3.19-rc1 net: less interrupt masking in NAPI

BUG In review Oded Gabbay 4 participants 11 messages Jan 10, 2015 – Jan 12, 2015

Subsystems (3)

Networking Network drivers Atlx Ethernet drivers

lore

Share & tools

lore

Oded Gabbay

author

2015-01-10 20:39 UTC #0 lore raw

Hi,

Commit d75b1ade567ffab085e8adbbdacf0092d10cd09c breaks my "Qualcomm Atheros AR8161 Gigabit Ethernet (rev 10)" Ethernet controller, which is handled by the alx network driver.

ogabbay@odedg-ubuntu:~$ lspci -s 01:00.0 -k
01:00.0 Ethernet controller:
	Qualcomm Atheros AR8161 Gigabit Ethernet (rev 10)
	Subsystem: Qualcomm Atheros Device 1071
	Kernel driver in use: alx

I have this controller on a mobile platform of AMD APU Kaveri, which I use to test amdkfd, and from 3.19-rc1 the network stopped working when trying to transfer files through scp or nfs.

I bisected the kernel (from 3.18.0 to 3.19-rc1) and reached this commit.

Here is the log of the bisect:

git bisect start
# bad: [97bf6af1f928216fd6c5a66e8a57bfa95a659672] Linux 3.19-rc1 git bisect bad 97bf6af1f928216fd6c5a66e8a57bfa95a659672

# good: [b2776bf7149bddd1f4161f14f79520f17fc1d71d] Linux 3.18 git bisect good b2776bf7149bddd1f4161f14f79520f17fc1d71d

# bad: [70e71ca0af244f48a5dcf56dc435243792e3a495] Merge
git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next
git bisect bad 70e71ca0af244f48a5dcf56dc435243792e3a495

# good: [e28870f9b3e92cd3570925089c6bb789c2603bc4] Merge tag 'backlight-for-linus-3.19' of
git://git.kernel.org/pub/scm/linux/kernel/git/lee/backlight
git bisect good e28870f9b3e92cd3570925089c6bb789c2603bc4

# bad: [450fa21942fe2c37f0c9f52d1a33bbc081eee288] sh_eth: Remove redundant alignment adjustment
git bisect bad 450fa21942fe2c37f0c9f52d1a33bbc081eee288

# bad: [5c8d19da950861d0482abc0ac3481acca34b008f] e100e: use netdev_rss_key_fill() helper
git bisect bad 5c8d19da950861d0482abc0ac3481acca34b008f

# good: [bf515fb11ab539c76d04f0e3c5216ed41f41d81f] Merge tag 'mac80211-next-for-john-2014-11-04' of
git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211-next
git bisect good bf515fb11ab539c76d04f0e3c5216ed41f41d81f

# bad: [2c99cd914d4fed9160d98849c9dd38034616768e] Merge branch 'amd-xgbe-next' git bisect bad 2c99cd914d4fed9160d98849c9dd38034616768e

# good: [3d762a0f0ab9cb4a6b5993db3ce56c92f9f90ab2] net: dsa: Add support for reading switch registers with ethtool
git bisect good 3d762a0f0ab9cb4a6b5993db3ce56c92f9f90ab2

# bad: [8ce0c8254f15229aa99fc6c04141f28c446e5f8c] Merge branch 'master' of
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net-next
git bisect bad 8ce0c8254f15229aa99fc6c04141f28c446e5f8c

# good: [f0c65567b3c1b23f79e8a49139580a3872a68d1f] Merge branch 'sunvnet-multi-tx-queue'
git bisect good f0c65567b3c1b23f79e8a49139580a3872a68d1f

# bad: [547f2735c20023d7b50a791b1b17cacb652e9237] Merge branch 'mlx4-next' git bisect bad 547f2735c20023d7b50a791b1b17cacb652e9237

# good: [4cdb1e2e3d3495423db558d3bb7ed11d66aabce7] net: shrink struct softnet_data
git bisect good 4cdb1e2e3d3495423db558d3bb7ed11d66aabce7

# bad: [0a98455666ec87378148a1dde97f1ce5baf75a64] net/mlx4_core: Protect port type setting by mutex
git bisect bad 0a98455666ec87378148a1dde97f1ce5baf75a64

# bad: [6e8066999800d90d52af5c84ac49ebf683d14cdc] net/mlx4_core: Prevent VF from changing port configuration
git bisect bad 6e8066999800d90d52af5c84ac49ebf683d14cdc

# bad: [d75b1ade567ffab085e8adbbdacf0092d10cd09c] net: less interrupt masking in NAPI
git bisect bad d75b1ade567ffab085e8adbbdacf0092d10cd09c

# first bad commit: [d75b1ade567ffab085e8adbbdacf0092d10cd09c]
net: less interrupt masking in NAPI

Could you please solve this issue as it renders my board quite useless.

If you need more info, please ask.

Thanks,
Oded

Eric Dumazet

2015-01-10 20:58 UTC #1 lore raw

Quoted 79 lines · Oded Gabbay

Hi,
Commit d75b1ade567ffab085e8adbbdacf0092d10cd09c breaks my "Qualcomm Atheros AR8161 Gigabit Ethernet (rev 10)" Ethernet controller, which is handled by the alx network driver.
ogabbay@odedg-ubuntu:~$ lspci -s 01:00.0 -k
01:00.0 Ethernet controller:
	Qualcomm Atheros AR8161 Gigabit Ethernet (rev 10)
	Subsystem: Qualcomm Atheros Device 1071
	Kernel driver in use: alx
I have this controller on a mobile platform of AMD APU Kaveri, which I use to test amdkfd, and from 3.19-rc1 the network stopped working when trying to transfer files through scp or nfs.
I bisected the kernel (from 3.18.0 to 3.19-rc1) and reached this commit.
Here is the log of the bisect:
git bisect start
# bad: [97bf6af1f928216fd6c5a66e8a57bfa95a659672] Linux 3.19-rc1 git bisect bad 97bf6af1f928216fd6c5a66e8a57bfa95a659672
# good: [b2776bf7149bddd1f4161f14f79520f17fc1d71d] Linux 3.18 git bisect good b2776bf7149bddd1f4161f14f79520f17fc1d71d
# bad: [70e71ca0af244f48a5dcf56dc435243792e3a495] Merge
git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next
git bisect bad 70e71ca0af244f48a5dcf56dc435243792e3a495
# good: [e28870f9b3e92cd3570925089c6bb789c2603bc4] Merge tag 'backlight-for-linus-3.19' of
git://git.kernel.org/pub/scm/linux/kernel/git/lee/backlight
git bisect good e28870f9b3e92cd3570925089c6bb789c2603bc4
# bad: [450fa21942fe2c37f0c9f52d1a33bbc081eee288] sh_eth: Remove redundant alignment adjustment
git bisect bad 450fa21942fe2c37f0c9f52d1a33bbc081eee288
# bad: [5c8d19da950861d0482abc0ac3481acca34b008f] e100e: use netdev_rss_key_fill() helper
git bisect bad 5c8d19da950861d0482abc0ac3481acca34b008f
# good: [bf515fb11ab539c76d04f0e3c5216ed41f41d81f] Merge tag 'mac80211-next-for-john-2014-11-04' of
git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211-next
git bisect good bf515fb11ab539c76d04f0e3c5216ed41f41d81f
# bad: [2c99cd914d4fed9160d98849c9dd38034616768e] Merge branch 'amd-xgbe-next' git bisect bad 2c99cd914d4fed9160d98849c9dd38034616768e
# good: [3d762a0f0ab9cb4a6b5993db3ce56c92f9f90ab2] net: dsa: Add support for reading switch registers with ethtool
git bisect good 3d762a0f0ab9cb4a6b5993db3ce56c92f9f90ab2
# bad: [8ce0c8254f15229aa99fc6c04141f28c446e5f8c] Merge branch 'master' of
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net-next
git bisect bad 8ce0c8254f15229aa99fc6c04141f28c446e5f8c
# good: [f0c65567b3c1b23f79e8a49139580a3872a68d1f] Merge branch 'sunvnet-multi-tx-queue'
git bisect good f0c65567b3c1b23f79e8a49139580a3872a68d1f
# bad: [547f2735c20023d7b50a791b1b17cacb652e9237] Merge branch 'mlx4-next' git bisect bad 547f2735c20023d7b50a791b1b17cacb652e9237
# good: [4cdb1e2e3d3495423db558d3bb7ed11d66aabce7] net: shrink struct softnet_data
git bisect good 4cdb1e2e3d3495423db558d3bb7ed11d66aabce7
# bad: [0a98455666ec87378148a1dde97f1ce5baf75a64] net/mlx4_core: Protect port type setting by mutex
git bisect bad 0a98455666ec87378148a1dde97f1ce5baf75a64
# bad: [6e8066999800d90d52af5c84ac49ebf683d14cdc] net/mlx4_core: Prevent VF from changing port configuration
git bisect bad 6e8066999800d90d52af5c84ac49ebf683d14cdc

Quoted 9 lines

# bad: [d75b1ade567ffab085e8adbbdacf0092d10cd09c] net: less interrupt masking in NAPI
git bisect bad d75b1ade567ffab085e8adbbdacf0092d10cd09c
# first bad commit: [d75b1ade567ffab085e8adbbdacf0092d10cd09c]
net: less interrupt masking in NAPI
Could you please solve this issue as it renders my board quite useless.

Thanks for the report and bisection !

Could you try following fix ?

drivers/net/ethernet/atheros/alx/main.c +1 -1

index e398eda07298..209c40765e0d 100644
@@ -272,7 +272,7 @@ static int alx_poll(struct napi_struct *napi, int budget)
 		   alx_clean_rx_irq(alx, budget);
 
 	if (!complete)
-		return 1;
+		return budget;
 
 	napi_complete(&alx->napi);

Eric Dumazet

2015-01-10 21:10 UTC #2 lore raw

Quoted 102 lines · Eric Dumazet

Quoted 79 lines · Oded Gabbay
Hi,
Commit d75b1ade567ffab085e8adbbdacf0092d10cd09c breaks my "Qualcomm Atheros AR8161 Gigabit Ethernet (rev 10)" Ethernet controller, which is handled by the alx network driver.
ogabbay@odedg-ubuntu:~$ lspci -s 01:00.0 -k
01:00.0 Ethernet controller:
	Qualcomm Atheros AR8161 Gigabit Ethernet (rev 10)
	Subsystem: Qualcomm Atheros Device 1071
	Kernel driver in use: alx
I have this controller on a mobile platform of AMD APU Kaveri, which I use to test amdkfd, and from 3.19-rc1 the network stopped working when trying to transfer files through scp or nfs.
I bisected the kernel (from 3.18.0 to 3.19-rc1) and reached this commit.
Here is the log of the bisect:
git bisect start
# bad: [97bf6af1f928216fd6c5a66e8a57bfa95a659672] Linux 3.19-rc1 git bisect bad 97bf6af1f928216fd6c5a66e8a57bfa95a659672
# good: [b2776bf7149bddd1f4161f14f79520f17fc1d71d] Linux 3.18 git bisect good b2776bf7149bddd1f4161f14f79520f17fc1d71d
# bad: [70e71ca0af244f48a5dcf56dc435243792e3a495] Merge
git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next
git bisect bad 70e71ca0af244f48a5dcf56dc435243792e3a495
# good: [e28870f9b3e92cd3570925089c6bb789c2603bc4] Merge tag 'backlight-for-linus-3.19' of
git://git.kernel.org/pub/scm/linux/kernel/git/lee/backlight
git bisect good e28870f9b3e92cd3570925089c6bb789c2603bc4
# bad: [450fa21942fe2c37f0c9f52d1a33bbc081eee288] sh_eth: Remove redundant alignment adjustment
git bisect bad 450fa21942fe2c37f0c9f52d1a33bbc081eee288
# bad: [5c8d19da950861d0482abc0ac3481acca34b008f] e100e: use netdev_rss_key_fill() helper
git bisect bad 5c8d19da950861d0482abc0ac3481acca34b008f
# good: [bf515fb11ab539c76d04f0e3c5216ed41f41d81f] Merge tag 'mac80211-next-for-john-2014-11-04' of
git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211-next
git bisect good bf515fb11ab539c76d04f0e3c5216ed41f41d81f
# bad: [2c99cd914d4fed9160d98849c9dd38034616768e] Merge branch 'amd-xgbe-next' git bisect bad 2c99cd914d4fed9160d98849c9dd38034616768e
# good: [3d762a0f0ab9cb4a6b5993db3ce56c92f9f90ab2] net: dsa: Add support for reading switch registers with ethtool
git bisect good 3d762a0f0ab9cb4a6b5993db3ce56c92f9f90ab2
# bad: [8ce0c8254f15229aa99fc6c04141f28c446e5f8c] Merge branch 'master' of
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net-next
git bisect bad 8ce0c8254f15229aa99fc6c04141f28c446e5f8c
# good: [f0c65567b3c1b23f79e8a49139580a3872a68d1f] Merge branch 'sunvnet-multi-tx-queue'
git bisect good f0c65567b3c1b23f79e8a49139580a3872a68d1f
# bad: [547f2735c20023d7b50a791b1b17cacb652e9237] Merge branch 'mlx4-next' git bisect bad 547f2735c20023d7b50a791b1b17cacb652e9237
# good: [4cdb1e2e3d3495423db558d3bb7ed11d66aabce7] net: shrink struct softnet_data
git bisect good 4cdb1e2e3d3495423db558d3bb7ed11d66aabce7
# bad: [0a98455666ec87378148a1dde97f1ce5baf75a64] net/mlx4_core: Protect port type setting by mutex
git bisect bad 0a98455666ec87378148a1dde97f1ce5baf75a64
# bad: [6e8066999800d90d52af5c84ac49ebf683d14cdc] net/mlx4_core: Prevent VF from changing port configuration
git bisect bad 6e8066999800d90d52af5c84ac49ebf683d14cdc
Quoted 9 lines
# bad: [d75b1ade567ffab085e8adbbdacf0092d10cd09c] net: less interrupt masking in NAPI
git bisect bad d75b1ade567ffab085e8adbbdacf0092d10cd09c
# first bad commit: [d75b1ade567ffab085e8adbbdacf0092d10cd09c]
net: less interrupt masking in NAPI
Could you please solve this issue as it renders my board quite useless.
Thanks for the report and bisection !
Could you try following fix ?
drivers/net/ethernet/atheros/alx/main.c +0 -0
index e398eda07298..209c40765e0d 100644

drivers/net/ethernet/atheros/alx/main.c static int alx_poll(struct napi_struct *napi, int budget) · line 278

Quoted 8 lines

drivers/net/ethernet/atheros/alx/main.c +1 -1

@@ -272,7 +272,7 @@ static int alx_poll(struct napi_struct *napi, int budget)
 		   alx_clean_rx_irq(alx, budget);
 
 	if (!complete)
-		return 1;
+		return budget;
 
 	napi_complete(&alx->napi);

BTW this driver has other issues :

complete = alx_clean_tx_irq(alx) &&

            alx_clean_rx_irq(alx, budget);

Means that under TX completion pressure (alx_clean_tx_irq(alx) return false), we never dequeue packets from RX rings.

Oded Gabbay

author

2015-01-10 21:30 UTC #3 lore raw

In reply to Eric Dumazet

Quoted 103 lines · Eric Dumazet

Quoted 79 lines · Oded Gabbay
Hi,
Commit d75b1ade567ffab085e8adbbdacf0092d10cd09c breaks my "Qualcomm Atheros AR8161 Gigabit Ethernet (rev 10)" Ethernet controller, which is handled by the alx network driver.
ogabbay@odedg-ubuntu:~$ lspci -s 01:00.0 -k
01:00.0 Ethernet controller:
	Qualcomm Atheros AR8161 Gigabit Ethernet (rev 10)
	Subsystem: Qualcomm Atheros Device 1071
	Kernel driver in use: alx
I have this controller on a mobile platform of AMD APU Kaveri, which I use to test amdkfd, and from 3.19-rc1 the network stopped working when trying to transfer files through scp or nfs.
I bisected the kernel (from 3.18.0 to 3.19-rc1) and reached this commit.
Here is the log of the bisect:
git bisect start
# bad: [97bf6af1f928216fd6c5a66e8a57bfa95a659672] Linux 3.19-rc1 git bisect bad 97bf6af1f928216fd6c5a66e8a57bfa95a659672
# good: [b2776bf7149bddd1f4161f14f79520f17fc1d71d] Linux 3.18 git bisect good b2776bf7149bddd1f4161f14f79520f17fc1d71d
# bad: [70e71ca0af244f48a5dcf56dc435243792e3a495] Merge
git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next
git bisect bad 70e71ca0af244f48a5dcf56dc435243792e3a495
# good: [e28870f9b3e92cd3570925089c6bb789c2603bc4] Merge tag 'backlight-for-linus-3.19' of
git://git.kernel.org/pub/scm/linux/kernel/git/lee/backlight
git bisect good e28870f9b3e92cd3570925089c6bb789c2603bc4
# bad: [450fa21942fe2c37f0c9f52d1a33bbc081eee288] sh_eth: Remove redundant alignment adjustment
git bisect bad 450fa21942fe2c37f0c9f52d1a33bbc081eee288
# bad: [5c8d19da950861d0482abc0ac3481acca34b008f] e100e: use netdev_rss_key_fill() helper
git bisect bad 5c8d19da950861d0482abc0ac3481acca34b008f
# good: [bf515fb11ab539c76d04f0e3c5216ed41f41d81f] Merge tag 'mac80211-next-for-john-2014-11-04' of
git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211-next
git bisect good bf515fb11ab539c76d04f0e3c5216ed41f41d81f
# bad: [2c99cd914d4fed9160d98849c9dd38034616768e] Merge branch 'amd-xgbe-next' git bisect bad 2c99cd914d4fed9160d98849c9dd38034616768e
# good: [3d762a0f0ab9cb4a6b5993db3ce56c92f9f90ab2] net: dsa: Add support for reading switch registers with ethtool
git bisect good 3d762a0f0ab9cb4a6b5993db3ce56c92f9f90ab2
# bad: [8ce0c8254f15229aa99fc6c04141f28c446e5f8c] Merge branch 'master' of
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net-next
git bisect bad 8ce0c8254f15229aa99fc6c04141f28c446e5f8c
# good: [f0c65567b3c1b23f79e8a49139580a3872a68d1f] Merge branch 'sunvnet-multi-tx-queue'
git bisect good f0c65567b3c1b23f79e8a49139580a3872a68d1f
# bad: [547f2735c20023d7b50a791b1b17cacb652e9237] Merge branch 'mlx4-next' git bisect bad 547f2735c20023d7b50a791b1b17cacb652e9237
# good: [4cdb1e2e3d3495423db558d3bb7ed11d66aabce7] net: shrink struct softnet_data
git bisect good 4cdb1e2e3d3495423db558d3bb7ed11d66aabce7
# bad: [0a98455666ec87378148a1dde97f1ce5baf75a64] net/mlx4_core: Protect port type setting by mutex
git bisect bad 0a98455666ec87378148a1dde97f1ce5baf75a64
# bad: [6e8066999800d90d52af5c84ac49ebf683d14cdc] net/mlx4_core: Prevent VF from changing port configuration
git bisect bad 6e8066999800d90d52af5c84ac49ebf683d14cdc
Quoted 9 lines
# bad: [d75b1ade567ffab085e8adbbdacf0092d10cd09c] net: less interrupt masking in NAPI
git bisect bad d75b1ade567ffab085e8adbbdacf0092d10cd09c
# first bad commit: [d75b1ade567ffab085e8adbbdacf0092d10cd09c]
net: less interrupt masking in NAPI
Could you please solve this issue as it renders my board quite useless.
Thanks for the report and bisection !
Could you try following fix ?
drivers/net/ethernet/atheros/alx/main.c +0 -0
index e398eda07298..209c40765e0d 100644

drivers/net/ethernet/atheros/alx/main.c static int alx_poll(struct napi_struct *napi, int budget) · line 278

Quoted 8 lines

drivers/net/ethernet/atheros/alx/main.c +1 -1

@@ -272,7 +272,7 @@ static int alx_poll(struct napi_struct *napi, int budget)
 		   alx_clean_rx_irq(alx, budget);
 
 	if (!complete)
-		return 1;
+		return budget;
 
 	napi_complete(&alx->napi);

Yes, no problem.
I will update on the result.

Oded

Oded Gabbay

author

2015-01-10 21:43 UTC #4 lore raw

Quoted 117 lines · Oded Gabbay

Quoted 103 lines · Eric Dumazet
Quoted 79 lines · Oded Gabbay
Hi,
Commit d75b1ade567ffab085e8adbbdacf0092d10cd09c breaks my "Qualcomm Atheros AR8161 Gigabit Ethernet (rev 10)" Ethernet controller, which is handled by the alx network driver.
ogabbay@odedg-ubuntu:~$ lspci -s 01:00.0 -k
01:00.0 Ethernet controller:
	Qualcomm Atheros AR8161 Gigabit Ethernet (rev 10)
	Subsystem: Qualcomm Atheros Device 1071
	Kernel driver in use: alx
I have this controller on a mobile platform of AMD APU Kaveri, which I use to test amdkfd, and from 3.19-rc1 the network stopped working when trying to transfer files through scp or nfs.
I bisected the kernel (from 3.18.0 to 3.19-rc1) and reached this commit.
Here is the log of the bisect:
git bisect start
# bad: [97bf6af1f928216fd6c5a66e8a57bfa95a659672] Linux 3.19-rc1 git bisect bad 97bf6af1f928216fd6c5a66e8a57bfa95a659672
# good: [b2776bf7149bddd1f4161f14f79520f17fc1d71d] Linux 3.18 git bisect good b2776bf7149bddd1f4161f14f79520f17fc1d71d
# bad: [70e71ca0af244f48a5dcf56dc435243792e3a495] Merge
git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-next
git bisect bad 70e71ca0af244f48a5dcf56dc435243792e3a495
# good: [e28870f9b3e92cd3570925089c6bb789c2603bc4] Merge tag 'backlight-for-linus-3.19' of
git://git.kernel.org/pub/scm/linux/kernel/git/lee/backlight
git bisect good e28870f9b3e92cd3570925089c6bb789c2603bc4
# bad: [450fa21942fe2c37f0c9f52d1a33bbc081eee288] sh_eth: Remove redundant alignment adjustment
git bisect bad 450fa21942fe2c37f0c9f52d1a33bbc081eee288
# bad: [5c8d19da950861d0482abc0ac3481acca34b008f] e100e: use netdev_rss_key_fill() helper
git bisect bad 5c8d19da950861d0482abc0ac3481acca34b008f
# good: [bf515fb11ab539c76d04f0e3c5216ed41f41d81f] Merge tag 'mac80211-next-for-john-2014-11-04' of
git://git.kernel.org/pub/scm/linux/kernel/git/jberg/mac80211-next
git bisect good bf515fb11ab539c76d04f0e3c5216ed41f41d81f
# bad: [2c99cd914d4fed9160d98849c9dd38034616768e] Merge branch 'amd-xgbe-next' git bisect bad 2c99cd914d4fed9160d98849c9dd38034616768e
# good: [3d762a0f0ab9cb4a6b5993db3ce56c92f9f90ab2] net: dsa: Add support for reading switch registers with ethtool
git bisect good 3d762a0f0ab9cb4a6b5993db3ce56c92f9f90ab2
# bad: [8ce0c8254f15229aa99fc6c04141f28c446e5f8c] Merge branch 'master' of
git://git.kernel.org/pub/scm/linux/kernel/git/jkirsher/net-next
git bisect bad 8ce0c8254f15229aa99fc6c04141f28c446e5f8c
# good: [f0c65567b3c1b23f79e8a49139580a3872a68d1f] Merge branch 'sunvnet-multi-tx-queue'
git bisect good f0c65567b3c1b23f79e8a49139580a3872a68d1f
# bad: [547f2735c20023d7b50a791b1b17cacb652e9237] Merge branch 'mlx4-next' git bisect bad 547f2735c20023d7b50a791b1b17cacb652e9237
# good: [4cdb1e2e3d3495423db558d3bb7ed11d66aabce7] net: shrink struct softnet_data
git bisect good 4cdb1e2e3d3495423db558d3bb7ed11d66aabce7
# bad: [0a98455666ec87378148a1dde97f1ce5baf75a64] net/mlx4_core: Protect port type setting by mutex
git bisect bad 0a98455666ec87378148a1dde97f1ce5baf75a64
# bad: [6e8066999800d90d52af5c84ac49ebf683d14cdc] net/mlx4_core: Prevent VF from changing port configuration
git bisect bad 6e8066999800d90d52af5c84ac49ebf683d14cdc
Quoted 9 lines
# bad: [d75b1ade567ffab085e8adbbdacf0092d10cd09c] net: less interrupt masking in NAPI
git bisect bad d75b1ade567ffab085e8adbbdacf0092d10cd09c
# first bad commit: [d75b1ade567ffab085e8adbbdacf0092d10cd09c]
net: less interrupt masking in NAPI
Could you please solve this issue as it renders my board quite useless.
Thanks for the report and bisection !
Could you try following fix ?
drivers/net/ethernet/atheros/alx/main.c +0 -0
index e398eda07298..209c40765e0d 100644
drivers/net/ethernet/atheros/alx/main.c static int alx_poll(struct napi_struct *napi, int budget) · line 278
Quoted 8 lines
drivers/net/ethernet/atheros/alx/main.c +1 -1
@@ -272,7 +272,7 @@ static int alx_poll(struct napi_struct *napi, int budget)
 		   alx_clean_rx_irq(alx, budget);
 
 	if (!complete)
-		return 1;
+		return budget;
 
 	napi_complete(&alx->napi);
 
Yes, no problem.
I will update on the result.

Quoted 7 lines

Oded
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

Hi Eric,
Your patch fixed the problem.
Thanks for the quick help!

Oded

Eric Dumazet

2015-01-10 21:50 UTC #5 lore raw

In reply to Oded Gabbay

Quoted 2 lines · Oded Gabbay

Yes, no problem.
I will update on the result.

Please try this more complete patch, solving the TX pressure problem as well, and not lying about NAPI budget. thanks !

drivers/net/ethernet/atheros/alx/main.c +12 ~~-10~~

index e398eda07298..5f05b387c0a7 100644
@@ -184,15 +184,16 @@ static void alx_schedule_reset(struct alx_priv *alx)
 	schedule_work(&alx->reset_wk);
 }
 
-static bool alx_clean_rx_irq(struct alx_priv *alx, int budget)
+static int alx_clean_rx_irq(struct alx_priv *alx, int budget)
 {
 	struct alx_rx_queue *rxq = &alx->rxq;
 	struct alx_rrd *rrd;
 	struct alx_buffer *rxb;
 	struct sk_buff *skb;
 	u16 length, rfd_cleaned = 0;
+	int work = 0;
 
-	while (budget > 0) {
+	while (work < budget) {
 		rrd = &rxq->rrd[rxq->rrd_read_idx];
 		if (!(rrd->word3 & cpu_to_le32(1 << RRD_UPDATED_SHIFT)))
 			break;

@@ -243,7 +244,7 @@ static bool alx_clean_rx_irq(struct alx_priv *alx, int budget)
 		}
 
 		napi_gro_receive(&alx->napi, skb);
-		budget--;
+		work++;
 
 next_pkt:
 		if (++rxq->read_idx == alx->rx_ringsz)

@@ -258,21 +259,22 @@ next_pkt:
 	if (rfd_cleaned)
 		alx_refill_rx_ring(alx, GFP_ATOMIC);
 
-	return budget > 0;
+	return work;
 }
 
 static int alx_poll(struct napi_struct *napi, int budget)
 {
 	struct alx_priv *alx = container_of(napi, struct alx_priv, napi);
 	struct alx_hw *hw = &alx->hw;
-	bool complete = true;
 	unsigned long flags;
+	bool tx_complete;
+	int work;
 
-	complete = alx_clean_tx_irq(alx) &&
-		   alx_clean_rx_irq(alx, budget);
+	tx_complete = alx_clean_tx_irq(alx);
+	work = alx_clean_rx_irq(alx, budget);
 
-	if (!complete)
-		return 1;
+	if (!tx_complete || work == budget)
+		return budget;
 
 	napi_complete(&alx->napi);

@@ -284,7 +286,7 @@ static int alx_poll(struct napi_struct *napi, int budget)
 
 	alx_post_write(hw);
 
-	return 0;
+	return work;
 }
 
 static irqreturn_t alx_intr_handle(struct alx_priv *alx, u32 intr)

Oded Gabbay

author

2015-01-10 22:05 UTC #6 lore raw

Quoted 73 lines · Eric Dumazet

Quoted 2 lines · Oded Gabbay

Yes, no problem.
I will update on the result.

Please try this more complete patch, solving the TX pressure problem as well, and not lying about NAPI budget. thanks !

drivers/net/ethernet/atheros/alx/main.c +11 -9

index e398eda07298..5f05b387c0a7 100644
@@ -184,15 +184,16 @@ static void alx_schedule_reset(struct alx_priv *alx)
 	schedule_work(&alx->reset_wk);
 }
 
-static bool alx_clean_rx_irq(struct alx_priv *alx, int budget)
+static int alx_clean_rx_irq(struct alx_priv *alx, int budget)
 {
 	struct alx_rx_queue *rxq = &alx->rxq;
 	struct alx_rrd *rrd;
 	struct alx_buffer *rxb;
 	struct sk_buff *skb;
 	u16 length, rfd_cleaned = 0;
+	int work = 0;
 
-	while (budget > 0) {
+	while (work < budget) {
 		rrd = &rxq->rrd[rxq->rrd_read_idx];
 		if (!(rrd->word3 & cpu_to_le32(1 << RRD_UPDATED_SHIFT)))
 			break;

@@ -243,7 +244,7 @@ static bool alx_clean_rx_irq(struct alx_priv *alx, int budget)
 		}
 
 		napi_gro_receive(&alx->napi, skb);
-		budget--;
+		work++;
 
 next_pkt:
 		if (++rxq->read_idx == alx->rx_ringsz)

@@ -258,21 +259,22 @@ next_pkt:
 	if (rfd_cleaned)
 		alx_refill_rx_ring(alx, GFP_ATOMIC);
 
-	return budget > 0;
+	return work;
 }
 
 static int alx_poll(struct napi_struct *napi, int budget)
 {
 	struct alx_priv *alx = container_of(napi, struct alx_priv, napi);
 	struct alx_hw *hw = &alx->hw;
-	bool complete = true;
 	unsigned long flags;
+	bool tx_complete;
+	int work;
 
-	complete = alx_clean_tx_irq(alx) &&
-		   alx_clean_rx_irq(alx, budget);
+	tx_complete = alx_clean_tx_irq(alx);
+	work = alx_clean_rx_irq(alx, budget);
 
-	if (!complete)
-		return 1;
+	if (!tx_complete || work == budget)
+		return budget;
 
 	napi_complete(&alx->napi);

drivers/net/ethernet/atheros/alx/main.c static int alx_poll(struct napi_struct *napi, int budget) · line 292

Quoted 8 lines

drivers/net/ethernet/atheros/alx/main.c +1 -1

@@ -284,7 +286,7 @@ static int alx_poll(struct napi_struct *napi, int budget)
 
 	alx_post_write(hw);
 
-	return 0;
+	return work;
 }
 
 static irqreturn_t alx_intr_handle(struct alx_priv *alx, u32 intr)

Hi,
Checked it and its working.
Thanks again.
Please note that I only use this Ethernet controller for NFS and SCP, it is not used for development of networking software of any kind, so maybe a more extensive testing is needed.

Oded

Oded Gabbay

author Tested-by

2015-01-10 22:08 UTC #7 lore raw

Quoted 91 lines · Oded Gabbay

Quoted 73 lines · Eric Dumazet

Quoted 2 lines · Oded Gabbay

Yes, no problem.
I will update on the result.

Please try this more complete patch, solving the TX pressure problem as well, and not lying about NAPI budget. thanks !

drivers/net/ethernet/atheros/alx/main.c +11 -9

index e398eda07298..5f05b387c0a7 100644
@@ -184,15 +184,16 @@ static void alx_schedule_reset(struct alx_priv *alx)
 	schedule_work(&alx->reset_wk);
 }
 
-static bool alx_clean_rx_irq(struct alx_priv *alx, int budget)
+static int alx_clean_rx_irq(struct alx_priv *alx, int budget)
 {
 	struct alx_rx_queue *rxq = &alx->rxq;
 	struct alx_rrd *rrd;
 	struct alx_buffer *rxb;
 	struct sk_buff *skb;
 	u16 length, rfd_cleaned = 0;
+	int work = 0;
 
-	while (budget > 0) {
+	while (work < budget) {
 		rrd = &rxq->rrd[rxq->rrd_read_idx];
 		if (!(rrd->word3 & cpu_to_le32(1 << RRD_UPDATED_SHIFT)))
 			break;

@@ -243,7 +244,7 @@ static bool alx_clean_rx_irq(struct alx_priv *alx, int budget)
 		}
 
 		napi_gro_receive(&alx->napi, skb);
-		budget--;
+		work++;
 
 next_pkt:
 		if (++rxq->read_idx == alx->rx_ringsz)

@@ -258,21 +259,22 @@ next_pkt:
 	if (rfd_cleaned)
 		alx_refill_rx_ring(alx, GFP_ATOMIC);
 
-	return budget > 0;
+	return work;
 }
 
 static int alx_poll(struct napi_struct *napi, int budget)
 {
 	struct alx_priv *alx = container_of(napi, struct alx_priv, napi);
 	struct alx_hw *hw = &alx->hw;
-	bool complete = true;
 	unsigned long flags;
+	bool tx_complete;
+	int work;
 
-	complete = alx_clean_tx_irq(alx) &&
-		   alx_clean_rx_irq(alx, budget);
+	tx_complete = alx_clean_tx_irq(alx);
+	work = alx_clean_rx_irq(alx, budget);
 
-	if (!complete)
-		return 1;
+	if (!tx_complete || work == budget)
+		return budget;
 
 	napi_complete(&alx->napi);

drivers/net/ethernet/atheros/alx/main.c static int alx_poll(struct napi_struct *napi, int budget) · line 292

Quoted 8 lines

drivers/net/ethernet/atheros/alx/main.c +1 -1

@@ -284,7 +286,7 @@ static int alx_poll(struct napi_struct *napi, int budget)
 
 	alx_post_write(hw);
 
-	return 0;
+	return work;
 }
 
 static irqreturn_t alx_intr_handle(struct alx_priv *alx, u32 intr)

Quoted 7 lines

Oded
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/

But in any case, you can add:

Tested-by Oded Gabbay <[email protected]>

Oded

Eric Dumazet

2015-01-11 18:32 UTC #8 lore raw

PATCH 1/1 alx: fix alx_poll()

In reply to Eric Dumazet

From: Eric Dumazet <[email protected]>

Commit d75b1ade567f ("net: less interrupt masking in NAPI") uncovered wrong alx_poll() behavior.

A NAPI poll() handler is supposed to return exactly the budget when/if napi_complete() has not been called.

It is also supposed to return number of frames that were received, so that netdev_budget can have a meaning.

Also, in case of TX pressure, we still have to dequeue received packets : alx_clean_rx_irq() has to be called even if alx_clean_tx_irq(alx) returns false, otherwise device is half duplex.

Signed-off-by Eric Dumazet <[email protected]>

Fixes d75b1ade567f ("net: less interrupt masking in NAPI")

Reported-by Oded Gabbay <[email protected]>

Bisected-by Oded Gabbay <[email protected]>

Tested-by Oded Gabbay <[email protected]>

---
 drivers/net/ethernet/atheros/alx/main.c |   24 +++++++++++-----------
 1 file changed, 13 insertions(+), 11 deletions(-)

drivers/net/ethernet/atheros/alx/main.c +13 ~~-11~~

index e398eda07298..c8af3ce3ea38 100644
@@ -184,15 +184,16 @@ static void alx_schedule_reset(struct alx_priv *alx)
 	schedule_work(&alx->reset_wk);
 }
 
-static bool alx_clean_rx_irq(struct alx_priv *alx, int budget)
+static int alx_clean_rx_irq(struct alx_priv *alx, int budget)
 {
 	struct alx_rx_queue *rxq = &alx->rxq;
 	struct alx_rrd *rrd;
 	struct alx_buffer *rxb;
 	struct sk_buff *skb;
 	u16 length, rfd_cleaned = 0;
+	int work = 0;
 
-	while (budget > 0) {
+	while (work < budget) {
 		rrd = &rxq->rrd[rxq->rrd_read_idx];
 		if (!(rrd->word3 & cpu_to_le32(1 << RRD_UPDATED_SHIFT)))
 			break;

@@ -203,7 +204,7 @@ static bool alx_clean_rx_irq(struct alx_priv *alx, int budget)
 		    ALX_GET_FIELD(le32_to_cpu(rrd->word0),
 				  RRD_NOR) != 1) {
 			alx_schedule_reset(alx);
-			return 0;
+			return work;
 		}
 
 		rxb = &rxq->bufs[rxq->read_idx];

@@ -243,7 +244,7 @@ static bool alx_clean_rx_irq(struct alx_priv *alx, int budget)
 		}
 
 		napi_gro_receive(&alx->napi, skb);
-		budget--;
+		work++;
 
 next_pkt:
 		if (++rxq->read_idx == alx->rx_ringsz)

@@ -258,21 +259,22 @@ next_pkt:
 	if (rfd_cleaned)
 		alx_refill_rx_ring(alx, GFP_ATOMIC);
 
-	return budget > 0;
+	return work;
 }
 
 static int alx_poll(struct napi_struct *napi, int budget)
 {
 	struct alx_priv *alx = container_of(napi, struct alx_priv, napi);
 	struct alx_hw *hw = &alx->hw;
-	bool complete = true;
 	unsigned long flags;
+	bool tx_complete;
+	int work;
 
-	complete = alx_clean_tx_irq(alx) &&
-		   alx_clean_rx_irq(alx, budget);
+	tx_complete = alx_clean_tx_irq(alx);
+	work = alx_clean_rx_irq(alx, budget);
 
-	if (!complete)
-		return 1;
+	if (!tx_complete || work == budget)
+		return budget;
 
 	napi_complete(&alx->napi);

@@ -284,7 +286,7 @@ static int alx_poll(struct napi_struct *napi, int budget)
 
 	alx_post_write(hw);
 
-	return 0;
+	return work;
 }
 
 static irqreturn_t alx_intr_handle(struct alx_priv *alx, u32 intr)

David Miller

maintainer

2015-01-12 02:46 UTC #9 lore raw

Re: [PATCH net] alx: fix alx_poll()

From: Eric Dumazet <[email protected]>
Date: Sun, 11 Jan 2015 10:32:18 -0800

Quoted 15 lines

From: Eric Dumazet <[email protected]>
Commit d75b1ade567f ("net: less interrupt masking in NAPI") uncovered wrong alx_poll() behavior.
A NAPI poll() handler is supposed to return exactly the budget when/if napi_complete() has not been called.
It is also supposed to return number of frames that were received, so that netdev_budget can have a meaning.
Also, in case of TX pressure, we still have to dequeue received packets : alx_clean_rx_irq() has to be called even if alx_clean_tx_irq(alx) returns false, otherwise device is half duplex.

Quoted 5 lines

Signed-off-by Eric Dumazet <[email protected]>
Fixes d75b1ade567f ("net: less interrupt masking in NAPI")
Reported-by Oded Gabbay <[email protected]>
Bisected-by Oded Gabbay <[email protected]>
Tested-by Oded Gabbay <[email protected]>

Applied and queued up for -stable, thanks Eric.

Johannes Berg

2015-01-12 08:55 UTC #10 lore raw

Re: [PATCH net] alx: fix alx_poll()

In reply to Eric Dumazet

Quoted 2 lines · Eric Dumazet

Commit d75b1ade567f ("net: less interrupt masking in NAPI") uncovered wrong alx_poll() behavior.

[...]

Thanks Eric and Oded! :)

johannes