bnxt_en: Update for net-next
Michael Chan
maintainerv4 cover bnxt_en: Update for net-next
This series starts with 2 patches to support firmware crash dump. The driver allocates the required DMA memory ahead of time for firmware to store the crash dump if and when it crashes. Patch 3 adds priority and TPID for the .ndo_set_vf_vlan() callback. Note that this was rejected and reverted last year and it is being re-submitted after recent changes in the guidelines. The remaining patches are MSIX related. Legacy interrupt is no longer supported by firmware so we remove the support in the driver. We then convert to use the newer kernel APIs to allocate and enable MSIX vectors. The last patch adds support for dynamic MSIX.
v4:
Simplify patch #9 based on feedback from Michal Swiatkowski.
Link to v3:
https://lore.kernel.org/netdev/[email protected]/
v3:
Some changes to patch #1, #2, and #8 based on feedback from Przemek Kitszel and internal review. I'm keeping Simon's Reviewed-by tags since the changes are small.
Link to v2:
https://lore.kernel.org/netdev/[email protected]/
v2:
Only patch #4 is updated to fix a memory leakage reported by Simon. The changelog of some of the MSIX patches have been updated based on feedback from Bjorn Helgaas.
Link to v1:
https://lore.kernel.org/netdev/[email protected]/
Michael Chan (6):
bnxt_en: Deprecate support for legacy INTX mode
bnxt_en: Remove BNXT_FLAG_USING_MSIX flag
bnxt_en: Remove register mapping to support INTX
bnxt_en: Replace deprecated PCI MSIX APIs
bnxt_en: Allocate the max bp->irq_tbl size for dynamic msix allocation
bnxt_en: Support dynamic MSIX
Sreekanth Reddy (1):
bnxt_en: Support QOS and TPID settings for the SRIOV VLAN
Vikas Gupta (2):
bnxt_en: add support for storing crash dump into host memory
bnxt_en: add support for retrieving crash dump using ethtool
drivers/net/ethernet/broadcom/bnxt/bnxt.c | 310 +++++++++--------- drivers/net/ethernet/broadcom/bnxt/bnxt.h | 8 +- .../ethernet/broadcom/bnxt/bnxt_coredump.c | 98 +++++- .../ethernet/broadcom/bnxt/bnxt_coredump.h | 8 + .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 13 +- .../net/ethernet/broadcom/bnxt/bnxt_sriov.c | 29 +- 6 files changed, 279 insertions(+), 187 deletions(-)
Signature · 1 line
2.30.1
Michael Chan
maintainerPATCH v4 1/9 bnxt_en: add support for storing crash dump into host memory
From: Vikas Gupta <[email protected]>
Newer firmware supports automatic DMA of crash dump to host memory when it crashes. If the feature is supported, allocate the required memory using the existing context memory infrastructure. Communicate the page table containing the DMA addresses to the firmware.
Reviewed-by Somnath Kotur <[email protected]>
Reviewed-by Pavan Chebbi <[email protected]>
Reviewed-by Andy Gospodarek <[email protected]>
Reviewed-by Simon Horman <[email protected]>
Signed-off-by Vikas Gupta <[email protected]>
Signed-off-by Michael Chan <[email protected]>
---
v3:
1. Remove one-time use BNXT_SET_CRASHDUMP_PAGE_ATTR() macro
2. Optimize memory reallocation if the existing pages can support the new memory size
3. Fix suspend/resume code paths
--- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 92 +++++++++++++++++++ drivers/net/ethernet/broadcom/bnxt/bnxt.h | 3 + .../ethernet/broadcom/bnxt/bnxt_coredump.c | 18 +++- .../ethernet/broadcom/bnxt/bnxt_coredump.h | 8 ++ 4 files changed, 116 insertions(+), 5 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +92 -0
@@ -69,6 +69,7 @@#include "bnxt_tc.h" #include "bnxt_devlink.h" #include "bnxt_debugfs.h" +#include "bnxt_coredump.h" #include "bnxt_hwmon.h" #define BNXT_TX_TIMEOUT (5 * HZ)
@@ -8946,6 +8947,80 @@ static int bnxt_alloc_ctx_mem(struct bnxt *bp)return 0 ; } +static int bnxt_hwrm_crash_dump_mem_cfg (struct bnxt *bp) +{ +struct hwrm_dbg_crashdump_medium_cfg_input *req; +u16 page_attr; +int rc; + +if (!(bp->fw_dbg_cap &DBG_QCAPS_RESP_FLAGS_CRASHDUMP_HOST_DDR )) +return 0 ; + + rc =hwrm_req_init (bp, req,HWRM_DBG_CRASHDUMP_MEDIUM_CFG ); +if (rc) +return rc; + +if (BNXT_PAGE_SIZE ==0x2000 ) + page_attr =DBG_CRASHDUMP_MEDIUM_CFG_REQ_PG_SIZE_PG_8K ; +else if (BNXT_PAGE_SIZE ==0x10000 ) + page_attr =DBG_CRASHDUMP_MEDIUM_CFG_REQ_PG_SIZE_PG_64K ; +else + page_attr =DBG_CRASHDUMP_MEDIUM_CFG_REQ_PG_SIZE_PG_4K ; + req->pg_size_lvl =cpu_to_le16 (page_attr | + bp->fw_crash_mem->ring_mem.depth); + req->pbl =cpu_to_le64 (bp->fw_crash_mem->ring_mem.pg_tbl_map); + req->size =cpu_to_le32 (bp->fw_crash_len); + req->output_dest_flags =cpu_to_le16 (BNXT_DBG_CR_DUMP_MDM_CFG_DDR ); +return hwrm_req_send (bp, req); +} + +static void bnxt_free_crash_dump_mem (struct bnxt *bp) +{ +if (bp->fw_crash_mem) { +bnxt_free_ctx_pg_tbls (bp, bp->fw_crash_mem); +kfree (bp->fw_crash_mem); + bp->fw_crash_mem =NULL ; + } +} + +static int bnxt_alloc_crash_dump_mem (struct bnxt *bp) +{ +u32 mem_size =0 ; +int rc; + +if (!(bp->fw_dbg_cap &DBG_QCAPS_RESP_FLAGS_CRASHDUMP_HOST_DDR )) +return 0 ; + + rc =bnxt_hwrm_get_dump_len (bp,BNXT_DUMP_CRASH , &mem_size); +if (rc) +return rc; + + mem_size =round_up (mem_size,4 ); + +/* keep and use the existing pages */ +if (bp->fw_crash_mem && + mem_size <= bp->fw_crash_mem->nr_pages *BNXT_PAGE_SIZE ) + goto alloc_done; + +if (bp->fw_crash_mem) +bnxt_free_ctx_pg_tbls (bp, bp->fw_crash_mem); +else + bp->fw_crash_mem =kzalloc (sizeof (*bp->fw_crash_mem), +GFP_KERNEL ); +if (!bp->fw_crash_mem) +return -ENOMEM ; + + rc =bnxt_alloc_ctx_pg_tbls (bp, bp->fw_crash_mem, mem_size,1 ,NULL ); +if (rc) { +bnxt_free_crash_dump_mem (bp); +return rc; + } + +alloc_done: + bp->fw_crash_len = mem_size; +return 0 ; +} +int bnxt_hwrm_func_resc_qcaps (struct bnxt *bp,bool all) {struct hwrm_func_resource_qcaps_output *resp;
@@ -13986,6 +14061,19 @@ static int bnxt_fw_init_one_p2(struct bnxt *bp)if (rc)return -ENODEV ; + rc =bnxt_alloc_crash_dump_mem (bp); +if (rc) +netdev_warn (bp->dev,"crash dump mem alloc failure rc: %d\n" , + rc); +if (!rc) { + rc =bnxt_hwrm_crash_dump_mem_cfg (bp); +if (rc) { +bnxt_free_crash_dump_mem (bp); +netdev_warn (bp->dev, +"hwrm crash dump mem failure rc: %d\n" , rc); + } + } +if (bnxt_fw_pre_resv_vnics (bp)) bp->fw_cap |=BNXT_FW_CAP_PRE_RESV_VNICS ;
@@ -15294,6 +15382,7 @@ static void bnxt_remove_one(struct pci_dev *pdev) bp->fw_health =NULL ;bnxt_cleanup_pci (bp);bnxt_free_ctx_mem (bp); +bnxt_free_crash_dump_mem (bp);kfree (bp->rss_indir_tbl); bp->rss_indir_tbl =NULL ;bnxt_free_port_stats (bp);
@@ -15933,6 +16022,7 @@ static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent) bp->fw_health =NULL ;bnxt_cleanup_pci (bp);bnxt_free_ctx_mem (bp); +bnxt_free_crash_dump_mem (bp);kfree (bp->rss_indir_tbl); bp->rss_indir_tbl =NULL ;
@@ -16024,6 +16114,8 @@ static int bnxt_resume(struct device *device) rc = -ENODEV ; goto resume_exit; } +if (bp->fw_crash_mem) +bnxt_hwrm_crash_dump_mem_cfg (bp);bnxt_get_wol_settings (bp);if (netif_running (dev)) {
drivers/net/ethernet/broadcom/bnxt/bnxt.h +3 -0
@@ -2649,6 +2649,9 @@ struct bnxt {#endif u32 thermal_threshold_type;enum board_idx board_idx; + +struct bnxt_ctx_pg_info *fw_crash_mem; +u32 fw_crash_len; };#define BNXT_NUM_RX_RING_STATS 8
drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c +13 -5
@@ -385,7 +385,7 @@ int bnxt_get_coredump(struct bnxt *bp, u16 dump_type, void *buf, u32 *dump_len) } } -static int bnxt_hwrm_get_dump_len (struct bnxt *bp,u16 dump_type,u32 *dump_len) +int bnxt_hwrm_get_dump_len (struct bnxt *bp,u16 dump_type,u32 *dump_len) {struct hwrm_dbg_qcfg_output *resp;struct hwrm_dbg_qcfg_input *req;
@@ -395,7 +395,8 @@ static int bnxt_hwrm_get_dump_len(struct bnxt *bp, u16 dump_type, u32 *dump_len)return -EOPNOTSUPP ;if (dump_type ==BNXT_DUMP_CRASH && - !(bp->fw_dbg_cap &DBG_QCAPS_RESP_FLAGS_CRASHDUMP_SOC_DDR )) + !(bp->fw_dbg_cap &DBG_QCAPS_RESP_FLAGS_CRASHDUMP_SOC_DDR || + (bp->fw_dbg_cap &DBG_QCAPS_RESP_FLAGS_CRASHDUMP_HOST_DDR )))return -EOPNOTSUPP ; rc =hwrm_req_init (bp, req,HWRM_DBG_QCFG );
@@ -403,8 +404,12 @@ static int bnxt_hwrm_get_dump_len(struct bnxt *bp, u16 dump_type, u32 *dump_len)return rc; req->fid =cpu_to_le16 (0xffff ); -if (dump_type ==BNXT_DUMP_CRASH ) - req->flags =cpu_to_le16 (DBG_QCFG_REQ_FLAGS_CRASHDUMP_SIZE_FOR_DEST_DEST_SOC_DDR ); +if (dump_type ==BNXT_DUMP_CRASH ) { +if (bp->fw_dbg_cap &DBG_QCAPS_RESP_FLAGS_CRASHDUMP_SOC_DDR ) + req->flags =cpu_to_le16 (BNXT_DBG_FL_CR_DUMP_SIZE_SOC ); +else + req->flags =cpu_to_le16 (BNXT_DBG_FL_CR_DUMP_SIZE_HOST ); + } resp =hwrm_req_hold (bp, req); rc =hwrm_req_send (bp, req);
@@ -412,7 +417,10 @@ static int bnxt_hwrm_get_dump_len(struct bnxt *bp, u16 dump_type, u32 *dump_len) goto get_dump_len_exit;if (dump_type ==BNXT_DUMP_CRASH ) { - *dump_len =le32_to_cpu (resp->crashdump_size); +if (bp->fw_dbg_cap &DBG_QCAPS_RESP_FLAGS_CRASHDUMP_SOC_DDR ) + *dump_len =BNXT_CRASH_DUMP_LEN ; +else + *dump_len =le32_to_cpu (resp->crashdump_size); }else {/* Driver adds coredump header and "HWRM_VER_GET response" * segment additionally to coredump.
drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.h +8 -0
@@ -111,7 +111,15 @@ struct hwrm_dbg_cmn_output {#define HWRM_DBG_CMN_FLAGS_MORE 1 }; +#define BNXT_DBG_FL_CR_DUMP_SIZE_SOC \ + DBG_QCFG_REQ_FLAGS_CRASHDUMP_SIZE_FOR_DEST_DEST_SOC_DDR +#define BNXT_DBG_FL_CR_DUMP_SIZE_HOST \ + DBG_QCFG_REQ_FLAGS_CRASHDUMP_SIZE_FOR_DEST_DEST_HOST_DDR +#define BNXT_DBG_CR_DUMP_MDM_CFG_DDR \ + DBG_CRASHDUMP_MEDIUM_CFG_REQ_TYPE_DDR +int bnxt_get_coredump (struct bnxt *bp,u16 dump_type,void *buf,u32 *dump_len); +int bnxt_hwrm_get_dump_len (struct bnxt *bp,u16 dump_type,u32 *dump_len);u32 bnxt_get_coredump_length (struct bnxt *bp,u16 dump_type);#endif
Signature · 1 line
2.30.1
Michael Chan
maintainerPATCH v4 2/9 bnxt_en: add support for retrieving crash dump using ethtool
In reply to Michael ChanFrom: Vikas Gupta <[email protected]>
Add support for retrieving crash dump using ethtool -w on the supported interface.
Reviewed-by Somnath Kotur <[email protected]>
Reviewed-by Pavan Chebbi <[email protected]>
Reviewed-by Andy Gospodarek <[email protected]>
Reviewed-by Simon Horman <[email protected]>
Signed-off-by Vikas Gupta <[email protected]>
Signed-off-by Michael Chan <[email protected]>
---
v3:
1. Use -ENOENT
2. Optimize bnxt_crash_dump_avail()
--- .../ethernet/broadcom/bnxt/bnxt_coredump.c | 80 +++++++++++++++++-- .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 13 ++- 2 files changed, 84 insertions(+), 9 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c +74 -6
@@ -372,14 +372,75 @@ static int __bnxt_get_coredump(struct bnxt *bp, void *buf, u32 *dump_len)return rc; } +static u32 bnxt_copy_crash_data (struct bnxt_ring_mem_info *rmem,void *buf, +u32 dump_len) +{ +u32 data_copied =0 ; +u32 data_len; +int i; + +for (i =0 ; i < rmem->nr_pages; i++) { + data_len = rmem->page_size; +if (data_copied + data_len > dump_len) + data_len = dump_len - data_copied; +memcpy (buf + data_copied, rmem->pg_arr[i], data_len); + data_copied += data_len; +if (data_copied >= dump_len) +break ; + } +return data_copied; +} + +static int bnxt_copy_crash_dump (struct bnxt *bp,void *buf,u32 dump_len) +{ +struct bnxt_ring_mem_info *rmem; +u32 offset =0 ; + +if (!bp->fw_crash_mem) +return -ENOENT ; + + rmem = &bp->fw_crash_mem->ring_mem; + +if (rmem->depth >1 ) { +int i; + +for (i =0 ; i < rmem->nr_pages; i++) { +struct bnxt_ctx_pg_info *pg_tbl; + + pg_tbl = bp->fw_crash_mem->ctx_pg_tbl[i]; + offset +=bnxt_copy_crash_data (&pg_tbl->ring_mem, + buf + offset, + dump_len - offset); +if (offset >= dump_len) +break ; + } + }else { +bnxt_copy_crash_data (rmem, buf, dump_len); + } + +return 0 ; +} + +static bool bnxt_crash_dump_avail (struct bnxt *bp) +{ +u32 sig =0 ; + +/* First 4 bytes(signature) of crash dump is always non-zero */ +bnxt_copy_crash_dump (bp, &sig,sizeof (sig)); +return !!sig; +} +int bnxt_get_coredump (struct bnxt *bp,u16 dump_type,void *buf,u32 *dump_len) {if (dump_type ==BNXT_DUMP_CRASH ) { +if (bp->fw_dbg_cap &DBG_QCAPS_RESP_FLAGS_CRASHDUMP_HOST_DDR ) +return bnxt_copy_crash_dump (bp, buf, *dump_len);#ifdef CONFIG_TEE_BNXT_FW -return tee_bnxt_copy_coredump (buf,0 , *dump_len); -#else -return -EOPNOTSUPP ; + else if (bp->fw_dbg_cap & DBG_QCAPS_RESP_FLAGS_CRASHDUMP_SOC_DDR) + return tee_bnxt_copy_coredump(buf, 0, *dump_len);#endif + else + return -EOPNOTSUPP; }else {return __bnxt_get_coredump (bp, buf, dump_len); }
@@ -442,10 +503,17 @@ u32 bnxt_get_coredump_length(struct bnxt *bp, u16 dump_type) {u32 len =0 ; +if (dump_type ==BNXT_DUMP_CRASH && + bp->fw_dbg_cap &DBG_QCAPS_RESP_FLAGS_CRASHDUMP_HOST_DDR && + bp->fw_crash_mem) { +if (!bnxt_crash_dump_avail (bp)) +return 0 ; + +return bp->fw_crash_len; + } +if (bnxt_hwrm_get_dump_len (bp, dump_type, &len)) { -if (dump_type ==BNXT_DUMP_CRASH ) - len =BNXT_CRASH_DUMP_LEN ; -else +if (dump_type !=BNXT_DUMP_CRASH )__bnxt_get_coredump (bp,NULL , &len); }return len;
drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c +10 -3
@@ -4989,9 +4989,16 @@ static int bnxt_set_dump(struct net_device *dev, struct ethtool_dump *dump)return -EINVAL ; } -if (!IS_ENABLED (CONFIG_TEE_BNXT_FW ) && dump->flag ==BNXT_DUMP_CRASH ) { -netdev_info (dev,"Cannot collect crash dump as TEE_BNXT_FW config option is not enabled.\n" ); -return -EOPNOTSUPP ; +if (dump->flag ==BNXT_DUMP_CRASH ) { +if (bp->fw_dbg_cap &DBG_QCAPS_RESP_FLAGS_CRASHDUMP_SOC_DDR && + (!IS_ENABLED (CONFIG_TEE_BNXT_FW ))) { +netdev_info (dev, +"Cannot collect crash dump as TEE_BNXT_FW config option is not enabled.\n" ); +return -EOPNOTSUPP ; + }else if (!(bp->fw_dbg_cap &DBG_QCAPS_RESP_FLAGS_CRASHDUMP_HOST_DDR )) { +netdev_info (dev,"Crash dump collection from host memory is not supported on this interface.\n" ); +return -EOPNOTSUPP ; + } } bp->dump_flag = dump->flag;
Signature · 1 line
2.30.1
Jakub Kicinski
maintainerRe: [PATCH net-next v4 2/9] bnxt_en: add support for retrieving crash dump using ethtool
drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c static int __bnxt_get_coredump(struct bnxt *bp, void *buf, u32 *dump_len) · line 385
Quoted 2 lines · On Wed, 28 Aug 2024 11:32:28 -0700 Michael Chan wrote:
drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c +2
-0+if (data_copied + data_len > dump_len) + data_len = dump_len - data_copied;
Let's see how clever the min() police is.
Michael Chan
maintainerPATCH v4 3/9 bnxt_en: Support QOS and TPID settings for the SRIOV VLAN
In reply to Michael ChanFrom: Sreekanth Reddy <[email protected]>
With recent changes in the .ndo_set_vf_*() guidelines, resubmitting this patch that was reverted eariler in 2023:
c27153682eac ("Revert "bnxt_en: Support QOS and TPID settings for the SRIOV VLAN")
Add these missing settings in the .ndo_set_vf_vlan() method. Older firmware does not support the TPID setting so check for proper support.
Remove the unused BNXT_VF_QOS flag.
Reviewed-by Simon Horman <[email protected]>
Signed-off-by Sreekanth Reddy <[email protected]>
Signed-off-by Michael Chan <[email protected]>
--- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 2 ++ drivers/net/ethernet/broadcom/bnxt/bnxt.h | 1 - .../net/ethernet/broadcom/bnxt/bnxt_sriov.c | 24 ++++++++++--------- 3 files changed, 15 insertions(+), 12 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +2 -0
@@ -9196,6 +9196,8 @@ static int __bnxt_hwrm_func_qcaps(struct bnxt *bp) bp->fw_cap |=BNXT_FW_CAP_HOT_RESET_IF ;if (BNXT_PF (bp) && (flags_ext &FUNC_QCAPS_RESP_FLAGS_EXT_FW_LIVEPATCH_SUPPORTED )) bp->fw_cap |=BNXT_FW_CAP_LIVEPATCH ; +if (BNXT_PF (bp) && (flags_ext &FUNC_QCAPS_RESP_FLAGS_EXT_DFLT_VLAN_TPID_PCP_SUPPORTED )) + bp->fw_cap |=BNXT_FW_CAP_DFLT_VLAN_TPID_PCP ;if (flags_ext &FUNC_QCAPS_RESP_FLAGS_EXT_BS_V2_SUPPORTED ) bp->fw_cap |=BNXT_FW_CAP_BACKING_STORE_V2 ;if (flags_ext &FUNC_QCAPS_RESP_FLAGS_EXT_TX_COAL_CMPL_CAP )
drivers/net/ethernet/broadcom/bnxt/bnxt.h +0 -1
@@ -1356,7 +1356,6 @@ struct bnxt_vf_info {u16 vlan;u16 func_qcfg_flags;u32 flags; -#define BNXT_VF_QOS 0x1#define BNXT_VF_SPOOFCHK 0x2#define BNXT_VF_LINK_FORCED 0x4#define BNXT_VF_LINK_UP 0x8
drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c +13 -11
@@ -15,6 +15,7 @@#include <linux/if_vlan.h> #include <linux/interrupt.h> #include <linux/etherdevice.h> +#include <net/dcbnl.h> #include "bnxt_hsi.h" #include "bnxt.h" #include "bnxt_hwrm.h"
@@ -196,11 +197,8 @@ int bnxt_get_vf_config(struct net_device *dev, int vf_id,memcpy (&ivi->mac, vf->vf_mac_addr,ETH_ALEN ); ivi->max_tx_rate = vf->max_tx_rate; ivi->min_tx_rate = vf->min_tx_rate; - ivi->vlan = vf->vlan; -if (vf->flags &BNXT_VF_QOS ) - ivi->qos = vf->vlan >>VLAN_PRIO_SHIFT ; -else - ivi->qos =0 ; + ivi->vlan = vf->vlan &VLAN_VID_MASK ; + ivi->qos = vf->vlan >>VLAN_PRIO_SHIFT ; ivi->spoofchk = !!(vf->flags &BNXT_VF_SPOOFCHK ); ivi->trusted =bnxt_is_trusted_vf (bp, vf);if (!(vf->flags &BNXT_VF_LINK_FORCED ))
@@ -256,21 +254,21 @@ int bnxt_set_vf_vlan(struct net_device *dev, int vf_id, u16 vlan_id, u8 qos,if (bp->hwrm_spec_code <0x10201 )return -ENOTSUPP ; -if (vlan_proto !=htons (ETH_P_8021Q )) +if (vlan_proto !=htons (ETH_P_8021Q ) && + (vlan_proto !=htons (ETH_P_8021AD ) || + !(bp->fw_cap &BNXT_FW_CAP_DFLT_VLAN_TPID_PCP )))return -EPROTONOSUPPORT ; rc =bnxt_vf_ndo_prep (bp, vf_id);if (rc)return rc; -/* TODO: needed to implement proper handling of user priority, -* currently fail the command if there is valid priority -*/ -if (vlan_id >4095 || qos) +if (vlan_id >=VLAN_N_VID || qos >=IEEE_8021Q_MAX_PRIORITIES || + (!vlan_id && qos))return -EINVAL ; vf = &bp->pf.vf[vf_id]; - vlan_tag = vlan_id; + vlan_tag = vlan_id | (u16 )qos <<VLAN_PRIO_SHIFT ;if (vlan_tag == vf->vlan)return 0 ;
@@ -279,6 +277,10 @@ int bnxt_set_vf_vlan(struct net_device *dev, int vf_id, u16 vlan_id, u8 qos, req->fid =cpu_to_le16 (vf->fw_fid); req->dflt_vlan =cpu_to_le16 (vlan_tag); req->enables =cpu_to_le32 (FUNC_CFG_REQ_ENABLES_DFLT_VLAN ); +if (bp->fw_cap &BNXT_FW_CAP_DFLT_VLAN_TPID_PCP ) { + req->enables |=cpu_to_le32 (FUNC_CFG_REQ_ENABLES_TPID ); + req->tpid = vlan_proto; + } rc =hwrm_req_send (bp, req);if (!rc) vf->vlan = vlan_tag;
Signature · 1 line
2.30.1
Michael Chan
maintainerPATCH v4 4/9 bnxt_en: Deprecate support for legacy INTX mode
In reply to Michael ChanFirmware has deprecated support for legacy INTX in 2022 (since v2.27) and INTX hasn't been tested for many years before that. INTX was only used as a fallback mechansim in case MSIX wasn't available. MSIX is always supported by all firmware. If MSIX capability in PCI config space is not found during probe, abort.
Reviewed-by Hongguang Gao <[email protected]>
Reviewed-by Kalesh AP <[email protected]>
Reviewed-by Somnath Kotur <[email protected]>
Reviewed-by Simon Horman <[email protected]>
Signed-off-by Michael Chan <[email protected]>
---
v2: Fix memory leak, update changelog
--- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 43 +++++------------------ drivers/net/ethernet/broadcom/bnxt/bnxt.h | 1 - 2 files changed, 8 insertions(+), 36 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +8 -35
@@ -10770,7 +10770,7 @@ static int bnxt_get_num_msix(struct bnxt *bp)return bnxt_nq_rings_in_use (bp); } -static int bnxt_init_msix (struct bnxt *bp) +static int bnxt_init_int_mode (struct bnxt *bp) {int i, total_vecs, max, rc =0 , min =1 , ulp_msix, tx_cp;struct msix_entry *msix_ent;
@@ -10828,7 +10828,7 @@ static int bnxt_init_msix(struct bnxt *bp)return 0 ; msix_setup_exit: -netdev_err (bp->dev,"bnxt_init_msix err: %x\n" , rc); +netdev_err (bp->dev,"bnxt_init_int_mode err: %x\n" , rc);kfree (bp->irq_tbl); bp->irq_tbl =NULL ;pci_disable_msix (bp->pdev);
@@ -10836,35 +10836,6 @@ static int bnxt_init_msix(struct bnxt *bp)return rc; } -static int bnxt_init_inta (struct bnxt *bp) -{ - bp->irq_tbl =kzalloc (sizeof (struct bnxt_irq ),GFP_KERNEL ); -if (!bp->irq_tbl) -return -ENOMEM ; - - bp->total_irqs =1 ; - bp->rx_nr_rings =1 ; - bp->tx_nr_rings =1 ; - bp->cp_nr_rings =1 ; - bp->flags |=BNXT_FLAG_SHARED_RINGS ; - bp->irq_tbl[0 ].vector = bp->pdev->irq; -return 0 ; -} - -static int bnxt_init_int_mode (struct bnxt *bp) -{ -int rc = -ENODEV ; - -if (bp->flags &BNXT_FLAG_MSIX_CAP ) - rc =bnxt_init_msix (bp); - -if (!(bp->flags &BNXT_FLAG_USING_MSIX ) &&BNXT_PF (bp)) { -/* fallback to INTA */ - rc =bnxt_init_inta (bp); - } -return rc; -} -static void bnxt_clear_int_mode (struct bnxt *bp) {if (bp->flags &BNXT_FLAG_USING_MSIX )
@@ -12910,7 +12881,7 @@ bool bnxt_rfs_capable(struct bnxt *bp, bool new_rss_ctx) !BNXT_SUPPORTS_NTUPLE_VNIC(bp))return bnxt_rfs_supported (bp); -if (!(bp->flags &BNXT_FLAG_MSIX_CAP ) || !bnxt_can_reserve_rings (bp) || !bp->rx_nr_rings) +if (!bnxt_can_reserve_rings (bp) || !bp->rx_nr_rings)return false; hwr.grp = bp->rx_nr_rings;
@@ -15772,6 +15743,11 @@ static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)if (pci_is_bridge (pdev))return -ENODEV ; +if (!pdev->msix_cap) { +dev_err (&pdev->dev,"MSIX capability not found, aborting\n" ); +return -ENODEV ; + } +/* Clear any pending DMA transactions from crash kernel * while loading driver in capture kernel. */
@@ -15798,9 +15774,6 @@ static int bnxt_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)if (BNXT_PF (bp))SET_NETDEV_DEVLINK_PORT (dev, &bp->dl_port); -if (pdev->msix_cap) - bp->flags |=BNXT_FLAG_MSIX_CAP ; - rc =bnxt_init_board (pdev, dev);if (rc <0 ) goto init_err_free;
drivers/net/ethernet/broadcom/bnxt/bnxt.h +0 -1
@@ -2200,7 +2200,6 @@ struct bnxt {#define BNXT_FLAG_AGG_RINGS (BNXT_FLAG_JUMBO | BNXT_FLAG_GRO | \ BNXT_FLAG_LRO)#define BNXT_FLAG_USING_MSIX 0x40 -#define BNXT_FLAG_MSIX_CAP 0x80#define BNXT_FLAG_RFS 0x100#define BNXT_FLAG_SHARED_RINGS 0x200#define BNXT_FLAG_PORT_STATS 0x400
Signature · 1 line
2.30.1
Michael Chan
maintainerPATCH v4 5/9 bnxt_en: Remove BNXT_FLAG_USING_MSIX flag
In reply to Michael ChanNow that we only support MSIX, the BNXT_FLAG_USING_MSIX is always true. Remove it and any if conditions checking for it. Remove the INTX handler and associated logic.
Reviewed-by Kalesh AP <[email protected]>
Reviewed-by Somnath Kotur <[email protected]>
Reviewed-by Hongguang Gao <[email protected]>
Reviewed-by Simon Horman <[email protected]>
Signed-off-by Michael Chan <[email protected]>
--- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 99 ++++--------------- drivers/net/ethernet/broadcom/bnxt/bnxt.h | 1 - .../net/ethernet/broadcom/bnxt/bnxt_sriov.c | 5 - 3 files changed, 17 insertions(+), 88 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +17 -82
@@ -2854,34 +2854,6 @@ static inline int bnxt_has_work(struct bnxt *bp, struct bnxt_cp_ring_info *cpr)return TX_CMP_VALID (txcmp, raw_cons); } -static irqreturn_t bnxt_inta (int irq,void *dev_instance) -{ -struct bnxt_napi *bnapi = dev_instance; -struct bnxt *bp = bnapi->bp; -struct bnxt_cp_ring_info *cpr = &bnapi->cp_ring; -u32 cons =RING_CMP (cpr->cp_raw_cons); -u32 int_status; - -prefetch (&cpr->cp_desc_ring[CP_RING (cons)][CP_IDX (cons)]); - -if (!bnxt_has_work (bp, cpr)) { - int_status =readl (bp->bar0 +BNXT_CAG_REG_LEGACY_INT_STATUS ); -/* return if erroneous interrupt */ -if (!(int_status & (0x10000 << cpr->cp_ring_struct.fw_ring_id))) -return IRQ_NONE ; - } - -/* disable ring IRQ */ -BNXT_CP_DB_IRQ_DIS (cpr->cp_db.doorbell); - -/* Return here if interrupt is shared and is disabled. */ -if (unlikely (atomic_read (&bp->intr_sem) !=0 )) -return IRQ_HANDLED ; - -napi_schedule (&bnapi->napi); -return IRQ_HANDLED ; -} -static int __bnxt_poll_work (struct bnxt *bp,struct bnxt_cp_ring_info *cpr,int budget) {
@@ -6876,15 +6848,14 @@ static int hwrm_ring_alloc_send_msg(struct bnxt *bp, req->cq_handle =cpu_to_le64 (ring->handle); req->enables |=cpu_to_le32 (RING_ALLOC_REQ_ENABLES_NQ_RING_ID_VALID ); - }else if (bp->flags &BNXT_FLAG_USING_MSIX ) { + }else { req->int_mode =RING_ALLOC_REQ_INT_MODE_MSIX ; }break ;case HWRM_RING_ALLOC_NQ : req->ring_type =RING_ALLOC_REQ_RING_TYPE_NQ ; req->length =cpu_to_le32 (bp->cp_ring_mask +1 ); -if (bp->flags &BNXT_FLAG_USING_MSIX ) - req->int_mode =RING_ALLOC_REQ_INT_MODE_MSIX ; + req->int_mode =RING_ALLOC_REQ_INT_MODE_MSIX ;break ;default : netdev_err(bp->dev, "hwrm alloc invalid ring type %d\n",
@@ -10653,20 +10624,6 @@ static void bnxt_setup_msix(struct bnxt *bp) } } -static void bnxt_setup_inta (struct bnxt *bp) -{ -const int len =sizeof (bp->irq_tbl[0 ].name); - -if (bp->num_tc) { -netdev_reset_tc (bp->dev); - bp->num_tc =0 ; - } - -snprintf (bp->irq_tbl[0 ].name, len,"%s-%s-%d" , bp->dev->name,"TxRx" , -0 ); - bp->irq_tbl[0 ].handler = bnxt_inta; -} -static int bnxt_init_int_mode (struct bnxt *bp);static int bnxt_setup_int_mode (struct bnxt *bp)
@@ -10679,10 +10636,7 @@ static int bnxt_setup_int_mode(struct bnxt *bp)return rc ?: -ENODEV ; } -if (bp->flags &BNXT_FLAG_USING_MSIX ) -bnxt_setup_msix (bp); -else -bnxt_setup_inta (bp); +bnxt_setup_msix (bp); rc =bnxt_set_real_num_queues (bp);return rc;
@@ -10823,7 +10777,6 @@ static int bnxt_init_int_mode(struct bnxt *bp) rc = -ENOMEM ; goto msix_setup_exit; } - bp->flags |=BNXT_FLAG_USING_MSIX ;kfree (msix_ent);return 0 ;
@@ -10838,12 +10791,10 @@ static int bnxt_init_int_mode(struct bnxt *bp)static void bnxt_clear_int_mode (struct bnxt *bp) { -if (bp->flags &BNXT_FLAG_USING_MSIX ) -pci_disable_msix (bp->pdev); +pci_disable_msix (bp->pdev);kfree (bp->irq_tbl); bp->irq_tbl =NULL ; - bp->flags &= ~BNXT_FLAG_USING_MSIX ; }int bnxt_reserve_rings (struct bnxt *bp,bool irq_re_init)
@@ -10941,9 +10892,6 @@ static int bnxt_request_irq(struct bnxt *bp)#ifdef CONFIG_RFS_ACCEL rmap = bp->dev->rx_cpu_rmap;#endif -if (!(bp->flags &BNXT_FLAG_USING_MSIX )) - flags =IRQF_SHARED ; -for (i =0 , j =0 ; i < bp->cp_nr_rings; i++) {int map_idx =bnxt_cp_num_to_irq_num (bp, i);struct bnxt_irq *irq = &bp->irq_tbl[map_idx];
@@ -11008,29 +10956,22 @@ static void bnxt_del_napi(struct bnxt *bp)static void bnxt_init_napi (struct bnxt *bp) { -int i; +int (*poll_fn)(struct napi_struct *,int ) = bnxt_poll;unsigned int cp_nr_rings = bp->cp_nr_rings;struct bnxt_napi *bnapi; +int i; -if (bp->flags &BNXT_FLAG_USING_MSIX ) { -int (*poll_fn)(struct napi_struct *,int ) = bnxt_poll; - -if (bp->flags &BNXT_FLAG_CHIP_P5_PLUS ) - poll_fn = bnxt_poll_p5; -else if (BNXT_CHIP_TYPE_NITRO_A0 (bp)) - cp_nr_rings--; -for (i =0 ; i < cp_nr_rings; i++) { - bnapi = bp->bnapi[i]; -netif_napi_add (bp->dev, &bnapi->napi, poll_fn); - } -if (BNXT_CHIP_TYPE_NITRO_A0 (bp)) { - bnapi = bp->bnapi[cp_nr_rings]; -netif_napi_add (bp->dev, &bnapi->napi, - bnxt_poll_nitroa0); - } - }else { - bnapi = bp->bnapi[0 ]; -netif_napi_add (bp->dev, &bnapi->napi, bnxt_poll); +if (bp->flags &BNXT_FLAG_CHIP_P5_PLUS ) + poll_fn = bnxt_poll_p5; +else if (BNXT_CHIP_TYPE_NITRO_A0 (bp)) + cp_nr_rings--; +for (i =0 ; i < cp_nr_rings; i++) { + bnapi = bp->bnapi[i]; +netif_napi_add (bp->dev, &bnapi->napi, poll_fn); + } +if (BNXT_CHIP_TYPE_NITRO_A0 (bp)) { + bnapi = bp->bnapi[cp_nr_rings]; +netif_napi_add (bp->dev, &bnapi->napi, bnxt_poll_nitroa0); } }
@@ -12149,12 +12090,6 @@ static int __bnxt_open_nic(struct bnxt *bp, bool irq_re_init, bool link_re_init) rc =bnxt_reserve_rings (bp, irq_re_init);if (rc)return rc; -if ((bp->flags &BNXT_FLAG_RFS ) && - !(bp->flags &BNXT_FLAG_USING_MSIX )) { -/* disable RFS if falling back to INTA */ - bp->dev->hw_features &= ~NETIF_F_NTUPLE ; - bp->flags &= ~BNXT_FLAG_RFS ; - } rc =bnxt_alloc_mem (bp, irq_re_init);if (rc) {
drivers/net/ethernet/broadcom/bnxt/bnxt.h +0 -1
@@ -2199,7 +2199,6 @@ struct bnxt {#define BNXT_FLAG_STRIP_VLAN 0x20#define BNXT_FLAG_AGG_RINGS (BNXT_FLAG_JUMBO | BNXT_FLAG_GRO | \ BNXT_FLAG_LRO) -#define BNXT_FLAG_USING_MSIX 0x40#define BNXT_FLAG_RFS 0x100#define BNXT_FLAG_SHARED_RINGS 0x200#define BNXT_FLAG_PORT_STATS 0x400
drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c +0 -5
@@ -902,11 +902,6 @@ int bnxt_sriov_configure(struct pci_dev *pdev, int num_vfs)struct net_device *dev =pci_get_drvdata (pdev);struct bnxt *bp =netdev_priv (dev); -if (!(bp->flags &BNXT_FLAG_USING_MSIX )) { -netdev_warn (dev,"Not allow SRIOV if the irq mode is not MSIX\n" ); -return 0 ; - } -rtnl_lock ();if (!netif_running (dev)) {netdev_warn (dev,"Reject SRIOV config request since if is down!\n" );
Signature · 1 line
2.30.1
Michael Chan
maintainerPATCH v4 6/9 bnxt_en: Remove register mapping to support INTX
In reply to Michael ChanIn legacy INTX mode, a register is mapped so that the INTX handler can read it to determine if the NIC is the source of the interrupt. This and all the related macros are no longer needed now that INTX is no longer supported.
Reviewed-by Somnath Kotur <[email protected]>
Reviewed-by Hongguang Gao <[email protected]>
Reviewed-by Simon Horman <[email protected]>
Signed-off-by Michael Chan <[email protected]>
---
v2: Update changelog
--- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 19 ------------------- drivers/net/ethernet/broadcom/bnxt/bnxt.h | 2 -- 2 files changed, 21 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +0 -19
@@ -302,10 +302,6 @@ static bool bnxt_vf_pciid(enum board_idx idx)#define DB_CP_REARM_FLAGS (DB_KEY_CP | DB_IDX_VALID)#define DB_CP_FLAGS (DB_KEY_CP | DB_IDX_VALID | DB_IRQ_DIS) -#define DB_CP_IRQ_DIS_FLAGS (DB_KEY_CP | DB_IRQ_DIS) - -#define BNXT_CP_DB_IRQ_DIS (db) \ - writel(DB_CP_IRQ_DIS_FLAGS, db)#define BNXT_DB_CQ (db, idx) \ writel(DB_CP_FLAGS | DB_RING_IDX(db, idx), (db)->doorbell)
@@ -11959,20 +11955,6 @@ static int bnxt_update_phy_setting(struct bnxt *bp)return rc; } -/* Common routine to pre-map certain register block to different GRC window. -* A PF has 16 4K windows and a VF has 4 4K windows. However, only 15 windows -* in PF and 3 windows in VF that can be customized to map in different -* register blocks. -*/ -static void bnxt_preset_reg_win (struct bnxt *bp) -{ -if (BNXT_PF (bp)) { -/* CAG registers map to GRC window #4 */ -writel (BNXT_CAG_REG_BASE , - bp->bar0 +BNXT_GRCPF_REG_WINDOW_BASE_OUT +12 ); - } -} -static int bnxt_init_dflt_ring_mode (struct bnxt *bp);static int bnxt_reinit_after_abort (struct bnxt *bp)
@@ -12077,7 +12059,6 @@ static int __bnxt_open_nic(struct bnxt *bp, bool irq_re_init, bool link_re_init) {int rc =0 ; -bnxt_preset_reg_win (bp);netif_carrier_off (bp->dev);if (irq_re_init) {/* Reserve rings now if none were reserved at driver probe. */
drivers/net/ethernet/broadcom/bnxt/bnxt.h +0 -2
@@ -1755,8 +1755,6 @@ struct bnxt_test_info {#define BNXT_GRCPF_REG_CHIMP_COMM 0x0#define BNXT_GRCPF_REG_CHIMP_COMM_TRIGGER 0x100#define BNXT_GRCPF_REG_WINDOW_BASE_OUT 0x400 -#define BNXT_CAG_REG_LEGACY_INT_STATUS 0x4014 -#define BNXT_CAG_REG_BASE 0x300000#define BNXT_GRC_REG_STATUS_P5 0x520
Signature · 1 line
2.30.1
Michael Chan
maintainerPATCH v4 7/9 bnxt_en: Replace deprecated PCI MSIX APIs
In reply to Michael ChanUse the new pci_alloc_irq_vectors() and pci_free_irq_vectors() to replace the deprecated pci_enable_msix_range() and pci_disable_msix().
Reviewed-by Somnath Kotur <[email protected]>
Reviewed-by Pavan Chebbi <[email protected]>
Reviewed-by Simon Horman <[email protected]>
Reviewed-by Przemek Kitszel <[email protected]>
Signed-off-by Michael Chan <[email protected]>
--- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 21 +++++---------------- 1 file changed, 5 insertions(+), 16 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +5 -16
@@ -10723,7 +10723,6 @@ static int bnxt_get_num_msix(struct bnxt *bp)static int bnxt_init_int_mode (struct bnxt *bp) {int i, total_vecs, max, rc =0 , min =1 , ulp_msix, tx_cp; -struct msix_entry *msix_ent; total_vecs =bnxt_get_num_msix (bp); max =bnxt_get_max_func_irqs (bp);
@@ -10733,19 +10732,11 @@ static int bnxt_init_int_mode(struct bnxt *bp)if (!total_vecs)return 0 ; - msix_ent =kcalloc (total_vecs,sizeof (struct msix_entry ),GFP_KERNEL ); -if (!msix_ent) -return -ENOMEM ; - -for (i =0 ; i < total_vecs; i++) { - msix_ent[i].entry = i; - msix_ent[i].vector =0 ; - } -if (!(bp->flags &BNXT_FLAG_SHARED_RINGS )) min =2 ; - total_vecs =pci_enable_msix_range (bp->pdev, msix_ent, min, total_vecs); + total_vecs =pci_alloc_irq_vectors (bp->pdev, min, total_vecs, +PCI_IRQ_MSIX ); ulp_msix =bnxt_get_ulp_msix_num (bp);if (total_vecs <0 || total_vecs < ulp_msix) { rc = -ENODEV ;
@@ -10755,7 +10746,7 @@ static int bnxt_init_int_mode(struct bnxt *bp) bp->irq_tbl =kcalloc (total_vecs,sizeof (struct bnxt_irq ),GFP_KERNEL );if (bp->irq_tbl) {for (i =0 ; i < total_vecs; i++) - bp->irq_tbl[i].vector = msix_ent[i].vector; + bp->irq_tbl[i].vector =pci_irq_vector (bp->pdev, i); bp->total_irqs = total_vecs;/* Trim rings based upon num of vectors allocated */
@@ -10773,21 +10764,19 @@ static int bnxt_init_int_mode(struct bnxt *bp) rc = -ENOMEM ; goto msix_setup_exit; } -kfree (msix_ent);return 0 ; msix_setup_exit:netdev_err (bp->dev,"bnxt_init_int_mode err: %x\n" , rc);kfree (bp->irq_tbl); bp->irq_tbl =NULL ; -pci_disable_msix (bp->pdev); -kfree (msix_ent); +pci_free_irq_vectors (bp->pdev);return rc; }static void bnxt_clear_int_mode (struct bnxt *bp) { -pci_disable_msix (bp->pdev); +pci_free_irq_vectors (bp->pdev);kfree (bp->irq_tbl); bp->irq_tbl =NULL ;
Signature · 1 line
2.30.1
Michael Chan
maintainerPATCH v4 8/9 bnxt_en: Allocate the max bp->irq_tbl size for dynamic msix allocation
In reply to Michael ChanIf dynamic MSIX allocation is supported, additional MSIX can be allocated at run-time without reinitializing the existing MSIX entries. The first step to support this dynamic scheme is to allocate a large enough bp->irq_tbl if dynamic allocation is supported.
Reviewed-by Hongguang Gao <[email protected]>
Reviewed-by Somnath Kotur <[email protected]>
Reviewed-by Simon Horman <[email protected]>
Reviewed-by Przemek Kitszel <[email protected]>
Signed-off-by Michael Chan <[email protected]>
---
v3: Use sizeof(*bp->irq_tbl)
v2: Fix typo in changelog
--- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +5 -2
@@ -10722,7 +10722,7 @@ static int bnxt_get_num_msix(struct bnxt *bp)static int bnxt_init_int_mode (struct bnxt *bp) { -int i, total_vecs, max, rc =0 , min =1 , ulp_msix, tx_cp; +int i, total_vecs, max, rc =0 , min =1 , ulp_msix, tx_cp, tbl_size; total_vecs =bnxt_get_num_msix (bp); max =bnxt_get_max_func_irqs (bp);
@@ -10743,7 +10743,10 @@ static int bnxt_init_int_mode(struct bnxt *bp) goto msix_setup_exit; } - bp->irq_tbl =kcalloc (total_vecs,sizeof (struct bnxt_irq ),GFP_KERNEL ); + tbl_size = total_vecs; +if (pci_msix_can_alloc_dyn (bp->pdev)) + tbl_size = max; + bp->irq_tbl =kcalloc (tbl_size,sizeof (*bp->irq_tbl),GFP_KERNEL );if (bp->irq_tbl) {for (i =0 ; i < total_vecs; i++) bp->irq_tbl[i].vector =pci_irq_vector (bp->pdev, i);
Signature · 1 line
2.30.1
Michael Chan
maintainerPATCH v4 9/9 bnxt_en: Support dynamic MSIX
In reply to Michael ChanA range of MSIX vectors are allocated at initialization for the number needed for RocE and L2. During run-time, if the user increases or decreases the number of L2 rings, all the MSIX vectors have to be freed and a new range has to be allocated. This is not optimal and causes disruptions to RoCE traffic every time there is a change in L2 MSIX.
If the system supports dynamic MSIX allocations, use dynamic allocation to add new L2 MSIX vectors or free unneeded L2 MSIX vectors. RoCE traffic is not affected using this scheme.
Reviewed-by Hongguang Gao <[email protected]>
Reviewed-by Somnath Kotur <[email protected]>
Reviewed-by Simon Horman <[email protected]>
Reviewed-by Przemek Kitszel <[email protected]>
Signed-off-by Michael Chan <[email protected]>
---
Cc: Michal Swiatkowski <[email protected]>
v4: Simplify adding and deleting MSIX
v2: Fix typo in changelog
--- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 37 +++++++++++++++++++++-- 1 file changed, 34 insertions(+), 3 deletions(-)
drivers/net/ethernet/broadcom/bnxt/bnxt.c +34 -3
@@ -10622,6 +10622,30 @@ static void bnxt_setup_msix(struct bnxt *bp)static int bnxt_init_int_mode (struct bnxt *bp); +static int bnxt_change_msix (struct bnxt *bp,int total) +{ +struct msi_map map; +int i; + +/* add MSIX to the end if needed */ +for (i = bp->total_irqs; i < total; i++) { + map =pci_msix_alloc_irq_at (bp->pdev, i,NULL ); +if (map.index <0 ) +return bp->total_irqs; + bp->irq_tbl[i].vector = map.virq; + bp->total_irqs++; + } + +/* trim MSIX from the end if needed */ +for (i = bp->total_irqs; i > total; i--) { + map.index = i -1 ; + map.virq = bp->irq_tbl[i -1 ].vector; +pci_msix_free_irq (bp->pdev, map); + bp->total_irqs--; + } +return bp->total_irqs; +} +static int bnxt_setup_int_mode (struct bnxt *bp) {int rc;
@@ -10788,6 +10812,7 @@ static void bnxt_clear_int_mode(struct bnxt *bp)int bnxt_reserve_rings (struct bnxt *bp,bool irq_re_init) {bool irq_cleared = false; +bool irq_change = false;int tcs = bp->num_tc;int irqs_required;int rc;
@@ -10806,15 +10831,21 @@ int bnxt_reserve_rings(struct bnxt *bp, bool irq_re_init) }if (irq_re_init &&BNXT_NEW_RM (bp) && irqs_required != bp->total_irqs) { -bnxt_ulp_irq_stop (bp); -bnxt_clear_int_mode (bp); - irq_cleared = true; + irq_change = true; +if (!pci_msix_can_alloc_dyn (bp->pdev)) { +bnxt_ulp_irq_stop (bp); +bnxt_clear_int_mode (bp); + irq_cleared = true; + } } rc =__bnxt_reserve_rings (bp);if (irq_cleared) {if (!rc) rc =bnxt_init_int_mode (bp);bnxt_ulp_irq_restart (bp, rc); + }else if (irq_change && !rc) { +if (bnxt_change_msix (bp, irqs_required) != irqs_required) + rc = -ENOSPC ; }if (rc) {netdev_err (bp->dev,"ring reservation/IRQ init failure rc: %d\n" , rc);
Signature · 1 line
2.30.1
Re: [PATCH net-next v4 9/9] bnxt_en: Support dynamic MSIX
Quoted 94 lines · Michael Chan
A range of MSIX vectors are allocated at initialization for the number needed for RocE and L2. During run-time, if the user increases or decreases the number of L2 rings, all the MSIX vectors have to be freed and a new range has to be allocated. This is not optimal and causes disruptions to RoCE traffic every time there is a change in L2 MSIX.
If the system supports dynamic MSIX allocations, use dynamic allocation to add new L2 MSIX vectors or free unneeded L2 MSIX vectors. RoCE traffic is not affected using this scheme.
Reviewed-by Hongguang Gao <[email protected]>Reviewed-by Somnath Kotur <[email protected]>Reviewed-by Simon Horman <[email protected]>Reviewed-by Przemek Kitszel <[email protected]>Signed-off-by Michael Chan <[email protected]>---
Cc: Michal Swiatkowski <[email protected]>
v4: Simplify adding and deleting MSIX
v2: Fix typo in changelog--- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 37 +++++++++++++++++++++-- 1 file changed, 34 insertions(+), 3 deletions(-)drivers/net/ethernet/broadcom/bnxt/bnxt.c +34
-3@@ -10622,6 +10622,30 @@ static void bnxt_setup_msix(struct bnxt *bp)static int bnxt_init_int_mode (struct bnxt *bp); +static int bnxt_change_msix (struct bnxt *bp,int total) +{ +struct msi_map map; +int i; + +/* add MSIX to the end if needed */ +for (i = bp->total_irqs; i < total; i++) { + map =pci_msix_alloc_irq_at (bp->pdev, i,NULL ); +if (map.index <0 ) +return bp->total_irqs; + bp->irq_tbl[i].vector = map.virq; + bp->total_irqs++; + } + +/* trim MSIX from the end if needed */ +for (i = bp->total_irqs; i > total; i--) { + map.index = i -1 ; + map.virq = bp->irq_tbl[i -1 ].vector; +pci_msix_free_irq (bp->pdev, map); + bp->total_irqs--; + } +return bp->total_irqs; +} +static int bnxt_setup_int_mode (struct bnxt *bp) {int rc;@@ -10788,6 +10812,7 @@ static void bnxt_clear_int_mode(struct bnxt *bp)int bnxt_reserve_rings (struct bnxt *bp,bool irq_re_init) {bool irq_cleared = false; +bool irq_change = false;int tcs = bp->num_tc;int irqs_required;int rc;@@ -10806,15 +10831,21 @@ int bnxt_reserve_rings(struct bnxt *bp, bool irq_re_init) }if (irq_re_init &&BNXT_NEW_RM (bp) && irqs_required != bp->total_irqs) { -bnxt_ulp_irq_stop (bp); -bnxt_clear_int_mode (bp); - irq_cleared = true; + irq_change = true; +if (!pci_msix_can_alloc_dyn (bp->pdev)) { +bnxt_ulp_irq_stop (bp); +bnxt_clear_int_mode (bp); + irq_cleared = true; + } } rc =__bnxt_reserve_rings (bp);if (irq_cleared) {if (!rc) rc =bnxt_init_int_mode (bp);bnxt_ulp_irq_restart (bp, rc); + }else if (irq_change && !rc) { +if (bnxt_change_msix (bp, irqs_required) != irqs_required) + rc = -ENOSPC ; }if (rc) {netdev_err (bp->dev,"ring reservation/IRQ init failure rc: %d\n" , rc);Signature · 1 line
2.30.1
Thanks,
Reviewed-by Michal Swiatkowski <[email protected]>
Quoted 1 line
Re: [PATCH net-next v4 9/9] bnxt_en: Support dynamic MSIX
In reply to Michael ChanQuoted 29 lines · Michael Chan
A range of MSIX vectors are allocated at initialization for the number needed for RocE and L2. During run-time, if the user increases or decreases the number of L2 rings, all the MSIX vectors have to be freed and a new range has to be allocated. This is not optimal and causes disruptions to RoCE traffic every time there is a change in L2 MSIX.
If the system supports dynamic MSIX allocations, use dynamic allocation to add new L2 MSIX vectors or free unneeded L2 MSIX vectors. RoCE traffic is not affected using this scheme.
Reviewed-by Hongguang Gao <[email protected]>Reviewed-by Somnath Kotur <[email protected]>Reviewed-by Simon Horman <[email protected]>Reviewed-by Przemek Kitszel <[email protected]>Signed-off-by Michael Chan <[email protected]>---
Cc: Michal Swiatkowski <[email protected]>
v4: Simplify adding and deleting MSIX
v2: Fix typo in changelog--- drivers/net/ethernet/broadcom/bnxt/bnxt.c | 37 +++++++++++++++++++++-- 1 file changed, 34 insertions(+), 3 deletions(-)drivers/net/ethernet/broadcom/bnxt/bnxt.c +0
-0
drivers/net/ethernet/broadcom/bnxt/bnxt.c static void bnxt_setup_msix(struct bnxt *bp) · line 10622
Quoted 1 line
drivers/net/ethernet/broadcom/bnxt/bnxt.c +0
-0@@ -10622,6 +10622,30 @@ static void bnxt_setup_msix(struct bnxt *bp)
Thanks,
Reviewed-by Michal Swiatkowski <[email protected]>
Jakub Kicinski
maintainerQuoted 10 lines · On Wed, 28 Aug 2024 11:32:26 -0700 Michael Chan wrote:
This series starts with 2 patches to support firmware crash dump. The driver allocates the required DMA memory ahead of time for firmware to store the crash dump if and when it crashes. Patch 3 adds priority and TPID for the .ndo_set_vf_vlan() callback. Note that this was rejected and reverted last year and it is being re-submitted after recent changes in the guidelines. The remaining patches are MSIX related. Legacy interrupt is no longer supported by firmware so we remove the support in the driver. We then convert to use the newer kernel APIs to allocate and enable MSIX vectors. The last patch adds support for dynamic MSIX.
Some disturbance in the force on kernel.org, their bot isn't responding. Applied a few hours ago, thanks!