktlen(
__in efx_nic_t *enp,
__in uint8_t *buffer,
__out uint16_t *lengthp);
static void
siena_rx_qpost(
__in efx_rxq_t *erp,
__in_ecount(ndescs) efsys_dma_addr_t *addrp,
__in size_t size,
__in unsigned int ndescs,
__in unsigned int completed,
__in unsigned int added);
static void
siena_rx_qpush(
__in efx_rxq_t *erp,
__in unsigned int added,
__inout unsigned int *pushedp);
#if EFSYS_OPT_RX_PACKED_STREAM
static void
siena_rx_qpush_ps_credits(
__in efx_rxq_t *erp);
static __checkReturn uint8_t *
siena_rx_qps_packet_info(
__in efx_rxq_t *erp,
__in uint8_t *buffer,
__in uint32_t buffer_length,
__in uint32_t current_offset,
__out uint16_t *lengthp,
__out uint32_t *next_offsetp,
__out uint32_t *timestamp);
#endif
static __checkReturn efx_rc_t
siena_rx_qflush(
__in efx_rxq_t *erp);
static void
siena_rx_qenable(
__in efx_rxq_t *erp);
static __checkReturn efx_rc_t
siena_rx_qcreate(
__in efx_nic_t *enp,
__in unsigned int index,
__in unsigned int label,
__in efx_rxq_type_t type,
__in_opt const efx_rxq_type_data_t *type_data,
__in efsys_mem_t *esmp,
__in size_t ndescs,
__in uint32_t id,
__in unsigned int flags,
__in efx_evq_t *eep,
__in efx_rxq_t *erp);
static void
siena_rx_qdestroy(
__in efx_rxq_t *erp);
#endif /* EFSYS_OPT_SIENA */
#if EFSYS_OPT_SIENA
static const efx_rx_ops_t __efx_rx_siena_ops = {
siena_rx_init, /* erxo_init */
siena_rx_fini, /* erxo_fini */
#if EFSYS_OPT_RX_SCATTER
siena_rx_scatter_enable, /* erxo_scatter_enable */
#endif
#if EFSYS_OPT_RX_SCALE
NULL, /* erxo_scale_context_alloc */
NULL, /* erxo_scale_context_free */
siena_rx_scale_mode_set, /* erxo_scale_mode_set */
siena_rx_scale_key_set, /* erxo_scale_key_set */
siena_rx_scale_tbl_set, /* erxo_scale_tbl_set */
siena_rx_prefix_hash, /* erxo_prefix_hash */
#endif
siena_rx_prefix_pktlen, /* erxo_prefix_pktlen */
siena_rx_qpost, /* erxo_qpost */
siena_rx_qpush, /* erxo_qpush */
#if EFSYS_OPT_RX_PACKED_STREAM
siena_rx_qpush_ps_credits, /* erxo_qpush_ps_credits */
siena_rx_qps_packet_info, /* erxo_qps_packet_info */
#endif
siena_rx_qflush, /* erxo_qflush */
siena_rx_qenable, /* erxo_qenable */
siena_rx_qcreate, /* erxo_qcreate */
siena_rx_qdestroy, /* erxo_qdestroy */
};
#endif /* EFSYS_OPT_SIENA */
#if EFSYS_OPT_HUNTINGTON || EFSYS_OPT_MEDFORD || EFSYS_OPT_MEDFORD2
static const efx_rx_ops_t __efx_rx_ef10_ops = {
ef10_rx_init, /* erxo_init */
ef10_rx_fini, /* erxo_fini */
#if EFSYS_OPT_RX_SCATTER
ef10_rx_scatter_enable, /* erxo_scatter_enable */
#endif
#if EFSYS_OPT_RX_SCALE
ef10_rx_scale_context_alloc, /* erxo_scale_context_alloc */
ef10_rx_scale_context_free, /* erxo_scale_context_free */
ef10_rx_scale_mode_set, /* erxo_scale_mode_set */
ef10_rx_scale_key_set, /* erxo_scale_key_set */
ef10_rx_scale_tbl_set, /* erxo_scale_tbl_set */
ef10_rx_prefix_hash, /* erxo_prefix_hash */
#endif
ef10_rx_prefix_pktlen, /* erxo_prefix_pktlen */
ef10_rx_qpost, /* erxo_qpost */
ef10_rx_qpush, /* erxo_qpush */
#if EFSYS_OPT_RX_PACKED_STREAM
ef10_rx_qpush_ps_credits, /* erxo_qpush_ps_credits */
ef10_rx_qps_packet_info, /* erxo_qps_packet_info */
#endif
ef10_rx_qflush, /* erxo_qflush */
ef10_rx_qenable, /* erxo_qenable */
ef10_rx_qcreate, /* erxo_qcreate */
ef10_rx_qdestroy, /* erxo_qdestroy */
};
#endif /* EFSYS_OPT_HUNTINGTON || EFSYS_OPT_MEDFORD || EFSYS_OPT_MEDFORD2 */
__checkReturn efx_rc_t
efx_rx_init(
__inout efx_nic_t *enp)
{
const efx_rx_ops_t *erxop;
efx_rc_t rc;
EFSYS_ASSERT3U(enp->en_magic, ==, EFX_NIC_MAGIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_NIC);
if (!(enp->en_mod_flags & EFX_MOD_EV)) {
rc = EINVAL;
goto fail1;
}
if (enp->en_mod_flags & EFX_MOD_RX) {
rc = EINVAL;
goto fail2;
}
switch (enp->en_family) {
#if EFSYS_OPT_SIENA
case EFX_FAMILY_SIENA:
erxop = &__efx_rx_siena_ops;
break;
#endif /* EFSYS_OPT_SIENA */
#if EFSYS_OPT_HUNTINGTON
case EFX_FAMILY_HUNTINGTON:
erxop = &__efx_rx_ef10_ops;
break;
#endif /* EFSYS_OPT_HUNTINGTON */
#if EFSYS_OPT_MEDFORD
case EFX_FAMILY_MEDFORD:
erxop = &__efx_rx_ef10_ops;
break;
#endif /* EFSYS_OPT_MEDFORD */
#if EFSYS_OPT_MEDFORD2
case EFX_FAMILY_MEDFORD2:
erxop = &__efx_rx_ef10_ops;
break;
#endif /* EFSYS_OPT_MEDFORD2 */
default:
EFSYS_ASSERT(0);
rc = ENOTSUP;
goto fail3;
}
if ((rc = erxop->erxo_init(enp)) != 0)
goto fail4;
enp->en_erxop = erxop;
enp->en_mod_flags |= EFX_MOD_RX;
return (0);
fail4:
EFSYS_PROBE(fail4);
fail3:
EFSYS_PROBE(fail3);
fail2:
EFSYS_PROBE(fail2);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
enp->en_erxop = NULL;
enp->en_mod_flags &= ~EFX_MOD_RX;
return (rc);
}
void
efx_rx_fini(
__in efx_nic_t *enp)
{
const efx_rx_ops_t *erxop = enp->en_erxop;
EFSYS_ASSERT3U(enp->en_magic, ==, EFX_NIC_MAGIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_NIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_RX);
EFSYS_ASSERT3U(enp->en_rx_qcount, ==, 0);
erxop->erxo_fini(enp);
enp->en_erxop = NULL;
enp->en_mod_flags &= ~EFX_MOD_RX;
}
#if EFSYS_OPT_RX_SCATTER
__checkReturn efx_rc_t
efx_rx_scatter_enable(
__in efx_nic_t *enp,
__in unsigned int buf_size)
{
const efx_rx_ops_t *erxop = enp->en_erxop;
efx_rc_t rc;
EFSYS_ASSERT3U(enp->en_magic, ==, EFX_NIC_MAGIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_RX);
if ((rc = erxop->erxo_scatter_enable(enp, buf_size)) != 0)
goto fail1;
return (0);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
#endif /* EFSYS_OPT_RX_SCATTER */
#if EFSYS_OPT_RX_SCALE
__checkReturn efx_rc_t
efx_rx_scale_hash_flags_get(
__in efx_nic_t *enp,
__in efx_rx_hash_alg_t hash_alg,
__out_ecount_part(max_nflags, *nflagsp) unsigned int *flagsp,
__in unsigned int max_nflags,
__out unsigned int *nflagsp)
{
efx_nic_cfg_t *encp = &enp->en_nic_cfg;
unsigned int nflags = 0;
efx_rc_t rc;
if (flagsp == NULL || nflagsp == NULL) {
rc = EINVAL;
goto fail1;
}
if ((encp->enc_rx_scale_hash_alg_mask & (1U << hash_alg)) == 0) {
nflags = 0;
goto done;
}
/* Helper to add flags word to flags array without buffer overflow */
#define INSERT_FLAGS(_flags) \
do { \
if (nflags >= max_nflags) { \
rc = E2BIG; \
goto fail2; \
} \
*(flagsp + nflags) = (_flags); \
nflags++; \
\
_NOTE(CONSTANTCONDITION) \
} while (B_FALSE)
if (encp->enc_rx_scale_l4_hash_supported != B_FALSE) {
INSERT_FLAGS(EFX_RX_HASH(IPV4_TCP, 4TUPLE));
INSERT_FLAGS(EFX_RX_HASH(IPV6_TCP, 4TUPLE));
}
if ((encp->enc_rx_scale_l4_hash_supported != B_FALSE) &&
(encp->enc_rx_scale_additional_modes_supported != B_FALSE)) {
INSERT_FLAGS(EFX_RX_HASH(IPV4_TCP, 2TUPLE_DST));
INSERT_FLAGS(EFX_RX_HASH(IPV4_TCP, 2TUPLE_SRC));
INSERT_FLAGS(EFX_RX_HASH(IPV6_TCP, 2TUPLE_DST));
INSERT_FLAGS(EFX_RX_HASH(IPV6_TCP, 2TUPLE_SRC));
INSERT_FLAGS(EFX_RX_HASH(IPV4_UDP, 4TUPLE));
INSERT_FLAGS(EFX_RX_HASH(IPV4_UDP, 2TUPLE_DST));
INSERT_FLAGS(EFX_RX_HASH(IPV4_UDP, 2TUPLE_SRC));
INSERT_FLAGS(EFX_RX_HASH(IPV6_UDP, 4TUPLE));
INSERT_FLAGS(EFX_RX_HASH(IPV6_UDP, 2TUPLE_DST));
INSERT_FLAGS(EFX_RX_HASH(IPV6_UDP, 2TUPLE_SRC));
}
INSERT_FLAGS(EFX_RX_HASH(IPV4_TCP, 2TUPLE));
INSERT_FLAGS(EFX_RX_HASH(IPV6_TCP, 2TUPLE));
INSERT_FLAGS(EFX_RX_HASH(IPV4, 2TUPLE));
INSERT_FLAGS(EFX_RX_HASH(IPV6, 2TUPLE));
if (encp->enc_rx_scale_additional_modes_supported != B_FALSE) {
INSERT_FLAGS(EFX_RX_HASH(IPV4_TCP, 1TUPLE_DST));
INSERT_FLAGS(EFX_RX_HASH(IPV4_TCP, 1TUPLE_SRC));
INSERT_FLAGS(EFX_RX_HASH(IPV6_TCP, 1TUPLE_DST));
INSERT_FLAGS(EFX_RX_HASH(IPV6_TCP, 1TUPLE_SRC));
INSERT_FLAGS(EFX_RX_HASH(IPV4_UDP, 2TUPLE));
INSERT_FLAGS(EFX_RX_HASH(IPV4_UDP, 1TUPLE_DST));
INSERT_FLAGS(EFX_RX_HASH(IPV4_UDP, 1TUPLE_SRC));
INSERT_FLAGS(EFX_RX_HASH(IPV6_UDP, 2TUPLE));
INSERT_FLAGS(EFX_RX_HASH(IPV6_UDP, 1TUPLE_DST));
INSERT_FLAGS(EFX_RX_HASH(IPV6_UDP, 1TUPLE_SRC));
INSERT_FLAGS(EFX_RX_HASH(IPV4, 1TUPLE_DST));
INSERT_FLAGS(EFX_RX_HASH(IPV4, 1TUPLE_SRC));
INSERT_FLAGS(EFX_RX_HASH(IPV6, 1TUPLE_DST));
INSERT_FLAGS(EFX_RX_HASH(IPV6, 1TUPLE_SRC));
}
INSERT_FLAGS(EFX_RX_HASH(IPV4_TCP, DISABLE));
INSERT_FLAGS(EFX_RX_HASH(IPV6_TCP, DISABLE));
INSERT_FLAGS(EFX_RX_HASH(IPV4_UDP, DISABLE));
INSERT_FLAGS(EFX_RX_HASH(IPV6_UDP, DISABLE));
INSERT_FLAGS(EFX_RX_HASH(IPV4, DISABLE));
INSERT_FLAGS(EFX_RX_HASH(IPV6, DISABLE));
#undef INSERT_FLAGS
done:
*nflagsp = nflags;
return (0);
fail2:
EFSYS_PROBE(fail2);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
__checkReturn efx_rc_t
efx_rx_hash_default_support_get(
__in efx_nic_t *enp,
__out efx_rx_hash_support_t *supportp)
{
efx_rc_t rc;
EFSYS_ASSERT3U(enp->en_magic, ==, EFX_NIC_MAGIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_RX);
if (supportp == NULL) {
rc = EINVAL;
goto fail1;
}
/*
* Report the hashing support the client gets by default if it
* does not allocate an RSS context itself.
*/
*supportp = enp->en_hash_support;
return (0);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
__checkReturn efx_rc_t
efx_rx_scale_default_support_get(
__in efx_nic_t *enp,
__out efx_rx_scale_context_type_t *typep)
{
efx_rc_t rc;
EFSYS_ASSERT3U(enp->en_magic, ==, EFX_NIC_MAGIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_RX);
if (typep == NULL) {
rc = EINVAL;
goto fail1;
}
/*
* Report the RSS support the client gets by default if it
* does not allocate an RSS context itself.
*/
*typep = enp->en_rss_context_type;
return (0);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
#endif /* EFSYS_OPT_RX_SCALE */
#if EFSYS_OPT_RX_SCALE
__checkReturn efx_rc_t
efx_rx_scale_context_alloc(
__in efx_nic_t *enp,
__in efx_rx_scale_context_type_t type,
__in uint32_t num_queues,
__out uint32_t *rss_contextp)
{
const efx_rx_ops_t *erxop = enp->en_erxop;
efx_rc_t rc;
EFSYS_ASSERT3U(enp->en_magic, ==, EFX_NIC_MAGIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_RX);
if (erxop->erxo_scale_context_alloc == NULL) {
rc = ENOTSUP;
goto fail1;
}
if ((rc = erxop->erxo_scale_context_alloc(enp, type,
num_queues, rss_contextp)) != 0) {
goto fail2;
}
return (0);
fail2:
EFSYS_PROBE(fail2);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
#endif /* EFSYS_OPT_RX_SCALE */
#if EFSYS_OPT_RX_SCALE
__checkReturn efx_rc_t
efx_rx_scale_context_free(
__in efx_nic_t *enp,
__in uint32_t rss_context)
{
const efx_rx_ops_t *erxop = enp->en_erxop;
efx_rc_t rc;
EFSYS_ASSERT3U(enp->en_magic, ==, EFX_NIC_MAGIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_RX);
if (erxop->erxo_scale_context_free == NULL) {
rc = ENOTSUP;
goto fail1;
}
if ((rc = erxop->erxo_scale_context_free(enp, rss_context)) != 0)
goto fail2;
return (0);
fail2:
EFSYS_PROBE(fail2);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
#endif /* EFSYS_OPT_RX_SCALE */
#if EFSYS_OPT_RX_SCALE
__checkReturn efx_rc_t
efx_rx_scale_mode_set(
__in efx_nic_t *enp,
__in uint32_t rss_context,
__in efx_rx_hash_alg_t alg,
__in efx_rx_hash_type_t type,
__in boolean_t insert)
{
sfxge(4): improve handling of legacy RSS hash flags Client drivers may use either legacy flags, for example, EFX_RX_HASH_TCPIPV4, or generalised flags, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE), to configure RSS hash. The libefx is able to recognise what scheme is used. Legacy flags may be consumed directly by a chip-specific handler to configure the NIC, that is, on EF10, these flags can be used to fill in legacy RSS mode field in MCDI request. Generalised flags can also be directly used in EF10-specific handler as they are fully compatible with additional fields of the same MCDI request. Legacy flags undergo conversion to generalised flags before they are consumed by a chip-specific handler. This conversion is used to make sure that chip-specific handlers expect only generalised flags in the input for the sake of clarity of the code. Depending on firmware capabilities, a chip-specififc handler either supplies the input to the NIC directly, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE) flag will enable 4 bits in RSS_CONTEXT_SET_FLAGS_IN_TCP_IPV4_RSS_MODE field on EF10, or takes the opportunity to translate the input to enable bits which don't map to the generic flag, like setting RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_TCPV4_EN on EF10 when the firmware claims no support for additional modes. However, this approach has introduced a severe problem which can be reproduced with ultra-low-latency firmware variant. In order to enable IP hash, EF10-specific handler requires the user to request 2-tuple hash for IP-other, TCP and UDP traffic classes, unconditionally. In example, IPv4 hash can be enabled using the following input: EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4_UDP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). At the same time, on ultra-low-latency firmware, the common code will never report support for any UDP tuple to the client driver. That is, in the same example, the driver will use EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). This input will not be recognised by EF10-specific handler, and RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_IPV4_EN bit will not be set in the MCDI request. In order to solve the problem, the patch removes conversion code from chip-specific handlers and adds appropriate code to convert EFX_RX_HASH() flags to their legacy counterparts to the common scale mode set function. If the firmware does not support additional modes, the function will convert generalised flags to legacy flags correctly without any demand for UDP flags and pass the result to a chip-specific handler. Submitted by: Ivan Malov <ivan.malov at oktetlabs.ru> Sponsored by: Solarflare Communications, Inc. Differential Revision: https://reviews.freebsd.org/D18269
2018-11-30 07:07:31 +00:00
efx_nic_cfg_t *encp = &enp->en_nic_cfg;
const efx_rx_ops_t *erxop = enp->en_erxop;
efx_rx_hash_type_t type_check;
unsigned int i;
efx_rc_t rc;
EFSYS_ASSERT3U(enp->en_magic, ==, EFX_NIC_MAGIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_RX);
/*
* Legacy flags and modern bits cannot be
* used at the same time in the hash type.
*/
if ((type & EFX_RX_HASH_LEGACY_MASK) &&
(type & ~EFX_RX_HASH_LEGACY_MASK)) {
rc = EINVAL;
goto fail1;
}
/*
sfxge(4): improve handling of legacy RSS hash flags Client drivers may use either legacy flags, for example, EFX_RX_HASH_TCPIPV4, or generalised flags, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE), to configure RSS hash. The libefx is able to recognise what scheme is used. Legacy flags may be consumed directly by a chip-specific handler to configure the NIC, that is, on EF10, these flags can be used to fill in legacy RSS mode field in MCDI request. Generalised flags can also be directly used in EF10-specific handler as they are fully compatible with additional fields of the same MCDI request. Legacy flags undergo conversion to generalised flags before they are consumed by a chip-specific handler. This conversion is used to make sure that chip-specific handlers expect only generalised flags in the input for the sake of clarity of the code. Depending on firmware capabilities, a chip-specififc handler either supplies the input to the NIC directly, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE) flag will enable 4 bits in RSS_CONTEXT_SET_FLAGS_IN_TCP_IPV4_RSS_MODE field on EF10, or takes the opportunity to translate the input to enable bits which don't map to the generic flag, like setting RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_TCPV4_EN on EF10 when the firmware claims no support for additional modes. However, this approach has introduced a severe problem which can be reproduced with ultra-low-latency firmware variant. In order to enable IP hash, EF10-specific handler requires the user to request 2-tuple hash for IP-other, TCP and UDP traffic classes, unconditionally. In example, IPv4 hash can be enabled using the following input: EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4_UDP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). At the same time, on ultra-low-latency firmware, the common code will never report support for any UDP tuple to the client driver. That is, in the same example, the driver will use EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). This input will not be recognised by EF10-specific handler, and RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_IPV4_EN bit will not be set in the MCDI request. In order to solve the problem, the patch removes conversion code from chip-specific handlers and adds appropriate code to convert EFX_RX_HASH() flags to their legacy counterparts to the common scale mode set function. If the firmware does not support additional modes, the function will convert generalised flags to legacy flags correctly without any demand for UDP flags and pass the result to a chip-specific handler. Submitted by: Ivan Malov <ivan.malov at oktetlabs.ru> Sponsored by: Solarflare Communications, Inc. Differential Revision: https://reviews.freebsd.org/D18269
2018-11-30 07:07:31 +00:00
* If RSS hash type is represented by additional bits
* in the value, the latter need to be verified since
* not all bit combinations are valid RSS modes. Also,
* depending on the firmware, some valid combinations
* may be unsupported. Discern additional bits in the
* type value and try to recognise valid combinations.
* If some bits remain unrecognised, report the error.
*/
sfxge(4): improve handling of legacy RSS hash flags Client drivers may use either legacy flags, for example, EFX_RX_HASH_TCPIPV4, or generalised flags, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE), to configure RSS hash. The libefx is able to recognise what scheme is used. Legacy flags may be consumed directly by a chip-specific handler to configure the NIC, that is, on EF10, these flags can be used to fill in legacy RSS mode field in MCDI request. Generalised flags can also be directly used in EF10-specific handler as they are fully compatible with additional fields of the same MCDI request. Legacy flags undergo conversion to generalised flags before they are consumed by a chip-specific handler. This conversion is used to make sure that chip-specific handlers expect only generalised flags in the input for the sake of clarity of the code. Depending on firmware capabilities, a chip-specififc handler either supplies the input to the NIC directly, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE) flag will enable 4 bits in RSS_CONTEXT_SET_FLAGS_IN_TCP_IPV4_RSS_MODE field on EF10, or takes the opportunity to translate the input to enable bits which don't map to the generic flag, like setting RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_TCPV4_EN on EF10 when the firmware claims no support for additional modes. However, this approach has introduced a severe problem which can be reproduced with ultra-low-latency firmware variant. In order to enable IP hash, EF10-specific handler requires the user to request 2-tuple hash for IP-other, TCP and UDP traffic classes, unconditionally. In example, IPv4 hash can be enabled using the following input: EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4_UDP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). At the same time, on ultra-low-latency firmware, the common code will never report support for any UDP tuple to the client driver. That is, in the same example, the driver will use EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). This input will not be recognised by EF10-specific handler, and RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_IPV4_EN bit will not be set in the MCDI request. In order to solve the problem, the patch removes conversion code from chip-specific handlers and adds appropriate code to convert EFX_RX_HASH() flags to their legacy counterparts to the common scale mode set function. If the firmware does not support additional modes, the function will convert generalised flags to legacy flags correctly without any demand for UDP flags and pass the result to a chip-specific handler. Submitted by: Ivan Malov <ivan.malov at oktetlabs.ru> Sponsored by: Solarflare Communications, Inc. Differential Revision: https://reviews.freebsd.org/D18269
2018-11-30 07:07:31 +00:00
type_check = type & ~EFX_RX_HASH_LEGACY_MASK;
if (type_check != 0) {
unsigned int type_flags[EFX_RX_HASH_NFLAGS];
unsigned int type_nflags;
sfxge(4): improve handling of legacy RSS hash flags Client drivers may use either legacy flags, for example, EFX_RX_HASH_TCPIPV4, or generalised flags, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE), to configure RSS hash. The libefx is able to recognise what scheme is used. Legacy flags may be consumed directly by a chip-specific handler to configure the NIC, that is, on EF10, these flags can be used to fill in legacy RSS mode field in MCDI request. Generalised flags can also be directly used in EF10-specific handler as they are fully compatible with additional fields of the same MCDI request. Legacy flags undergo conversion to generalised flags before they are consumed by a chip-specific handler. This conversion is used to make sure that chip-specific handlers expect only generalised flags in the input for the sake of clarity of the code. Depending on firmware capabilities, a chip-specififc handler either supplies the input to the NIC directly, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE) flag will enable 4 bits in RSS_CONTEXT_SET_FLAGS_IN_TCP_IPV4_RSS_MODE field on EF10, or takes the opportunity to translate the input to enable bits which don't map to the generic flag, like setting RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_TCPV4_EN on EF10 when the firmware claims no support for additional modes. However, this approach has introduced a severe problem which can be reproduced with ultra-low-latency firmware variant. In order to enable IP hash, EF10-specific handler requires the user to request 2-tuple hash for IP-other, TCP and UDP traffic classes, unconditionally. In example, IPv4 hash can be enabled using the following input: EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4_UDP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). At the same time, on ultra-low-latency firmware, the common code will never report support for any UDP tuple to the client driver. That is, in the same example, the driver will use EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). This input will not be recognised by EF10-specific handler, and RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_IPV4_EN bit will not be set in the MCDI request. In order to solve the problem, the patch removes conversion code from chip-specific handlers and adds appropriate code to convert EFX_RX_HASH() flags to their legacy counterparts to the common scale mode set function. If the firmware does not support additional modes, the function will convert generalised flags to legacy flags correctly without any demand for UDP flags and pass the result to a chip-specific handler. Submitted by: Ivan Malov <ivan.malov at oktetlabs.ru> Sponsored by: Solarflare Communications, Inc. Differential Revision: https://reviews.freebsd.org/D18269
2018-11-30 07:07:31 +00:00
rc = efx_rx_scale_hash_flags_get(enp, alg, type_flags,
EFX_ARRAY_SIZE(type_flags), &type_nflags);
if (rc != 0)
goto fail2;
sfxge(4): improve handling of legacy RSS hash flags Client drivers may use either legacy flags, for example, EFX_RX_HASH_TCPIPV4, or generalised flags, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE), to configure RSS hash. The libefx is able to recognise what scheme is used. Legacy flags may be consumed directly by a chip-specific handler to configure the NIC, that is, on EF10, these flags can be used to fill in legacy RSS mode field in MCDI request. Generalised flags can also be directly used in EF10-specific handler as they are fully compatible with additional fields of the same MCDI request. Legacy flags undergo conversion to generalised flags before they are consumed by a chip-specific handler. This conversion is used to make sure that chip-specific handlers expect only generalised flags in the input for the sake of clarity of the code. Depending on firmware capabilities, a chip-specififc handler either supplies the input to the NIC directly, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE) flag will enable 4 bits in RSS_CONTEXT_SET_FLAGS_IN_TCP_IPV4_RSS_MODE field on EF10, or takes the opportunity to translate the input to enable bits which don't map to the generic flag, like setting RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_TCPV4_EN on EF10 when the firmware claims no support for additional modes. However, this approach has introduced a severe problem which can be reproduced with ultra-low-latency firmware variant. In order to enable IP hash, EF10-specific handler requires the user to request 2-tuple hash for IP-other, TCP and UDP traffic classes, unconditionally. In example, IPv4 hash can be enabled using the following input: EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4_UDP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). At the same time, on ultra-low-latency firmware, the common code will never report support for any UDP tuple to the client driver. That is, in the same example, the driver will use EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). This input will not be recognised by EF10-specific handler, and RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_IPV4_EN bit will not be set in the MCDI request. In order to solve the problem, the patch removes conversion code from chip-specific handlers and adds appropriate code to convert EFX_RX_HASH() flags to their legacy counterparts to the common scale mode set function. If the firmware does not support additional modes, the function will convert generalised flags to legacy flags correctly without any demand for UDP flags and pass the result to a chip-specific handler. Submitted by: Ivan Malov <ivan.malov at oktetlabs.ru> Sponsored by: Solarflare Communications, Inc. Differential Revision: https://reviews.freebsd.org/D18269
2018-11-30 07:07:31 +00:00
for (i = 0; i < type_nflags; ++i) {
if ((type_check & type_flags[i]) == type_flags[i])
type_check &= ~(type_flags[i]);
}
sfxge(4): improve handling of legacy RSS hash flags Client drivers may use either legacy flags, for example, EFX_RX_HASH_TCPIPV4, or generalised flags, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE), to configure RSS hash. The libefx is able to recognise what scheme is used. Legacy flags may be consumed directly by a chip-specific handler to configure the NIC, that is, on EF10, these flags can be used to fill in legacy RSS mode field in MCDI request. Generalised flags can also be directly used in EF10-specific handler as they are fully compatible with additional fields of the same MCDI request. Legacy flags undergo conversion to generalised flags before they are consumed by a chip-specific handler. This conversion is used to make sure that chip-specific handlers expect only generalised flags in the input for the sake of clarity of the code. Depending on firmware capabilities, a chip-specififc handler either supplies the input to the NIC directly, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE) flag will enable 4 bits in RSS_CONTEXT_SET_FLAGS_IN_TCP_IPV4_RSS_MODE field on EF10, or takes the opportunity to translate the input to enable bits which don't map to the generic flag, like setting RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_TCPV4_EN on EF10 when the firmware claims no support for additional modes. However, this approach has introduced a severe problem which can be reproduced with ultra-low-latency firmware variant. In order to enable IP hash, EF10-specific handler requires the user to request 2-tuple hash for IP-other, TCP and UDP traffic classes, unconditionally. In example, IPv4 hash can be enabled using the following input: EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4_UDP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). At the same time, on ultra-low-latency firmware, the common code will never report support for any UDP tuple to the client driver. That is, in the same example, the driver will use EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). This input will not be recognised by EF10-specific handler, and RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_IPV4_EN bit will not be set in the MCDI request. In order to solve the problem, the patch removes conversion code from chip-specific handlers and adds appropriate code to convert EFX_RX_HASH() flags to their legacy counterparts to the common scale mode set function. If the firmware does not support additional modes, the function will convert generalised flags to legacy flags correctly without any demand for UDP flags and pass the result to a chip-specific handler. Submitted by: Ivan Malov <ivan.malov at oktetlabs.ru> Sponsored by: Solarflare Communications, Inc. Differential Revision: https://reviews.freebsd.org/D18269
2018-11-30 07:07:31 +00:00
if (type_check != 0) {
rc = EINVAL;
goto fail3;
}
}
/*
sfxge(4): improve handling of legacy RSS hash flags Client drivers may use either legacy flags, for example, EFX_RX_HASH_TCPIPV4, or generalised flags, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE), to configure RSS hash. The libefx is able to recognise what scheme is used. Legacy flags may be consumed directly by a chip-specific handler to configure the NIC, that is, on EF10, these flags can be used to fill in legacy RSS mode field in MCDI request. Generalised flags can also be directly used in EF10-specific handler as they are fully compatible with additional fields of the same MCDI request. Legacy flags undergo conversion to generalised flags before they are consumed by a chip-specific handler. This conversion is used to make sure that chip-specific handlers expect only generalised flags in the input for the sake of clarity of the code. Depending on firmware capabilities, a chip-specififc handler either supplies the input to the NIC directly, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE) flag will enable 4 bits in RSS_CONTEXT_SET_FLAGS_IN_TCP_IPV4_RSS_MODE field on EF10, or takes the opportunity to translate the input to enable bits which don't map to the generic flag, like setting RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_TCPV4_EN on EF10 when the firmware claims no support for additional modes. However, this approach has introduced a severe problem which can be reproduced with ultra-low-latency firmware variant. In order to enable IP hash, EF10-specific handler requires the user to request 2-tuple hash for IP-other, TCP and UDP traffic classes, unconditionally. In example, IPv4 hash can be enabled using the following input: EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4_UDP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). At the same time, on ultra-low-latency firmware, the common code will never report support for any UDP tuple to the client driver. That is, in the same example, the driver will use EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). This input will not be recognised by EF10-specific handler, and RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_IPV4_EN bit will not be set in the MCDI request. In order to solve the problem, the patch removes conversion code from chip-specific handlers and adds appropriate code to convert EFX_RX_HASH() flags to their legacy counterparts to the common scale mode set function. If the firmware does not support additional modes, the function will convert generalised flags to legacy flags correctly without any demand for UDP flags and pass the result to a chip-specific handler. Submitted by: Ivan Malov <ivan.malov at oktetlabs.ru> Sponsored by: Solarflare Communications, Inc. Differential Revision: https://reviews.freebsd.org/D18269
2018-11-30 07:07:31 +00:00
* Translate EFX_RX_HASH() flags to their legacy counterparts
* provided that the FW claims no support for additional modes.
*/
sfxge(4): improve handling of legacy RSS hash flags Client drivers may use either legacy flags, for example, EFX_RX_HASH_TCPIPV4, or generalised flags, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE), to configure RSS hash. The libefx is able to recognise what scheme is used. Legacy flags may be consumed directly by a chip-specific handler to configure the NIC, that is, on EF10, these flags can be used to fill in legacy RSS mode field in MCDI request. Generalised flags can also be directly used in EF10-specific handler as they are fully compatible with additional fields of the same MCDI request. Legacy flags undergo conversion to generalised flags before they are consumed by a chip-specific handler. This conversion is used to make sure that chip-specific handlers expect only generalised flags in the input for the sake of clarity of the code. Depending on firmware capabilities, a chip-specififc handler either supplies the input to the NIC directly, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE) flag will enable 4 bits in RSS_CONTEXT_SET_FLAGS_IN_TCP_IPV4_RSS_MODE field on EF10, or takes the opportunity to translate the input to enable bits which don't map to the generic flag, like setting RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_TCPV4_EN on EF10 when the firmware claims no support for additional modes. However, this approach has introduced a severe problem which can be reproduced with ultra-low-latency firmware variant. In order to enable IP hash, EF10-specific handler requires the user to request 2-tuple hash for IP-other, TCP and UDP traffic classes, unconditionally. In example, IPv4 hash can be enabled using the following input: EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4_UDP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). At the same time, on ultra-low-latency firmware, the common code will never report support for any UDP tuple to the client driver. That is, in the same example, the driver will use EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). This input will not be recognised by EF10-specific handler, and RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_IPV4_EN bit will not be set in the MCDI request. In order to solve the problem, the patch removes conversion code from chip-specific handlers and adds appropriate code to convert EFX_RX_HASH() flags to their legacy counterparts to the common scale mode set function. If the firmware does not support additional modes, the function will convert generalised flags to legacy flags correctly without any demand for UDP flags and pass the result to a chip-specific handler. Submitted by: Ivan Malov <ivan.malov at oktetlabs.ru> Sponsored by: Solarflare Communications, Inc. Differential Revision: https://reviews.freebsd.org/D18269
2018-11-30 07:07:31 +00:00
if (encp->enc_rx_scale_additional_modes_supported == B_FALSE) {
efx_rx_hash_type_t t_ipv4 = EFX_RX_HASH(IPV4, 2TUPLE) |
EFX_RX_HASH(IPV4_TCP, 2TUPLE);
efx_rx_hash_type_t t_ipv6 = EFX_RX_HASH(IPV6, 2TUPLE) |
EFX_RX_HASH(IPV6_TCP, 2TUPLE);
efx_rx_hash_type_t t_ipv4_tcp = EFX_RX_HASH(IPV4_TCP, 4TUPLE);
efx_rx_hash_type_t t_ipv6_tcp = EFX_RX_HASH(IPV6_TCP, 4TUPLE);
if ((type & t_ipv4) == t_ipv4)
type |= EFX_RX_HASH_IPV4;
if ((type & t_ipv6) == t_ipv6)
type |= EFX_RX_HASH_IPV6;
if (encp->enc_rx_scale_l4_hash_supported == B_TRUE) {
if ((type & t_ipv4_tcp) == t_ipv4_tcp)
type |= EFX_RX_HASH_TCPIPV4;
if ((type & t_ipv6_tcp) == t_ipv6_tcp)
type |= EFX_RX_HASH_TCPIPV6;
}
type &= EFX_RX_HASH_LEGACY_MASK;
}
if (erxop->erxo_scale_mode_set != NULL) {
if ((rc = erxop->erxo_scale_mode_set(enp, rss_context, alg,
type, insert)) != 0)
goto fail4;
}
return (0);
fail4:
EFSYS_PROBE(fail4);
fail3:
EFSYS_PROBE(fail3);
fail2:
EFSYS_PROBE(fail2);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
#endif /* EFSYS_OPT_RX_SCALE */
#if EFSYS_OPT_RX_SCALE
__checkReturn efx_rc_t
efx_rx_scale_key_set(
__in efx_nic_t *enp,
__in uint32_t rss_context,
__in_ecount(n) uint8_t *key,
__in size_t n)
{
const efx_rx_ops_t *erxop = enp->en_erxop;
efx_rc_t rc;
EFSYS_ASSERT3U(enp->en_magic, ==, EFX_NIC_MAGIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_RX);
if ((rc = erxop->erxo_scale_key_set(enp, rss_context, key, n)) != 0)
goto fail1;
return (0);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
#endif /* EFSYS_OPT_RX_SCALE */
#if EFSYS_OPT_RX_SCALE
__checkReturn efx_rc_t
efx_rx_scale_tbl_set(
__in efx_nic_t *enp,
__in uint32_t rss_context,
__in_ecount(n) unsigned int *table,
__in size_t n)
{
const efx_rx_ops_t *erxop = enp->en_erxop;
efx_rc_t rc;
EFSYS_ASSERT3U(enp->en_magic, ==, EFX_NIC_MAGIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_RX);
if ((rc = erxop->erxo_scale_tbl_set(enp, rss_context, table, n)) != 0)
goto fail1;
return (0);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
#endif /* EFSYS_OPT_RX_SCALE */
void
efx_rx_qpost(
__in efx_rxq_t *erp,
__in_ecount(ndescs) efsys_dma_addr_t *addrp,
__in size_t size,
__in unsigned int ndescs,
__in unsigned int completed,
__in unsigned int added)
{
efx_nic_t *enp = erp->er_enp;
const efx_rx_ops_t *erxop = enp->en_erxop;
EFSYS_ASSERT3U(erp->er_magic, ==, EFX_RXQ_MAGIC);
erxop->erxo_qpost(erp, addrp, size, ndescs, completed, added);
}
#if EFSYS_OPT_RX_PACKED_STREAM
void
efx_rx_qpush_ps_credits(
__in efx_rxq_t *erp)
{
efx_nic_t *enp = erp->er_enp;
const efx_rx_ops_t *erxop = enp->en_erxop;
EFSYS_ASSERT3U(erp->er_magic, ==, EFX_RXQ_MAGIC);
erxop->erxo_qpush_ps_credits(erp);
}
__checkReturn uint8_t *
efx_rx_qps_packet_info(
__in efx_rxq_t *erp,
__in uint8_t *buffer,
__in uint32_t buffer_length,
__in uint32_t current_offset,
__out uint16_t *lengthp,
__out uint32_t *next_offsetp,
__out uint32_t *timestamp)
{
efx_nic_t *enp = erp->er_enp;
const efx_rx_ops_t *erxop = enp->en_erxop;
return (erxop->erxo_qps_packet_info(erp, buffer,
buffer_length, current_offset, lengthp,
next_offsetp, timestamp));
}
#endif /* EFSYS_OPT_RX_PACKED_STREAM */
void
efx_rx_qpush(
__in efx_rxq_t *erp,
__in unsigned int added,
__inout unsigned int *pushedp)
{
efx_nic_t *enp = erp->er_enp;
const efx_rx_ops_t *erxop = enp->en_erxop;
EFSYS_ASSERT3U(erp->er_magic, ==, EFX_RXQ_MAGIC);
erxop->erxo_qpush(erp, added, pushedp);
}
__checkReturn efx_rc_t
efx_rx_qflush(
__in efx_rxq_t *erp)
{
efx_nic_t *enp = erp->er_enp;
const efx_rx_ops_t *erxop = enp->en_erxop;
efx_rc_t rc;
EFSYS_ASSERT3U(erp->er_magic, ==, EFX_RXQ_MAGIC);
if ((rc = erxop->erxo_qflush(erp)) != 0)
goto fail1;
return (0);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
void
efx_rx_qenable(
__in efx_rxq_t *erp)
{
efx_nic_t *enp = erp->er_enp;
const efx_rx_ops_t *erxop = enp->en_erxop;
EFSYS_ASSERT3U(erp->er_magic, ==, EFX_RXQ_MAGIC);
erxop->erxo_qenable(erp);
}
static __checkReturn efx_rc_t
efx_rx_qcreate_internal(
__in efx_nic_t *enp,
__in unsigned int index,
__in unsigned int label,
__in efx_rxq_type_t type,
__in_opt const efx_rxq_type_data_t *type_data,
__in efsys_mem_t *esmp,
__in size_t ndescs,
__in uint32_t id,
__in unsigned int flags,
__in efx_evq_t *eep,
__deref_out efx_rxq_t **erpp)
{
const efx_rx_ops_t *erxop = enp->en_erxop;
efx_rxq_t *erp;
efx_rc_t rc;
EFSYS_ASSERT3U(enp->en_magic, ==, EFX_NIC_MAGIC);
EFSYS_ASSERT3U(enp->en_mod_flags, &, EFX_MOD_RX);
/* Allocate an RXQ object */
EFSYS_KMEM_ALLOC(enp->en_esip, sizeof (efx_rxq_t), erp);
if (erp == NULL) {
rc = ENOMEM;
goto fail1;
}
erp->er_magic = EFX_RXQ_MAGIC;
erp->er_enp = enp;
erp->er_index = index;
erp->er_mask = ndescs - 1;
erp->er_esmp = esmp;
if ((rc = erxop->erxo_qcreate(enp, index, label, type, type_data, esmp,
ndescs, id, flags, eep, erp)) != 0)
goto fail2;
enp->en_rx_qcount++;
*erpp = erp;
return (0);
fail2:
EFSYS_PROBE(fail2);
EFSYS_KMEM_FREE(enp->en_esip, sizeof (efx_rxq_t), erp);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
__checkReturn efx_rc_t
efx_rx_qcreate(
__in efx_nic_t *enp,
__in unsigned int index,
__in unsigned int label,
__in efx_rxq_type_t type,
__in efsys_mem_t *esmp,
__in size_t ndescs,
__in uint32_t id,
__in unsigned int flags,
__in efx_evq_t *eep,
__deref_out efx_rxq_t **erpp)
{
return efx_rx_qcreate_internal(enp, index, label, type, NULL,
esmp, ndescs, id, flags, eep, erpp);
}
#if EFSYS_OPT_RX_PACKED_STREAM
__checkReturn efx_rc_t
efx_rx_qcreate_packed_stream(
__in efx_nic_t *enp,
__in unsigned int index,
__in unsigned int label,
__in uint32_t ps_buf_size,
__in efsys_mem_t *esmp,
__in size_t ndescs,
__in efx_evq_t *eep,
__deref_out efx_rxq_t **erpp)
{
efx_rxq_type_data_t type_data;
memset(&type_data, 0, sizeof (type_data));
type_data.ertd_packed_stream.eps_buf_size = ps_buf_size;
return efx_rx_qcreate_internal(enp, index, label,
EFX_RXQ_TYPE_PACKED_STREAM, &type_data, esmp, ndescs,
0 /* id unused on EF10 */, EFX_RXQ_FLAG_NONE, eep, erpp);
}
#endif
#if EFSYS_OPT_RX_ES_SUPER_BUFFER
__checkReturn efx_rc_t
efx_rx_qcreate_es_super_buffer(
__in efx_nic_t *enp,
__in unsigned int index,
__in unsigned int label,
__in uint32_t n_bufs_per_desc,
__in uint32_t max_dma_len,
__in uint32_t buf_stride,
__in uint32_t hol_block_timeout,
__in efsys_mem_t *esmp,
__in size_t ndescs,
__in unsigned int flags,
__in efx_evq_t *eep,
__deref_out efx_rxq_t **erpp)
{
efx_rc_t rc;
efx_rxq_type_data_t type_data;
if (hol_block_timeout > EFX_RXQ_ES_SUPER_BUFFER_HOL_BLOCK_MAX) {
rc = EINVAL;
goto fail1;
}
memset(&type_data, 0, sizeof (type_data));
type_data.ertd_es_super_buffer.eessb_bufs_per_desc = n_bufs_per_desc;
type_data.ertd_es_super_buffer.eessb_max_dma_len = max_dma_len;
type_data.ertd_es_super_buffer.eessb_buf_stride = buf_stride;
type_data.ertd_es_super_buffer.eessb_hol_block_timeout =
hol_block_timeout;
rc = efx_rx_qcreate_internal(enp, index, label,
EFX_RXQ_TYPE_ES_SUPER_BUFFER, &type_data, esmp, ndescs,
0 /* id unused on EF10 */, flags, eep, erpp);
if (rc != 0)
goto fail2;
return (0);
fail2:
EFSYS_PROBE(fail2);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
#endif
void
efx_rx_qdestroy(
__in efx_rxq_t *erp)
{
efx_nic_t *enp = erp->er_enp;
const efx_rx_ops_t *erxop = enp->en_erxop;
EFSYS_ASSERT3U(erp->er_magic, ==, EFX_RXQ_MAGIC);
erxop->erxo_qdestroy(erp);
}
__checkReturn efx_rc_t
efx_pseudo_hdr_pkt_length_get(
__in efx_rxq_t *erp,
__in uint8_t *buffer,
__out uint16_t *lengthp)
{
efx_nic_t *enp = erp->er_enp;
const efx_rx_ops_t *erxop = enp->en_erxop;
EFSYS_ASSERT3U(erp->er_magic, ==, EFX_RXQ_MAGIC);
return (erxop->erxo_prefix_pktlen(enp, buffer, lengthp));
}
#if EFSYS_OPT_RX_SCALE
__checkReturn uint32_t
efx_pseudo_hdr_hash_get(
__in efx_rxq_t *erp,
__in efx_rx_hash_alg_t func,
__in uint8_t *buffer)
{
efx_nic_t *enp = erp->er_enp;
const efx_rx_ops_t *erxop = enp->en_erxop;
EFSYS_ASSERT3U(erp->er_magic, ==, EFX_RXQ_MAGIC);
EFSYS_ASSERT3U(enp->en_hash_support, ==, EFX_RX_HASH_AVAILABLE);
return (erxop->erxo_prefix_hash(enp, func, buffer));
}
#endif /* EFSYS_OPT_RX_SCALE */
#if EFSYS_OPT_SIENA
static __checkReturn efx_rc_t
siena_rx_init(
__in efx_nic_t *enp)
{
efx_oword_t oword;
unsigned int index;
EFX_BAR_READO(enp, FR_AZ_RX_CFG_REG, &oword);
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_DESC_PUSH_EN, 0);
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_HASH_ALG, 0);
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_IP_HASH, 0);
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_TCP_SUP, 0);
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_HASH_INSRT_HDR, 0);
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_USR_BUF_SIZE, 0x3000 / 32);
EFX_BAR_WRITEO(enp, FR_AZ_RX_CFG_REG, &oword);
/* Zero the RSS table */
for (index = 0; index < FR_BZ_RX_INDIRECTION_TBL_ROWS;
index++) {
EFX_ZERO_OWORD(oword);
EFX_BAR_TBL_WRITEO(enp, FR_BZ_RX_INDIRECTION_TBL,
index, &oword, B_TRUE);
}
#if EFSYS_OPT_RX_SCALE
/* The RSS key and indirection table are writable. */
enp->en_rss_context_type = EFX_RX_SCALE_EXCLUSIVE;
/* Hardware can insert RX hash with/without RSS */
enp->en_hash_support = EFX_RX_HASH_AVAILABLE;
#endif /* EFSYS_OPT_RX_SCALE */
return (0);
}
#if EFSYS_OPT_RX_SCATTER
static __checkReturn efx_rc_t
siena_rx_scatter_enable(
__in efx_nic_t *enp,
__in unsigned int buf_size)
{
unsigned int nbuf32;
efx_oword_t oword;
efx_rc_t rc;
nbuf32 = buf_size / 32;
IF ((NBUF32 == 0) ||
(nbuf32 >= (1 << FRF_BZ_RX_USR_BUF_SIZE_WIDTH)) ||
((buf_size % 32) != 0)) {
rc = EINVAL;
goto fail1;
}
if (enp->en_rx_qcount > 0) {
rc = EBUSY;
goto fail2;
}
/* Set scatter buffer size */
EFX_BAR_READO(enp, FR_AZ_RX_CFG_REG, &oword);
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_USR_BUF_SIZE, nbuf32);
EFX_BAR_WRITEO(enp, FR_AZ_RX_CFG_REG, &oword);
/* Enable scatter for packets not matching a filter */
EFX_BAR_READO(enp, FR_AZ_RX_FILTER_CTL_REG, &oword);
EFX_SET_OWORD_FIELD(oword, FRF_BZ_SCATTER_ENBL_NO_MATCH_Q, 1);
EFX_BAR_WRITEO(enp, FR_AZ_RX_FILTER_CTL_REG, &oword);
return (0);
fail2:
EFSYS_PROBE(fail2);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
#endif /* EFSYS_OPT_RX_SCATTER */
#define EFX_RX_LFSR_HASH(_enp, _insert) \
do { \
efx_oword_t oword; \
\
EFX_BAR_READO((_enp), FR_AZ_RX_CFG_REG, &oword); \
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_HASH_ALG, 0); \
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_IP_HASH, 0); \
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_TCP_SUP, 0); \
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_HASH_INSRT_HDR, \
(_insert) ? 1 : 0); \
EFX_BAR_WRITEO((_enp), FR_AZ_RX_CFG_REG, &oword); \
\
if ((_enp)->en_family == EFX_FAMILY_SIENA) { \
EFX_BAR_READO((_enp), FR_CZ_RX_RSS_IPV6_REG3, \
&oword); \
EFX_SET_OWORD_FIELD(oword, \
FRF_CZ_RX_RSS_IPV6_THASH_ENABLE, 0); \
EFX_BAR_WRITEO((_enp), FR_CZ_RX_RSS_IPV6_REG3, \
&oword); \
} \
\
_NOTE(CONSTANTCONDITION) \
} while (B_FALSE)
#define EFX_RX_TOEPLITZ_IPV4_HASH(_enp, _insert, _ip, _tcp) \
do { \
efx_oword_t oword; \
\
EFX_BAR_READO((_enp), FR_AZ_RX_CFG_REG, &oword); \
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_HASH_ALG, 1); \
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_IP_HASH, \
(_ip) ? 1 : 0); \
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_TCP_SUP, \
(_tcp) ? 0 : 1); \
EFX_SET_OWORD_FIELD(oword, FRF_BZ_RX_HASH_INSRT_HDR, \
(_insert) ? 1 : 0); \
EFX_BAR_WRITEO((_enp), FR_AZ_RX_CFG_REG, &oword); \
\
_NOTE(CONSTANTCONDITION) \
} while (B_FALSE)
#define EFX_RX_TOEPLITZ_IPV6_HASH(_enp, _ip, _tcp, _rc) \
do { \
efx_oword_t oword; \
\
EFX_BAR_READO((_enp), FR_CZ_RX_RSS_IPV6_REG3, &oword); \
EFX_SET_OWORD_FIELD(oword, \
FRF_CZ_RX_RSS_IPV6_THASH_ENABLE, 1); \
EFX_SET_OWORD_FIELD(oword, \
FRF_CZ_RX_RSS_IPV6_IP_THASH_ENABLE, (_ip) ? 1 : 0); \
EFX_SET_OWORD_FIELD(oword, \
FRF_CZ_RX_RSS_IPV6_TCP_SUPPRESS, (_tcp) ? 0 : 1); \
EFX_BAR_WRITEO((_enp), FR_CZ_RX_RSS_IPV6_REG3, &oword); \
\
(_rc) = 0; \
\
_NOTE(CONSTANTCONDITION) \
} while (B_FALSE)
#if EFSYS_OPT_RX_SCALE
static __checkReturn efx_rc_t
siena_rx_scale_mode_set(
__in efx_nic_t *enp,
__in uint32_t rss_context,
__in efx_rx_hash_alg_t alg,
__in efx_rx_hash_type_t type,
__in boolean_t insert)
{
efx_rc_t rc;
if (rss_context != EFX_RSS_CONTEXT_DEFAULT) {
rc = EINVAL;
goto fail1;
}
switch (alg) {
case EFX_RX_HASHALG_LFSR:
EFX_RX_LFSR_HASH(enp, insert);
break;
case EFX_RX_HASHALG_TOEPLITZ:
EFX_RX_TOEPLITZ_IPV4_HASH(enp, insert,
sfxge(4): improve handling of legacy RSS hash flags Client drivers may use either legacy flags, for example, EFX_RX_HASH_TCPIPV4, or generalised flags, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE), to configure RSS hash. The libefx is able to recognise what scheme is used. Legacy flags may be consumed directly by a chip-specific handler to configure the NIC, that is, on EF10, these flags can be used to fill in legacy RSS mode field in MCDI request. Generalised flags can also be directly used in EF10-specific handler as they are fully compatible with additional fields of the same MCDI request. Legacy flags undergo conversion to generalised flags before they are consumed by a chip-specific handler. This conversion is used to make sure that chip-specific handlers expect only generalised flags in the input for the sake of clarity of the code. Depending on firmware capabilities, a chip-specififc handler either supplies the input to the NIC directly, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE) flag will enable 4 bits in RSS_CONTEXT_SET_FLAGS_IN_TCP_IPV4_RSS_MODE field on EF10, or takes the opportunity to translate the input to enable bits which don't map to the generic flag, like setting RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_TCPV4_EN on EF10 when the firmware claims no support for additional modes. However, this approach has introduced a severe problem which can be reproduced with ultra-low-latency firmware variant. In order to enable IP hash, EF10-specific handler requires the user to request 2-tuple hash for IP-other, TCP and UDP traffic classes, unconditionally. In example, IPv4 hash can be enabled using the following input: EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4_UDP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). At the same time, on ultra-low-latency firmware, the common code will never report support for any UDP tuple to the client driver. That is, in the same example, the driver will use EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). This input will not be recognised by EF10-specific handler, and RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_IPV4_EN bit will not be set in the MCDI request. In order to solve the problem, the patch removes conversion code from chip-specific handlers and adds appropriate code to convert EFX_RX_HASH() flags to their legacy counterparts to the common scale mode set function. If the firmware does not support additional modes, the function will convert generalised flags to legacy flags correctly without any demand for UDP flags and pass the result to a chip-specific handler. Submitted by: Ivan Malov <ivan.malov at oktetlabs.ru> Sponsored by: Solarflare Communications, Inc. Differential Revision: https://reviews.freebsd.org/D18269
2018-11-30 07:07:31 +00:00
(type & EFX_RX_HASH_IPV4) ? B_TRUE : B_FALSE,
(type & EFX_RX_HASH_TCPIPV4) ? B_TRUE : B_FALSE);
EFX_RX_TOEPLITZ_IPV6_HASH(enp,
sfxge(4): improve handling of legacy RSS hash flags Client drivers may use either legacy flags, for example, EFX_RX_HASH_TCPIPV4, or generalised flags, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE), to configure RSS hash. The libefx is able to recognise what scheme is used. Legacy flags may be consumed directly by a chip-specific handler to configure the NIC, that is, on EF10, these flags can be used to fill in legacy RSS mode field in MCDI request. Generalised flags can also be directly used in EF10-specific handler as they are fully compatible with additional fields of the same MCDI request. Legacy flags undergo conversion to generalised flags before they are consumed by a chip-specific handler. This conversion is used to make sure that chip-specific handlers expect only generalised flags in the input for the sake of clarity of the code. Depending on firmware capabilities, a chip-specififc handler either supplies the input to the NIC directly, for example, EFX_RX_HASH(IPV4_TCP, 4TUPLE) flag will enable 4 bits in RSS_CONTEXT_SET_FLAGS_IN_TCP_IPV4_RSS_MODE field on EF10, or takes the opportunity to translate the input to enable bits which don't map to the generic flag, like setting RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_TCPV4_EN on EF10 when the firmware claims no support for additional modes. However, this approach has introduced a severe problem which can be reproduced with ultra-low-latency firmware variant. In order to enable IP hash, EF10-specific handler requires the user to request 2-tuple hash for IP-other, TCP and UDP traffic classes, unconditionally. In example, IPv4 hash can be enabled using the following input: EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4_UDP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). At the same time, on ultra-low-latency firmware, the common code will never report support for any UDP tuple to the client driver. That is, in the same example, the driver will use EFX_RX_HASH(IPV4_TCP, 2TUPLE) | EFX_RX_HASH(IPV4, 2TUPLE). This input will not be recognised by EF10-specific handler, and RSS_CONTEXT_SET_FLAGS_IN_TOEPLITZ_IPV4_EN bit will not be set in the MCDI request. In order to solve the problem, the patch removes conversion code from chip-specific handlers and adds appropriate code to convert EFX_RX_HASH() flags to their legacy counterparts to the common scale mode set function. If the firmware does not support additional modes, the function will convert generalised flags to legacy flags correctly without any demand for UDP flags and pass the result to a chip-specific handler. Submitted by: Ivan Malov <ivan.malov at oktetlabs.ru> Sponsored by: Solarflare Communications, Inc. Differential Revision: https://reviews.freebsd.org/D18269
2018-11-30 07:07:31 +00:00
(type & EFX_RX_HASH_IPV6) ? B_TRUE : B_FALSE,
(type & EFX_RX_HASH_TCPIPV6) ? B_TRUE : B_FALSE,
rc);
if (rc != 0)
goto fail2;
break;
default:
rc = EINVAL;
goto fail3;
}
return (0);
fail3:
EFSYS_PROBE(fail3);
fail2:
EFSYS_PROBE(fail2);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
EFX_RX_LFSR_HASH(enp, B_FALSE);
return (rc);
}
#endif
#if EFSYS_OPT_RX_SCALE
static __checkReturn efx_rc_t
siena_rx_scale_key_set(
__in efx_nic_t *enp,
__in uint32_t rss_context,
__in_ecount(n) uint8_t *key,
__in size_t n)
{
efx_oword_t oword;
unsigned int byte;
unsigned int offset;
efx_rc_t rc;
if (rss_context != EFX_RSS_CONTEXT_DEFAULT) {
rc = EINVAL;
goto fail1;
}
byte = 0;
/* Write Toeplitz IPv4 hash key */
EFX_ZERO_OWORD(oword);
for (offset = (FRF_BZ_RX_RSS_TKEY_LBN + FRF_BZ_RX_RSS_TKEY_WIDTH) / 8;
offset > 0 && byte < n;
--offset)
oword.eo_u8[offset - 1] = key[byte++];
EFX_BAR_WRITEO(enp, FR_BZ_RX_RSS_TKEY_REG, &oword);
byte = 0;
/* Verify Toeplitz IPv4 hash key */
EFX_BAR_READO(enp, FR_BZ_RX_RSS_TKEY_REG, &oword);
for (offset = (FRF_BZ_RX_RSS_TKEY_LBN + FRF_BZ_RX_RSS_TKEY_WIDTH) / 8;
offset > 0 && byte < n;
--offset) {
if (oword.eo_u8[offset - 1] != key[byte++]) {
rc = EFAULT;
goto fail2;
}
}
if ((enp->en_features & EFX_FEATURE_IPV6) == 0)
goto done;
byte = 0;
/* Write Toeplitz IPv6 hash key 3 */
EFX_BAR_READO(enp, FR_CZ_RX_RSS_IPV6_REG3, &oword);
for (offset = (FRF_CZ_RX_RSS_IPV6_TKEY_HI_LBN +
FRF_CZ_RX_RSS_IPV6_TKEY_HI_WIDTH) / 8;
offset > 0 && byte < n;
--offset)
oword.eo_u8[offset - 1] = key[byte++];
EFX_BAR_WRITEO(enp, FR_CZ_RX_RSS_IPV6_REG3, &oword);
/* Write Toeplitz IPv6 hash key 2 */
EFX_ZERO_OWORD(oword);
for (offset = (FRF_CZ_RX_RSS_IPV6_TKEY_MID_LBN +
FRF_CZ_RX_RSS_IPV6_TKEY_MID_WIDTH) / 8;
offset > 0 && byte < n;
--offset)
oword.eo_u8[offset - 1] = key[byte++];
EFX_BAR_WRITEO(enp, FR_CZ_RX_RSS_IPV6_REG2, &oword);
/* Write Toeplitz IPv6 hash key 1 */
EFX_ZERO_OWORD(oword);
for (offset = (FRF_CZ_RX_RSS_IPV6_TKEY_LO_LBN +
FRF_CZ_RX_RSS_IPV6_TKEY_LO_WIDTH) / 8;
offset > 0 && byte < n;
--offset)
oword.eo_u8[offset - 1] = key[byte++];
EFX_BAR_WRITEO(enp, FR_CZ_RX_RSS_IPV6_REG1, &oword);
byte = 0;
/* Verify Toeplitz IPv6 hash key 3 */
EFX_BAR_READO(enp, FR_CZ_RX_RSS_IPV6_REG3, &oword);
for (offset = (FRF_CZ_RX_RSS_IPV6_TKEY_HI_LBN +
FRF_CZ_RX_RSS_IPV6_TKEY_HI_WIDTH) / 8;
offset > 0 && byte < n;
--offset) {
if (oword.eo_u8[offset - 1] != key[byte++]) {
rc = EFAULT;
goto fail3;
}
}
/* Verify Toeplitz IPv6 hash key 2 */
EFX_BAR_READO(enp, FR_CZ_RX_RSS_IPV6_REG2, &oword);
for (offset = (FRF_CZ_RX_RSS_IPV6_TKEY_MID_LBN +
FRF_CZ_RX_RSS_IPV6_TKEY_MID_WIDTH) / 8;
offset > 0 && byte < n;
--offset) {
if (oword.eo_u8[offset - 1] != key[byte++]) {
rc = EFAULT;
goto fail4;
}
}
/* Verify Toeplitz IPv6 hash key 1 */
EFX_BAR_READO(enp, FR_CZ_RX_RSS_IPV6_REG1, &oword);
for (offset = (FRF_CZ_RX_RSS_IPV6_TKEY_LO_LBN +
FRF_CZ_RX_RSS_IPV6_TKEY_LO_WIDTH) / 8;
offset > 0 && byte < n;
--offset) {
if (oword.eo_u8[offset - 1] != key[byte++]) {
rc = EFAULT;
goto fail5;
}
}
done:
return (0);
fail5:
EFSYS_PROBE(fail5);
fail4:
EFSYS_PROBE(fail4);
fail3:
EFSYS_PROBE(fail3);
fail2:
EFSYS_PROBE(fail2);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
#endif
#if EFSYS_OPT_RX_SCALE
static __checkReturn efx_rc_t
siena_rx_scale_tbl_set(
__in efx_nic_t *enp,
__in uint32_t rss_context,
__in_ecount(n) unsigned int *table,
__in size_t n)
{
efx_oword_t oword;
int index;
efx_rc_t rc;
EFX_STATIC_ASSERT(EFX_RSS_TBL_SIZE == FR_BZ_RX_INDIRECTION_TBL_ROWS);
EFX_STATIC_ASSERT(EFX_MAXRSS == (1 << FRF_BZ_IT_QUEUE_WIDTH));
if (rss_context != EFX_RSS_CONTEXT_DEFAULT) {
rc = EINVAL;
goto fail1;
}
if (n > FR_BZ_RX_INDIRECTION_TBL_ROWS) {
rc = EINVAL;
goto fail2;
}
for (index = 0; index < FR_BZ_RX_INDIRECTION_TBL_ROWS; index++) {
uint32_t byte;
/* Calculate the entry to place in the table */
byte = (n > 0) ? (uint32_t)table[index % n] : 0;
EFSYS_PROBE2(table, int, index, uint32_t, byte);
EFX_POPULATE_OWORD_1(oword, FRF_BZ_IT_QUEUE, byte);
/* Write the table */
EFX_BAR_TBL_WRITEO(enp, FR_BZ_RX_INDIRECTION_TBL,
index, &oword, B_TRUE);
}
for (index = FR_BZ_RX_INDIRECTION_TBL_ROWS - 1; index >= 0; --index) {
uint32_t byte;
/* Determine if we're starting a new batch */
byte = (n > 0) ? (uint32_t)table[index % n] : 0;
/* Read the table */
EFX_BAR_TBL_READO(enp, FR_BZ_RX_INDIRECTION_TBL,
index, &oword, B_TRUE);
/* Verify the entry */
if (EFX_OWORD_FIELD(oword, FRF_BZ_IT_QUEUE) != byte) {
rc = EFAULT;
goto fail3;
}
}
return (0);
fail3:
EFSYS_PROBE(fail3);
fail2:
EFSYS_PROBE(fail2);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
#endif
/*
* Falcon/Siena pseudo-header
* --------------------------
*
* Receive packets are prefixed by an optional 16 byte pseudo-header.
* The pseudo-header is a byte array of one of the forms:
*
* 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
* xx.xx.xx.xx.xx.xx.xx.xx.xx.xx.xx.xx.TT.TT.TT.TT
* xx.xx.xx.xx.xx.xx.xx.xx.xx.xx.xx.xx.xx.xx.LL.LL
*
* where:
* TT.TT.TT.TT Toeplitz hash (32-bit big-endian)
* LL.LL LFSR hash (16-bit big-endian)
*/
#if EFSYS_OPT_RX_SCALE
static __checkReturn uint32_t
siena_rx_prefix_hash(
__in efx_nic_t *enp,
__in efx_rx_hash_alg_t func,
__in uint8_t *buffer)
{
_NOTE(ARGUNUSED(enp))
switch (func) {
case EFX_RX_HASHALG_TOEPLITZ:
return ((buffer[12] << 24) |
(buffer[13] << 16) |
(buffer[14] << 8) |
buffer[15]);
case EFX_RX_HASHALG_LFSR:
return ((buffer[14] << 8) | buffer[15]);
default:
EFSYS_ASSERT(0);
return (0);
}
}
#endif /* EFSYS_OPT_RX_SCALE */
static __checkReturn efx_rc_t
siena_rx_prefix_pktlen(
__in efx_nic_t *enp,
__in uint8_t *buffer,
__out uint16_t *lengthp)
{
_NOTE(ARGUNUSED(enp, buffer, lengthp))
/* Not supported by Falcon/Siena hardware */
EFSYS_ASSERT(0);
return (ENOTSUP);
}
static void
siena_rx_qpost(
__in efx_rxq_t *erp,
__in_ecount(ndescs) efsys_dma_addr_t *addrp,
__in size_t size,
__in unsigned int ndescs,
__in unsigned int completed,
__in unsigned int added)
{
efx_qword_t qword;
unsigned int i;
unsigned int offset;
unsigned int id;
/* The client driver must not overfill the queue */
EFSYS_ASSERT3U(added - completed + ndescs, <=,
EFX_RXQ_LIMIT(erp->er_mask + 1));
id = added & (erp->er_mask);
for (i = 0; i < ndescs; i++) {
EFSYS_PROBE4(rx_post, unsigned int, erp->er_index,
unsigned int, id, efsys_dma_addr_t, addrp[i],
size_t, size);
EFX_POPULATE_QWORD_3(qword,
FSF_AZ_RX_KER_BUF_SIZE, (uint32_t)(size),
FSF_AZ_RX_KER_BUF_ADDR_DW0,
(uint32_t)(addrp[i] & 0xffffffff),
FSF_AZ_RX_KER_BUF_ADDR_DW1,
(uint32_t)(addrp[i] >> 32));
offset = id * sizeof (efx_qword_t);
EFSYS_MEM_WRITEQ(erp->er_esmp, offset, &qword);
id = (id + 1) & (erp->er_mask);
}
}
static void
siena_rx_qpush(
__in efx_rxq_t *erp,
__in unsigned int added,
__inout unsigned int *pushedp)
{
efx_nic_t *enp = erp->er_enp;
unsigned int pushed = *pushedp;
uint32_t wptr;
efx_oword_t oword;
efx_dword_t dword;
/* All descriptors are pushed */
*pushedp = added;
/* Push the populated descriptors out */
wptr = added & erp->er_mask;
EFX_POPULATE_OWORD_1(oword, FRF_AZ_RX_DESC_WPTR, wptr);
/* Only write the third DWORD */
EFX_POPULATE_DWORD_1(dword,
EFX_DWORD_0, EFX_OWORD_FIELD(oword, EFX_DWORD_3));
/* Guarantee ordering of memory (descriptors) and PIO (doorbell) */
EFX_DMA_SYNC_QUEUE_FOR_DEVICE(erp->er_esmp, erp->er_mask + 1,
wptr, pushed & erp->er_mask);
EFSYS_PIO_WRITE_BARRIER();
EFX_BAR_TBL_WRITED3(enp, FR_BZ_RX_DESC_UPD_REGP0,
erp->er_index, &dword, B_FALSE);
}
#if EFSYS_OPT_RX_PACKED_STREAM
static void
siena_rx_qpush_ps_credits(
__in efx_rxq_t *erp)
{
/* Not supported by Siena hardware */
EFSYS_ASSERT(0);
}
static uint8_t *
siena_rx_qps_packet_info(
__in efx_rxq_t *erp,
__in uint8_t *buffer,
__in uint32_t buffer_length,
__in uint32_t current_offset,
__out uint16_t *lengthp,
__out uint32_t *next_offsetp,
__out uint32_t *timestamp)
{
/* Not supported by Siena hardware */
EFSYS_ASSERT(0);
return (NULL);
}
#endif /* EFSYS_OPT_RX_PACKED_STREAM */
static __checkReturn efx_rc_t
siena_rx_qflush(
__in efx_rxq_t *erp)
{
efx_nic_t *enp = erp->er_enp;
efx_oword_t oword;
uint32_t label;
label = erp->er_index;
/* Flush the queue */
EFX_POPULATE_OWORD_2(oword, FRF_AZ_RX_FLUSH_DESCQ_CMD, 1,
FRF_AZ_RX_FLUSH_DESCQ, label);
EFX_BAR_WRITEO(enp, FR_AZ_RX_FLUSH_DESCQ_REG, &oword);
return (0);
}
static void
siena_rx_qenable(
__in efx_rxq_t *erp)
{
efx_nic_t *enp = erp->er_enp;
efx_oword_t oword;
EFSYS_ASSERT3U(erp->er_magic, ==, EFX_RXQ_MAGIC);
EFX_BAR_TBL_READO(enp, FR_AZ_RX_DESC_PTR_TBL,
erp->er_index, &oword, B_TRUE);
EFX_SET_OWORD_FIELD(oword, FRF_AZ_RX_DC_HW_RPTR, 0);
EFX_SET_OWORD_FIELD(oword, FRF_AZ_RX_DESCQ_HW_RPTR, 0);
EFX_SET_OWORD_FIELD(oword, FRF_AZ_RX_DESCQ_EN, 1);
EFX_BAR_TBL_WRITEO(enp, FR_AZ_RX_DESC_PTR_TBL,
erp->er_index, &oword, B_TRUE);
}
static __checkReturn efx_rc_t
siena_rx_qcreate(
__in efx_nic_t *enp,
__in unsigned int index,
__in unsigned int label,
__in efx_rxq_type_t type,
__in_opt const efx_rxq_type_data_t *type_data,
__in efsys_mem_t *esmp,
__in size_t ndescs,
__in uint32_t id,
__in unsigned int flags,
__in efx_evq_t *eep,
__in efx_rxq_t *erp)
{
efx_nic_cfg_t *encp = &(enp->en_nic_cfg);
efx_oword_t oword;
uint32_t size;
boolean_t jumbo = B_FALSE;
efx_rc_t rc;
_NOTE(ARGUNUSED(esmp))
_NOTE(ARGUNUSED(type_data))
EFX_STATIC_ASSERT(EFX_EV_RX_NLABELS ==
(1 << FRF_AZ_RX_DESCQ_LABEL_WIDTH));
EFSYS_ASSERT3U(label, <, EFX_EV_RX_NLABELS);
EFSYS_ASSERT3U(enp->en_rx_qcount + 1, <, encp->enc_rxq_limit);
EFX_STATIC_ASSERT(ISP2(EFX_RXQ_MAXNDESCS));
EFX_STATIC_ASSERT(ISP2(EFX_RXQ_MINNDESCS));
if (!ISP2(ndescs) ||
(ndescs < EFX_RXQ_MINNDESCS) || (ndescs > EFX_RXQ_MAXNDESCS)) {
rc = EINVAL;
goto fail1;
}
if (index >= encp->enc_rxq_limit) {
rc = EINVAL;
goto fail2;
}
for (size = 0; (1 << size) <= (EFX_RXQ_MAXNDESCS / EFX_RXQ_MINNDESCS);
size++)
if ((1 << size) == (int)(ndescs / EFX_RXQ_MINNDESCS))
break;
if (id + (1 << size) >= encp->enc_buftbl_limit) {
rc = EINVAL;
goto fail3;
}
switch (type) {
case EFX_RXQ_TYPE_DEFAULT:
break;
default:
rc = EINVAL;
goto fail4;
}
if (flags & EFX_RXQ_FLAG_SCATTER) {
#if EFSYS_OPT_RX_SCATTER
jumbo = B_TRUE;
#else
rc = EINVAL;
goto fail5;
#endif /* EFSYS_OPT_RX_SCATTER */
}
/* Set up the new descriptor queue */
EFX_POPULATE_OWORD_7(oword,
FRF_AZ_RX_DESCQ_BUF_BASE_ID, id,
FRF_AZ_RX_DESCQ_EVQ_ID, eep->ee_index,
FRF_AZ_RX_DESCQ_OWNER_ID, 0,
FRF_AZ_RX_DESCQ_LABEL, label,
FRF_AZ_RX_DESCQ_SIZE, size,
FRF_AZ_RX_DESCQ_TYPE, 0,
FRF_AZ_RX_DESCQ_JUMBO, jumbo);
EFX_BAR_TBL_WRITEO(enp, FR_AZ_RX_DESC_PTR_TBL,
erp->er_index, &oword, B_TRUE);
return (0);
#if !EFSYS_OPT_RX_SCATTER
fail5:
EFSYS_PROBE(fail5);
#endif
fail4:
EFSYS_PROBE(fail4);
fail3:
EFSYS_PROBE(fail3);
fail2:
EFSYS_PROBE(fail2);
fail1:
EFSYS_PROBE1(fail1, efx_rc_t, rc);
return (rc);
}
static void
siena_rx_qdestroy(
__in efx_rxq_t *erp)
{
efx_nic_t *enp = erp->er_enp;
efx_oword_t oword;
EFSYS_ASSERT(enp->en_rx_qcount != 0);
--enp->en_rx_qcount;
/* Purge descriptor queue */
EFX_ZERO_OWORD(oword);
EFX_BAR_TBL_WRITEO(enp, FR_AZ_RX_DESC_PTR_TBL,
erp->er_index, &oword, B_TRUE);
/* Free the RXQ object */
EFSYS_KMEM_FREE(enp->en_esip, sizeof (efx_rxq_t), erp);
}
static void
siena_rx_fini(
__in efx_nic_t *enp)
{
_NOTE(ARGUNUSED(enp))
}
#endif /* EFSYS_OPT_SIENA */