[net,v2,6/7] octeontx2-af: Skip PFs if not enabled

Message ID 20230407122344.4059-7-saikrishnag@marvell.com
State New
Headers
Series octeontx2: Miscellaneous fixes |

Commit Message

Sai Krishna Gajula April 7, 2023, 12:23 p.m. UTC
  From: Ratheesh Kannoth <rkannoth@marvell.com>

Skip mbox initialization of disabled PFs. Firmware configures PFs
and allocate mbox resources etc. Linux should configure particular
PFs, which ever are enabled by firmware.

Fixes: 9bdc47a6e328 ("octeontx2-af: Mbox communication support btw AF and it's VFs")
Signed-off-by: Ratheesh Kannoth <rkannoth@marvell.com>
Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
Signed-off-by: Sai Krishna <saikrishnag@marvell.com>
---
 .../net/ethernet/marvell/octeontx2/af/mbox.c  |  5 ++-
 .../net/ethernet/marvell/octeontx2/af/mbox.h  |  3 +-
 .../net/ethernet/marvell/octeontx2/af/rvu.c   | 38 ++++++++++++++++---
 3 files changed, 39 insertions(+), 7 deletions(-)
  

Comments

Simon Horman April 8, 2023, 2:56 p.m. UTC | #1
On Fri, Apr 07, 2023 at 05:53:43PM +0530, Sai Krishna wrote:
> From: Ratheesh Kannoth <rkannoth@marvell.com>
> 
> Skip mbox initialization of disabled PFs. Firmware configures PFs
> and allocate mbox resources etc. Linux should configure particular
> PFs, which ever are enabled by firmware.
> 
> Fixes: 9bdc47a6e328 ("octeontx2-af: Mbox communication support btw AF and it's VFs")
> Signed-off-by: Ratheesh Kannoth <rkannoth@marvell.com>
> Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
> Signed-off-by: Sai Krishna <saikrishnag@marvell.com>

...

> @@ -2343,8 +2349,27 @@ static int rvu_mbox_init(struct rvu *rvu, struct mbox_wq_info *mw,
>  	int err = -EINVAL, i, dir, dir_up;
>  	void __iomem *reg_base;
>  	struct rvu_work *mwork;
> +	unsigned long *pf_bmap;
>  	void **mbox_regions;
>  	const char *name;
> +	u64 cfg;
> +
> +	pf_bmap = devm_kcalloc(rvu->dev, BITS_TO_LONGS(num), sizeof(long), GFP_KERNEL);

Hi Sai and Ratheesh,

I am a little confused about the lifecycle of pf_bmap.
It is a local variable of this function.
But it is allocated using devm_kcalloc(), so it will persist for
the life of the device.

Also, I note that rvu_mbox_init() has too call sites.
So is there a situation where a pf_bmap is allocated more than once
for the same rvu->dev instance?

It seems to me it would be more appropriate to allocate bf_bmap using
kcalloc() take clear to free it before leaving rvu_mbox_init(),
as is already done for mbox_regions.

> +	if (!pf_bmap)
> +		return -ENOMEM;

...
  
Sai Krishna Gajula April 14, 2023, 11:58 a.m. UTC | #2
Please see inline,

> -----Original Message-----
> From: Simon Horman <simon.horman@corigine.com>
> Sent: Saturday, April 8, 2023 8:26 PM
> To: Sai Krishna Gajula <saikrishnag@marvell.com>
> Cc: davem@davemloft.net; edumazet@google.com; kuba@kernel.org;
> pabeni@redhat.com; netdev@vger.kernel.org; linux-
> kernel@vger.kernel.org; Sunil Kovvuri Goutham <sgoutham@marvell.com>;
> Geethasowjanya Akula <gakula@marvell.com>; richardcochran@gmail.com;
> Linu Cherian <lcherian@marvell.com>; Jerin Jacob Kollanukkaran
> <jerinj@marvell.com>; Hariprasad Kelam <hkelam@marvell.com>;
> Subbaraya Sundeep Bhatta <sbhatta@marvell.com>; Ratheesh Kannoth
> <rkannoth@marvell.com>
> Subject: Re: [net PATCH v2 6/7] octeontx2-af: Skip PFs if not enabled
> 
> On Fri, Apr 07, 2023 at 05:53:43PM +0530, Sai Krishna wrote:
> > From: Ratheesh Kannoth <rkannoth@marvell.com>
> >
> > Skip mbox initialization of disabled PFs. Firmware configures PFs and
> > allocate mbox resources etc. Linux should configure particular PFs,
> > which ever are enabled by firmware.
> >
> > Fixes: 9bdc47a6e328 ("octeontx2-af: Mbox communication support btw AF
> > and it's VFs")
> > Signed-off-by: Ratheesh Kannoth <rkannoth@marvell.com>
> > Signed-off-by: Sunil Kovvuri Goutham <sgoutham@marvell.com>
> > Signed-off-by: Sai Krishna <saikrishnag@marvell.com>
> 
> ...
> 
> > @@ -2343,8 +2349,27 @@ static int rvu_mbox_init(struct rvu *rvu, struct
> mbox_wq_info *mw,
> >  	int err = -EINVAL, i, dir, dir_up;
> >  	void __iomem *reg_base;
> >  	struct rvu_work *mwork;
> > +	unsigned long *pf_bmap;
> >  	void **mbox_regions;
> >  	const char *name;
> > +	u64 cfg;
> > +
> > +	pf_bmap = devm_kcalloc(rvu->dev, BITS_TO_LONGS(num),
> sizeof(long),
> > +GFP_KERNEL);
> 
> Hi Sai and Ratheesh,
> 
> I am a little confused about the lifecycle of pf_bmap.
> It is a local variable of this function.
> But it is allocated using devm_kcalloc(), so it will persist for the life of the
> device.
> 
> Also, I note that rvu_mbox_init() has too call sites.
> So is there a situation where a pf_bmap is allocated more than once for the
> same rvu->dev instance?
> 
> It seems to me it would be more appropriate to allocate bf_bmap using
> kcalloc() take clear to free it before leaving rvu_mbox_init(), as is already
> done for mbox_regions.
> 

We will prepare v3 patch according to the suggested changes using kcalloc().

> > +	if (!pf_bmap)
> > +		return -ENOMEM;
> 
> ...
  

Patch

diff --git a/drivers/net/ethernet/marvell/octeontx2/af/mbox.c b/drivers/net/ethernet/marvell/octeontx2/af/mbox.c
index 2898931d5260..9690ac01f02c 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/mbox.c
+++ b/drivers/net/ethernet/marvell/octeontx2/af/mbox.c
@@ -157,7 +157,7 @@  EXPORT_SYMBOL(otx2_mbox_init);
  */
 int otx2_mbox_regions_init(struct otx2_mbox *mbox, void **hwbase,
 			   struct pci_dev *pdev, void *reg_base,
-			   int direction, int ndevs)
+			   int direction, int ndevs, unsigned long *pf_bmap)
 {
 	struct otx2_mbox_dev *mdev;
 	int devid, err;
@@ -169,6 +169,9 @@  int otx2_mbox_regions_init(struct otx2_mbox *mbox, void **hwbase,
 	mbox->hwbase = hwbase[0];
 
 	for (devid = 0; devid < ndevs; devid++) {
+		if (!test_bit(devid, pf_bmap))
+			continue;
+
 		mdev = &mbox->dev[devid];
 		mdev->mbase = hwbase[devid];
 		mdev->hwbase = hwbase[devid];
diff --git a/drivers/net/ethernet/marvell/octeontx2/af/mbox.h b/drivers/net/ethernet/marvell/octeontx2/af/mbox.h
index 0ce533848536..26636a4d7dcc 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/mbox.h
+++ b/drivers/net/ethernet/marvell/octeontx2/af/mbox.h
@@ -96,9 +96,10 @@  void otx2_mbox_destroy(struct otx2_mbox *mbox);
 int otx2_mbox_init(struct otx2_mbox *mbox, void __force *hwbase,
 		   struct pci_dev *pdev, void __force *reg_base,
 		   int direction, int ndevs);
+
 int otx2_mbox_regions_init(struct otx2_mbox *mbox, void __force **hwbase,
 			   struct pci_dev *pdev, void __force *reg_base,
-			   int direction, int ndevs);
+			   int direction, int ndevs, unsigned long *bmap);
 void otx2_mbox_msg_send(struct otx2_mbox *mbox, int devid);
 int otx2_mbox_wait_for_rsp(struct otx2_mbox *mbox, int devid);
 int otx2_mbox_busy_poll_for_rsp(struct otx2_mbox *mbox, int devid);
diff --git a/drivers/net/ethernet/marvell/octeontx2/af/rvu.c b/drivers/net/ethernet/marvell/octeontx2/af/rvu.c
index 8683ce57ed3f..61c658fa3f28 100644
--- a/drivers/net/ethernet/marvell/octeontx2/af/rvu.c
+++ b/drivers/net/ethernet/marvell/octeontx2/af/rvu.c
@@ -2282,7 +2282,7 @@  static inline void rvu_afvf_mbox_up_handler(struct work_struct *work)
 }
 
 static int rvu_get_mbox_regions(struct rvu *rvu, void **mbox_addr,
-				int num, int type)
+				int num, int type, unsigned long *pf_bmap)
 {
 	struct rvu_hwinfo *hw = rvu->hw;
 	int region;
@@ -2294,6 +2294,9 @@  static int rvu_get_mbox_regions(struct rvu *rvu, void **mbox_addr,
 	 */
 	if (type == TYPE_AFVF) {
 		for (region = 0; region < num; region++) {
+			if (!test_bit(region, pf_bmap))
+				continue;
+
 			if (hw->cap.per_pf_mbox_regs) {
 				bar4 = rvu_read64(rvu, BLKADDR_RVUM,
 						  RVU_AF_PFX_BAR4_ADDR(0)) +
@@ -2315,6 +2318,9 @@  static int rvu_get_mbox_regions(struct rvu *rvu, void **mbox_addr,
 	 * RVU_AF_PF_BAR4_ADDR register.
 	 */
 	for (region = 0; region < num; region++) {
+		if (!test_bit(region, pf_bmap))
+			continue;
+
 		if (hw->cap.per_pf_mbox_regs) {
 			bar4 = rvu_read64(rvu, BLKADDR_RVUM,
 					  RVU_AF_PFX_BAR4_ADDR(region));
@@ -2343,8 +2349,27 @@  static int rvu_mbox_init(struct rvu *rvu, struct mbox_wq_info *mw,
 	int err = -EINVAL, i, dir, dir_up;
 	void __iomem *reg_base;
 	struct rvu_work *mwork;
+	unsigned long *pf_bmap;
 	void **mbox_regions;
 	const char *name;
+	u64 cfg;
+
+	pf_bmap = devm_kcalloc(rvu->dev, BITS_TO_LONGS(num), sizeof(long), GFP_KERNEL);
+	if (!pf_bmap)
+		return -ENOMEM;
+
+	/* RVU VFs */
+	if (type == TYPE_AFVF)
+		bitmap_set(pf_bmap, 0, num);
+
+	if (type == TYPE_AFPF) {
+		/* Mark enabled PFs in bitmap */
+		for (i = 0; i < num; i++) {
+			cfg = rvu_read64(rvu, BLKADDR_RVUM, RVU_PRIV_PFX_CFG(i));
+			if (cfg & BIT_ULL(20))
+				set_bit(i, pf_bmap);
+		}
+	}
 
 	mbox_regions = kcalloc(num, sizeof(void *), GFP_KERNEL);
 	if (!mbox_regions)
@@ -2356,7 +2381,7 @@  static int rvu_mbox_init(struct rvu *rvu, struct mbox_wq_info *mw,
 		dir = MBOX_DIR_AFPF;
 		dir_up = MBOX_DIR_AFPF_UP;
 		reg_base = rvu->afreg_base;
-		err = rvu_get_mbox_regions(rvu, mbox_regions, num, TYPE_AFPF);
+		err = rvu_get_mbox_regions(rvu, mbox_regions, num, TYPE_AFPF, pf_bmap);
 		if (err)
 			goto free_regions;
 		break;
@@ -2365,7 +2390,7 @@  static int rvu_mbox_init(struct rvu *rvu, struct mbox_wq_info *mw,
 		dir = MBOX_DIR_PFVF;
 		dir_up = MBOX_DIR_PFVF_UP;
 		reg_base = rvu->pfreg_base;
-		err = rvu_get_mbox_regions(rvu, mbox_regions, num, TYPE_AFVF);
+		err = rvu_get_mbox_regions(rvu, mbox_regions, num, TYPE_AFVF, pf_bmap);
 		if (err)
 			goto free_regions;
 		break;
@@ -2396,16 +2421,19 @@  static int rvu_mbox_init(struct rvu *rvu, struct mbox_wq_info *mw,
 	}
 
 	err = otx2_mbox_regions_init(&mw->mbox, mbox_regions, rvu->pdev,
-				     reg_base, dir, num);
+				     reg_base, dir, num, pf_bmap);
 	if (err)
 		goto exit;
 
 	err = otx2_mbox_regions_init(&mw->mbox_up, mbox_regions, rvu->pdev,
-				     reg_base, dir_up, num);
+				     reg_base, dir_up, num, pf_bmap);
 	if (err)
 		goto exit;
 
 	for (i = 0; i < num; i++) {
+		if (!test_bit(i, pf_bmap))
+			continue;
+
 		mwork = &mw->mbox_wrk[i];
 		mwork->rvu = rvu;
 		INIT_WORK(&mwork->work, mbox_handler);