| Message ID | 20250315062516.788528-1-richard120310@gmail.com (mailing list archive) |
|---|---|
| State | New |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | net/smc: Reduce size of smc_wr_tx_tasklet_fn | expand |
On 2025-03-15 14:25:16, I Hsin Cheng wrote: >The variable "polled" in smc_wr_tx_tasklet_fn is a counter to determine >whether the loop has been executed for the first time. Refactor the type >of "polled" from "int" to "bool" can reduce the size of generated code >size by 12 bytes shown with the test below > >$ ./scripts/bloat-o-meter vmlinux_old vmlinux_new >add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-12 (-12) >Function old new delta >smc_wr_tx_tasklet_fn 1076 1064 -12 >Total: Before=24795091, After=24795079, chg -0.00% > >In some configuration, the compiler will complain this function for >exceeding 1024 bytes for function stack, this change can at least reduce >the size by 12 bytes within manner. > >Signed-off-by: I Hsin Cheng <richard120310@gmail.com> Reviewed-by: Dust Li <dust.li@linux.alibaba.com> Best regards, Dust >--- > net/smc/smc_wr.c | 8 ++++---- > 1 file changed, 4 insertions(+), 4 deletions(-) > >diff --git a/net/smc/smc_wr.c b/net/smc/smc_wr.c >index b04a21b8c511..3cc435ed7fde 100644 >--- a/net/smc/smc_wr.c >+++ b/net/smc/smc_wr.c >@@ -138,14 +138,14 @@ static void smc_wr_tx_tasklet_fn(struct tasklet_struct *t) > struct smc_ib_device *dev = from_tasklet(dev, t, send_tasklet); > struct ib_wc wc[SMC_WR_MAX_POLL_CQE]; > int i = 0, rc; >- int polled = 0; >+ bool polled = false; > > again: >- polled++; >+ polled = !polled; > do { > memset(&wc, 0, sizeof(wc)); > rc = ib_poll_cq(dev->roce_cq_send, SMC_WR_MAX_POLL_CQE, wc); >- if (polled == 1) { >+ if (polled) { > ib_req_notify_cq(dev->roce_cq_send, > IB_CQ_NEXT_COMP | > IB_CQ_REPORT_MISSED_EVENTS); >@@ -155,7 +155,7 @@ static void smc_wr_tx_tasklet_fn(struct tasklet_struct *t) > for (i = 0; i < rc; i++) > smc_wr_tx_process_cqe(&wc[i]); > } while (rc > 0); >- if (polled == 1) >+ if (polled) > goto again; > } > >-- >2.43.0 >
On 15.03.25 07:25, I Hsin Cheng wrote: > The variable "polled" in smc_wr_tx_tasklet_fn is a counter to determine > whether the loop has been executed for the first time. Refactor the type > of "polled" from "int" to "bool" can reduce the size of generated code > size by 12 bytes shown with the test below > > $ ./scripts/bloat-o-meter vmlinux_old vmlinux_new > add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-12 (-12) > Function old new delta > smc_wr_tx_tasklet_fn 1076 1064 -12 > Total: Before=24795091, After=24795079, chg -0.00% > > In some configuration, the compiler will complain this function for > exceeding 1024 bytes for function stack, this change can at least reduce > the size by 12 bytes within manner. > The code itself looks good. However, I’m curious about the specific situation where the compiler complained. Also, compared to exceeding the function stack limit by 1024 bytes, I don’t see how saving 12 bytes would bring any significant benefit. Thanks, Wenjia
On Mon, Mar 17, 2025 at 12:22:46PM +0100, Wenjia Zhang wrote: > > > On 15.03.25 07:25, I Hsin Cheng wrote: > > The variable "polled" in smc_wr_tx_tasklet_fn is a counter to determine > > whether the loop has been executed for the first time. Refactor the type > > of "polled" from "int" to "bool" can reduce the size of generated code > > size by 12 bytes shown with the test below > > > > $ ./scripts/bloat-o-meter vmlinux_old vmlinux_new > > add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-12 (-12) > > Function old new delta > > smc_wr_tx_tasklet_fn 1076 1064 -12 > > Total: Before=24795091, After=24795079, chg -0.00% > > > > In some configuration, the compiler will complain this function for > > exceeding 1024 bytes for function stack, this change can at least reduce > > the size by 12 bytes within manner. > > > The code itself looks good. However, I’m curious about the specific > situation where the compiler complained. Also, compared to exceeding the > function stack limit by 1024 bytes, I don’t see how saving 12 bytes would > bring any significant benefit. The patch description doesn't make sense: bloat-a-meter prints the _text size_ difference of two kernels, which really has nothing to do with potential stack size savings. If there are any changes in stack size with this patch is unknown; at least if you rely only on the patch description. You may want to have a look at scripts/stackusage and scripts/stackdelta.
diff --git a/net/smc/smc_wr.c b/net/smc/smc_wr.c index b04a21b8c511..3cc435ed7fde 100644 --- a/net/smc/smc_wr.c +++ b/net/smc/smc_wr.c @@ -138,14 +138,14 @@ static void smc_wr_tx_tasklet_fn(struct tasklet_struct *t) struct smc_ib_device *dev = from_tasklet(dev, t, send_tasklet); struct ib_wc wc[SMC_WR_MAX_POLL_CQE]; int i = 0, rc; - int polled = 0; + bool polled = false; again: - polled++; + polled = !polled; do { memset(&wc, 0, sizeof(wc)); rc = ib_poll_cq(dev->roce_cq_send, SMC_WR_MAX_POLL_CQE, wc); - if (polled == 1) { + if (polled) { ib_req_notify_cq(dev->roce_cq_send, IB_CQ_NEXT_COMP | IB_CQ_REPORT_MISSED_EVENTS); @@ -155,7 +155,7 @@ static void smc_wr_tx_tasklet_fn(struct tasklet_struct *t) for (i = 0; i < rc; i++) smc_wr_tx_process_cqe(&wc[i]); } while (rc > 0); - if (polled == 1) + if (polled) goto again; }
The variable "polled" in smc_wr_tx_tasklet_fn is a counter to determine whether the loop has been executed for the first time. Refactor the type of "polled" from "int" to "bool" can reduce the size of generated code size by 12 bytes shown with the test below $ ./scripts/bloat-o-meter vmlinux_old vmlinux_new add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-12 (-12) Function old new delta smc_wr_tx_tasklet_fn 1076 1064 -12 Total: Before=24795091, After=24795079, chg -0.00% In some configuration, the compiler will complain this function for exceeding 1024 bytes for function stack, this change can at least reduce the size by 12 bytes within manner. Signed-off-by: I Hsin Cheng <richard120310@gmail.com> --- net/smc/smc_wr.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)