Skip to contentSkip to conversation

RDMA/rxe: Bump up default maximum values used via uverbs

PATCH Applied Rao Shoaib 3 participants 11 messages Sep 15, 2021 – Oct 4, 2021

Subsystems (2)
lore
Share & tools
lore

Rao Shoaib

author
#0 lore raw

PATCH 1/1 RDMA/rxe: Bump up default maximum values used via uverbs

In our internal testing we have found that default maximum values are too small. Ideally there should be no limits, but since maximum values are reported via ibv_query_device, we have to return some value. So, the default maximums have been changed to large values.

---

Resubmitting the patch after applying Bob's latest patches and testing using via rping.

 drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
 1 file changed, 16 insertions(+), 14 deletions(-)
drivers/infiniband/sw/rxe/rxe_param.h +16 -14
index 742e6ec93686..d703a54df4ec 100644
@@ -9,6 +9,8 @@
 
 #include <uapi/rdma/rdma_user_rxe.h>
 
+#define DEFAULT_MAX_VALUE (1 << 20)
+
 static inline enum ib_mtu rxe_mtu_int_to_enum(int mtu)
 {
 	if (mtu < 256)
@@ -37,7 +39,7 @@ static inline enum ib_mtu eth_mtu_int_to_enum(int mtu)
 enum rxe_device_param {
 	RXE_MAX_MR_SIZE			= -1ull,
 	RXE_PAGE_SIZE_CAP		= 0xfffff000,
-	RXE_MAX_QP_WR			= 0x4000,
+	RXE_MAX_QP_WR			= DEFAULT_MAX_VALUE,
 	RXE_DEVICE_CAP_FLAGS		= IB_DEVICE_BAD_PKEY_CNTR
 					| IB_DEVICE_BAD_QKEY_CNTR
 					| IB_DEVICE_AUTO_PATH_MIG
@@ -58,42 +60,42 @@ enum rxe_device_param {
 	RXE_MAX_INLINE_DATA		= RXE_MAX_WQE_SIZE -
 					  sizeof(struct rxe_send_wqe),
 	RXE_MAX_SGE_RD			= 32,
-	RXE_MAX_CQ			= 16384,
+	RXE_MAX_CQ			= DEFAULT_MAX_VALUE,
 	RXE_MAX_LOG_CQE			= 15,
-	RXE_MAX_PD			= 0x7ffc,
+	RXE_MAX_PD			= DEFAULT_MAX_VALUE,
 	RXE_MAX_QP_RD_ATOM		= 128,
 	RXE_MAX_RES_RD_ATOM		= 0x3f000,
 	RXE_MAX_QP_INIT_RD_ATOM		= 128,
 	RXE_MAX_MCAST_GRP		= 8192,
 	RXE_MAX_MCAST_QP_ATTACH		= 56,
 	RXE_MAX_TOT_MCAST_QP_ATTACH	= 0x70000,
-	RXE_MAX_AH			= 100,
-	RXE_MAX_SRQ_WR			= 0x4000,
+	RXE_MAX_AH			= DEFAULT_MAX_VALUE,
+	RXE_MAX_SRQ_WR			= DEFAULT_MAX_VALUE,
 	RXE_MIN_SRQ_WR			= 1,
 	RXE_MAX_SRQ_SGE			= 27,
 	RXE_MIN_SRQ_SGE			= 1,
 	RXE_MAX_FMR_PAGE_LIST_LEN	= 512,
-	RXE_MAX_PKEYS			= 1,
+	RXE_MAX_PKEYS			= 64,
 	RXE_LOCAL_CA_ACK_DELAY		= 15,
 
-	RXE_MAX_UCONTEXT		= 512,
+	RXE_MAX_UCONTEXT		= DEFAULT_MAX_VALUE,
 
 	RXE_NUM_PORT			= 1,
 
-	RXE_MAX_QP			= 0x10000,
 	RXE_MIN_QP_INDEX		= 16,
-	RXE_MAX_QP_INDEX		= 0x00020000,
+	RXE_MAX_QP_INDEX		= DEFAULT_MAX_VALUE,
+	RXE_MAX_QP			= DEFAULT_MAX_VALUE - RXE_MIN_QP_INDEX,
 
-	RXE_MAX_SRQ			= 0x00001000,
 	RXE_MIN_SRQ_INDEX		= 0x00020001,
-	RXE_MAX_SRQ_INDEX		= 0x00040000,
+	RXE_MAX_SRQ_INDEX		= DEFAULT_MAX_VALUE,
+	RXE_MAX_SRQ			= DEFAULT_MAX_VALUE - RXE_MIN_SRQ_INDEX,
 
-	RXE_MAX_MR			= 0x00001000,
-	RXE_MAX_MW			= 0x00001000,
 	RXE_MIN_MR_INDEX		= 0x00000001,
-	RXE_MAX_MR_INDEX		= 0x00010000,
+	RXE_MAX_MR_INDEX		= DEFAULT_MAX_VALUE,
+	RXE_MAX_MR			= DEFAULT_MAX_VALUE - RXE_MIN_MR_INDEX,
 	RXE_MIN_MW_INDEX		= 0x00010001,
 	RXE_MAX_MW_INDEX		= 0x00020000,
+	RXE_MAX_MW			= 0x00001000,
 
 	RXE_MAX_PKT_PER_ACK		= 64,
 
Signature · 1 line
2.27.0
maintainer
#1 lore raw
Quoted 7 lines · Rao Shoaib

In our internal testing we have found that default maximum values are too small. Ideally there should be no limits, but since maximum values are reported via ibv_query_device, we have to return some value. So, the default maximums have been changed to large values.

Quoted 8 lines

---

Resubmitting the patch after applying Bob's latest patches and testing using via rping.

 drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
 1 file changed, 16 insertions(+), 14 deletions(-)

So are we good with this? Bob? Zhu?

drivers/infiniband/sw/rxe/rxe_param.h enum rxe_device_param { · line 95

Quoted 3 lines
drivers/infiniband/sw/rxe/rxe_param.h +2 -1
-	RXE_MAX_MR_INDEX		= 0x00010000,
+	RXE_MAX_MR_INDEX		= DEFAULT_MAX_VALUE,
+	RXE_MAX_MR			= DEFAULT_MAX_VALUE - RXE_MIN_MR_INDEX,

Bob, were you saying this was what needed to be bigger to pass blktests??

Jason

maintainer
#2 lore raw
Quoted 12 lines · Jason Gunthorpe
Quoted 10 lines · Rao Shoaib

In our internal testing we have found that default maximum values are too small. Ideally there should be no limits, but since maximum values are reported via ibv_query_device, we have to return some value. So, the default maximums have been changed to large values.

---

Quoted 7 lines
Quoted 5 lines

Resubmitting the patch after applying Bob's latest patches and testing using via rping.

 drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
 1 file changed, 16 insertions(+), 14 deletions(-)

So are we good with this? Bob? Zhu?

I have already checked this commit. And I have found 2 problems with this commit.
This commit changes many MAXs.
And now rxe is not stable enough. Not sure this commit will cause the new problems.

Zhu Yanjun

Quoted 9 lines

drivers/infiniband/sw/rxe/rxe_param.h enum rxe_device_param { · line 95

Quoted 4 lines
drivers/infiniband/sw/rxe/rxe_param.h +2 -1
-     RXE_MAX_MR_INDEX                = 0x00010000,
+     RXE_MAX_MR_INDEX                = DEFAULT_MAX_VALUE,
+     RXE_MAX_MR                      = DEFAULT_MAX_VALUE - RXE_MIN_MR_INDEX,

Bob, were you saying this was what needed to be bigger to pass blktests??

Jason

author
#3 lore raw
Quoted 17 lines · Zhu Yanjun
Quoted 16 lines · Jason Gunthorpe
Quoted 15 lines · Rao Shoaib

In our internal testing we have found that default maximum values are too small. Ideally there should be no limits, but since maximum values are reported via ibv_query_device, we have to return some value. So, the default maximums have been changed to large values.

---

Resubmitting the patch after applying Bob's latest patches and testing using via rping.

  drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
  1 file changed, 16 insertions(+), 14 deletions(-)
Quoted 8 lines
Quoted 1 line

So are we good with this? Bob? Zhu?

I have already checked this commit. And I have found 2 problems with this commit.
This commit changes many MAXs.
And now rxe is not stable enough. Not sure this commit will cause the new problems.

Zhu Yanjun

Hi Zhu,

A generic statement without any technical data does not help. As far as I am aware, currently there are no outstanding issues. If there are, please provide data that clearly shows that the issue is caused by this patch.

Thanks you.

Shoaib

Quoted 7 lines
Quoted 7 lines

drivers/infiniband/sw/rxe/rxe_param.h enum rxe_device_param { · line 95

Quoted 3 lines
drivers/infiniband/sw/rxe/rxe_param.h +2 -1
-     RXE_MAX_MR_INDEX                = 0x00010000,
+     RXE_MAX_MR_INDEX                = DEFAULT_MAX_VALUE,
+     RXE_MAX_MR                      = DEFAULT_MAX_VALUE - RXE_MIN_MR_INDEX,

Bob, were you saying this was what needed to be bigger to pass blktests??

Jason

maintainer
#4 lore raw
Quoted 27 lines · Shoaib Rao
Quoted 24 lines · Zhu Yanjun
Quoted 11 lines · Jason Gunthorpe
Quoted 10 lines · Rao Shoaib

In our internal testing we have found that default maximum values are too small. Ideally there should be no limits, but since maximum values are reported via ibv_query_device, we have to return some value. So, the default maximums have been changed to large values.

---

Quoted 6 lines
Quoted 5 lines

Resubmitting the patch after applying Bob's latest patches and testing using via rping.

  drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
  1 file changed, 16 insertions(+), 14 deletions(-)

So are we good with this? Bob? Zhu?

I have already checked this commit. And I have found 2 problems with this commit.
This commit changes many MAXs.
And now rxe is not stable enough. Not sure this commit will cause the new problems.

Quoted 8 lines
Quoted 1 line

Zhu Yanjun

Hi Zhu,

A generic statement without any technical data does not help. As far as I am aware, currently there are no outstanding issues. If there are, please provide data that clearly shows that the issue is caused by this patch.

Hi, Shoaib

With this commit, I found 2 problems.
This is why I suspect that this commit will introduce risks.

Before a commit is sent to the upstream, please make full tests with it.

Zhu Yanjun

Quoted 12 lines

Thanks you.

Shoaib

Quoted 7 lines
Quoted 7 lines

drivers/infiniband/sw/rxe/rxe_param.h enum rxe_device_param { · line 95

Quoted 3 lines
drivers/infiniband/sw/rxe/rxe_param.h +2 -1
-     RXE_MAX_MR_INDEX                = 0x00010000,
+     RXE_MAX_MR_INDEX                = DEFAULT_MAX_VALUE,
+     RXE_MAX_MR                      = DEFAULT_MAX_VALUE - RXE_MIN_MR_INDEX,

Bob, were you saying this was what needed to be bigger to pass blktests??

Jason

author
#5 lore raw
Quoted 30 lines · Zhu Yanjun
Quoted 29 lines · Shoaib Rao
Quoted 17 lines · Zhu Yanjun
Quoted 16 lines · Jason Gunthorpe
Quoted 15 lines · Rao Shoaib

In our internal testing we have found that default maximum values are too small. Ideally there should be no limits, but since maximum values are reported via ibv_query_device, we have to return some value. So, the default maximums have been changed to large values.

---

Resubmitting the patch after applying Bob's latest patches and testing using via rping.

   drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
   1 file changed, 16 insertions(+), 14 deletions(-)
Quoted 8 lines
Quoted 1 line

So are we good with this? Bob? Zhu?

I have already checked this commit. And I have found 2 problems with this commit.
This commit changes many MAXs.
And now rxe is not stable enough. Not sure this commit will cause the new problems.

Zhu Yanjun

Hi Zhu,

Quoted 8 lines
Quoted 4 lines

A generic statement without any technical data does not help. As far as I am aware, currently there are no outstanding issues. If there are, please provide data that clearly shows that the issue is caused by this patch.

Hi, Shoaib

With this commit, I found 2 problems.
This is why I suspect that this commit will introduce risks.

Hi Zhu,

I did full testing before I sent the patch, that is how I found that rping did not work. What are the issues that you found? How to I reproduce those issues?

Shoaib

Quoted 16 lines

Before a commit is sent to the upstream, please make full tests with it.

Zhu Yanjun

Quoted 11 lines

Thanks you.

Shoaib

Quoted 7 lines
Quoted 7 lines

drivers/infiniband/sw/rxe/rxe_param.h enum rxe_device_param { · line 95

Quoted 3 lines
drivers/infiniband/sw/rxe/rxe_param.h +2 -1
-     RXE_MAX_MR_INDEX                = 0x00010000,
+     RXE_MAX_MR_INDEX                = DEFAULT_MAX_VALUE,
+     RXE_MAX_MR                      = DEFAULT_MAX_VALUE - RXE_MIN_MR_INDEX,

Bob, were you saying this was what needed to be bigger to pass blktests??

Jason

maintainer
#6 lore raw
Quoted 39 lines · Shoaib Rao
Quoted 36 lines · Zhu Yanjun
Quoted 26 lines · Shoaib Rao
Quoted 24 lines · Zhu Yanjun
Quoted 11 lines · Jason Gunthorpe
Quoted 10 lines · Rao Shoaib

In our internal testing we have found that default maximum values are too small. Ideally there should be no limits, but since maximum values are reported via ibv_query_device, we have to return some value. So, the default maximums have been changed to large values.

---

Quoted 6 lines
Quoted 5 lines

Resubmitting the patch after applying Bob's latest patches and testing using via rping.

   drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
   1 file changed, 16 insertions(+), 14 deletions(-)

So are we good with this? Bob? Zhu?

I have already checked this commit. And I have found 2 problems with this commit.
This commit changes many MAXs.
And now rxe is not stable enough. Not sure this commit will cause the new problems.

Quoted 7 lines
Quoted 1 line

Zhu Yanjun

Hi Zhu,

A generic statement without any technical data does not help. As far as I am aware, currently there are no outstanding issues. If there are, please provide data that clearly shows that the issue is caused by this patch.

Hi, Shoaib

Quoted 8 lines
Quoted 2 lines

With this commit, I found 2 problems.
This is why I suspect that this commit will introduce risks.

Hi Zhu,

I did full testing before I sent the patch, that is how I found that rping did not work. What are the issues that you found? How to I reproduce those issues?

Sorry. What tests do you make?

Do you make tests with the followings:

1. your commit + latest kernel <------rping------- > 5.10 stable kernel
2. your commit + latest kernel < ------rping------- > 5.11 stable kernel
3. your commit + latest kernel < ------rping------- > 5.12 stable kernel
4. your commit + latest kernel < ------rping------- > 5.13 stable kernel
5. your commit + latest kernel < ------rping------- > 5.14 stable kernel
6. rdma-core tests with your commit + latest kernel

Zhu Yanjun

Quoted 19 lines

Shoaib

Quoted 16 lines

Before a commit is sent to the upstream, please make full tests with it.

Zhu Yanjun

Quoted 11 lines

Thanks you.

Shoaib

Quoted 7 lines
Quoted 7 lines

drivers/infiniband/sw/rxe/rxe_param.h enum rxe_device_param { · line 95

Quoted 3 lines
drivers/infiniband/sw/rxe/rxe_param.h +2 -1
-     RXE_MAX_MR_INDEX                = 0x00010000,
+     RXE_MAX_MR_INDEX                = DEFAULT_MAX_VALUE,
+     RXE_MAX_MR                      = DEFAULT_MAX_VALUE - RXE_MIN_MR_INDEX,

Bob, were you saying this was what needed to be bigger to pass blktests??

Jason

author
#7 lore raw
Quoted 49 lines · Zhu Yanjun
Quoted 37 lines · Shoaib Rao
Quoted 35 lines · Zhu Yanjun
Quoted 25 lines · Shoaib Rao
Quoted 24 lines · Zhu Yanjun
Quoted 11 lines · Jason Gunthorpe
Quoted 10 lines · Rao Shoaib

In our internal testing we have found that default maximum values are too small. Ideally there should be no limits, but since maximum values are reported via ibv_query_device, we have to return some value. So, the default maximums have been changed to large values.

---

Quoted 6 lines
Quoted 5 lines

Resubmitting the patch after applying Bob's latest patches and testing using via rping.

    drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
    1 file changed, 16 insertions(+), 14 deletions(-)

So are we good with this? Bob? Zhu?

I have already checked this commit. And I have found 2 problems with this commit.
This commit changes many MAXs.
And now rxe is not stable enough. Not sure this commit will cause the new problems.

Quoted 7 lines
Quoted 1 line

Zhu Yanjun

Hi Zhu,

A generic statement without any technical data does not help. As far as I am aware, currently there are no outstanding issues. If there are, please provide data that clearly shows that the issue is caused by this patch.

Hi, Shoaib

Quoted 7 lines
Quoted 2 lines

With this commit, I found 2 problems.
This is why I suspect that this commit will introduce risks.

Hi Zhu,

I did full testing before I sent the patch, that is how I found that rping did not work. What are the issues that you found? How to I reproduce those issues?

Sorry. What tests do you make?

Do you make tests with the followings:

Quoted 8 lines

1. your commit + latest kernel <------rping------- > 5.10 stable kernel
2. your commit + latest kernel < ------rping------- > 5.11 stable kernel
3. your commit + latest kernel < ------rping------- > 5.12 stable kernel
4. your commit + latest kernel < ------rping------- > 5.13 stable kernel
5. your commit + latest kernel < ------rping------- > 5.14 stable kernel
6. rdma-core tests with your commit + latest kernel

Zhu Yanjun

Hi Zhu,

With all due respect, I am a little surprised by the special treatment being given to this rather simple patch. Normally comments to a patch are submitted with technical data to back them up. In this case none have been provided. If we are going by gut feelings, then there are more involved patches that are being accepted without any tests as listed above.

My initial patch increased the value of 3 variables but the community suggested that values of other variables should be increased as well. The discussion happened on this mailing list and no objections were raised.

The two issues that you mention were mechanical issues, (one reported by you and one by the kernel bot) that have been fixed. They had nothing to do with the values being increased. As I have said I have tested the patch several times, most recently about a week or so ago with Bob's latest change.

I am not keen on changing the values of any other parameters, but the 3 that were contained in my original patch. The link to those patches is

https://www.spinics.net/lists/linux-rdma/msg103570.html

Please let me know if those are acceptable. They have been tested *extensively* running several different kinds of Oracle DB workloads.

Regards,

Shoaib

Quoted 18 lines
Quoted 18 lines

Shoaib

Quoted 15 lines

Before a commit is sent to the upstream, please make full tests with it.

Zhu Yanjun

Quoted 11 lines

Thanks you.

Shoaib

Quoted 7 lines
Quoted 7 lines

Review

Quoted 3 lines
(quoted diff) +2 -1
-     RXE_MAX_MR_INDEX                = 0x00010000,
+     RXE_MAX_MR_INDEX                = DEFAULT_MAX_VALUE,
+     RXE_MAX_MR                      = DEFAULT_MAX_VALUE - RXE_MIN_MR_INDEX,

Bob, were you saying this was what needed to be bigger to pass blktests??

Jason

maintainer
#8 lore raw
Quoted 61 lines · Shoaib Rao
Quoted 49 lines · Zhu Yanjun
Quoted 37 lines · Shoaib Rao
Quoted 35 lines · Zhu Yanjun
Quoted 25 lines · Shoaib Rao
Quoted 24 lines · Zhu Yanjun
Quoted 11 lines · Jason Gunthorpe
Quoted 10 lines · Rao Shoaib

In our internal testing we have found that default maximum values are too small. Ideally there should be no limits, but since maximum values are reported via ibv_query_device, we have to return some value. So, the default maximums have been changed to large values.

---

Quoted 6 lines
Quoted 5 lines

Resubmitting the patch after applying Bob's latest patches and testing using via rping.

    drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
    1 file changed, 16 insertions(+), 14 deletions(-)

So are we good with this? Bob? Zhu?

I have already checked this commit. And I have found 2 problems with this commit.
This commit changes many MAXs.
And now rxe is not stable enough. Not sure this commit will cause the new problems.

Quoted 7 lines
Quoted 1 line

Zhu Yanjun

Hi Zhu,

A generic statement without any technical data does not help. As far as I am aware, currently there are no outstanding issues. If there are, please provide data that clearly shows that the issue is caused by this patch.

Hi, Shoaib

Quoted 7 lines
Quoted 2 lines

With this commit, I found 2 problems.
This is why I suspect that this commit will introduce risks.

Hi Zhu,

I did full testing before I sent the patch, that is how I found that rping did not work. What are the issues that you found? How to I reproduce those issues?

Sorry. What tests do you make?

Do you make tests with the followings:

Quoted 8 lines

1. your commit + latest kernel <------rping------- > 5.10 stable kernel
2. your commit + latest kernel < ------rping------- > 5.11 stable kernel
3. your commit + latest kernel < ------rping------- > 5.12 stable kernel
4. your commit + latest kernel < ------rping------- > 5.13 stable kernel
5. your commit + latest kernel < ------rping------- > 5.14 stable kernel
6. rdma-core tests with your commit + latest kernel

Zhu Yanjun

Quoted 7 lines

Hi Zhu,

With all due respect, I am a little surprised by the special treatment being given to this rather simple patch. Normally comments to a patch are submitted with technical data to back them up. In this case none have been provided. If we are going by gut feelings, then there are more involved patches that are being accepted without any tests as listed above.

All the commits that I reviewed will pass the mentioned tests. But your commit failed to pass the tests at least 2 times. So I suggest that you make the above tests in your local hosts. This will save us time.
I do not treat your commit specially.

Quoted 10 lines

My initial patch increased the value of 3 variables but the community suggested that values of other variables should be increased as well. The discussion happened on this mailing list and no objections were raised.

The two issues that you mention were mechanical issues, (one reported by you and one by the kernel bot) that have been fixed. They had nothing to do with the values being increased. As I have said I have tested the patch several times, most recently about a week or so ago with Bob's latest change.

I do not know how you tested your patch in your host. If your tests include backward compatibility and rdma-core tests that I mentioned in the above mail, I am fine with this commit.

Zhu Yanjun

Quoted 31 lines

I am not keen on changing the values of any other parameters, but the 3 that were contained in my original patch. The link to those patches is

https://www.spinics.net/lists/linux-rdma/msg103570.html

Please let me know if those are acceptable. They have been tested *extensively* running several different kinds of Oracle DB workloads.

Regards,

Shoaib

Quoted 18 lines
Quoted 18 lines

Shoaib

Quoted 15 lines

Before a commit is sent to the upstream, please make full tests with it.

Zhu Yanjun

Quoted 11 lines

Thanks you.

Shoaib

Quoted 7 lines
Quoted 7 lines

Review

Quoted 3 lines
(quoted diff) +2 -1
-     RXE_MAX_MR_INDEX                = 0x00010000,
+     RXE_MAX_MR_INDEX                = DEFAULT_MAX_VALUE,
+     RXE_MAX_MR                      = DEFAULT_MAX_VALUE - RXE_MIN_MR_INDEX,

Bob, were you saying this was what needed to be bigger to pass blktests??

Jason

author
#9 lore raw
Quoted 80 lines · Zhu Yanjun
Quoted 57 lines · Shoaib Rao
Quoted 55 lines · Zhu Yanjun
Quoted 36 lines · Shoaib Rao
Quoted 35 lines · Zhu Yanjun
Quoted 25 lines · Shoaib Rao
Quoted 24 lines · Zhu Yanjun
Quoted 11 lines · Jason Gunthorpe
Quoted 10 lines · Rao Shoaib

In our internal testing we have found that default maximum values are too small. Ideally there should be no limits, but since maximum values are reported via ibv_query_device, we have to return some value. So, the default maximums have been changed to large values.

---

Quoted 6 lines
Quoted 5 lines

Resubmitting the patch after applying Bob's latest patches and testing using via rping.

     drivers/infiniband/sw/rxe/rxe_param.h | 30 ++++++++++++++-------------
     1 file changed, 16 insertions(+), 14 deletions(-)

So are we good with this? Bob? Zhu?

I have already checked this commit. And I have found 2 problems with this commit.
This commit changes many MAXs.
And now rxe is not stable enough. Not sure this commit will cause the new problems.

Quoted 7 lines
Quoted 1 line

Zhu Yanjun

Hi Zhu,

A generic statement without any technical data does not help. As far as I am aware, currently there are no outstanding issues. If there are, please provide data that clearly shows that the issue is caused by this patch.

Hi, Shoaib

Quoted 7 lines
Quoted 2 lines

With this commit, I found 2 problems.
This is why I suspect that this commit will introduce risks.

Hi Zhu,

I did full testing before I sent the patch, that is how I found that rping did not work. What are the issues that you found? How to I reproduce those issues?

Sorry. What tests do you make?

Do you make tests with the followings:

1. your commit + latest kernel <------rping------- > 5.10 stable kernel
2. your commit + latest kernel < ------rping------- > 5.11 stable kernel
3. your commit + latest kernel < ------rping------- > 5.12 stable kernel
4. your commit + latest kernel < ------rping------- > 5.13 stable kernel
5. your commit + latest kernel < ------rping------- > 5.14 stable kernel
6. rdma-core tests with your commit + latest kernel

Quoted 8 lines
Quoted 1 line

Zhu Yanjun

Hi Zhu,

With all due respect, I am a little surprised by the special treatment being given to this rather simple patch. Normally comments to a patch are submitted with technical data to back them up. In this case none have been provided. If we are going by gut feelings, then there are more involved patches that are being accepted without any tests as listed above.

All the commits that I reviewed will pass the mentioned tests. But your commit failed to pass the tests at least 2 times. So I suggest that you make the above tests in your local hosts. This will save us time.
I do not treat your commit specially.

Quoted 9 lines

My initial patch increased the value of 3 variables but the community suggested that values of other variables should be increased as well. The discussion happened on this mailing list and no objections were raised.

The two issues that you mention were mechanical issues, (one reported by you and one by the kernel bot) that have been fixed. They had nothing to do with the values being increased. As I have said I have tested the patch several times, most recently about a week or so ago with Bob's latest change.

Quoted 5 lines

I do not know how you tested your patch in your host. If your tests include backward compatibility and rdma-core tests that I mentioned in the above mail, I am fine with this commit.

Zhu Yanjun

Thanks. I did test backward compatibility. If any issues are found in the future I will fix them.

Regards,

Shoaib

Quoted 30 lines
Quoted 30 lines

I am not keen on changing the values of any other parameters, but the 3 that were contained in my original patch. The link to those patches is

https://urldefense.com/v3/__https://www.spinics.net/lists/linux-rdma/msg103570.html__;!!ACWV5N9M2RV99hQ!c_05bC74v_nDObwKFnI2y3b5EuwvlSIq8hKf_4iXStmPVUk6qxw1_Jii_AC7oiui$

Please let me know if those are acceptable. They have been tested *extensively* running several different kinds of Oracle DB workloads.

Regards,

Shoaib

Quoted 17 lines
Quoted 17 lines

Shoaib

Quoted 15 lines

Before a commit is sent to the upstream, please make full tests with it.

Zhu Yanjun

Quoted 11 lines

Thanks you.

Shoaib

Quoted 7 lines
Quoted 7 lines

Review

Quoted 3 lines
(quoted diff) +2 -1
-     RXE_MAX_MR_INDEX                = 0x00010000,
+     RXE_MAX_MR_INDEX                = DEFAULT_MAX_VALUE,
+     RXE_MAX_MR                      = DEFAULT_MAX_VALUE - RXE_MIN_MR_INDEX,

Bob, were you saying this was what needed to be bigger to pass blktests??

Jason

maintainer Applied
#10 lore raw
In reply to Rao Shoaib
Quoted 7 lines · Rao Shoaib

In our internal testing we have found that default maximum values are too small. Ideally there should be no limits, but since maximum values are reported via ibv_query_device, we have to return some value. So, the default maximums have been changed to large values.

Quoted 5 lines

---

Resubmitting the patch after applying Bob's latest patches and testing using via rping.

Since nobody points to something wrong with this and Bob says it fixes a blktests problem..

Applied to for-next, thanks

Jason