* [PATCH v9 0/7] igb: packet-split descriptors support [not found] <CGME20230525153743eucas1p145858f3572c2d53687377a4e879d1e0b@eucas1p1.samsung.com> @ 2023-05-25 15:37 ` Tomasz Dzieciol [not found] ` <CGME20230525153743eucas1p2cfe25bb5c0bfc59064189d58b328a537@eucas1p2.samsung.com> ` (7 more replies) 0 siblings, 8 replies; 11+ messages in thread From: Tomasz Dzieciol @ 2023-05-25 15:37 UTC (permalink / raw) To: qemu-devel, akihiko.odaki Cc: sriram.yagnaraman, jasowang, k.kwiecien, m.sochacki Based-on: <20230523024339.50875-1-akihiko.odaki@daynix.com> ("[PATCH v5 00/48] igb: Fix for DPDK") Purposes of this series of patches: * introduce packet-split RX descriptors support. This feature is used by Linux VF driver for MTU values from 2048. * refactor RX descriptor handling for introduction of packet-split RX descriptors support * fix descriptors flags handling In addition to comments from previous review endianess issues in igb_write_adv_ps_rx_descr were fixed. Tomasz Dzieciol (7): igb: remove TCP ACK detection igb: rename E1000E_RingInfo_st igb: RX descriptors guest writting refactoring igb: RX payload guest writting refactoring igb: add IPv6 extended headers traffic detection igb: packet-split descriptors support e1000e: rename e1000e_ba_state and e1000e_write_hdr_to_rx_buffers hw/net/e1000e_core.c | 78 +++-- hw/net/igb_core.c | 740 ++++++++++++++++++++++++++++----------- hw/net/igb_regs.h | 20 +- hw/net/trace-events | 6 +- tests/qtest/libqos/igb.c | 5 + 5 files changed, 602 insertions(+), 247 deletions(-) -- 2.25.1 ^ permalink raw reply [flat|nested] 11+ messages in thread
[parent not found: <CGME20230525153743eucas1p2cfe25bb5c0bfc59064189d58b328a537@eucas1p2.samsung.com>]
* [PATCH v9 1/7] igb: remove TCP ACK detection [not found] ` <CGME20230525153743eucas1p2cfe25bb5c0bfc59064189d58b328a537@eucas1p2.samsung.com> @ 2023-05-25 15:37 ` Tomasz Dzieciol 0 siblings, 0 replies; 11+ messages in thread From: Tomasz Dzieciol @ 2023-05-25 15:37 UTC (permalink / raw) To: qemu-devel, akihiko.odaki Cc: sriram.yagnaraman, jasowang, k.kwiecien, m.sochacki TCP ACK detection is no longer present in igb. Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com> --- hw/net/igb_core.c | 5 ----- 1 file changed, 5 deletions(-) diff --git a/hw/net/igb_core.c b/hw/net/igb_core.c index d00b1caa6a..e927c51061 100644 --- a/hw/net/igb_core.c +++ b/hw/net/igb_core.c @@ -1327,11 +1327,6 @@ igb_build_rx_metadata(IGBCore *core, trace_e1000e_rx_metadata_ip_id(*ip_id); } - if (l4hdr_proto == ETH_L4_HDR_PROTO_TCP && net_rx_pkt_is_tcp_ack(pkt)) { - *status_flags |= E1000_RXD_STAT_ACK; - trace_e1000e_rx_metadata_ack(); - } - if (pkt_info) { *pkt_info = rss_info->enabled ? rss_info->type : 0; -- 2.25.1 ^ permalink raw reply related [flat|nested] 11+ messages in thread
[parent not found: <CGME20230525153744eucas1p208318603e67e11c9ab3f0a6556b8d3a4@eucas1p2.samsung.com>]
* [PATCH v9 2/7] igb: rename E1000E_RingInfo_st [not found] ` <CGME20230525153744eucas1p208318603e67e11c9ab3f0a6556b8d3a4@eucas1p2.samsung.com> @ 2023-05-25 15:37 ` Tomasz Dzieciol 0 siblings, 0 replies; 11+ messages in thread From: Tomasz Dzieciol @ 2023-05-25 15:37 UTC (permalink / raw) To: qemu-devel, akihiko.odaki Cc: sriram.yagnaraman, jasowang, k.kwiecien, m.sochacki Rename E1000E_RingInfo_st and E1000E_RingInfo according to qemu typdefs guide. Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com> --- hw/net/e1000e_core.c | 34 +++++++++++++++++----------------- hw/net/igb_core.c | 42 +++++++++++++++++++++--------------------- 2 files changed, 38 insertions(+), 38 deletions(-) diff --git a/hw/net/e1000e_core.c b/hw/net/e1000e_core.c index 9f185d099c..f6b628eaef 100644 --- a/hw/net/e1000e_core.c +++ b/hw/net/e1000e_core.c @@ -810,24 +810,24 @@ e1000e_txdesc_writeback(E1000ECore *core, dma_addr_t base, return e1000e_tx_wb_interrupt_cause(core, queue_idx); } -typedef struct E1000E_RingInfo_st { +typedef struct E1000ERingInfo { int dbah; int dbal; int dlen; int dh; int dt; int idx; -} E1000E_RingInfo; +} E1000ERingInfo; static inline bool -e1000e_ring_empty(E1000ECore *core, const E1000E_RingInfo *r) +e1000e_ring_empty(E1000ECore *core, const E1000ERingInfo *r) { return core->mac[r->dh] == core->mac[r->dt] || core->mac[r->dt] >= core->mac[r->dlen] / E1000_RING_DESC_LEN; } static inline uint64_t -e1000e_ring_base(E1000ECore *core, const E1000E_RingInfo *r) +e1000e_ring_base(E1000ECore *core, const E1000ERingInfo *r) { uint64_t bah = core->mac[r->dbah]; uint64_t bal = core->mac[r->dbal]; @@ -836,13 +836,13 @@ e1000e_ring_base(E1000ECore *core, const E1000E_RingInfo *r) } static inline uint64_t -e1000e_ring_head_descr(E1000ECore *core, const E1000E_RingInfo *r) +e1000e_ring_head_descr(E1000ECore *core, const E1000ERingInfo *r) { return e1000e_ring_base(core, r) + E1000_RING_DESC_LEN * core->mac[r->dh]; } static inline void -e1000e_ring_advance(E1000ECore *core, const E1000E_RingInfo *r, uint32_t count) +e1000e_ring_advance(E1000ECore *core, const E1000ERingInfo *r, uint32_t count) { core->mac[r->dh] += count; @@ -852,7 +852,7 @@ e1000e_ring_advance(E1000ECore *core, const E1000E_RingInfo *r, uint32_t count) } static inline uint32_t -e1000e_ring_free_descr_num(E1000ECore *core, const E1000E_RingInfo *r) +e1000e_ring_free_descr_num(E1000ECore *core, const E1000ERingInfo *r) { trace_e1000e_ring_free_space(r->idx, core->mac[r->dlen], core->mac[r->dh], core->mac[r->dt]); @@ -871,19 +871,19 @@ e1000e_ring_free_descr_num(E1000ECore *core, const E1000E_RingInfo *r) } static inline bool -e1000e_ring_enabled(E1000ECore *core, const E1000E_RingInfo *r) +e1000e_ring_enabled(E1000ECore *core, const E1000ERingInfo *r) { return core->mac[r->dlen] > 0; } static inline uint32_t -e1000e_ring_len(E1000ECore *core, const E1000E_RingInfo *r) +e1000e_ring_len(E1000ECore *core, const E1000ERingInfo *r) { return core->mac[r->dlen]; } typedef struct E1000E_TxRing_st { - const E1000E_RingInfo *i; + const E1000ERingInfo *i; struct e1000e_tx *tx; } E1000E_TxRing; @@ -896,7 +896,7 @@ e1000e_mq_queue_idx(int base_reg_idx, int reg_idx) static inline void e1000e_tx_ring_init(E1000ECore *core, E1000E_TxRing *txr, int idx) { - static const E1000E_RingInfo i[E1000E_NUM_QUEUES] = { + static const E1000ERingInfo i[E1000E_NUM_QUEUES] = { { TDBAH, TDBAL, TDLEN, TDH, TDT, 0 }, { TDBAH1, TDBAL1, TDLEN1, TDH1, TDT1, 1 } }; @@ -908,13 +908,13 @@ e1000e_tx_ring_init(E1000ECore *core, E1000E_TxRing *txr, int idx) } typedef struct E1000E_RxRing_st { - const E1000E_RingInfo *i; + const E1000ERingInfo *i; } E1000E_RxRing; static inline void e1000e_rx_ring_init(E1000ECore *core, E1000E_RxRing *rxr, int idx) { - static const E1000E_RingInfo i[E1000E_NUM_QUEUES] = { + static const E1000ERingInfo i[E1000E_NUM_QUEUES] = { { RDBAH0, RDBAL0, RDLEN0, RDH0, RDT0, 0 }, { RDBAH1, RDBAL1, RDLEN1, RDH1, RDT1, 1 } }; @@ -930,7 +930,7 @@ e1000e_start_xmit(E1000ECore *core, const E1000E_TxRing *txr) dma_addr_t base; struct e1000_tx_desc desc; bool ide = false; - const E1000E_RingInfo *txi = txr->i; + const E1000ERingInfo *txi = txr->i; uint32_t cause = E1000_ICS_TXQE; if (!(core->mac[TCTL] & E1000_TCTL_EN)) { @@ -960,7 +960,7 @@ e1000e_start_xmit(E1000ECore *core, const E1000E_TxRing *txr) } static bool -e1000e_has_rxbufs(E1000ECore *core, const E1000E_RingInfo *r, +e1000e_has_rxbufs(E1000ECore *core, const E1000ERingInfo *r, size_t total_size) { uint32_t bufs = e1000e_ring_free_descr_num(core, r); @@ -1460,7 +1460,7 @@ e1000e_update_rx_stats(E1000ECore *core, size_t pkt_size, size_t pkt_fcs_size) } static inline bool -e1000e_rx_descr_threshold_hit(E1000ECore *core, const E1000E_RingInfo *rxi) +e1000e_rx_descr_threshold_hit(E1000ECore *core, const E1000ERingInfo *rxi) { return e1000e_ring_free_descr_num(core, rxi) == e1000e_ring_len(core, rxi) >> core->rxbuf_min_shift; @@ -1521,7 +1521,7 @@ e1000e_write_packet_to_guest(E1000ECore *core, struct NetRxPkt *pkt, struct iovec *iov = net_rx_pkt_get_iovec(pkt); size_t size = net_rx_pkt_get_total_len(pkt); size_t total_size = size + e1000x_fcs_len(core->mac); - const E1000E_RingInfo *rxi; + const E1000ERingInfo *rxi; size_t ps_hdr_len = 0; bool do_ps = e1000e_do_ps(core, pkt, &ps_hdr_len); bool is_first = true; diff --git a/hw/net/igb_core.c b/hw/net/igb_core.c index e927c51061..051980b4f5 100644 --- a/hw/net/igb_core.c +++ b/hw/net/igb_core.c @@ -694,24 +694,24 @@ static uint32_t igb_rx_wb_eic(IGBCore *core, int queue_idx) return (ent & E1000_IVAR_VALID) ? BIT(ent & 0x1f) : 0; } -typedef struct E1000E_RingInfo_st { +typedef struct E1000ERingInfo { int dbah; int dbal; int dlen; int dh; int dt; int idx; -} E1000E_RingInfo; +} E1000ERingInfo; static inline bool -igb_ring_empty(IGBCore *core, const E1000E_RingInfo *r) +igb_ring_empty(IGBCore *core, const E1000ERingInfo *r) { return core->mac[r->dh] == core->mac[r->dt] || core->mac[r->dt] >= core->mac[r->dlen] / E1000_RING_DESC_LEN; } static inline uint64_t -igb_ring_base(IGBCore *core, const E1000E_RingInfo *r) +igb_ring_base(IGBCore *core, const E1000ERingInfo *r) { uint64_t bah = core->mac[r->dbah]; uint64_t bal = core->mac[r->dbal]; @@ -720,13 +720,13 @@ igb_ring_base(IGBCore *core, const E1000E_RingInfo *r) } static inline uint64_t -igb_ring_head_descr(IGBCore *core, const E1000E_RingInfo *r) +igb_ring_head_descr(IGBCore *core, const E1000ERingInfo *r) { return igb_ring_base(core, r) + E1000_RING_DESC_LEN * core->mac[r->dh]; } static inline void -igb_ring_advance(IGBCore *core, const E1000E_RingInfo *r, uint32_t count) +igb_ring_advance(IGBCore *core, const E1000ERingInfo *r, uint32_t count) { core->mac[r->dh] += count; @@ -736,7 +736,7 @@ igb_ring_advance(IGBCore *core, const E1000E_RingInfo *r, uint32_t count) } static inline uint32_t -igb_ring_free_descr_num(IGBCore *core, const E1000E_RingInfo *r) +igb_ring_free_descr_num(IGBCore *core, const E1000ERingInfo *r) { trace_e1000e_ring_free_space(r->idx, core->mac[r->dlen], core->mac[r->dh], core->mac[r->dt]); @@ -755,13 +755,13 @@ igb_ring_free_descr_num(IGBCore *core, const E1000E_RingInfo *r) } static inline bool -igb_ring_enabled(IGBCore *core, const E1000E_RingInfo *r) +igb_ring_enabled(IGBCore *core, const E1000ERingInfo *r) { return core->mac[r->dlen] > 0; } typedef struct IGB_TxRing_st { - const E1000E_RingInfo *i; + const E1000ERingInfo *i; struct igb_tx *tx; } IGB_TxRing; @@ -774,7 +774,7 @@ igb_mq_queue_idx(int base_reg_idx, int reg_idx) static inline void igb_tx_ring_init(IGBCore *core, IGB_TxRing *txr, int idx) { - static const E1000E_RingInfo i[IGB_NUM_QUEUES] = { + static const E1000ERingInfo i[IGB_NUM_QUEUES] = { { TDBAH0, TDBAL0, TDLEN0, TDH0, TDT0, 0 }, { TDBAH1, TDBAL1, TDLEN1, TDH1, TDT1, 1 }, { TDBAH2, TDBAL2, TDLEN2, TDH2, TDT2, 2 }, @@ -800,13 +800,13 @@ igb_tx_ring_init(IGBCore *core, IGB_TxRing *txr, int idx) } typedef struct E1000E_RxRing_st { - const E1000E_RingInfo *i; + const E1000ERingInfo *i; } E1000E_RxRing; static inline void igb_rx_ring_init(IGBCore *core, E1000E_RxRing *rxr, int idx) { - static const E1000E_RingInfo i[IGB_NUM_QUEUES] = { + static const E1000ERingInfo i[IGB_NUM_QUEUES] = { { RDBAH0, RDBAL0, RDLEN0, RDH0, RDT0, 0 }, { RDBAH1, RDBAL1, RDLEN1, RDH1, RDT1, 1 }, { RDBAH2, RDBAL2, RDLEN2, RDH2, RDT2, 2 }, @@ -833,7 +833,7 @@ igb_rx_ring_init(IGBCore *core, E1000E_RxRing *rxr, int idx) static uint32_t igb_txdesc_writeback(IGBCore *core, dma_addr_t base, union e1000_adv_tx_desc *tx_desc, - const E1000E_RingInfo *txi) + const E1000ERingInfo *txi) { PCIDevice *d; uint32_t cmd_type_len = le32_to_cpu(tx_desc->read.cmd_type_len); @@ -866,7 +866,7 @@ igb_txdesc_writeback(IGBCore *core, dma_addr_t base, } static inline bool -igb_tx_enabled(IGBCore *core, const E1000E_RingInfo *txi) +igb_tx_enabled(IGBCore *core, const E1000ERingInfo *txi) { bool vmdq = core->mac[MRQC] & 1; uint16_t qn = txi->idx; @@ -883,7 +883,7 @@ igb_start_xmit(IGBCore *core, const IGB_TxRing *txr) PCIDevice *d; dma_addr_t base; union e1000_adv_tx_desc desc; - const E1000E_RingInfo *txi = txr->i; + const E1000ERingInfo *txi = txr->i; uint32_t eic = 0; if (!igb_tx_enabled(core, txi)) { @@ -918,7 +918,7 @@ igb_start_xmit(IGBCore *core, const IGB_TxRing *txr) } static uint32_t -igb_rxbufsize(IGBCore *core, const E1000E_RingInfo *r) +igb_rxbufsize(IGBCore *core, const E1000ERingInfo *r) { uint32_t srrctl = core->mac[E1000_SRRCTL(r->idx) >> 2]; uint32_t bsizepkt = srrctl & E1000_SRRCTL_BSIZEPKT_MASK; @@ -930,7 +930,7 @@ igb_rxbufsize(IGBCore *core, const E1000E_RingInfo *r) } static bool -igb_has_rxbufs(IGBCore *core, const E1000E_RingInfo *r, size_t total_size) +igb_has_rxbufs(IGBCore *core, const E1000ERingInfo *r, size_t total_size) { uint32_t bufs = igb_ring_free_descr_num(core, r); uint32_t bufsize = igb_rxbufsize(core, r); @@ -1522,7 +1522,7 @@ igb_write_to_rx_buffers(IGBCore *core, } static void -igb_update_rx_stats(IGBCore *core, const E1000E_RingInfo *rxi, +igb_update_rx_stats(IGBCore *core, const E1000ERingInfo *rxi, size_t pkt_size, size_t pkt_fcs_size) { eth_pkt_types_e pkt_type = net_rx_pkt_get_packet_type(core->rx_pkt); @@ -1540,7 +1540,7 @@ igb_update_rx_stats(IGBCore *core, const E1000E_RingInfo *rxi, } static inline bool -igb_rx_descr_threshold_hit(IGBCore *core, const E1000E_RingInfo *rxi) +igb_rx_descr_threshold_hit(IGBCore *core, const E1000ERingInfo *rxi) { return igb_ring_free_descr_num(core, rxi) == ((core->mac[E1000_SRRCTL(rxi->idx) >> 2] >> 20) & 31) * 16; @@ -1562,7 +1562,7 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, struct iovec *iov = net_rx_pkt_get_iovec(pkt); size_t size = net_rx_pkt_get_total_len(pkt); size_t total_size = size + e1000x_fcs_len(core->mac); - const E1000E_RingInfo *rxi = rxr->i; + const E1000ERingInfo *rxi = rxr->i; size_t bufsize = igb_rxbufsize(core, rxi); d = pcie_sriov_get_vf_at_index(core->owner, rxi->idx % 8); @@ -1643,7 +1643,7 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, } static bool -igb_rx_strip_vlan(IGBCore *core, const E1000E_RingInfo *rxi) +igb_rx_strip_vlan(IGBCore *core, const E1000ERingInfo *rxi) { if (core->mac[MRQC] & 1) { uint16_t pool = rxi->idx % IGB_NUM_VM_POOLS; -- 2.25.1 ^ permalink raw reply related [flat|nested] 11+ messages in thread
[parent not found: <CGME20230525153744eucas1p1844ba0a10563caff4bb0f92c82f30f7d@eucas1p1.samsung.com>]
* [PATCH v9 3/7] igb: RX descriptors guest writting refactoring [not found] ` <CGME20230525153744eucas1p1844ba0a10563caff4bb0f92c82f30f7d@eucas1p1.samsung.com> @ 2023-05-25 15:37 ` Tomasz Dzieciol 0 siblings, 0 replies; 11+ messages in thread From: Tomasz Dzieciol @ 2023-05-25 15:37 UTC (permalink / raw) To: qemu-devel, akihiko.odaki Cc: sriram.yagnaraman, jasowang, k.kwiecien, m.sochacki Refactoring is done in preparation for support of multiple advanced descriptors RX modes, especially packet-split modes. Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com> --- hw/net/igb_core.c | 170 +++++++++++++++++++++++--------------------- hw/net/igb_regs.h | 10 +-- hw/net/trace-events | 4 +- 3 files changed, 96 insertions(+), 88 deletions(-) diff --git a/hw/net/igb_core.c b/hw/net/igb_core.c index 051980b4f5..3a6df0d55e 100644 --- a/hw/net/igb_core.c +++ b/hw/net/igb_core.c @@ -1281,15 +1281,11 @@ igb_verify_csum_in_sw(IGBCore *core, } static void -igb_build_rx_metadata(IGBCore *core, - struct NetRxPkt *pkt, - bool is_eop, - const E1000E_RSSInfo *rss_info, uint16_t etqf, bool ts, - uint16_t *pkt_info, uint16_t *hdr_info, - uint32_t *rss, - uint32_t *status_flags, - uint16_t *ip_id, - uint16_t *vlan_tag) +igb_build_rx_metadata_common(IGBCore *core, + struct NetRxPkt *pkt, + bool is_eop, + uint32_t *status_flags, + uint16_t *vlan_tag) { struct virtio_net_hdr *vhdr; bool hasip4, hasip6, csum_valid; @@ -1298,7 +1294,6 @@ igb_build_rx_metadata(IGBCore *core, *status_flags = E1000_RXD_STAT_DD; /* No additional metadata needed for non-EOP descriptors */ - /* TODO: EOP apply only to status so don't skip whole function. */ if (!is_eop) { goto func_exit; } @@ -1315,59 +1310,6 @@ igb_build_rx_metadata(IGBCore *core, trace_e1000e_rx_metadata_vlan(*vlan_tag); } - /* Packet parsing results */ - if ((core->mac[RXCSUM] & E1000_RXCSUM_PCSD) != 0) { - if (rss_info->enabled) { - *rss = cpu_to_le32(rss_info->hash); - trace_igb_rx_metadata_rss(*rss); - } - } else if (hasip4) { - *status_flags |= E1000_RXD_STAT_IPIDV; - *ip_id = cpu_to_le16(net_rx_pkt_get_ip_id(pkt)); - trace_e1000e_rx_metadata_ip_id(*ip_id); - } - - if (pkt_info) { - *pkt_info = rss_info->enabled ? rss_info->type : 0; - - if (etqf < 8) { - *pkt_info |= (BIT(11) | etqf) << 4; - } else { - if (hasip4) { - *pkt_info |= E1000_ADVRXD_PKT_IP4; - } - - if (hasip6) { - *pkt_info |= E1000_ADVRXD_PKT_IP6; - } - - switch (l4hdr_proto) { - case ETH_L4_HDR_PROTO_TCP: - *pkt_info |= E1000_ADVRXD_PKT_TCP; - break; - - case ETH_L4_HDR_PROTO_UDP: - *pkt_info |= E1000_ADVRXD_PKT_UDP; - break; - - case ETH_L4_HDR_PROTO_SCTP: - *pkt_info |= E1000_ADVRXD_PKT_SCTP; - break; - - default: - break; - } - } - } - - if (hdr_info) { - *hdr_info = 0; - } - - if (ts) { - *status_flags |= BIT(16); - } - /* RX CSO information */ if (hasip6 && (core->mac[RFCTL] & E1000_RFCTL_IPV6_XSUM_DIS)) { trace_e1000e_rx_metadata_ipv6_sum_disabled(); @@ -1423,43 +1365,108 @@ func_exit: static inline void igb_write_lgcy_rx_descr(IGBCore *core, struct e1000_rx_desc *desc, struct NetRxPkt *pkt, - const E1000E_RSSInfo *rss_info, uint16_t etqf, bool ts, + const E1000E_RSSInfo *rss_info, uint16_t length) { - uint32_t status_flags, rss; - uint16_t ip_id; + uint32_t status_flags; assert(!rss_info->enabled); + + memset(desc, 0, sizeof(*desc)); desc->length = cpu_to_le16(length); - desc->csum = 0; + igb_build_rx_metadata_common(core, pkt, pkt != NULL, + &status_flags, + &desc->special); - igb_build_rx_metadata(core, pkt, pkt != NULL, - rss_info, etqf, ts, - NULL, NULL, &rss, - &status_flags, &ip_id, - &desc->special); desc->errors = (uint8_t) (le32_to_cpu(status_flags) >> 24); desc->status = (uint8_t) le32_to_cpu(status_flags); } +static uint16_t +igb_rx_desc_get_packet_type(IGBCore *core, struct NetRxPkt *pkt, uint16_t etqf) +{ + uint16_t pkt_type; + bool hasip4, hasip6; + EthL4HdrProto l4hdr_proto; + + if (etqf < 8) { + pkt_type = BIT(11) | etqf; + return pkt_type; + } + + net_rx_pkt_get_protocols(pkt, &hasip4, &hasip6, &l4hdr_proto); + + if (hasip6 && !(core->mac[RFCTL] & E1000_RFCTL_IPV6_DIS)) { + pkt_type = E1000_ADVRXD_PKT_IP6; + } else if (hasip4) { + pkt_type = E1000_ADVRXD_PKT_IP4; + } else { + pkt_type = 0; + } + + switch (l4hdr_proto) { + case ETH_L4_HDR_PROTO_TCP: + pkt_type |= E1000_ADVRXD_PKT_TCP; + break; + case ETH_L4_HDR_PROTO_UDP: + pkt_type |= E1000_ADVRXD_PKT_UDP; + break; + case ETH_L4_HDR_PROTO_SCTP: + pkt_type |= E1000_ADVRXD_PKT_SCTP; + break; + default: + break; + } + + return pkt_type; +} + static inline void igb_write_adv_rx_descr(IGBCore *core, union e1000_adv_rx_desc *desc, struct NetRxPkt *pkt, const E1000E_RSSInfo *rss_info, uint16_t etqf, bool ts, uint16_t length) { + bool hasip4, hasip6; + EthL4HdrProto l4hdr_proto; + uint16_t rss_type = 0, pkt_type; + bool eop = (pkt != NULL); + uint32_t adv_desc_status_error = 0; memset(&desc->wb, 0, sizeof(desc->wb)); desc->wb.upper.length = cpu_to_le16(length); + igb_build_rx_metadata_common(core, pkt, eop, + &desc->wb.upper.status_error, + &desc->wb.upper.vlan); + + if (!eop) { + return; + } + + net_rx_pkt_get_protocols(pkt, &hasip4, &hasip6, &l4hdr_proto); + + if ((core->mac[RXCSUM] & E1000_RXCSUM_PCSD) != 0) { + if (rss_info->enabled) { + desc->wb.lower.hi_dword.rss = cpu_to_le32(rss_info->hash); + rss_type = rss_info->type; + trace_igb_rx_metadata_rss(desc->wb.lower.hi_dword.rss, rss_type); + } + } else if (hasip4) { + adv_desc_status_error |= E1000_RXD_STAT_IPIDV; + desc->wb.lower.hi_dword.csum_ip.ip_id = + cpu_to_le16(net_rx_pkt_get_ip_id(pkt)); + trace_e1000e_rx_metadata_ip_id( + desc->wb.lower.hi_dword.csum_ip.ip_id); + } + + if (ts) { + adv_desc_status_error |= BIT(16); + } - igb_build_rx_metadata(core, pkt, pkt != NULL, - rss_info, etqf, ts, - &desc->wb.lower.lo_dword.pkt_info, - &desc->wb.lower.lo_dword.hdr_info, - &desc->wb.lower.hi_dword.rss, - &desc->wb.upper.status_error, - &desc->wb.lower.hi_dword.csum_ip.ip_id, - &desc->wb.upper.vlan); + pkt_type = igb_rx_desc_get_packet_type(core, pkt, etqf); + trace_e1000e_rx_metadata_pkt_type(pkt_type); + desc->wb.lower.lo_dword.pkt_info = cpu_to_le16(rss_type | (pkt_type << 4)); + desc->wb.upper.status_error |= cpu_to_le32(adv_desc_status_error); } static inline void @@ -1468,8 +1475,7 @@ igb_write_rx_descr(IGBCore *core, union e1000_rx_desc_union *desc, uint16_t etqf, bool ts, uint16_t length) { if (igb_rx_use_legacy_descriptor(core)) { - igb_write_lgcy_rx_descr(core, &desc->legacy, pkt, rss_info, - etqf, ts, length); + igb_write_lgcy_rx_descr(core, &desc->legacy, pkt, rss_info, length); } else { igb_write_adv_rx_descr(core, &desc->adv, pkt, rss_info, etqf, ts, length); diff --git a/hw/net/igb_regs.h b/hw/net/igb_regs.h index 82ff195dfc..71a8833229 100644 --- a/hw/net/igb_regs.h +++ b/hw/net/igb_regs.h @@ -692,11 +692,11 @@ union e1000_adv_rx_desc { #define E1000_STATUS_NUM_VFS_SHIFT 14 -#define E1000_ADVRXD_PKT_IP4 BIT(4) -#define E1000_ADVRXD_PKT_IP6 BIT(6) -#define E1000_ADVRXD_PKT_TCP BIT(8) -#define E1000_ADVRXD_PKT_UDP BIT(9) -#define E1000_ADVRXD_PKT_SCTP BIT(10) +#define E1000_ADVRXD_PKT_IP4 BIT(0) +#define E1000_ADVRXD_PKT_IP6 BIT(2) +#define E1000_ADVRXD_PKT_TCP BIT(4) +#define E1000_ADVRXD_PKT_UDP BIT(5) +#define E1000_ADVRXD_PKT_SCTP BIT(6) static inline uint8_t igb_ivar_entry_rx(uint8_t i) { diff --git a/hw/net/trace-events b/hw/net/trace-events index e4a98b2c7d..159df09014 100644 --- a/hw/net/trace-events +++ b/hw/net/trace-events @@ -279,7 +279,7 @@ igb_link_set_ext_params(bool asd_check, bool speed_select_bypass, bool pfrstd) " igb_rx_desc_buff_size(uint32_t b) "buffer size: %u" igb_rx_desc_buff_write(uint64_t addr, uint16_t offset, const void* source, uint32_t len) "addr: 0x%"PRIx64", offset: %u, from: %p, length: %u" -igb_rx_metadata_rss(uint32_t rss) "RSS data: 0x%X" +igb_rx_metadata_rss(uint32_t rss, uint16_t rss_pkt_type) "RSS data: rss: 0x%X, rss_pkt_type: 0x%X" igb_irq_icr_clear_gpie_nsicr(void) "Clearing ICR on read due to GPIE.NSICR enabled" igb_irq_set_iam(uint32_t icr) "Update IAM: 0x%x" @@ -294,6 +294,8 @@ igb_irq_eitr_set(uint32_t eitr_num, uint32_t val) "EITR[%u] = 0x%x" igb_set_pfmailbox(uint32_t vf_num, uint32_t val) "PFMailbox[%d]: 0x%x" igb_set_vfmailbox(uint32_t vf_num, uint32_t val) "VFMailbox[%d]: 0x%x" +igb_wrn_rx_desc_modes_not_supp(int desc_type) "Not supported descriptor type: %d" + # igbvf.c igbvf_wrn_io_addr_unknown(uint64_t addr) "IO unknown register 0x%"PRIx64 -- 2.25.1 ^ permalink raw reply related [flat|nested] 11+ messages in thread
[parent not found: <CGME20230525153745eucas1p1cd6d947ed0cd46a8fdf637b514c237ff@eucas1p1.samsung.com>]
* [PATCH v9 4/7] igb: RX payload guest writting refactoring [not found] ` <CGME20230525153745eucas1p1cd6d947ed0cd46a8fdf637b514c237ff@eucas1p1.samsung.com> @ 2023-05-25 15:37 ` Tomasz Dzieciol 2023-05-26 23:54 ` Akihiko Odaki 0 siblings, 1 reply; 11+ messages in thread From: Tomasz Dzieciol @ 2023-05-25 15:37 UTC (permalink / raw) To: qemu-devel, akihiko.odaki Cc: sriram.yagnaraman, jasowang, k.kwiecien, m.sochacki Refactoring is done in preparation for support of multiple advanced descriptors RX modes, especially packet-split modes. Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com> --- hw/net/e1000e_core.c | 18 ++-- hw/net/igb_core.c | 214 +++++++++++++++++++++++++-------------- tests/qtest/libqos/igb.c | 5 + 3 files changed, 151 insertions(+), 86 deletions(-) diff --git a/hw/net/e1000e_core.c b/hw/net/e1000e_core.c index f6b628eaef..f29e5e2897 100644 --- a/hw/net/e1000e_core.c +++ b/hw/net/e1000e_core.c @@ -1418,11 +1418,11 @@ e1000e_write_hdr_to_rx_buffers(E1000ECore *core, } static void -e1000e_write_to_rx_buffers(E1000ECore *core, - hwaddr ba[MAX_PS_BUFFERS], - e1000e_ba_state *bastate, - const char *data, - dma_addr_t data_len) +e1000e_write_payload_frag_to_rx_buffers(E1000ECore *core, + hwaddr ba[MAX_PS_BUFFERS], + e1000e_ba_state *bastate, + const char *data, + dma_addr_t data_len) { while (data_len > 0) { uint32_t cur_buf_len = core->rxbuf_sizes[bastate->cur_idx]; @@ -1594,8 +1594,10 @@ e1000e_write_packet_to_guest(E1000ECore *core, struct NetRxPkt *pkt, while (copy_size) { iov_copy = MIN(copy_size, iov->iov_len - iov_ofs); - e1000e_write_to_rx_buffers(core, ba, &bastate, - iov->iov_base + iov_ofs, iov_copy); + e1000e_write_payload_frag_to_rx_buffers(core, ba, &bastate, + iov->iov_base + + iov_ofs, + iov_copy); copy_size -= iov_copy; iov_ofs += iov_copy; @@ -1607,7 +1609,7 @@ e1000e_write_packet_to_guest(E1000ECore *core, struct NetRxPkt *pkt, if (desc_offset + desc_size >= total_size) { /* Simulate FCS checksum presence in the last descriptor */ - e1000e_write_to_rx_buffers(core, ba, &bastate, + e1000e_write_payload_frag_to_rx_buffers(core, ba, &bastate, (const char *) &fcs_pad, e1000x_fcs_len(core->mac)); } } diff --git a/hw/net/igb_core.c b/hw/net/igb_core.c index 3a6df0d55e..8e32f39ece 100644 --- a/hw/net/igb_core.c +++ b/hw/net/igb_core.c @@ -941,6 +941,14 @@ igb_has_rxbufs(IGBCore *core, const E1000ERingInfo *r, size_t total_size) bufsize; } +static uint32_t +igb_rxhdrbufsize(IGBCore *core, const E1000ERingInfo *r) +{ + uint32_t srrctl = core->mac[E1000_SRRCTL(r->idx) >> 2]; + return (srrctl & E1000_SRRCTL_BSIZEHDRSIZE_MASK) >> + E1000_SRRCTL_BSIZEHDRSIZE_SHIFT; +} + void igb_start_recv(IGBCore *core) { @@ -1231,6 +1239,21 @@ igb_read_adv_rx_descr(IGBCore *core, union e1000_adv_rx_desc *desc, *buff_addr = le64_to_cpu(desc->read.pkt_addr); } +typedef struct IGBPacketRxDMAState { + size_t size; + size_t total_size; + size_t ps_hdr_len; + size_t desc_size; + size_t desc_offset; + uint32_t rx_desc_packet_buf_size; + uint32_t rx_desc_header_buf_size; + struct iovec *iov; + size_t iov_ofs; + bool is_first; + uint16_t written; + hwaddr ba; +} IGBPacketRxDMAState; + static inline void igb_read_rx_descr(IGBCore *core, union e1000_rx_desc_union *desc, hwaddr *buff_addr) @@ -1514,19 +1537,6 @@ igb_pci_dma_write_rx_desc(IGBCore *core, PCIDevice *dev, dma_addr_t addr, } } -static void -igb_write_to_rx_buffers(IGBCore *core, - PCIDevice *d, - hwaddr ba, - uint16_t *written, - const char *data, - dma_addr_t data_len) -{ - trace_igb_rx_desc_buff_write(ba, *written, data, data_len); - pci_dma_write(d, ba + *written, data, data_len); - *written += data_len; -} - static void igb_update_rx_stats(IGBCore *core, const E1000ERingInfo *rxi, size_t pkt_size, size_t pkt_fcs_size) @@ -1552,6 +1562,93 @@ igb_rx_descr_threshold_hit(IGBCore *core, const E1000ERingInfo *rxi) ((core->mac[E1000_SRRCTL(rxi->idx) >> 2] >> 20) & 31) * 16; } +static void +igb_truncate_to_descriptor_size(IGBPacketRxDMAState *pdma_st, size_t *size) +{ + if (*size > pdma_st->rx_desc_packet_buf_size) { + *size = pdma_st->rx_desc_packet_buf_size; + } +} + +static void +igb_write_payload_frag_to_rx_buffers(IGBCore *core, + PCIDevice *d, + hwaddr ba, + uint16_t *written, + uint32_t cur_buf_len, + const char *data, + dma_addr_t data_len) +{ + trace_igb_rx_desc_buff_write(ba, *written, data, data_len); + pci_dma_write(d, ba + *written, data, data_len); + *written += data_len; +} + +static void +igb_write_payload_to_rx_buffers(IGBCore *core, + struct NetRxPkt *pkt, + PCIDevice *d, + IGBPacketRxDMAState *pdma_st, + size_t *copy_size) +{ + static const uint32_t fcs_pad; + size_t iov_copy; + + /* Copy packet payload */ + while (*copy_size) { + iov_copy = MIN(*copy_size, pdma_st->iov->iov_len - pdma_st->iov_ofs); + igb_write_payload_frag_to_rx_buffers(core, d, + pdma_st->ba, + &pdma_st->written, + pdma_st->rx_desc_packet_buf_size, + pdma_st->iov->iov_base + + pdma_st->iov_ofs, + iov_copy); + + *copy_size -= iov_copy; + pdma_st->iov_ofs += iov_copy; + if (pdma_st->iov_ofs == pdma_st->iov->iov_len) { + pdma_st->iov++; + pdma_st->iov_ofs = 0; + } + } + + if (pdma_st->desc_offset + pdma_st->desc_size >= pdma_st->total_size) { + /* Simulate FCS checksum presence in the last descriptor */ + igb_write_payload_frag_to_rx_buffers(core, d, + pdma_st->ba, + &pdma_st->written, + pdma_st->rx_desc_packet_buf_size, + (const char *) &fcs_pad, + e1000x_fcs_len(core->mac)); + } +} + +static void +igb_write_to_rx_buffers(IGBCore *core, + struct NetRxPkt *pkt, + PCIDevice *d, + IGBPacketRxDMAState *pdma_st) +{ + size_t copy_size; + + if (!pdma_st->ba) { + /* as per intel docs; skip descriptors with null buf addr */ + trace_e1000e_rx_null_descriptor(); + return; + } + + if (pdma_st->desc_offset >= pdma_st->size) { + return; + } + + pdma_st->desc_size = pdma_st->total_size - pdma_st->desc_offset; + igb_truncate_to_descriptor_size(pdma_st, &pdma_st->desc_size); + copy_size = pdma_st->size - pdma_st->desc_offset; + igb_truncate_to_descriptor_size(pdma_st, ©_size); + igb_write_payload_to_rx_buffers(core, pkt, d, pdma_st, ©_size); +} + static void igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, const E1000E_RxRing *rxr, @@ -1561,91 +1658,52 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, PCIDevice *d; dma_addr_t base; union e1000_rx_desc_union desc; - size_t desc_size; - size_t desc_offset = 0; - size_t iov_ofs = 0; - - struct iovec *iov = net_rx_pkt_get_iovec(pkt); - size_t size = net_rx_pkt_get_total_len(pkt); - size_t total_size = size + e1000x_fcs_len(core->mac); - const E1000ERingInfo *rxi = rxr->i; - size_t bufsize = igb_rxbufsize(core, rxi); - + const E1000ERingInfo *rxi; + size_t rx_desc_len; + + IGBPacketRxDMAState pdma_st = {0}; + pdma_st.is_first = true; + pdma_st.size = net_rx_pkt_get_total_len(pkt); + pdma_st.total_size = pdma_st.size + e1000x_fcs_len(core->mac); + + rxi = rxr->i; + rx_desc_len = core->rx_desc_len; + pdma_st.rx_desc_packet_buf_size = igb_rxbufsize(core, rxi); + pdma_st.rx_desc_header_buf_size = igb_rxhdrbufsize(core, rxi); + pdma_st.iov = net_rx_pkt_get_iovec(pkt); d = pcie_sriov_get_vf_at_index(core->owner, rxi->idx % 8); if (!d) { d = core->owner; } do { - hwaddr ba; - uint16_t written = 0; + pdma_st.written = 0; bool is_last = false; - desc_size = total_size - desc_offset; - - if (desc_size > bufsize) { - desc_size = bufsize; - } - if (igb_ring_empty(core, rxi)) { return; } base = igb_ring_head_descr(core, rxi); + pci_dma_read(d, base, &desc, rx_desc_len); + trace_e1000e_rx_descr(rxi->idx, base, rx_desc_len); - pci_dma_read(d, base, &desc, core->rx_desc_len); - - trace_e1000e_rx_descr(rxi->idx, base, core->rx_desc_len); - - igb_read_rx_descr(core, &desc, &ba); - - if (ba) { - if (desc_offset < size) { - static const uint32_t fcs_pad; - size_t iov_copy; - size_t copy_size = size - desc_offset; - if (copy_size > bufsize) { - copy_size = bufsize; - } - - /* Copy packet payload */ - while (copy_size) { - iov_copy = MIN(copy_size, iov->iov_len - iov_ofs); - - igb_write_to_rx_buffers(core, d, ba, &written, - iov->iov_base + iov_ofs, iov_copy); + igb_read_rx_descr(core, &desc, &pdma_st.ba); - copy_size -= iov_copy; - iov_ofs += iov_copy; - if (iov_ofs == iov->iov_len) { - iov++; - iov_ofs = 0; - } - } - - if (desc_offset + desc_size >= total_size) { - /* Simulate FCS checksum presence in the last descriptor */ - igb_write_to_rx_buffers(core, d, ba, &written, - (const char *) &fcs_pad, e1000x_fcs_len(core->mac)); - } - } - } else { /* as per intel docs; skip descriptors with null buf addr */ - trace_e1000e_rx_null_descriptor(); - } - desc_offset += desc_size; - if (desc_offset >= total_size) { + igb_write_to_rx_buffers(core, pkt, d, &pdma_st); + pdma_st.desc_offset += pdma_st.desc_size; + if (pdma_st.desc_offset >= pdma_st.total_size) { is_last = true; } igb_write_rx_descr(core, &desc, is_last ? core->rx_pkt : NULL, - rss_info, etqf, ts, written); - igb_pci_dma_write_rx_desc(core, d, base, &desc, core->rx_desc_len); - - igb_ring_advance(core, rxi, core->rx_desc_len / E1000_MIN_RX_DESC_LEN); - - } while (desc_offset < total_size); + rss_info, etqf, ts, pdma_st.written); + igb_pci_dma_write_rx_desc(core, d, base, &desc, rx_desc_len); + igb_ring_advance(core, rxi, + rx_desc_len / E1000_MIN_RX_DESC_LEN); + } while (pdma_st.desc_offset < pdma_st.total_size); - igb_update_rx_stats(core, rxi, size, total_size); + igb_update_rx_stats(core, rxi, pdma_st.size, pdma_st.total_size); } static bool diff --git a/tests/qtest/libqos/igb.c b/tests/qtest/libqos/igb.c index a603468beb..f40c4ec4cd 100644 --- a/tests/qtest/libqos/igb.c +++ b/tests/qtest/libqos/igb.c @@ -109,6 +109,11 @@ static void igb_pci_start_hw(QOSGraphObject *obj) E1000_RAH_AV | E1000_RAH_POOL_1 | le16_to_cpu(*(uint16_t *)(address + 4))); + /* Set supported receive descriptor mode */ + e1000e_macreg_write(&d->e1000e, + E1000_SRRCTL(0), + E1000_SRRCTL_DESCTYPE_ADV_ONEBUF); + /* Enable receive */ e1000e_macreg_write(&d->e1000e, E1000_RFCTL, E1000_RFCTL_EXTEN); e1000e_macreg_write(&d->e1000e, E1000_RCTL, E1000_RCTL_EN); -- 2.25.1 ^ permalink raw reply related [flat|nested] 11+ messages in thread
* Re: [PATCH v9 4/7] igb: RX payload guest writting refactoring 2023-05-25 15:37 ` [PATCH v9 4/7] igb: RX payload " Tomasz Dzieciol @ 2023-05-26 23:54 ` Akihiko Odaki 0 siblings, 0 replies; 11+ messages in thread From: Akihiko Odaki @ 2023-05-26 23:54 UTC (permalink / raw) To: Tomasz Dzieciol, qemu-devel Cc: sriram.yagnaraman, jasowang, k.kwiecien, m.sochacki On 2023/05/26 0:37, Tomasz Dzieciol wrote: > Refactoring is done in preparation for support of multiple advanced > descriptors RX modes, especially packet-split modes. > > Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com> > --- > hw/net/e1000e_core.c | 18 ++-- > hw/net/igb_core.c | 214 +++++++++++++++++++++++++-------------- > tests/qtest/libqos/igb.c | 5 + > 3 files changed, 151 insertions(+), 86 deletions(-) > > diff --git a/hw/net/e1000e_core.c b/hw/net/e1000e_core.c > index f6b628eaef..f29e5e2897 100644 > --- a/hw/net/e1000e_core.c > +++ b/hw/net/e1000e_core.c > @@ -1418,11 +1418,11 @@ e1000e_write_hdr_to_rx_buffers(E1000ECore *core, > } > > static void > -e1000e_write_to_rx_buffers(E1000ECore *core, > - hwaddr ba[MAX_PS_BUFFERS], > - e1000e_ba_state *bastate, > - const char *data, > - dma_addr_t data_len) > +e1000e_write_payload_frag_to_rx_buffers(E1000ECore *core, > + hwaddr ba[MAX_PS_BUFFERS], > + e1000e_ba_state *bastate, > + const char *data, > + dma_addr_t data_len) > { > while (data_len > 0) { > uint32_t cur_buf_len = core->rxbuf_sizes[bastate->cur_idx]; > @@ -1594,8 +1594,10 @@ e1000e_write_packet_to_guest(E1000ECore *core, struct NetRxPkt *pkt, > while (copy_size) { > iov_copy = MIN(copy_size, iov->iov_len - iov_ofs); > > - e1000e_write_to_rx_buffers(core, ba, &bastate, > - iov->iov_base + iov_ofs, iov_copy); > + e1000e_write_payload_frag_to_rx_buffers(core, ba, &bastate, > + iov->iov_base + > + iov_ofs, > + iov_copy); > > copy_size -= iov_copy; > iov_ofs += iov_copy; > @@ -1607,7 +1609,7 @@ e1000e_write_packet_to_guest(E1000ECore *core, struct NetRxPkt *pkt, > > if (desc_offset + desc_size >= total_size) { > /* Simulate FCS checksum presence in the last descriptor */ > - e1000e_write_to_rx_buffers(core, ba, &bastate, > + e1000e_write_payload_frag_to_rx_buffers(core, ba, &bastate, > (const char *) &fcs_pad, e1000x_fcs_len(core->mac)); > } > } > diff --git a/hw/net/igb_core.c b/hw/net/igb_core.c > index 3a6df0d55e..8e32f39ece 100644 > --- a/hw/net/igb_core.c > +++ b/hw/net/igb_core.c > @@ -941,6 +941,14 @@ igb_has_rxbufs(IGBCore *core, const E1000ERingInfo *r, size_t total_size) > bufsize; > } > > +static uint32_t > +igb_rxhdrbufsize(IGBCore *core, const E1000ERingInfo *r) > +{ > + uint32_t srrctl = core->mac[E1000_SRRCTL(r->idx) >> 2]; > + return (srrctl & E1000_SRRCTL_BSIZEHDRSIZE_MASK) >> > + E1000_SRRCTL_BSIZEHDRSIZE_SHIFT; > +} > + > void > igb_start_recv(IGBCore *core) > { > @@ -1231,6 +1239,21 @@ igb_read_adv_rx_descr(IGBCore *core, union e1000_adv_rx_desc *desc, > *buff_addr = le64_to_cpu(desc->read.pkt_addr); > } > > +typedef struct IGBPacketRxDMAState { > + size_t size; > + size_t total_size; > + size_t ps_hdr_len; > + size_t desc_size; > + size_t desc_offset; > + uint32_t rx_desc_packet_buf_size; > + uint32_t rx_desc_header_buf_size; > + struct iovec *iov; > + size_t iov_ofs; > + bool is_first; > + uint16_t written; > + hwaddr ba; > +} IGBPacketRxDMAState; > + > static inline void > igb_read_rx_descr(IGBCore *core, union e1000_rx_desc_union *desc, > hwaddr *buff_addr) > @@ -1514,19 +1537,6 @@ igb_pci_dma_write_rx_desc(IGBCore *core, PCIDevice *dev, dma_addr_t addr, > } > } > > -static void > -igb_write_to_rx_buffers(IGBCore *core, > - PCIDevice *d, > - hwaddr ba, > - uint16_t *written, > - const char *data, > - dma_addr_t data_len) > -{ > - trace_igb_rx_desc_buff_write(ba, *written, data, data_len); > - pci_dma_write(d, ba + *written, data, data_len); > - *written += data_len; > -} > - > static void > igb_update_rx_stats(IGBCore *core, const E1000ERingInfo *rxi, > size_t pkt_size, size_t pkt_fcs_size) > @@ -1552,6 +1562,93 @@ igb_rx_descr_threshold_hit(IGBCore *core, const E1000ERingInfo *rxi) > ((core->mac[E1000_SRRCTL(rxi->idx) >> 2] >> 20) & 31) * 16; > } > > +static void > +igb_truncate_to_descriptor_size(IGBPacketRxDMAState *pdma_st, size_t *size) > +{ > + if (*size > pdma_st->rx_desc_packet_buf_size) { > + *size = pdma_st->rx_desc_packet_buf_size; > + } > +} > + > +static void > +igb_write_payload_frag_to_rx_buffers(IGBCore *core, > + PCIDevice *d, > + hwaddr ba, > + uint16_t *written, > + uint32_t cur_buf_len, > + const char *data, > + dma_addr_t data_len) > +{ > + trace_igb_rx_desc_buff_write(ba, *written, data, data_len); > + pci_dma_write(d, ba + *written, data, data_len); > + *written += data_len; > +} > + > +static void > +igb_write_payload_to_rx_buffers(IGBCore *core, > + struct NetRxPkt *pkt, > + PCIDevice *d, > + IGBPacketRxDMAState *pdma_st, > + size_t *copy_size) > +{ > + static const uint32_t fcs_pad; > + size_t iov_copy; > + > + /* Copy packet payload */ > + while (*copy_size) { > + iov_copy = MIN(*copy_size, pdma_st->iov->iov_len - pdma_st->iov_ofs); > + igb_write_payload_frag_to_rx_buffers(core, d, > + pdma_st->ba, > + &pdma_st->written, > + pdma_st->rx_desc_packet_buf_size, > + pdma_st->iov->iov_base + > + pdma_st->iov_ofs, > + iov_copy); > + > + *copy_size -= iov_copy; > + pdma_st->iov_ofs += iov_copy; > + if (pdma_st->iov_ofs == pdma_st->iov->iov_len) { > + pdma_st->iov++; > + pdma_st->iov_ofs = 0; > + } > + } > + > + if (pdma_st->desc_offset + pdma_st->desc_size >= pdma_st->total_size) { > + /* Simulate FCS checksum presence in the last descriptor */ > + igb_write_payload_frag_to_rx_buffers(core, d, > + pdma_st->ba, > + &pdma_st->written, > + pdma_st->rx_desc_packet_buf_size, > + (const char *) &fcs_pad, > + e1000x_fcs_len(core->mac)); > + } > +} > + > +static void > +igb_write_to_rx_buffers(IGBCore *core, > + struct NetRxPkt *pkt, > + PCIDevice *d, > + IGBPacketRxDMAState *pdma_st) > +{ > + size_t copy_size; > + > + if (!pdma_st->ba) { > + /* as per intel docs; skip descriptors with null buf addr */ > + trace_e1000e_rx_null_descriptor(); > + return; > + } > + > + if (pdma_st->desc_offset >= pdma_st->size) { > + return; > + } > + > + pdma_st->desc_size = pdma_st->total_size - pdma_st->desc_offset; > + igb_truncate_to_descriptor_size(pdma_st, &pdma_st->desc_size); > + copy_size = pdma_st->size - pdma_st->desc_offset; > + igb_truncate_to_descriptor_size(pdma_st, ©_size); > + igb_write_payload_to_rx_buffers(core, pkt, d, pdma_st, ©_size); > +} > + > static void > igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, > const E1000E_RxRing *rxr, > @@ -1561,91 +1658,52 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, > PCIDevice *d; > dma_addr_t base; > union e1000_rx_desc_union desc; > - size_t desc_size; > - size_t desc_offset = 0; > - size_t iov_ofs = 0; > - > - struct iovec *iov = net_rx_pkt_get_iovec(pkt); > - size_t size = net_rx_pkt_get_total_len(pkt); > - size_t total_size = size + e1000x_fcs_len(core->mac); > - const E1000ERingInfo *rxi = rxr->i; > - size_t bufsize = igb_rxbufsize(core, rxi); > - > + const E1000ERingInfo *rxi; > + size_t rx_desc_len; > + > + IGBPacketRxDMAState pdma_st = {0}; > + pdma_st.is_first = true; > + pdma_st.size = net_rx_pkt_get_total_len(pkt); > + pdma_st.total_size = pdma_st.size + e1000x_fcs_len(core->mac); > + > + rxi = rxr->i; > + rx_desc_len = core->rx_desc_len; > + pdma_st.rx_desc_packet_buf_size = igb_rxbufsize(core, rxi); > + pdma_st.rx_desc_header_buf_size = igb_rxhdrbufsize(core, rxi); > + pdma_st.iov = net_rx_pkt_get_iovec(pkt); > d = pcie_sriov_get_vf_at_index(core->owner, rxi->idx % 8); > if (!d) { > d = core->owner; > } > > do { > - hwaddr ba; > - uint16_t written = 0; > + pdma_st.written = 0; > bool is_last = false; > > - desc_size = total_size - desc_offset; > - > - if (desc_size > bufsize) { > - desc_size = bufsize; > - } > - > if (igb_ring_empty(core, rxi)) { > return; > } > > base = igb_ring_head_descr(core, rxi); > + pci_dma_read(d, base, &desc, rx_desc_len); > + trace_e1000e_rx_descr(rxi->idx, base, rx_desc_len); > > - pci_dma_read(d, base, &desc, core->rx_desc_len); > - > - trace_e1000e_rx_descr(rxi->idx, base, core->rx_desc_len); > - > - igb_read_rx_descr(core, &desc, &ba); > - > - if (ba) { > - if (desc_offset < size) { > - static const uint32_t fcs_pad; > - size_t iov_copy; > - size_t copy_size = size - desc_offset; > - if (copy_size > bufsize) { > - copy_size = bufsize; > - } > - > - /* Copy packet payload */ > - while (copy_size) { > - iov_copy = MIN(copy_size, iov->iov_len - iov_ofs); > - > - igb_write_to_rx_buffers(core, d, ba, &written, > - iov->iov_base + iov_ofs, iov_copy); > + igb_read_rx_descr(core, &desc, &pdma_st.ba); > > - copy_size -= iov_copy; > - iov_ofs += iov_copy; > - if (iov_ofs == iov->iov_len) { > - iov++; > - iov_ofs = 0; > - } > - } > - > - if (desc_offset + desc_size >= total_size) { > - /* Simulate FCS checksum presence in the last descriptor */ > - igb_write_to_rx_buffers(core, d, ba, &written, > - (const char *) &fcs_pad, e1000x_fcs_len(core->mac)); > - } > - } > - } else { /* as per intel docs; skip descriptors with null buf addr */ > - trace_e1000e_rx_null_descriptor(); > - } > - desc_offset += desc_size; > - if (desc_offset >= total_size) { > + igb_write_to_rx_buffers(core, pkt, d, &pdma_st); > + pdma_st.desc_offset += pdma_st.desc_size; > + if (pdma_st.desc_offset >= pdma_st.total_size) { > is_last = true; > } > > igb_write_rx_descr(core, &desc, is_last ? core->rx_pkt : NULL, > - rss_info, etqf, ts, written); > - igb_pci_dma_write_rx_desc(core, d, base, &desc, core->rx_desc_len); > - > - igb_ring_advance(core, rxi, core->rx_desc_len / E1000_MIN_RX_DESC_LEN); > - > - } while (desc_offset < total_size); > + rss_info, etqf, ts, pdma_st.written); > + igb_pci_dma_write_rx_desc(core, d, base, &desc, rx_desc_len); > + igb_ring_advance(core, rxi, > + rx_desc_len / E1000_MIN_RX_DESC_LEN); No need for a linebreak for this igb_ring_advance() call. > + } while (pdma_st.desc_offset < pdma_st.total_size); > > - igb_update_rx_stats(core, rxi, size, total_size); > + igb_update_rx_stats(core, rxi, pdma_st.size, pdma_st.total_size); > } > > static bool > diff --git a/tests/qtest/libqos/igb.c b/tests/qtest/libqos/igb.c > index a603468beb..f40c4ec4cd 100644 > --- a/tests/qtest/libqos/igb.c > +++ b/tests/qtest/libqos/igb.c > @@ -109,6 +109,11 @@ static void igb_pci_start_hw(QOSGraphObject *obj) > E1000_RAH_AV | E1000_RAH_POOL_1 | > le16_to_cpu(*(uint16_t *)(address + 4))); > > + /* Set supported receive descriptor mode */ > + e1000e_macreg_write(&d->e1000e, > + E1000_SRRCTL(0), > + E1000_SRRCTL_DESCTYPE_ADV_ONEBUF); > + > /* Enable receive */ > e1000e_macreg_write(&d->e1000e, E1000_RFCTL, E1000_RFCTL_EXTEN); > e1000e_macreg_write(&d->e1000e, E1000_RCTL, E1000_RCTL_EN); ^ permalink raw reply [flat|nested] 11+ messages in thread
[parent not found: <CGME20230525153745eucas1p22c62582ef16e6d6c9da9ab58243414f0@eucas1p2.samsung.com>]
* [PATCH v9 5/7] igb: add IPv6 extended headers traffic detection [not found] ` <CGME20230525153745eucas1p22c62582ef16e6d6c9da9ab58243414f0@eucas1p2.samsung.com> @ 2023-05-25 15:37 ` Tomasz Dzieciol 0 siblings, 0 replies; 11+ messages in thread From: Tomasz Dzieciol @ 2023-05-25 15:37 UTC (permalink / raw) To: qemu-devel, akihiko.odaki Cc: sriram.yagnaraman, jasowang, k.kwiecien, m.sochacki Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com> --- hw/net/igb_core.c | 4 +++- hw/net/igb_regs.h | 1 + 2 files changed, 4 insertions(+), 1 deletion(-) diff --git a/hw/net/igb_core.c b/hw/net/igb_core.c index 8e32f39ece..2bfae517fc 100644 --- a/hw/net/igb_core.c +++ b/hw/net/igb_core.c @@ -1420,7 +1420,9 @@ igb_rx_desc_get_packet_type(IGBCore *core, struct NetRxPkt *pkt, uint16_t etqf) net_rx_pkt_get_protocols(pkt, &hasip4, &hasip6, &l4hdr_proto); if (hasip6 && !(core->mac[RFCTL] & E1000_RFCTL_IPV6_DIS)) { - pkt_type = E1000_ADVRXD_PKT_IP6; + eth_ip6_hdr_info *ip6hdr_info = net_rx_pkt_get_ip6_info(pkt); + pkt_type = ip6hdr_info->has_ext_hdrs ? E1000_ADVRXD_PKT_IP6E : + E1000_ADVRXD_PKT_IP6; } else if (hasip4) { pkt_type = E1000_ADVRXD_PKT_IP4; } else { diff --git a/hw/net/igb_regs.h b/hw/net/igb_regs.h index 71a8833229..36763f2ff7 100644 --- a/hw/net/igb_regs.h +++ b/hw/net/igb_regs.h @@ -694,6 +694,7 @@ union e1000_adv_rx_desc { #define E1000_ADVRXD_PKT_IP4 BIT(0) #define E1000_ADVRXD_PKT_IP6 BIT(2) +#define E1000_ADVRXD_PKT_IP6E BIT(3) #define E1000_ADVRXD_PKT_TCP BIT(4) #define E1000_ADVRXD_PKT_UDP BIT(5) #define E1000_ADVRXD_PKT_SCTP BIT(6) -- 2.25.1 ^ permalink raw reply related [flat|nested] 11+ messages in thread
[parent not found: <CGME20230525153746eucas1p1957a3a91d90521a6d1c684595c848f52@eucas1p1.samsung.com>]
* [PATCH v9 6/7] igb: packet-split descriptors support [not found] ` <CGME20230525153746eucas1p1957a3a91d90521a6d1c684595c848f52@eucas1p1.samsung.com> @ 2023-05-25 15:37 ` Tomasz Dzieciol 2023-05-27 0:07 ` Akihiko Odaki 0 siblings, 1 reply; 11+ messages in thread From: Tomasz Dzieciol @ 2023-05-25 15:37 UTC (permalink / raw) To: qemu-devel, akihiko.odaki Cc: sriram.yagnaraman, jasowang, k.kwiecien, m.sochacki Packet-split descriptors are used by Linux VF driver for MTU values from 2048 Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com> --- hw/net/igb_core.c | 357 ++++++++++++++++++++++++++++++++++++++------ hw/net/igb_regs.h | 9 ++ hw/net/trace-events | 2 +- 3 files changed, 325 insertions(+), 43 deletions(-) diff --git a/hw/net/igb_core.c b/hw/net/igb_core.c index 2bfae517fc..c1a8dfbca8 100644 --- a/hw/net/igb_core.c +++ b/hw/net/igb_core.c @@ -267,6 +267,29 @@ igb_rx_use_legacy_descriptor(IGBCore *core) return false; } +typedef struct E1000ERingInfo { + int dbah; + int dbal; + int dlen; + int dh; + int dt; + int idx; +} E1000ERingInfo; + +static uint32_t +igb_rx_queue_desctyp_get(IGBCore *core, const E1000ERingInfo *r) +{ + return core->mac[E1000_SRRCTL(r->idx) >> 2] & E1000_SRRCTL_DESCTYPE_MASK; +} + +static bool +igb_rx_use_ps_descriptor(IGBCore *core, const E1000ERingInfo *r) +{ + uint32_t desctyp = igb_rx_queue_desctyp_get(core, r); + return desctyp == E1000_SRRCTL_DESCTYPE_HDR_SPLIT || + desctyp == E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS; +} + static inline bool igb_rss_enabled(IGBCore *core) { @@ -694,15 +717,6 @@ static uint32_t igb_rx_wb_eic(IGBCore *core, int queue_idx) return (ent & E1000_IVAR_VALID) ? BIT(ent & 0x1f) : 0; } -typedef struct E1000ERingInfo { - int dbah; - int dbal; - int dlen; - int dh; - int dt; - int idx; -} E1000ERingInfo; - static inline bool igb_ring_empty(IGBCore *core, const E1000ERingInfo *r) { @@ -1233,12 +1247,25 @@ igb_read_lgcy_rx_descr(IGBCore *core, struct e1000_rx_desc *desc, } static inline void -igb_read_adv_rx_descr(IGBCore *core, union e1000_adv_rx_desc *desc, - hwaddr *buff_addr) +igb_read_adv_rx_single_buf_descr(IGBCore *core, union e1000_adv_rx_desc *desc, + hwaddr *buff_addr) { *buff_addr = le64_to_cpu(desc->read.pkt_addr); } +static inline void +igb_read_adv_rx_split_buf_descr(IGBCore *core, union e1000_adv_rx_desc *desc, + hwaddr *buff_addr) +{ + buff_addr[0] = le64_to_cpu(desc->read.hdr_addr); + buff_addr[1] = le64_to_cpu(desc->read.pkt_addr); +} + +typedef struct IGBBAState { + uint16_t written[IGB_MAX_PS_BUFFERS]; + uint8_t cur_idx; +} IGBBAState; + typedef struct IGBPacketRxDMAState { size_t size; size_t total_size; @@ -1249,20 +1276,41 @@ typedef struct IGBPacketRxDMAState { uint32_t rx_desc_header_buf_size; struct iovec *iov; size_t iov_ofs; + bool do_ps; bool is_first; - uint16_t written; - hwaddr ba; + IGBBAState bastate; + hwaddr ba[IGB_MAX_PS_BUFFERS]; } IGBPacketRxDMAState; static inline void -igb_read_rx_descr(IGBCore *core, union e1000_rx_desc_union *desc, - hwaddr *buff_addr) +igb_read_rx_descr(IGBCore *core, + union e1000_rx_desc_union *desc, + IGBPacketRxDMAState *pdma_st, + const E1000ERingInfo *r) { + uint32_t desc_type; + if (igb_rx_use_legacy_descriptor(core)) { - igb_read_lgcy_rx_descr(core, &desc->legacy, buff_addr); - } else { - igb_read_adv_rx_descr(core, &desc->adv, buff_addr); + igb_read_lgcy_rx_descr(core, &desc->legacy, &pdma_st->ba[1]); + pdma_st->ba[0] = 0; + return; + } + + /* advanced header split descriptor */ + if (igb_rx_use_ps_descriptor(core, r)) { + igb_read_adv_rx_split_buf_descr(core, &desc->adv, &pdma_st->ba[0]); + return; + } + + /* descriptor replication modes not supported */ + desc_type = igb_rx_queue_desctyp_get(core, r); + if (desc_type != E1000_SRRCTL_DESCTYPE_ADV_ONEBUF) { + trace_igb_wrn_rx_desc_modes_not_supp(desc_type); } + + /* advanced single buffer descriptor */ + igb_read_adv_rx_single_buf_descr(core, &desc->adv, &pdma_st->ba[1]); + pdma_st->ba[0] = 0; } static void @@ -1405,6 +1453,13 @@ igb_write_lgcy_rx_descr(IGBCore *core, struct e1000_rx_desc *desc, desc->status = (uint8_t) le32_to_cpu(status_flags); } +static bool +igb_rx_ps_descriptor_split_always(IGBCore *core, const E1000ERingInfo *r) +{ + uint32_t desctyp = igb_rx_queue_desctyp_get(core, r); + return desctyp == E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS; +} + static uint16_t igb_rx_desc_get_packet_type(IGBCore *core, struct NetRxPkt *pkt, uint16_t etqf) { @@ -1494,16 +1549,64 @@ igb_write_adv_rx_descr(IGBCore *core, union e1000_adv_rx_desc *desc, desc->wb.upper.status_error |= cpu_to_le32(adv_desc_status_error); } +typedef struct IGBSplitDescriptorData { + bool sph; + bool hbo; + size_t hdr_len; +} IGBSplitDescriptorData; + static inline void -igb_write_rx_descr(IGBCore *core, union e1000_rx_desc_union *desc, - struct NetRxPkt *pkt, const E1000E_RSSInfo *rss_info, - uint16_t etqf, bool ts, uint16_t length) +igb_write_adv_ps_rx_descr(IGBCore *core, + union e1000_adv_rx_desc *desc, + struct NetRxPkt *pkt, + const E1000E_RSSInfo *rss_info, + const E1000ERingInfo *r, + uint16_t etqf, + bool ts, + IGBSplitDescriptorData *ps_desc_data, + IGBPacketRxDMAState *pdma_st) +{ + size_t pkt_len; + size_t hdr_len = ps_desc_data->hdr_len; + uint16_t hdr_info = 0; + + if (pdma_st->do_ps) { + pkt_len = pdma_st->bastate.written[1]; + } else { + pkt_len = pdma_st->bastate.written[0] + pdma_st->bastate.written[1]; + } + + igb_write_adv_rx_descr(core, desc, pkt, rss_info, etqf, ts, pkt_len); + + hdr_info = (hdr_len << E1000_ADVRXD_HDR_LEN_OFFSET) & + E1000_ADVRXD_ADV_HDR_LEN_MASK; + hdr_info |= ps_desc_data->sph ? E1000_ADVRXD_HDR_SPH : 0; + desc->wb.lower.lo_dword.hdr_info = cpu_to_le16(hdr_info); + + desc->wb.upper.status_error |= cpu_to_le32( + ps_desc_data->hbo ? E1000_ADVRXD_ST_ERR_HBO_OFFSET : 0); +} + +static inline void +igb_write_rx_descr(IGBCore *core, + union e1000_rx_desc_union *desc, + struct NetRxPkt *pkt, + const E1000E_RSSInfo *rss_info, + uint16_t etqf, + bool ts, + IGBSplitDescriptorData *ps_desc_data, + IGBPacketRxDMAState *pdma_st, + const E1000ERingInfo *r) { if (igb_rx_use_legacy_descriptor(core)) { - igb_write_lgcy_rx_descr(core, &desc->legacy, pkt, rss_info, length); + igb_write_lgcy_rx_descr(core, &desc->legacy, pkt, rss_info, + pdma_st->bastate.written[1]); + } else if (igb_rx_use_ps_descriptor(core, r)) { + igb_write_adv_ps_rx_descr(core, &desc->adv, pkt, rss_info, r, etqf, ts, + ps_desc_data, pdma_st); } else { igb_write_adv_rx_descr(core, &desc->adv, pkt, rss_info, - etqf, ts, length); + etqf, ts, pdma_st->bastate.written[1]); } } @@ -1564,26 +1667,182 @@ igb_rx_descr_threshold_hit(IGBCore *core, const E1000ERingInfo *rxi) ((core->mac[E1000_SRRCTL(rxi->idx) >> 2] >> 20) & 31) * 16; } +static bool +igb_do_ps(IGBCore *core, + const E1000ERingInfo *r, + struct NetRxPkt *pkt, + size_t *hdr_len, + IGBSplitDescriptorData *ps_desc_data) +{ + bool hasip4, hasip6; + EthL4HdrProto l4hdr_proto; + bool fragment; + bool split_always; + size_t bheader_size; + size_t total_pkt_len; + + if (!igb_rx_use_ps_descriptor(core, r)) { + return false; + } + + memset(ps_desc_data, 0, sizeof(IGBSplitDescriptorData)); + + total_pkt_len = net_rx_pkt_get_total_len(pkt); + bheader_size = igb_rxhdrbufsize(core, r); + split_always = igb_rx_ps_descriptor_split_always(core, r); + if (split_always && total_pkt_len <= bheader_size) { + *hdr_len = total_pkt_len; + ps_desc_data->hdr_len = total_pkt_len; + return true; + } + + net_rx_pkt_get_protocols(pkt, &hasip4, &hasip6, &l4hdr_proto); + + if (hasip4) { + fragment = net_rx_pkt_get_ip4_info(pkt)->fragment; + } else if (hasip6) { + fragment = net_rx_pkt_get_ip6_info(pkt)->fragment; + } else { + ps_desc_data->hdr_len = bheader_size; + goto header_not_handled; + } + + if (fragment && (core->mac[RFCTL] & E1000_RFCTL_IPFRSP_DIS)) { + ps_desc_data->hdr_len = bheader_size; + goto header_not_handled; + } + + /* no header splitting for SCTP */ + if (!fragment && (l4hdr_proto == ETH_L4_HDR_PROTO_UDP || + l4hdr_proto == ETH_L4_HDR_PROTO_TCP)) { + *hdr_len = net_rx_pkt_get_l5_hdr_offset(pkt); + } else { + *hdr_len = net_rx_pkt_get_l4_hdr_offset(pkt); + } + + ps_desc_data->sph = true; + ps_desc_data->hdr_len = *hdr_len; + + if (*hdr_len > bheader_size) { + ps_desc_data->hbo = true; + goto header_not_handled; + } + + return true; + +header_not_handled: + if (split_always) { + *hdr_len = bheader_size; + return true; + } + + return false; +} + static void igb_truncate_to_descriptor_size(IGBPacketRxDMAState *pdma_st, size_t *size) { - if (*size > pdma_st->rx_desc_packet_buf_size) { - *size = pdma_st->rx_desc_packet_buf_size; + if (pdma_st->do_ps && pdma_st->is_first) { + if (*size > pdma_st->rx_desc_packet_buf_size + pdma_st->ps_hdr_len) { + *size = pdma_st->rx_desc_packet_buf_size + pdma_st->ps_hdr_len; + } + } else { + if (*size > pdma_st->rx_desc_packet_buf_size) { + *size = pdma_st->rx_desc_packet_buf_size; + } } } +static inline void +igb_write_hdr_frag_to_rx_buffers(IGBCore *core, + PCIDevice *d, + IGBPacketRxDMAState *pdma_st, + const char *data, + dma_addr_t data_len) +{ + assert(data_len <= pdma_st->rx_desc_header_buf_size - + pdma_st->bastate.written[0]); + pci_dma_write(d, + pdma_st->ba[0] + pdma_st->bastate.written[0], + data, data_len); + pdma_st->bastate.written[0] += data_len; + pdma_st->bastate.cur_idx = 1; +} + +static void +igb_write_header_to_rx_buffers(IGBCore *core, + struct NetRxPkt *pkt, + PCIDevice *d, + IGBPacketRxDMAState *pdma_st, + size_t *copy_size) +{ + size_t iov_copy; + size_t ps_hdr_copied = 0; + + if (!pdma_st->is_first) { + /* Leave buffer 0 of each descriptor except first */ + /* empty */ + pdma_st->bastate.cur_idx = 1; + return; + } + + do { + iov_copy = MIN(pdma_st->ps_hdr_len - ps_hdr_copied, + pdma_st->iov->iov_len - pdma_st->iov_ofs); + + igb_write_hdr_frag_to_rx_buffers(core, d, pdma_st, + pdma_st->iov->iov_base, + iov_copy); + + *copy_size -= iov_copy; + ps_hdr_copied += iov_copy; + + pdma_st->iov_ofs += iov_copy; + if (pdma_st->iov_ofs == pdma_st->iov->iov_len) { + pdma_st->iov++; + pdma_st->iov_ofs = 0; + } + } while (ps_hdr_copied < pdma_st->ps_hdr_len); + + pdma_st->is_first = false; +} + static void igb_write_payload_frag_to_rx_buffers(IGBCore *core, PCIDevice *d, - hwaddr ba, - uint16_t *written, - uint32_t cur_buf_len, + IGBPacketRxDMAState *pdma_st, const char *data, dma_addr_t data_len) { - trace_igb_rx_desc_buff_write(ba, *written, data, data_len); - pci_dma_write(d, ba + *written, data, data_len); - *written += data_len; + while (data_len > 0) { + assert(pdma_st->bastate.cur_idx < IGB_MAX_PS_BUFFERS); + + uint32_t cur_buf_bytes_left = + pdma_st->rx_desc_packet_buf_size - + pdma_st->bastate.written[pdma_st->bastate.cur_idx]; + uint32_t bytes_to_write = MIN(data_len, cur_buf_bytes_left); + + trace_igb_rx_desc_buff_write( + pdma_st->bastate.cur_idx, + pdma_st->ba[pdma_st->bastate.cur_idx], + pdma_st->bastate.written[pdma_st->bastate.cur_idx], + data, + bytes_to_write); + + pci_dma_write(d, + pdma_st->ba[pdma_st->bastate.cur_idx] + + pdma_st->bastate.written[pdma_st->bastate.cur_idx], + data, bytes_to_write); + + pdma_st->bastate.written[pdma_st->bastate.cur_idx] += bytes_to_write; + data += bytes_to_write; + data_len -= bytes_to_write; + + if (pdma_st->bastate.written[pdma_st->bastate.cur_idx] == + pdma_st->rx_desc_packet_buf_size) { + pdma_st->bastate.cur_idx++; + } + } } static void @@ -1600,9 +1859,7 @@ igb_write_payload_to_rx_buffers(IGBCore *core, while (*copy_size) { iov_copy = MIN(*copy_size, pdma_st->iov->iov_len - pdma_st->iov_ofs); igb_write_payload_frag_to_rx_buffers(core, d, - pdma_st->ba, - &pdma_st->written, - pdma_st->rx_desc_packet_buf_size, + pdma_st, pdma_st->iov->iov_base + pdma_st->iov_ofs, iov_copy); @@ -1618,9 +1875,7 @@ igb_write_payload_to_rx_buffers(IGBCore *core, if (pdma_st->desc_offset + pdma_st->desc_size >= pdma_st->total_size) { /* Simulate FCS checksum presence in the last descriptor */ igb_write_payload_frag_to_rx_buffers(core, d, - pdma_st->ba, - &pdma_st->written, - pdma_st->rx_desc_packet_buf_size, + pdma_st, (const char *) &fcs_pad, e1000x_fcs_len(core->mac)); } @@ -1634,7 +1889,7 @@ igb_write_to_rx_buffers(IGBCore *core, { size_t copy_size; - if (!pdma_st->ba) { + if (!(pdma_st->ba)[1] || (pdma_st->do_ps && !(pdma_st->ba[0]))) { /* as per intel docs; skip descriptors with null buf addr */ trace_e1000e_rx_null_descriptor(); return; @@ -1648,6 +1903,14 @@ igb_write_to_rx_buffers(IGBCore *core, igb_truncate_to_descriptor_size(pdma_st, &pdma_st->desc_size); copy_size = pdma_st->size - pdma_st->desc_offset; igb_truncate_to_descriptor_size(pdma_st, ©_size); + + /* For PS mode copy the packet header first */ + if (pdma_st->do_ps) { + igb_write_header_to_rx_buffers(core, pkt, d, pdma_st, ©_size); + } else { + pdma_st->bastate.cur_idx = 1; + } + igb_write_payload_to_rx_buffers(core, pkt, d, pdma_st, ©_size); } @@ -1662,6 +1925,7 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, union e1000_rx_desc_union desc; const E1000ERingInfo *rxi; size_t rx_desc_len; + IGBSplitDescriptorData ps_desc_data; IGBPacketRxDMAState pdma_st = {0}; pdma_st.is_first = true; @@ -1678,8 +1942,12 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, d = core->owner; } + pdma_st.do_ps = igb_do_ps(core, rxi, pkt, + &pdma_st.ps_hdr_len, + &ps_desc_data); + do { - pdma_st.written = 0; + memset(&pdma_st.bastate, 0, sizeof(IGBBAState)); bool is_last = false; if (igb_ring_empty(core, rxi)) { @@ -1690,7 +1958,7 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, pci_dma_read(d, base, &desc, rx_desc_len); trace_e1000e_rx_descr(rxi->idx, base, rx_desc_len); - igb_read_rx_descr(core, &desc, &pdma_st.ba); + igb_read_rx_descr(core, &desc, &pdma_st, rxi); igb_write_to_rx_buffers(core, pkt, d, &pdma_st); pdma_st.desc_offset += pdma_st.desc_size; @@ -1698,8 +1966,13 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, is_last = true; } - igb_write_rx_descr(core, &desc, is_last ? core->rx_pkt : NULL, - rss_info, etqf, ts, pdma_st.written); + igb_write_rx_descr(core, &desc, + is_last ? pkt : NULL, + rss_info, + etqf, ts, + &ps_desc_data, + &pdma_st, + rxi); igb_pci_dma_write_rx_desc(core, d, base, &desc, rx_desc_len); igb_ring_advance(core, rxi, rx_desc_len / E1000_MIN_RX_DESC_LEN); diff --git a/hw/net/igb_regs.h b/hw/net/igb_regs.h index 36763f2ff7..ed7427b8fe 100644 --- a/hw/net/igb_regs.h +++ b/hw/net/igb_regs.h @@ -452,6 +452,7 @@ union e1000_adv_rx_desc { #define E1000_SRRCTL_BSIZEHDRSIZE_MASK 0x00000F00 #define E1000_SRRCTL_BSIZEHDRSIZE_SHIFT 2 /* Shift _left_ */ #define E1000_SRRCTL_DESCTYPE_ADV_ONEBUF 0x02000000 +#define E1000_SRRCTL_DESCTYPE_HDR_SPLIT 0x04000000 #define E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS 0x0A000000 #define E1000_SRRCTL_DESCTYPE_MASK 0x0E000000 #define E1000_SRRCTL_DROP_EN 0x80000000 @@ -699,6 +700,14 @@ union e1000_adv_rx_desc { #define E1000_ADVRXD_PKT_UDP BIT(5) #define E1000_ADVRXD_PKT_SCTP BIT(6) +#define IGB_MAX_PS_BUFFERS 2 + +#define E1000_ADVRXD_HDR_LEN_OFFSET (21 - 16) +#define E1000_ADVRXD_ADV_HDR_LEN_MASK ((BIT(10) - 1) << \ + E1000_ADVRXD_HDR_LEN_OFFSET) +#define E1000_ADVRXD_HDR_SPH BIT(15) +#define E1000_ADVRXD_ST_ERR_HBO_OFFSET BIT(3 + 20) + static inline uint8_t igb_ivar_entry_rx(uint8_t i) { return i < 8 ? i * 4 : (i - 8) * 4 + 2; diff --git a/hw/net/trace-events b/hw/net/trace-events index 159df09014..e97e9dc17b 100644 --- a/hw/net/trace-events +++ b/hw/net/trace-events @@ -277,7 +277,7 @@ igb_core_mdic_write_unhandled(uint32_t addr) "MDIC WRITE: PHY[%u] UNHANDLED" igb_link_set_ext_params(bool asd_check, bool speed_select_bypass, bool pfrstd) "Set extended link params: ASD check: %d, Speed select bypass: %d, PF reset done: %d" igb_rx_desc_buff_size(uint32_t b) "buffer size: %u" -igb_rx_desc_buff_write(uint64_t addr, uint16_t offset, const void* source, uint32_t len) "addr: 0x%"PRIx64", offset: %u, from: %p, length: %u" +igb_rx_desc_buff_write(uint8_t idx, uint64_t addr, uint16_t offset, const void* source, uint32_t len) "buffer %u, addr: 0x%"PRIx64", offset: %u, from: %p, length: %u" igb_rx_metadata_rss(uint32_t rss, uint16_t rss_pkt_type) "RSS data: rss: 0x%X, rss_pkt_type: 0x%X" -- 2.25.1 ^ permalink raw reply related [flat|nested] 11+ messages in thread
* Re: [PATCH v9 6/7] igb: packet-split descriptors support 2023-05-25 15:37 ` [PATCH v9 6/7] igb: packet-split descriptors support Tomasz Dzieciol @ 2023-05-27 0:07 ` Akihiko Odaki 0 siblings, 0 replies; 11+ messages in thread From: Akihiko Odaki @ 2023-05-27 0:07 UTC (permalink / raw) To: Tomasz Dzieciol, qemu-devel Cc: sriram.yagnaraman, jasowang, k.kwiecien, m.sochacki On 2023/05/26 0:37, Tomasz Dzieciol wrote: > Packet-split descriptors are used by Linux VF driver for MTU values from 2048 > > Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com> > --- > hw/net/igb_core.c | 357 ++++++++++++++++++++++++++++++++++++++------ > hw/net/igb_regs.h | 9 ++ > hw/net/trace-events | 2 +- > 3 files changed, 325 insertions(+), 43 deletions(-) > > diff --git a/hw/net/igb_core.c b/hw/net/igb_core.c > index 2bfae517fc..c1a8dfbca8 100644 > --- a/hw/net/igb_core.c > +++ b/hw/net/igb_core.c > @@ -267,6 +267,29 @@ igb_rx_use_legacy_descriptor(IGBCore *core) > return false; > } > > +typedef struct E1000ERingInfo { > + int dbah; > + int dbal; > + int dlen; > + int dh; > + int dt; > + int idx; > +} E1000ERingInfo; > + > +static uint32_t > +igb_rx_queue_desctyp_get(IGBCore *core, const E1000ERingInfo *r) > +{ > + return core->mac[E1000_SRRCTL(r->idx) >> 2] & E1000_SRRCTL_DESCTYPE_MASK; > +} > + > +static bool > +igb_rx_use_ps_descriptor(IGBCore *core, const E1000ERingInfo *r) > +{ > + uint32_t desctyp = igb_rx_queue_desctyp_get(core, r); > + return desctyp == E1000_SRRCTL_DESCTYPE_HDR_SPLIT || > + desctyp == E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS; > +} > + > static inline bool > igb_rss_enabled(IGBCore *core) > { > @@ -694,15 +717,6 @@ static uint32_t igb_rx_wb_eic(IGBCore *core, int queue_idx) > return (ent & E1000_IVAR_VALID) ? BIT(ent & 0x1f) : 0; > } > > -typedef struct E1000ERingInfo { > - int dbah; > - int dbal; > - int dlen; > - int dh; > - int dt; > - int idx; > -} E1000ERingInfo; > - > static inline bool > igb_ring_empty(IGBCore *core, const E1000ERingInfo *r) > { > @@ -1233,12 +1247,25 @@ igb_read_lgcy_rx_descr(IGBCore *core, struct e1000_rx_desc *desc, > } > > static inline void > -igb_read_adv_rx_descr(IGBCore *core, union e1000_adv_rx_desc *desc, > - hwaddr *buff_addr) > +igb_read_adv_rx_single_buf_descr(IGBCore *core, union e1000_adv_rx_desc *desc, > + hwaddr *buff_addr) > { > *buff_addr = le64_to_cpu(desc->read.pkt_addr); > } > > +static inline void > +igb_read_adv_rx_split_buf_descr(IGBCore *core, union e1000_adv_rx_desc *desc, > + hwaddr *buff_addr) > +{ > + buff_addr[0] = le64_to_cpu(desc->read.hdr_addr); > + buff_addr[1] = le64_to_cpu(desc->read.pkt_addr); > +} > + > +typedef struct IGBBAState { > + uint16_t written[IGB_MAX_PS_BUFFERS]; > + uint8_t cur_idx; > +} IGBBAState; > + > typedef struct IGBPacketRxDMAState { > size_t size; > size_t total_size; > @@ -1249,20 +1276,41 @@ typedef struct IGBPacketRxDMAState { > uint32_t rx_desc_header_buf_size; > struct iovec *iov; > size_t iov_ofs; > + bool do_ps; > bool is_first; > - uint16_t written; > - hwaddr ba; > + IGBBAState bastate; > + hwaddr ba[IGB_MAX_PS_BUFFERS]; > } IGBPacketRxDMAState; > > static inline void > -igb_read_rx_descr(IGBCore *core, union e1000_rx_desc_union *desc, > - hwaddr *buff_addr) > +igb_read_rx_descr(IGBCore *core, > + union e1000_rx_desc_union *desc, > + IGBPacketRxDMAState *pdma_st, > + const E1000ERingInfo *r) > { > + uint32_t desc_type; > + > if (igb_rx_use_legacy_descriptor(core)) { > - igb_read_lgcy_rx_descr(core, &desc->legacy, buff_addr); > - } else { > - igb_read_adv_rx_descr(core, &desc->adv, buff_addr); > + igb_read_lgcy_rx_descr(core, &desc->legacy, &pdma_st->ba[1]); > + pdma_st->ba[0] = 0; > + return; > + } > + > + /* advanced header split descriptor */ > + if (igb_rx_use_ps_descriptor(core, r)) { > + igb_read_adv_rx_split_buf_descr(core, &desc->adv, &pdma_st->ba[0]); > + return; > + } > + > + /* descriptor replication modes not supported */ > + desc_type = igb_rx_queue_desctyp_get(core, r); > + if (desc_type != E1000_SRRCTL_DESCTYPE_ADV_ONEBUF) { > + trace_igb_wrn_rx_desc_modes_not_supp(desc_type); > } > + > + /* advanced single buffer descriptor */ > + igb_read_adv_rx_single_buf_descr(core, &desc->adv, &pdma_st->ba[1]); > + pdma_st->ba[0] = 0; > } > > static void > @@ -1405,6 +1453,13 @@ igb_write_lgcy_rx_descr(IGBCore *core, struct e1000_rx_desc *desc, > desc->status = (uint8_t) le32_to_cpu(status_flags); > } > > +static bool > +igb_rx_ps_descriptor_split_always(IGBCore *core, const E1000ERingInfo *r) > +{ > + uint32_t desctyp = igb_rx_queue_desctyp_get(core, r); > + return desctyp == E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS; > +} > + > static uint16_t > igb_rx_desc_get_packet_type(IGBCore *core, struct NetRxPkt *pkt, uint16_t etqf) > { > @@ -1494,16 +1549,64 @@ igb_write_adv_rx_descr(IGBCore *core, union e1000_adv_rx_desc *desc, > desc->wb.upper.status_error |= cpu_to_le32(adv_desc_status_error); > } > > +typedef struct IGBSplitDescriptorData { > + bool sph; > + bool hbo; > + size_t hdr_len; > +} IGBSplitDescriptorData; > + > static inline void > -igb_write_rx_descr(IGBCore *core, union e1000_rx_desc_union *desc, > - struct NetRxPkt *pkt, const E1000E_RSSInfo *rss_info, > - uint16_t etqf, bool ts, uint16_t length) > +igb_write_adv_ps_rx_descr(IGBCore *core, > + union e1000_adv_rx_desc *desc, > + struct NetRxPkt *pkt, > + const E1000E_RSSInfo *rss_info, > + const E1000ERingInfo *r, > + uint16_t etqf, > + bool ts, > + IGBSplitDescriptorData *ps_desc_data, > + IGBPacketRxDMAState *pdma_st) > +{ > + size_t pkt_len; > + size_t hdr_len = ps_desc_data->hdr_len; > + uint16_t hdr_info = 0; > + > + if (pdma_st->do_ps) { > + pkt_len = pdma_st->bastate.written[1]; > + } else { > + pkt_len = pdma_st->bastate.written[0] + pdma_st->bastate.written[1]; > + } > + > + igb_write_adv_rx_descr(core, desc, pkt, rss_info, etqf, ts, pkt_len); > + > + hdr_info = (hdr_len << E1000_ADVRXD_HDR_LEN_OFFSET) & > + E1000_ADVRXD_ADV_HDR_LEN_MASK; > + hdr_info |= ps_desc_data->sph ? E1000_ADVRXD_HDR_SPH : 0; > + desc->wb.lower.lo_dword.hdr_info = cpu_to_le16(hdr_info); > + > + desc->wb.upper.status_error |= cpu_to_le32( > + ps_desc_data->hbo ? E1000_ADVRXD_ST_ERR_HBO_OFFSET : 0); > +} > + > +static inline void > +igb_write_rx_descr(IGBCore *core, > + union e1000_rx_desc_union *desc, > + struct NetRxPkt *pkt, > + const E1000E_RSSInfo *rss_info, > + uint16_t etqf, > + bool ts, > + IGBSplitDescriptorData *ps_desc_data, > + IGBPacketRxDMAState *pdma_st, > + const E1000ERingInfo *r) > { > if (igb_rx_use_legacy_descriptor(core)) { > - igb_write_lgcy_rx_descr(core, &desc->legacy, pkt, rss_info, length); > + igb_write_lgcy_rx_descr(core, &desc->legacy, pkt, rss_info, > + pdma_st->bastate.written[1]); > + } else if (igb_rx_use_ps_descriptor(core, r)) { > + igb_write_adv_ps_rx_descr(core, &desc->adv, pkt, rss_info, r, etqf, ts, > + ps_desc_data, pdma_st); > } else { > igb_write_adv_rx_descr(core, &desc->adv, pkt, rss_info, > - etqf, ts, length); > + etqf, ts, pdma_st->bastate.written[1]); > } > } > > @@ -1564,26 +1667,182 @@ igb_rx_descr_threshold_hit(IGBCore *core, const E1000ERingInfo *rxi) > ((core->mac[E1000_SRRCTL(rxi->idx) >> 2] >> 20) & 31) * 16; > } > > +static bool > +igb_do_ps(IGBCore *core, > + const E1000ERingInfo *r, > + struct NetRxPkt *pkt, > + size_t *hdr_len, > + IGBSplitDescriptorData *ps_desc_data) > +{ > + bool hasip4, hasip6; > + EthL4HdrProto l4hdr_proto; > + bool fragment; > + bool split_always; > + size_t bheader_size; > + size_t total_pkt_len; > + > + if (!igb_rx_use_ps_descriptor(core, r)) { > + return false; > + } > + > + memset(ps_desc_data, 0, sizeof(IGBSplitDescriptorData)); > + > + total_pkt_len = net_rx_pkt_get_total_len(pkt); > + bheader_size = igb_rxhdrbufsize(core, r); > + split_always = igb_rx_ps_descriptor_split_always(core, r); > + if (split_always && total_pkt_len <= bheader_size) { > + *hdr_len = total_pkt_len; > + ps_desc_data->hdr_len = total_pkt_len; > + return true; > + } > + > + net_rx_pkt_get_protocols(pkt, &hasip4, &hasip6, &l4hdr_proto); > + > + if (hasip4) { > + fragment = net_rx_pkt_get_ip4_info(pkt)->fragment; > + } else if (hasip6) { > + fragment = net_rx_pkt_get_ip6_info(pkt)->fragment; > + } else { > + ps_desc_data->hdr_len = bheader_size; > + goto header_not_handled; > + } > + > + if (fragment && (core->mac[RFCTL] & E1000_RFCTL_IPFRSP_DIS)) { > + ps_desc_data->hdr_len = bheader_size; > + goto header_not_handled; > + } > + > + /* no header splitting for SCTP */ > + if (!fragment && (l4hdr_proto == ETH_L4_HDR_PROTO_UDP || > + l4hdr_proto == ETH_L4_HDR_PROTO_TCP)) { > + *hdr_len = net_rx_pkt_get_l5_hdr_offset(pkt); > + } else { > + *hdr_len = net_rx_pkt_get_l4_hdr_offset(pkt); > + } > + > + ps_desc_data->sph = true; > + ps_desc_data->hdr_len = *hdr_len; > + > + if (*hdr_len > bheader_size) { > + ps_desc_data->hbo = true; > + goto header_not_handled; > + } > + > + return true; > + > +header_not_handled: > + if (split_always) { > + *hdr_len = bheader_size; > + return true; > + } > + > + return false; > +} > + > static void > igb_truncate_to_descriptor_size(IGBPacketRxDMAState *pdma_st, size_t *size) > { > - if (*size > pdma_st->rx_desc_packet_buf_size) { > - *size = pdma_st->rx_desc_packet_buf_size; > + if (pdma_st->do_ps && pdma_st->is_first) { > + if (*size > pdma_st->rx_desc_packet_buf_size + pdma_st->ps_hdr_len) { > + *size = pdma_st->rx_desc_packet_buf_size + pdma_st->ps_hdr_len; > + } > + } else { > + if (*size > pdma_st->rx_desc_packet_buf_size) { > + *size = pdma_st->rx_desc_packet_buf_size; > + } > } > } > > +static inline void > +igb_write_hdr_frag_to_rx_buffers(IGBCore *core, > + PCIDevice *d, > + IGBPacketRxDMAState *pdma_st, > + const char *data, > + dma_addr_t data_len) > +{ > + assert(data_len <= pdma_st->rx_desc_header_buf_size - > + pdma_st->bastate.written[0]); > + pci_dma_write(d, > + pdma_st->ba[0] + pdma_st->bastate.written[0], > + data, data_len); > + pdma_st->bastate.written[0] += data_len; > + pdma_st->bastate.cur_idx = 1; > +} > + > +static void > +igb_write_header_to_rx_buffers(IGBCore *core, > + struct NetRxPkt *pkt, > + PCIDevice *d, > + IGBPacketRxDMAState *pdma_st, > + size_t *copy_size) > +{ > + size_t iov_copy; > + size_t ps_hdr_copied = 0; > + > + if (!pdma_st->is_first) { > + /* Leave buffer 0 of each descriptor except first */ > + /* empty */ > + pdma_st->bastate.cur_idx = 1; > + return; > + } > + > + do { > + iov_copy = MIN(pdma_st->ps_hdr_len - ps_hdr_copied, > + pdma_st->iov->iov_len - pdma_st->iov_ofs); > + > + igb_write_hdr_frag_to_rx_buffers(core, d, pdma_st, > + pdma_st->iov->iov_base, > + iov_copy); > + > + *copy_size -= iov_copy; > + ps_hdr_copied += iov_copy; > + > + pdma_st->iov_ofs += iov_copy; > + if (pdma_st->iov_ofs == pdma_st->iov->iov_len) { > + pdma_st->iov++; > + pdma_st->iov_ofs = 0; > + } > + } while (ps_hdr_copied < pdma_st->ps_hdr_len); > + > + pdma_st->is_first = false; > +} > + > static void > igb_write_payload_frag_to_rx_buffers(IGBCore *core, > PCIDevice *d, > - hwaddr ba, > - uint16_t *written, > - uint32_t cur_buf_len, > + IGBPacketRxDMAState *pdma_st, > const char *data, > dma_addr_t data_len) > { > - trace_igb_rx_desc_buff_write(ba, *written, data, data_len); > - pci_dma_write(d, ba + *written, data, data_len); > - *written += data_len; > + while (data_len > 0) { > + assert(pdma_st->bastate.cur_idx < IGB_MAX_PS_BUFFERS); > + > + uint32_t cur_buf_bytes_left = > + pdma_st->rx_desc_packet_buf_size - > + pdma_st->bastate.written[pdma_st->bastate.cur_idx]; > + uint32_t bytes_to_write = MIN(data_len, cur_buf_bytes_left); > + > + trace_igb_rx_desc_buff_write( > + pdma_st->bastate.cur_idx, > + pdma_st->ba[pdma_st->bastate.cur_idx], > + pdma_st->bastate.written[pdma_st->bastate.cur_idx], > + data, > + bytes_to_write); > + > + pci_dma_write(d, > + pdma_st->ba[pdma_st->bastate.cur_idx] + > + pdma_st->bastate.written[pdma_st->bastate.cur_idx], > + data, bytes_to_write); > + > + pdma_st->bastate.written[pdma_st->bastate.cur_idx] += bytes_to_write; > + data += bytes_to_write; > + data_len -= bytes_to_write; > + > + if (pdma_st->bastate.written[pdma_st->bastate.cur_idx] == > + pdma_st->rx_desc_packet_buf_size) { > + pdma_st->bastate.cur_idx++; > + } > + } > } > > static void > @@ -1600,9 +1859,7 @@ igb_write_payload_to_rx_buffers(IGBCore *core, > while (*copy_size) { > iov_copy = MIN(*copy_size, pdma_st->iov->iov_len - pdma_st->iov_ofs); > igb_write_payload_frag_to_rx_buffers(core, d, > - pdma_st->ba, > - &pdma_st->written, > - pdma_st->rx_desc_packet_buf_size, > + pdma_st, > pdma_st->iov->iov_base + > pdma_st->iov_ofs, > iov_copy); > @@ -1618,9 +1875,7 @@ igb_write_payload_to_rx_buffers(IGBCore *core, > if (pdma_st->desc_offset + pdma_st->desc_size >= pdma_st->total_size) { > /* Simulate FCS checksum presence in the last descriptor */ > igb_write_payload_frag_to_rx_buffers(core, d, > - pdma_st->ba, > - &pdma_st->written, > - pdma_st->rx_desc_packet_buf_size, > + pdma_st, > (const char *) &fcs_pad, > e1000x_fcs_len(core->mac)); > } > @@ -1634,7 +1889,7 @@ igb_write_to_rx_buffers(IGBCore *core, > { > size_t copy_size; > > - if (!pdma_st->ba) { > + if (!(pdma_st->ba)[1] || (pdma_st->do_ps && !(pdma_st->ba[0]))) { > /* as per intel docs; skip descriptors with null buf addr */ > trace_e1000e_rx_null_descriptor(); > return; > @@ -1648,6 +1903,14 @@ igb_write_to_rx_buffers(IGBCore *core, > igb_truncate_to_descriptor_size(pdma_st, &pdma_st->desc_size); > copy_size = pdma_st->size - pdma_st->desc_offset; > igb_truncate_to_descriptor_size(pdma_st, ©_size); > + > + /* For PS mode copy the packet header first */ > + if (pdma_st->do_ps) { > + igb_write_header_to_rx_buffers(core, pkt, d, pdma_st, ©_size); > + } else { > + pdma_st->bastate.cur_idx = 1; > + } > + > igb_write_payload_to_rx_buffers(core, pkt, d, pdma_st, ©_size); > } > > @@ -1662,6 +1925,7 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, > union e1000_rx_desc_union desc; > const E1000ERingInfo *rxi; > size_t rx_desc_len; > + IGBSplitDescriptorData ps_desc_data; > > IGBPacketRxDMAState pdma_st = {0}; > pdma_st.is_first = true; > @@ -1678,8 +1942,12 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, > d = core->owner; > } > > + pdma_st.do_ps = igb_do_ps(core, rxi, pkt, > + &pdma_st.ps_hdr_len, > + &ps_desc_data); If you put ps_hdr_len in pdma_st, why don't you do so for ps_desc_data? > + > do { > - pdma_st.written = 0; > + memset(&pdma_st.bastate, 0, sizeof(IGBBAState)); > bool is_last = false; > > if (igb_ring_empty(core, rxi)) { > @@ -1690,7 +1958,7 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, > pci_dma_read(d, base, &desc, rx_desc_len); > trace_e1000e_rx_descr(rxi->idx, base, rx_desc_len); > > - igb_read_rx_descr(core, &desc, &pdma_st.ba); > + igb_read_rx_descr(core, &desc, &pdma_st, rxi); > > igb_write_to_rx_buffers(core, pkt, d, &pdma_st); > pdma_st.desc_offset += pdma_st.desc_size; > @@ -1698,8 +1966,13 @@ igb_write_packet_to_guest(IGBCore *core, struct NetRxPkt *pkt, > is_last = true; > } > > - igb_write_rx_descr(core, &desc, is_last ? core->rx_pkt : NULL, > - rss_info, etqf, ts, pdma_st.written); > + igb_write_rx_descr(core, &desc, > + is_last ? pkt : NULL, > + rss_info, > + etqf, ts, > + &ps_desc_data, > + &pdma_st, > + rxi); > igb_pci_dma_write_rx_desc(core, d, base, &desc, rx_desc_len); > igb_ring_advance(core, rxi, > rx_desc_len / E1000_MIN_RX_DESC_LEN); > diff --git a/hw/net/igb_regs.h b/hw/net/igb_regs.h > index 36763f2ff7..ed7427b8fe 100644 > --- a/hw/net/igb_regs.h > +++ b/hw/net/igb_regs.h > @@ -452,6 +452,7 @@ union e1000_adv_rx_desc { > #define E1000_SRRCTL_BSIZEHDRSIZE_MASK 0x00000F00 > #define E1000_SRRCTL_BSIZEHDRSIZE_SHIFT 2 /* Shift _left_ */ > #define E1000_SRRCTL_DESCTYPE_ADV_ONEBUF 0x02000000 > +#define E1000_SRRCTL_DESCTYPE_HDR_SPLIT 0x04000000 > #define E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS 0x0A000000 > #define E1000_SRRCTL_DESCTYPE_MASK 0x0E000000 > #define E1000_SRRCTL_DROP_EN 0x80000000 > @@ -699,6 +700,14 @@ union e1000_adv_rx_desc { > #define E1000_ADVRXD_PKT_UDP BIT(5) > #define E1000_ADVRXD_PKT_SCTP BIT(6) > > +#define IGB_MAX_PS_BUFFERS 2 > + > +#define E1000_ADVRXD_HDR_LEN_OFFSET (21 - 16) > +#define E1000_ADVRXD_ADV_HDR_LEN_MASK ((BIT(10) - 1) << \ > + E1000_ADVRXD_HDR_LEN_OFFSET) > +#define E1000_ADVRXD_HDR_SPH BIT(15) > +#define E1000_ADVRXD_ST_ERR_HBO_OFFSET BIT(3 + 20) > + > static inline uint8_t igb_ivar_entry_rx(uint8_t i) > { > return i < 8 ? i * 4 : (i - 8) * 4 + 2; > diff --git a/hw/net/trace-events b/hw/net/trace-events > index 159df09014..e97e9dc17b 100644 > --- a/hw/net/trace-events > +++ b/hw/net/trace-events > @@ -277,7 +277,7 @@ igb_core_mdic_write_unhandled(uint32_t addr) "MDIC WRITE: PHY[%u] UNHANDLED" > igb_link_set_ext_params(bool asd_check, bool speed_select_bypass, bool pfrstd) "Set extended link params: ASD check: %d, Speed select bypass: %d, PF reset done: %d" > > igb_rx_desc_buff_size(uint32_t b) "buffer size: %u" > -igb_rx_desc_buff_write(uint64_t addr, uint16_t offset, const void* source, uint32_t len) "addr: 0x%"PRIx64", offset: %u, from: %p, length: %u" > +igb_rx_desc_buff_write(uint8_t idx, uint64_t addr, uint16_t offset, const void* source, uint32_t len) "buffer %u, addr: 0x%"PRIx64", offset: %u, from: %p, length: %u" > > igb_rx_metadata_rss(uint32_t rss, uint16_t rss_pkt_type) "RSS data: rss: 0x%X, rss_pkt_type: 0x%X" > ^ permalink raw reply [flat|nested] 11+ messages in thread
[parent not found: <CGME20230525153746eucas1p2903b5d0ecf65994f9b68ddf039d3d881@eucas1p2.samsung.com>]
* [PATCH v9 7/7] e1000e: rename e1000e_ba_state and e1000e_write_hdr_to_rx_buffers [not found] ` <CGME20230525153746eucas1p2903b5d0ecf65994f9b68ddf039d3d881@eucas1p2.samsung.com> @ 2023-05-25 15:37 ` Tomasz Dzieciol 0 siblings, 0 replies; 11+ messages in thread From: Tomasz Dzieciol @ 2023-05-25 15:37 UTC (permalink / raw) To: qemu-devel, akihiko.odaki Cc: sriram.yagnaraman, jasowang, k.kwiecien, m.sochacki Rename e1000e_ba_state according and e1000e_write_hdr_to_rx_buffers for consistency with IGB. Signed-off-by: Tomasz Dzieciol <t.dzieciol@partner.samsung.com> --- hw/net/e1000e_core.c | 28 +++++++++++++++------------- 1 file changed, 15 insertions(+), 13 deletions(-) diff --git a/hw/net/e1000e_core.c b/hw/net/e1000e_core.c index f29e5e2897..bbeafd6fd7 100644 --- a/hw/net/e1000e_core.c +++ b/hw/net/e1000e_core.c @@ -1397,17 +1397,17 @@ e1000e_pci_dma_write_rx_desc(E1000ECore *core, dma_addr_t addr, } } -typedef struct e1000e_ba_state_st { +typedef struct E1000EBAState { uint16_t written[MAX_PS_BUFFERS]; uint8_t cur_idx; -} e1000e_ba_state; +} E1000EBAState; static inline void -e1000e_write_hdr_to_rx_buffers(E1000ECore *core, - hwaddr ba[MAX_PS_BUFFERS], - e1000e_ba_state *bastate, - const char *data, - dma_addr_t data_len) +e1000e_write_hdr_frag_to_rx_buffers(E1000ECore *core, + hwaddr ba[MAX_PS_BUFFERS], + E1000EBAState *bastate, + const char *data, + dma_addr_t data_len) { assert(data_len <= core->rxbuf_sizes[0] - bastate->written[0]); @@ -1420,7 +1420,7 @@ e1000e_write_hdr_to_rx_buffers(E1000ECore *core, static void e1000e_write_payload_frag_to_rx_buffers(E1000ECore *core, hwaddr ba[MAX_PS_BUFFERS], - e1000e_ba_state *bastate, + E1000EBAState *bastate, const char *data, dma_addr_t data_len) { @@ -1530,7 +1530,7 @@ e1000e_write_packet_to_guest(E1000ECore *core, struct NetRxPkt *pkt, do { hwaddr ba[MAX_PS_BUFFERS]; - e1000e_ba_state bastate = { { 0 } }; + E1000EBAState bastate = { { 0 } }; bool is_last = false; desc_size = total_size - desc_offset; @@ -1568,8 +1568,10 @@ e1000e_write_packet_to_guest(E1000ECore *core, struct NetRxPkt *pkt, iov_copy = MIN(ps_hdr_len - ps_hdr_copied, iov->iov_len - iov_ofs); - e1000e_write_hdr_to_rx_buffers(core, ba, &bastate, - iov->iov_base, iov_copy); + e1000e_write_hdr_frag_to_rx_buffers(core, ba, + &bastate, + iov->iov_base, + iov_copy); copy_size -= iov_copy; ps_hdr_copied += iov_copy; @@ -1585,8 +1587,8 @@ e1000e_write_packet_to_guest(E1000ECore *core, struct NetRxPkt *pkt, } else { /* Leave buffer 0 of each descriptor except first */ /* empty as per spec 7.1.5.1 */ - e1000e_write_hdr_to_rx_buffers(core, ba, &bastate, - NULL, 0); + e1000e_write_hdr_frag_to_rx_buffers(core, ba, &bastate, + NULL, 0); } } -- 2.25.1 ^ permalink raw reply related [flat|nested] 11+ messages in thread
* Re: [PATCH v9 0/7] igb: packet-split descriptors support 2023-05-25 15:37 ` [PATCH v9 0/7] igb: packet-split descriptors support Tomasz Dzieciol ` (6 preceding siblings ...) [not found] ` <CGME20230525153746eucas1p2903b5d0ecf65994f9b68ddf039d3d881@eucas1p2.samsung.com> @ 2023-05-27 0:10 ` Akihiko Odaki 7 siblings, 0 replies; 11+ messages in thread From: Akihiko Odaki @ 2023-05-27 0:10 UTC (permalink / raw) To: Tomasz Dzieciol, qemu-devel Cc: sriram.yagnaraman, jasowang, k.kwiecien, m.sochacki On 2023/05/26 0:37, Tomasz Dzieciol wrote: > Based-on: <20230523024339.50875-1-akihiko.odaki@daynix.com> > ("[PATCH v5 00/48] igb: Fix for DPDK") > > Purposes of this series of patches: > * introduce packet-split RX descriptors support. This feature is used by Linux > VF driver for MTU values from 2048. > * refactor RX descriptor handling for introduction of packet-split RX > descriptors support > * fix descriptors flags handling > > In addition to comments from previous review endianess issues in > igb_write_adv_ps_rx_descr were fixed. > > Tomasz Dzieciol (7): > igb: remove TCP ACK detection > igb: rename E1000E_RingInfo_st > igb: RX descriptors guest writting refactoring > igb: RX payload guest writting refactoring > igb: add IPv6 extended headers traffic detection > igb: packet-split descriptors support > e1000e: rename e1000e_ba_state and e1000e_write_hdr_to_rx_buffers > > hw/net/e1000e_core.c | 78 +++-- > hw/net/igb_core.c | 740 ++++++++++++++++++++++++++++----------- > hw/net/igb_regs.h | 20 +- > hw/net/trace-events | 6 +- > tests/qtest/libqos/igb.c | 5 + > 5 files changed, 602 insertions(+), 247 deletions(-) > Please rebase to the current master and remove Based-on: tag. This series should cleanly apply. I left some minor comments that does not involve semantic changes so check them out. I think I can give Reviewed-by: and Tested-by: for the next version. Thanks for keeping working on this. Regards, Akihiko Odaki ^ permalink raw reply [flat|nested] 11+ messages in thread
end of thread, other threads:[~2023-05-27 0:11 UTC | newest]
Thread overview: 11+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
[not found] <CGME20230525153743eucas1p145858f3572c2d53687377a4e879d1e0b@eucas1p1.samsung.com>
2023-05-25 15:37 ` [PATCH v9 0/7] igb: packet-split descriptors support Tomasz Dzieciol
[not found] ` <CGME20230525153743eucas1p2cfe25bb5c0bfc59064189d58b328a537@eucas1p2.samsung.com>
2023-05-25 15:37 ` [PATCH v9 1/7] igb: remove TCP ACK detection Tomasz Dzieciol
[not found] ` <CGME20230525153744eucas1p208318603e67e11c9ab3f0a6556b8d3a4@eucas1p2.samsung.com>
2023-05-25 15:37 ` [PATCH v9 2/7] igb: rename E1000E_RingInfo_st Tomasz Dzieciol
[not found] ` <CGME20230525153744eucas1p1844ba0a10563caff4bb0f92c82f30f7d@eucas1p1.samsung.com>
2023-05-25 15:37 ` [PATCH v9 3/7] igb: RX descriptors guest writting refactoring Tomasz Dzieciol
[not found] ` <CGME20230525153745eucas1p1cd6d947ed0cd46a8fdf637b514c237ff@eucas1p1.samsung.com>
2023-05-25 15:37 ` [PATCH v9 4/7] igb: RX payload " Tomasz Dzieciol
2023-05-26 23:54 ` Akihiko Odaki
[not found] ` <CGME20230525153745eucas1p22c62582ef16e6d6c9da9ab58243414f0@eucas1p2.samsung.com>
2023-05-25 15:37 ` [PATCH v9 5/7] igb: add IPv6 extended headers traffic detection Tomasz Dzieciol
[not found] ` <CGME20230525153746eucas1p1957a3a91d90521a6d1c684595c848f52@eucas1p1.samsung.com>
2023-05-25 15:37 ` [PATCH v9 6/7] igb: packet-split descriptors support Tomasz Dzieciol
2023-05-27 0:07 ` Akihiko Odaki
[not found] ` <CGME20230525153746eucas1p2903b5d0ecf65994f9b68ddf039d3d881@eucas1p2.samsung.com>
2023-05-25 15:37 ` [PATCH v9 7/7] e1000e: rename e1000e_ba_state and e1000e_write_hdr_to_rx_buffers Tomasz Dzieciol
2023-05-27 0:10 ` [PATCH v9 0/7] igb: packet-split descriptors support Akihiko Odaki
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).