[PATCH net-next v2 0/9] bnxt_en: Update for net-next

[PATCH net-next v2 0/9] bnxt_en: Update for net-next

[Michael Chan]

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.

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/20240713234339.70293-1-michael.chan@broadcom.com/

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     | 331 ++++++++++--------
 drivers/net/ethernet/broadcom/bnxt/bnxt.h     |   8 +-
 .../ethernet/broadcom/bnxt/bnxt_coredump.c    | 101 +++++-
 .../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, 303 insertions(+), 187 deletions(-)

-- 
2.30.1

[PATCH net-next v2 1/9] bnxt_en: add support for storing crash dump into host memory

[Michael Chan]

From: Vikas Gupta <vikas.gupta@broadcom.com>

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 <somnath.kotur@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Vikas Gupta <vikas.gupta@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 93 +++++++++++++++++++
 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, 117 insertions(+), 5 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 75e2cfed5769..017eefd78ba4 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -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 +8947,82 @@ static int bnxt_alloc_ctx_mem(struct bnxt *bp)
 	return 0;
 }
 
+#define BNXT_SET_CRASHDUMP_PAGE_ATTR(attr)				\
+do {									\
+	if (BNXT_PAGE_SIZE == 0x2000)					\
+		attr = DBG_CRASHDUMP_MEDIUM_CFG_REQ_PG_SIZE_PG_8K;	\
+	else if (BNXT_PAGE_SIZE == 0x10000)				\
+		attr = DBG_CRASHDUMP_MEDIUM_CFG_REQ_PG_SIZE_PG_64K;	\
+	else								\
+		attr = DBG_CRASHDUMP_MEDIUM_CFG_REQ_PG_SIZE_PG_4K;	\
+} while (0)
+
+static int bnxt_hwrm_crash_dump_mem_cfg(struct bnxt *bp)
+{
+	struct hwrm_dbg_crashdump_medium_cfg_input *req;
+	u16 page_attr = 0;
+	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;
+
+	BNXT_SET_CRASHDUMP_PAGE_ATTR(page_attr);
+	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_len = 0;
+		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);
+
+	if (bp->fw_crash_mem && mem_size == bp->fw_crash_len)
+		return 0;
+
+	bnxt_free_crash_dump_mem(bp);
+
+	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;
+	}
+
+	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 +14063,19 @@ 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 +15384,7 @@ 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 +16024,7 @@ 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;
 
@@ -15986,6 +16078,7 @@ static int bnxt_suspend(struct device *device)
 	bnxt_hwrm_func_drv_unrgtr(bp);
 	pci_disable_device(bp->pdev);
 	bnxt_free_ctx_mem(bp);
+	bnxt_free_crash_dump_mem(bp);
 	rtnl_unlock();
 	return rc;
 }
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index 62e637c5be31..2fa6572b6b1d 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -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
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c
index c06789882036..ebbad9ccab6a 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c
@@ -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.
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.h b/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.h
index b1a1b2fffb19..a76d5c281413 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.h
@@ -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
-- 
2.30.1

Re: [PATCH net-next v2 1/9] bnxt_en: add support for storing crash dump into host memory

[Przemek Kitszel]

On 8/16/24 23:28, Michael Chan wrote:
> From: Vikas Gupta <vikas.gupta@broadcom.com>
> 
> 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 <somnath.kotur@broadcom.com>
> Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
> Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
> Reviewed-by: Simon Horman <horms@kernel.org>
> Signed-off-by: Vikas Gupta <vikas.gupta@broadcom.com>
> Signed-off-by: Michael Chan <michael.chan@broadcom.com>
> ---
>   drivers/net/ethernet/broadcom/bnxt/bnxt.c     | 93 +++++++++++++++++++
>   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, 117 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
> index 75e2cfed5769..017eefd78ba4 100644
> --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
> +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
> @@ -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 +8947,82 @@ static int bnxt_alloc_ctx_mem(struct bnxt *bp)
>   	return 0;
>   }
>   
> +#define BNXT_SET_CRASHDUMP_PAGE_ATTR(attr)				\
> +do {									\
> +	if (BNXT_PAGE_SIZE == 0x2000)					\
> +		attr = DBG_CRASHDUMP_MEDIUM_CFG_REQ_PG_SIZE_PG_8K;	\
> +	else if (BNXT_PAGE_SIZE == 0x10000)				\
> +		attr = DBG_CRASHDUMP_MEDIUM_CFG_REQ_PG_SIZE_PG_64K;	\
> +	else								\
> +		attr = DBG_CRASHDUMP_MEDIUM_CFG_REQ_PG_SIZE_PG_4K;	\
> +} while (0)
> +
> +static int bnxt_hwrm_crash_dump_mem_cfg(struct bnxt *bp)
> +{
> +	struct hwrm_dbg_crashdump_medium_cfg_input *req;
> +	u16 page_attr = 0;

redundant assignement

> +	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;
> +
> +	BNXT_SET_CRASHDUMP_PAGE_ATTR(page_attr);

I would avoid introducing a macro for just one use here

> +	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_len = 0;
> +		bp->fw_crash_mem = NULL;

it should be sufficient to have just one variable cleared for the
purpose of "no crash mem allocated yet" detection

> +	}
> +}
> +
> +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);
> +
> +	if (bp->fw_crash_mem && mem_size == bp->fw_crash_len)
> +		return 0;
> +
> +	bnxt_free_crash_dump_mem(bp);

I would say it would be better to have the old buffer still allocated
in case of an allocation failure for the new one (like krealloc()).
Especially in case of shrink request.

> +
> +	bp->fw_crash_mem = kzalloc(sizeof(*bp->fw_crash_mem), GFP_KERNEL);

kmalloc(), assuming that you don't depend on the pages to be cleared

> +	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;
> +	}
> +
> +	bp->fw_crash_len = mem_size;
> +
> +	return 0;
> +}

Re: [PATCH net-next v2 1/9] bnxt_en: add support for storing crash dump into host memory

[Michael Chan]

[-- Attachment #1: Type: text/plain, Size: 1495 bytes --]

On Mon, Aug 19, 2024 at 3:00 AM Przemek Kitszel
<przemyslaw.kitszel@intel.com> wrote:
>
> On 8/16/24 23:28, Michael Chan wrote:
> > From: Vikas Gupta <vikas.gupta@broadcom.com>
> > +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);
> > +
> > +     if (bp->fw_crash_mem && mem_size == bp->fw_crash_len)
> > +             return 0;
> > +
> > +     bnxt_free_crash_dump_mem(bp);
>
> I would say it would be better to have the old buffer still allocated
> in case of an allocation failure for the new one (like krealloc()).
> Especially in case of shrink request.

Thanks for the review.  The original author Vikas is no longer working
for our group.  The FW crash dump memory is not just one buffer, but a
series of pages.  To resize these pages, we'll have to add a lot of
new code especially to cover the case that the page level can change.
A change in the memory size is rare.  It only happens when FW has
reset and the new running version requires a different crash dump
length.  So freeing it all and allocating again is a lot easier.  I'll
take a closer look to see if some small optimizations can be done
here.

[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4209 bytes --]

[PATCH net-next v2 2/9] bnxt_en: add support for retrieving crash dump using ethtool

[Michael Chan]

From: Vikas Gupta <vikas.gupta@broadcom.com>

Add support for retrieving crash dump using ethtool -w on the
supported interface.

Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Vikas Gupta <vikas.gupta@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
---
 .../ethernet/broadcom/bnxt/bnxt_coredump.c    | 83 +++++++++++++++++--
 .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 13 ++-
 2 files changed, 87 insertions(+), 9 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c
index ebbad9ccab6a..9ed915e4c618 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c
@@ -372,14 +372,78 @@ 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 -EEXIST;
+
+	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(u32));
+	if (!sig)
+		return false;
+
+	return true;
+}
+
 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 +506,17 @@ 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;
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
index 39eed5831e3a..265956c45ff5 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
@@ -4993,9 +4993,16 @@ 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;
-- 
2.30.1

Re: [PATCH net-next v2 2/9] bnxt_en: add support for retrieving crash dump using ethtool

[Przemek Kitszel]

On 8/16/24 23:28, Michael Chan wrote:
> From: Vikas Gupta <vikas.gupta@broadcom.com>
> 
> Add support for retrieving crash dump using ethtool -w on the
> supported interface.
> 
> Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
> Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
> Reviewed-by: Andy Gospodarek <andrew.gospodarek@broadcom.com>
> Reviewed-by: Simon Horman <horms@kernel.org>
> Signed-off-by: Vikas Gupta <vikas.gupta@broadcom.com>
> Signed-off-by: Michael Chan <michael.chan@broadcom.com>
> ---
>   .../ethernet/broadcom/bnxt/bnxt_coredump.c    | 83 +++++++++++++++++--
>   .../net/ethernet/broadcom/bnxt/bnxt_ethtool.c | 13 ++-
>   2 files changed, 87 insertions(+), 9 deletions(-)
> 

mostly nitpicks from me here

> diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c
> index ebbad9ccab6a..9ed915e4c618 100644
> --- a/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c
> +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_coredump.c
> @@ -372,14 +372,78 @@ 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)

==, but not big deal ;)

> +			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 -EEXIST;

I would interpret this as "dump already taken, no slot for a new one",
while you just don't have mem allocated for it, so:
ENOENT 2 No such file or directory
or similar

> +
> +	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(u32));

sizeof(sig)

> +	if (!sig)
> +		return false;
> +
> +	return true;

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 +506,17 @@ 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 &&

I would wrap bitwise-and with parenthesis, even if not mandated by the C
standard (the same in other places with combined expressions).

> +	    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;
> diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
> index 39eed5831e3a..265956c45ff5 100644
> --- a/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
> +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_ethtool.c
> @@ -4993,9 +4993,16 @@ 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;

Re: [PATCH net-next v2 2/9] bnxt_en: add support for retrieving crash dump using ethtool

[Jakub Kicinski]

On Mon, 19 Aug 2024 12:12:39 +0200 Przemek Kitszel wrote:
> > +	if (dump_type == BNXT_DUMP_CRASH &&
> > +	    bp->fw_dbg_cap & DBG_QCAPS_RESP_FLAGS_CRASHDUMP_HOST_DDR &&  
> 
> I would wrap bitwise-and with parenthesis, even if not mandated by the C
> standard (the same in other places with combined expressions).

Personal preference, I guess. I generally lean on the side of limiting
the brackets.

👍️ to other comments, thank for reviewing!

[PATCH net-next v2 3/9] bnxt_en: Support QOS and TPID settings for the SRIOV VLAN

[Michael Chan]

From: Sreekanth Reddy <sreekanth.reddy@broadcom.com>

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 <horms@kernel.org>
Signed-off-by: Sreekanth Reddy <sreekanth.reddy@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
---
 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(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 017eefd78ba4..149ec5bda477 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -9198,6 +9198,8 @@ 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)
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index 2fa6572b6b1d..aa2eb78a8763 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -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
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c
index 22898d3d088b..58bd84b59f0e 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c
@@ -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;
-- 
2.30.1

[PATCH net-next v2 4/9] bnxt_en: Deprecate support for legacy INTX mode

[Michael Chan]

Firmware 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 <hongguang.gao@broadcom.com>
Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
---
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(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 149ec5bda477..54d82e54456c 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -10772,7 +10772,7 @@ 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;
@@ -10830,7 +10830,7 @@ 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);
@@ -10838,35 +10838,6 @@ 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)
@@ -12912,7 +12883,7 @@ 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;
@@ -15774,6 +15745,11 @@ 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.
 	 */
@@ -15800,9 +15776,6 @@ 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;
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index aa2eb78a8763..e2e2986a976b 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -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
-- 
2.30.1

Re: [PATCH net-next v2 4/9] bnxt_en: Deprecate support for legacy INTX mode

[Simon Horman]

On Fri, Aug 16, 2024 at 02:28:27PM -0700, Michael Chan wrote:
> Firmware 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 <hongguang.gao@broadcom.com>
> Reviewed-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com>
> Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
> Signed-off-by: Michael Chan <michael.chan@broadcom.com>
> ---
> v2: Fix memory leak, update changelog

Hi Michael, Hongguang Gao, all,

Thanks for the update.

Reviewed-by: Simon Horman <horms@kernel.org>

[PATCH net-next v2 5/9] bnxt_en: Remove BNXT_FLAG_USING_MSIX flag

[Michael Chan]

Now 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 <kalesh-anakkur.purayil@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Hongguang Gao <hongguang.gao@broadcom.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
---
 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(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 54d82e54456c..8d3970ccdc21 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -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",
@@ -10655,20 +10626,6 @@ 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)
@@ -10681,10 +10638,7 @@ 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;
@@ -10825,7 +10779,6 @@ 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;
 
@@ -10840,12 +10793,10 @@ 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)
@@ -10943,9 +10894,6 @@ 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];
@@ -11010,29 +10958,22 @@ 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);
 	}
 }
 
@@ -12151,12 +12092,6 @@ 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) {
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index e2e2986a976b..043d2a849674 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -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
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c b/drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c
index 58bd84b59f0e..7bb8a5d74430 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c
@@ -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");
-- 
2.30.1

[PATCH net-next v2 6/9] bnxt_en: Remove register mapping to support INTX

[Michael Chan]

In 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 <somnath.kotur@broadcom.com>
Reviewed-by: Hongguang Gao <hongguang.gao@broadcom.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
---
v2: Update changelog
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 19 -------------------
 drivers/net/ethernet/broadcom/bnxt/bnxt.h |  2 --
 2 files changed, 21 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 8d3970ccdc21..0666bf0e5dc6 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -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)
@@ -11961,20 +11957,6 @@ 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)
@@ -12079,7 +12061,6 @@ 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. */
diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
index 043d2a849674..93e00215819a 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h
@@ -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
 
-- 
2.30.1

[PATCH net-next v2 7/9] bnxt_en: Replace deprecated PCI MSIX APIs

[Michael Chan]

Use 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 <somnath.kotur@broadcom.com>
Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 21 +++++----------------
 1 file changed, 5 insertions(+), 16 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 0666bf0e5dc6..5d0d40d78c37 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -10725,7 +10725,6 @@ 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);
@@ -10735,19 +10734,11 @@ 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;
@@ -10757,7 +10748,7 @@ 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 */
@@ -10775,21 +10766,19 @@ 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;
-- 
2.30.1

Re: [PATCH net-next v2 7/9] bnxt_en: Replace deprecated PCI MSIX APIs

[Przemek Kitszel]

On 8/16/24 23:28, Michael Chan wrote:
> Use 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 <somnath.kotur@broadcom.com>
> Reviewed-by: Pavan Chebbi <pavan.chebbi@broadcom.com>
> Reviewed-by: Simon Horman <horms@kernel.org>
> Signed-off-by: Michael Chan <michael.chan@broadcom.com>
> ---
>   drivers/net/ethernet/broadcom/bnxt/bnxt.c | 21 +++++----------------
>   1 file changed, 5 insertions(+), 16 deletions(-)
> 

very nice,
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>

[PATCH net-next v2 8/9] bnxt_en: Allocate the max bp->irq_tbl size for dynamic msix allocation

[Michael Chan]

If 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 <hongguang.gao@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
---
v2: Fix typo in changelog
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index 5d0d40d78c37..b969ace6b4d1 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -10724,7 +10724,7 @@ 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);
@@ -10745,7 +10745,10 @@ 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(struct bnxt_irq), GFP_KERNEL);
 	if (bp->irq_tbl) {
 		for (i = 0; i < total_vecs; i++)
 			bp->irq_tbl[i].vector = pci_irq_vector(bp->pdev, i);
-- 
2.30.1

Re: [PATCH net-next v2 8/9] bnxt_en: Allocate the max bp->irq_tbl size for dynamic msix allocation

[Przemek Kitszel]

On 8/16/24 23:28, Michael Chan wrote:
> If 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 <hongguang.gao@broadcom.com>
> Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
> Reviewed-by: Simon Horman <horms@kernel.org>
> Signed-off-by: Michael Chan <michael.chan@broadcom.com>
> ---
> v2: Fix typo in changelog
> ---
>   drivers/net/ethernet/broadcom/bnxt/bnxt.c | 7 +++++--
>   1 file changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
> index 5d0d40d78c37..b969ace6b4d1 100644
> --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
> +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
> @@ -10724,7 +10724,7 @@ 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);
> @@ -10745,7 +10745,10 @@ 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(struct bnxt_irq), GFP_KERNEL);

sizeof(*bp->irq_tbl)

but otherwise this is fine, so:
Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>

>   	if (bp->irq_tbl) {
>   		for (i = 0; i < total_vecs; i++)
>   			bp->irq_tbl[i].vector = pci_irq_vector(bp->pdev, i);

[PATCH net-next v2 9/9] bnxt_en: Support dynamic MSIX

[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 <hongguang.gao@broadcom.com>
Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Michael Chan <michael.chan@broadcom.com>
---
v2: Fix typo in changelog
---
 drivers/net/ethernet/broadcom/bnxt/bnxt.c | 57 +++++++++++++++++++++--
 1 file changed, 54 insertions(+), 3 deletions(-)

diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
index b969ace6b4d1..d702fc1d9ed5 100644
--- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c
+++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c
@@ -10624,6 +10624,43 @@ static void bnxt_setup_msix(struct bnxt *bp)
 
 static int bnxt_init_int_mode(struct bnxt *bp);
 
+static int bnxt_add_msix(struct bnxt *bp, int total)
+{
+	int i;
+
+	if (bp->total_irqs >= total)
+		return total;
+
+	for (i = bp->total_irqs; i < total; i++) {
+		struct msi_map map;
+
+		map = pci_msix_alloc_irq_at(bp->pdev, i, NULL);
+		if (map.index < 0)
+			break;
+		bp->irq_tbl[i].vector = map.virq;
+		bp->total_irqs++;
+	}
+	return bp->total_irqs;
+}
+
+static int bnxt_trim_msix(struct bnxt *bp, int total)
+{
+	int i;
+
+	if (bp->total_irqs <= total)
+		return total;
+
+	for (i = bp->total_irqs; i > total; i--) {
+		struct msi_map map;
+
+		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;
@@ -10790,6 +10827,7 @@ 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;
@@ -10808,15 +10846,28 @@ 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) {
+		int total;
+
+		if (irqs_required > bp->total_irqs)
+			total = bnxt_add_msix(bp, irqs_required);
+		else
+			total = bnxt_trim_msix(bp, irqs_required);
+
+		if (total != irqs_required)
+			rc = -ENOSPC;
 	}
 	if (rc) {
 		netdev_err(bp->dev, "ring reservation/IRQ init failure rc: %d\n", rc);
-- 
2.30.1

Re: [PATCH net-next v2 9/9] bnxt_en: Support dynamic MSIX

[Przemek Kitszel]

On 8/16/24 23:28, Michael Chan wrote:
> 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 <hongguang.gao@broadcom.com>
> Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com>
> Reviewed-by: Simon Horman <horms@kernel.org>
> Signed-off-by: Michael Chan <michael.chan@broadcom.com>
> ---
> v2: Fix typo in changelog
> ---
>   drivers/net/ethernet/broadcom/bnxt/bnxt.c | 57 +++++++++++++++++++++--
>   1 file changed, 54 insertions(+), 3 deletions(-)
> 

Reviewed-by: Przemek Kitszel <przemyslaw.kitszel@intel.com>