vDPA/ifcvf: match pointer check to use
Subsystems (2)
PATCH 1/1 vDPA/ifcvf: match pointer check to use
From: Tom Rix <[email protected]>
Clang static analysis reports this issue
ifcvf_main.c:49:4: warning: Called function
pointer is null (null dereference)
vf->vring->cb.callback(vring->cb.private);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The check
vring = &vf->vring[i];
if (vring->cb.callback)
Does not match the use. Change dereference so they match.
--- drivers/vdpa/ifcvf/ifcvf_main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/vdpa/ifcvf/ifcvf_main.c +1 -1
@@ -46,7 +46,7 @@ static irqreturn_t ifcvf_vqs_reused_intr_handler(int irq, void *arg)for (i =0 ; i < vf->nr_vring; i++) { vring = &vf->vring[i];if (vring->cb.callback) - vf->vring->cb.callback (vring->cb.private); + vring->cb.callback (vring->cb.private); }return IRQ_HANDLED ;
Signature · 1 line
2.26.3
Michael S. Tsirkin
maintainerQuoted 25 lines · On Tue, Mar 15, 2022 at 05:41:30AM -0700, [email protected] wrote:
From: Tom Rix <[email protected]>
Clang static analysis reports this issue
ifcvf_main.c:49:4: warning: Called function
pointer is null (null dereference)
vf->vring->cb.callback(vring->cb.private);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~The check
vring = &vf->vring[i];
if (vring->cb.callback)Does not match the use. Change dereference so they match.
Fixes 79333575b8bd ("vDPA/ifcvf: implement shared IRQ feature")Signed-off-by Tom Rix <[email protected]>--- drivers/vdpa/ifcvf/ifcvf_main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/vdpa/ifcvf/ifcvf_main.c +0
-0
drivers/vdpa/ifcvf/ifcvf_main.c static irqreturn_t ifcvf_vqs_reused_intr_handler(int irq, void *arg) · line 52
Quoted 8 lines
drivers/vdpa/ifcvf/ifcvf_main.c +1
-1@@ -46,7 +46,7 @@ static irqreturn_t ifcvf_vqs_reused_intr_handler(int irq, void *arg)for (i =0 ; i < vf->nr_vring; i++) { vring = &vf->vring[i];if (vring->cb.callback) - vf->vring->cb.callback (vring->cb.private); + vring->cb.callback (vring->cb.private); }return IRQ_HANDLED ;
Oh, absolutely. In fact vf->vring->cb.callback is just vf->vring[0].cb.callback so it's wrong for any ring except 0. Does not make sense.
So how did it work in testing then? No idea.
Zhu Lingshan, care to comment?
Quoted 2 lines
Signature · 1 line
2.26.3
Michael S. Tsirkin
maintainerQuoted 8 lines · On Tue, Mar 15, 2022 at 05:41:30AM -0700, [email protected] wrote:
From: Tom Rix <[email protected]>
Clang static analysis reports this issue
ifcvf_main.c:49:4: warning: Called function
pointer is null (null dereference)
vf->vring->cb.callback(vring->cb.private);
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Quoted 7 lines
The check
vring = &vf->vring[i];
if (vring->cb.callback)Does not match the use. Change dereference so they match.
Fixes 79333575b8bd ("vDPA/ifcvf: implement shared IRQ feature")
Thanks a lot! I squashed this into the offending patch - no point in breaking bisect. Pushed to linux. However I'm now
having second thoughts about applying that patchset - I'd like soma analysis explaining how this got through testing.
Quoted 21 lines
Signed-off-by Tom Rix <[email protected]>--- drivers/vdpa/ifcvf/ifcvf_main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/vdpa/ifcvf/ifcvf_main.c +1
-1@@ -46,7 +46,7 @@ static irqreturn_t ifcvf_vqs_reused_intr_handler(int irq, void *arg)for (i =0 ; i < vf->nr_vring; i++) { vring = &vf->vring[i];if (vring->cb.callback) - vf->vring->cb.callback (vring->cb.private); + vring->cb.callback (vring->cb.private); }return IRQ_HANDLED ;Signature · 1 line
2.26.3
Quoted 13 lines · Michael S. Tsirkin
Quoted 12 lines · On Tue, Mar 15, 2022 at 05:41:30AM -0700, [email protected] wrote:
From: Tom Rix <[email protected]>
Clang static analysis reports this issue
ifcvf_main.c:49:4: warning: Called function
pointer is null (null dereference)vf->vring->cb.callback(vring->cb.private); ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~The check
vring = &vf->vring[i]; if (vring->cb.callback)
Quoted 7 lines
Quoted 3 lines
Does not match the use. Change dereference so they match.
Fixes 79333575b8bd ("vDPA/ifcvf: implement shared IRQ feature")Thanks a lot! I squashed this into the offending patch - no point in breaking bisect. Pushed to linux. However I'm now
having second thoughts about applying that patchset - I'd like soma analysis explaining how this got through testing.
static analysis is something i do treewide.
There are currently ~2500 issues in linux-next, do not panic! many are false positives.
It is pretty easy to setup and once you have a baseline you can filter only your files.
Tom
Quoted 20 lines
Quoted 20 lines
Signed-off-by Tom Rix <[email protected]>--- drivers/vdpa/ifcvf/ifcvf_main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/vdpa/ifcvf/ifcvf_main.c +1
-1@@ -46,7 +46,7 @@ static irqreturn_t ifcvf_vqs_reused_intr_handler(int irq, void *arg)for (i =0 ; i < vf->nr_vring; i++) { vring = &vf->vring[i];if (vring->cb.callback) - vf->vring->cb.callback (vring->cb.private); + vring->cb.callback (vring->cb.private); }return IRQ_HANDLED ;Signature · 1 line
2.26.3
Michael S. Tsirkin
maintainerQuoted 25 lines · Tom Rix
Quoted 13 lines · Michael S. Tsirkin
Quoted 12 lines · On Tue, Mar 15, 2022 at 05:41:30AM -0700, [email protected] wrote:
From: Tom Rix <[email protected]>
Clang static analysis reports this issue
ifcvf_main.c:49:4: warning: Called function
pointer is null (null dereference)vf->vring->cb.callback(vring->cb.private); ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~The check
vring = &vf->vring[i]; if (vring->cb.callback)Quoted 7 lines
Quoted 3 lines
Does not match the use. Change dereference so they match.
Fixes 79333575b8bd ("vDPA/ifcvf: implement shared IRQ feature")Thanks a lot! I squashed this into the offending patch - no point in breaking bisect. Pushed to linux. However I'm now
having second thoughts about applying that patchset - I'd like soma analysis explaining how this got through testing.static analysis is something i do treewide.
Quoted 7 lines
There are currently ~2500 issues in linux-next, do not panic! many are false positives.
It is pretty easy to setup and once you have a baseline you can filter only your files.
Tom
Thanks for that info! I was actually directing this question to the contributor since the code does not look like it could have ever worked. I don't have the hardware in question myself.
Quoted 19 lines
Quoted 19 lines
Quoted 19 lines
Signed-off-by Tom Rix <[email protected]>--- drivers/vdpa/ifcvf/ifcvf_main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/vdpa/ifcvf/ifcvf_main.c +1
-1@@ -46,7 +46,7 @@ static irqreturn_t ifcvf_vqs_reused_intr_handler(int irq, void *arg)for (i =0 ; i < vf->nr_vring; i++) { vring = &vf->vring[i];if (vring->cb.callback) - vf->vring->cb.callback (vring->cb.private); + vring->cb.callback (vring->cb.private); }return IRQ_HANDLED ;Signature · 1 line
2.26.3
Quoted 27 lines · Michael S. Tsirkin
Quoted 26 lines · Tom Rix
Quoted 13 lines · Michael S. Tsirkin
Quoted 12 lines · On Tue, Mar 15, 2022 at 05:41:30AM -0700, [email protected] wrote:
From: Tom Rix <[email protected]>
Clang static analysis reports this issue
ifcvf_main.c:49:4: warning: Called functionpointer is null (null dereference) vf->vring->cb.callback(vring->cb.private); ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~The check
vring = &vf->vring[i]; if (vring->cb.callback)Quoted 7 lines
Quoted 3 lines
Does not match the use. Change dereference so they match.
Fixes 79333575b8bd ("vDPA/ifcvf: implement shared IRQ feature")Thanks a lot! I squashed this into the offending patch - no point in breaking bisect. Pushed to linux. However I'm now
having second thoughts about applying that patchset - I'd like soma analysis explaining how this got through testing.static analysis is something i do treewide.
There are currently ~2500 issues in linux-next, do not panic! many are false positives.
Quoted 7 lines
Quoted 4 lines
It is pretty easy to setup and once you have a baseline you can filter only your files.
Tom
Thanks for that info! I was actually directing this question to the contributor since the code does not look like it could have ever worked. I don't have the hardware in question myself.
Oh, that's my bad introducing this bug by typo, thanks Tom for fixing that!
In my previous testing, I tolerated the lower performance(only one queue working as you pointed out) due to the shared irq, and did not check mq status, sorry for that.
After fixing this issue, test again:
(1)set nvectors = 2 after allocate MSI vectors, force the queues share one vector. from /proc/interrupts, we can see:
[lszhu@localhost linux]$ cat /proc/interrupts | grep ifcvf
241: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2724424 0 0 0 0 0 0 0 0 0 IR-PCI-MSI 534528-edge ifcvf[0000:01:00.5]-vqs-reused-irq 242: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 IR-PCI-MSI 534529-edge ifcvf[0000:01:00.5]-config 251: 0 0 0 0 0 0 0 0 0 0 0 2693318 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 IR-PCI-MSI 536576-edge ifcvf[0000:01:00.6]-vqs-reused-irq 252: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 IR-PCI-MSI 536577-edge ifcvf[0000:01:00.6]-config
(2) after several rounds of scp from a VF to another, at the source side, ethtool -S shows(cut off, only tx):
localhost:/home/lszhu # ethtool -S eth1
NIC statistics:
tx_queue_0_packets: 437256
tx_queue_0_bytes: 629246017
tx_queue_0_xdp_tx: 0
tx_queue_0_xdp_tx_drops: 0
tx_queue_0_kicks: 34089
tx_queue_1_packets: 73
tx_queue_1_bytes: 96721
tx_queue_1_xdp_tx: 0
tx_queue_1_xdp_tx_drops: 0
tx_queue_1_kicks: 12
tx_queue_2_packets: 294647
tx_queue_2_bytes: 433949815
tx_queue_2_xdp_tx: 0
tx_queue_2_xdp_tx_drops: 0
tx_queue_2_kicks: 14948
tx_queue_3_packets: 20226451
tx_queue_3_bytes: 29735633548
tx_queue_3_xdp_tx: 0
tx_queue_3_xdp_tx_drops: 0
tx_queue_3_kicks: 1123675
so every queue carries traffic now, all enabled
Thanks,
Zhu Lingshan
Quoted 21 lines
Quoted 21 lines
Quoted 21 lines
Quoted 21 lines
Signed-off-by Tom Rix <[email protected]>--- drivers/vdpa/ifcvf/ifcvf_main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/vdpa/ifcvf/ifcvf_main.c +1
-1@@ -46,7 +46,7 @@ static irqreturn_t ifcvf_vqs_reused_intr_handler(int irq, void *arg)for (i =0 ; i < vf->nr_vring; i++) { vring = &vf->vring[i];if (vring->cb.callback) - vf->vring->cb.callback (vring->cb.private); + vring->cb.callback (vring->cb.private); }return IRQ_HANDLED ;Signature · 1 line
2.26.3