[dpdk-dev] [PATCH] net/ice: clear the HW tables when close device

Zhang, Qi Z qi.z.zhang at intel.com
Fri Nov 8 10:24:46 CET 2019



> -----Original Message-----
> From: Zhang, Qi Z
> Sent: Friday, November 8, 2019 2:51 PM
> To: Ye, Xiaolong <xiaolong.ye at intel.com>
> Cc: Yang, Qiming <qiming.yang at intel.com>; dev at dpdk.org
> Subject: RE: [PATCH] net/ice: clear the HW tables when close device
> 
> 
> 
> > -----Original Message-----
> > From: Ye, Xiaolong <xiaolong.ye at intel.com>
> > Sent: Friday, November 8, 2019 2:18 PM
> > To: Zhang, Qi Z <qi.z.zhang at intel.com>
> > Cc: Yang, Qiming <qiming.yang at intel.com>; dev at dpdk.org
> > Subject: Re: [PATCH] net/ice: clear the HW tables when close device
> >
> > Hi, Qi
> >
> > On 11/07, Qi Zhang wrote:
> > >Clear the HW tables during dev_close. Otherwise HW tables will not be
> > >initialized correctly after device reset.
> > >
> > >Fixes: a4c8c48fe3f4 ("net/ice: load OS default package")
> > >
> > >Signed-off-by: Qi Zhang <qi.z.zhang at intel.com>
> > >---
> > > drivers/net/ice/base/ice_flex_pipe.c | 55
> > >++++++++++++++++++++++++++++++++++++
> > > drivers/net/ice/base/ice_flex_pipe.h |  1 +
> > > drivers/net/ice/ice_ethdev.c         |  1 +
> > > 3 files changed, 57 insertions(+)
> > >
> > >diff --git a/drivers/net/ice/base/ice_flex_pipe.c
> > >b/drivers/net/ice/base/ice_flex_pipe.c
> > >index dd098f529..b816ca67f 100644
> > >--- a/drivers/net/ice/base/ice_flex_pipe.c
> > >+++ b/drivers/net/ice/base/ice_flex_pipe.c
> > >@@ -3618,6 +3618,61 @@ static void ice_init_flow_profs(struct ice_hw
> > >*hw, u8 blk_idx)  }
> > >
> > > /**
> > >+ * ice_clear_hw_tbls - clear HW tables and flow profiles
> > >+ * @hw: pointer to the hardware structure  */ void
> > >+ice_clear_hw_tbls(struct ice_hw *hw)
> >
> > What is the difference between this one and existing ice_free_hw_tbls,
> > can we use ice_free_hw_tbls directly?
> 
> The ice_clear_hw_tbls is designed for device reset scenario, it reset all the
> table, but not do free, so during next ice_init_hw_tbls, all the allocation steps
> can be skipped.
> Though, I guess ice_free_hw_tbls may also works, but Kernel driver use
> ice_clear_hw_tbls during reset, DPDK would better just follow the same
> implementation to avoid unnecessary evaluation effort.

After second thought, I think we should free the memory in dev_unit.to prevent memory leak.
So ice_free_hw_tbls should be the right fix,  ice_clear_hw_tbls should be used in the case when we try to optimize the dev_reset scenario which does not simply call dev_uninit/dev_init.
I will send a v2 for this.

Thanks
Qi
> 
> Regards
> Qi
> 
> 
> >
> > Thanks,
> > Xiaolong
> >
> > >+{
> > >+	u8 i;
> > >+
> > >+	for (i = 0; i < ICE_BLK_COUNT; i++) {
> > >+		struct ice_prof_redir *prof_redir = &hw->blk[i].prof_redir;
> > >+		struct ice_prof_tcam *prof = &hw->blk[i].prof;
> > >+		struct ice_xlt1 *xlt1 = &hw->blk[i].xlt1;
> > >+		struct ice_xlt2 *xlt2 = &hw->blk[i].xlt2;
> > >+		struct ice_es *es = &hw->blk[i].es;
> > >+
> > >+		if (hw->blk[i].is_list_init) {
> > >+			ice_free_prof_map(hw, i);
> > >+			ice_free_flow_profs(hw, i);
> > >+		}
> > >+
> > >+		ice_free_vsig_tbl(hw, (enum ice_block)i);
> > >+
> > >+		ice_memset(xlt1->ptypes, 0, xlt1->count * sizeof(*xlt1->ptypes),
> > >+			   ICE_NONDMA_MEM);
> > >+		ice_memset(xlt1->ptg_tbl, 0,
> > >+			   ICE_MAX_PTGS * sizeof(*xlt1->ptg_tbl),
> > >+			   ICE_NONDMA_MEM);
> > >+		ice_memset(xlt1->t, 0, xlt1->count * sizeof(*xlt1->t),
> > >+			   ICE_NONDMA_MEM);
> > >+
> > >+		ice_memset(xlt2->vsis, 0, xlt2->count * sizeof(*xlt2->vsis),
> > >+			   ICE_NONDMA_MEM);
> > >+		ice_memset(xlt2->vsig_tbl, 0,
> > >+			   xlt2->count * sizeof(*xlt2->vsig_tbl),
> > >+			   ICE_NONDMA_MEM);
> > >+		ice_memset(xlt2->t, 0, xlt2->count * sizeof(*xlt2->t),
> > >+			   ICE_NONDMA_MEM);
> > >+
> > >+		ice_memset(prof->t, 0, prof->count * sizeof(*prof->t),
> > >+			   ICE_NONDMA_MEM);
> > >+		ice_memset(prof_redir->t, 0,
> > >+			   prof_redir->count * sizeof(*prof_redir->t),
> > >+			   ICE_NONDMA_MEM);
> > >+
> > >+		ice_memset(es->t, 0, es->count * sizeof(*es->t),
> > >+			   ICE_NONDMA_MEM);
> > >+		ice_memset(es->ref_count, 0, es->count * sizeof(*es->ref_count),
> > >+			   ICE_NONDMA_MEM);
> > >+		ice_memset(es->written, 0, es->count * sizeof(*es->written),
> > >+			   ICE_NONDMA_MEM);
> > >+		ice_memset(es->mask_ena, 0, es->count * sizeof(*es->mask_ena),
> > >+			   ICE_NONDMA_MEM);
> > >+	}
> > >+}
> > >+
> > >+/**
> > >  * ice_init_hw_tbls - init hardware table memory
> > >  * @hw: pointer to the hardware structure
> > >  */
> > >diff --git a/drivers/net/ice/base/ice_flex_pipe.h
> > >b/drivers/net/ice/base/ice_flex_pipe.h
> > >index ee606af15..fa72e386d 100644
> > >--- a/drivers/net/ice/base/ice_flex_pipe.h
> > >+++ b/drivers/net/ice/base/ice_flex_pipe.h
> > >@@ -71,6 +71,7 @@ ice_copy_and_init_pkg(struct ice_hw *hw, const u8
> > >*buf, u32 len);  enum ice_status ice_init_hw_tbls(struct ice_hw *hw);
> > >void ice_free_seg(struct ice_hw *hw);  void ice_fill_blk_tbls(struct
> > >ice_hw *hw);
> > >+void ice_clear_hw_tbls(struct ice_hw *hw);
> > > void ice_free_hw_tbls(struct ice_hw *hw);  enum ice_status
> > >ice_rem_prof(struct ice_hw *hw, enum ice_block blk, u64 id); diff
> > >--git a/drivers/net/ice/ice_ethdev.c b/drivers/net/ice/ice_ethdev.c
> > >index d81eb5eeb..0b439cc0b 100644
> > >--- a/drivers/net/ice/ice_ethdev.c
> > >+++ b/drivers/net/ice/ice_ethdev.c
> > >@@ -2322,6 +2322,7 @@ ice_dev_close(struct rte_eth_dev *dev)
> > > 	rte_free(hw->port_info);
> > > 	hw->port_info = NULL;
> > > 	ice_shutdown_all_ctrlq(hw);
> > >+	ice_clear_hw_tbls(hw);
> > > 	rte_free(pf->proto_xtr);
> > > 	pf->proto_xtr = NULL;
> > >
> > >--
> > >2.13.6
> > >


More information about the dev mailing list