| Message ID | 1359575813-15703-1-git-send-email-mugunthanvnm@ti.com |
|---|---|
| State | Changes Requested, archived |
| Delegated to: | David Miller |
| Headers | show |
Op 30 jan. 2013, om 20:56 heeft Mugunthan V N <mugunthanvnm@ti.com> het volgende geschreven: > CPDMA interrupts are not properly acknowledged which leads to interrupt > storm, only cpdma interrupt 0 is acknowledged in Davinci CPDMA driver. > Changed cpdma_ctlr_eoi api to acknowledge 1 and 2 interrupts which are > used for rx and tx respectively. A brief inspection shows that this still isn't following the TRM, but Pantelis' patch does. Can you please fix this driver to follow the TRM and make it work on both PG1.0 and PG2.0 instead of papering over bugs instead of fixing them properly? -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 1/31/2013 1:33 AM, Koen Kooi wrote: > Op 30 jan. 2013, om 20:56 heeft Mugunthan V N <mugunthanvnm@ti.com> het volgende geschreven: > >> CPDMA interrupts are not properly acknowledged which leads to interrupt >> storm, only cpdma interrupt 0 is acknowledged in Davinci CPDMA driver. >> Changed cpdma_ctlr_eoi api to acknowledge 1 and 2 interrupts which are >> used for rx and tx respectively. > A brief inspection shows that this still isn't following the TRM, but Pantelis' patch does. Can you please fix this driver to follow the TRM and make it work on both PG1.0 and PG2.0 instead of papering over bugs instead of fixing them properly? Existing driver implementation is also complies with TRM. What Pantelis added additionally are non-napi implementation, handled cpdma processed tx and rx processing separately and renamed wr_reg as per TRM naming convention.. Also he has added a dummy reading tx/rx stat which is mentioned in TRM, but this stat is required only when using multichannel for data transfer. Current implementation of CPSW driver uses only channel 0 of Tx and Rx channels respectively for transmission and reading stat doesn't gets any effect in interrupt acknowledgment. Since both tx and rx are processed in same napi api, so i have added interrupt acknowledgment to the same existing api. Regards Mugunthan V N -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi On Jan 31, 2013, at 12:17 PM, Mugunthan V N wrote: > On 1/31/2013 1:33 AM, Koen Kooi wrote: >> Op 30 jan. 2013, om 20:56 heeft Mugunthan V N <mugunthanvnm@ti.com> het volgende geschreven: >> >>> CPDMA interrupts are not properly acknowledged which leads to interrupt >>> storm, only cpdma interrupt 0 is acknowledged in Davinci CPDMA driver. >>> Changed cpdma_ctlr_eoi api to acknowledge 1 and 2 interrupts which are >>> used for rx and tx respectively. >> A brief inspection shows that this still isn't following the TRM, but Pantelis' patch does. Can you please fix this driver to follow the TRM and make it work on both PG1.0 and PG2.0 instead of papering over bugs instead of fixing them properly? > Existing driver implementation is also complies with TRM. What Pantelis added > additionally are non-napi implementation, handled cpdma processed tx and rx > processing separately and renamed wr_reg as per TRM naming convention.. Also he > has added a dummy reading tx/rx stat which is mentioned in TRM, but this stat > is required only when using multichannel for data transfer. Current implementation > of CPSW driver uses only channel 0 of Tx and Rx channels respectively for transmission > and reading stat doesn't gets any effect in interrupt acknowledgment. > > Since both tx and rx are processed in same napi api, so i have added interrupt > acknowledgment to the same existing api. > First of all, this method of not needing to read the stat registers besides when using multichannel for data transfers is never described anywhere in any manual, or in the driver sources. Secondly, I find the method of ack-ing all interrupt sources problematic. Consider the following sequence Rx-interrupt ---> | | | IRQ handler | | schedules NAPI | | disables interrupts --> | cpsw_poll() | | handle Rx Tx-interrupt ---> |-------------------------|------------- | | ack Rx & Tx IRQ | | enable interrupts When will the Tx interrupt get handled? Is it guaranteed that the DMA logic will assert the Tx interrupt when the interrupts are enabled, even though the interrupt was previously acked? It is not clear in the TRM. Another problem that I see is that the other interrupts (MISC) are not supposed to be routed to the napi cpsw_poll() function at all. NAPI is for the tx/rx path as far as I know. > Regards > Mugunthan V N Regards -- Pantelis -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Pantelis On 1/31/2013 4:52 PM, Pantelis Antoniou wrote: > Hi > On Jan 31, 2013, at 12:17 PM, Mugunthan V N wrote: > >> On 1/31/2013 1:33 AM, Koen Kooi wrote: >>> Op 30 jan. 2013, om 20:56 heeft Mugunthan V N <mugunthanvnm@ti.com> het volgende geschreven: >>> >>>> CPDMA interrupts are not properly acknowledged which leads to interrupt >>>> storm, only cpdma interrupt 0 is acknowledged in Davinci CPDMA driver. >>>> Changed cpdma_ctlr_eoi api to acknowledge 1 and 2 interrupts which are >>>> used for rx and tx respectively. >>> A brief inspection shows that this still isn't following the TRM, but Pantelis' patch does. Can you please fix this driver to follow the TRM and make it work on both PG1.0 and PG2.0 instead of papering over bugs instead of fixing them properly? >> Existing driver implementation is also complies with TRM. What Pantelis added >> additionally are non-napi implementation, handled cpdma processed tx and rx >> processing separately and renamed wr_reg as per TRM naming convention.. Also he >> has added a dummy reading tx/rx stat which is mentioned in TRM, but this stat >> is required only when using multichannel for data transfer. Current implementation >> of CPSW driver uses only channel 0 of Tx and Rx channels respectively for transmission >> and reading stat doesn't gets any effect in interrupt acknowledgment. >> >> Since both tx and rx are processed in same napi api, so i have added interrupt >> acknowledgment to the same existing api. >> > > First of all, this method of not needing to read the stat registers besides when > using multichannel for data transfers is never described anywhere in any manual, > or in the driver sources. > > Secondly, I find the method of ack-ing all interrupt sources problematic. > Consider the following sequence > > Rx-interrupt ---> | | > | IRQ handler | > | schedules NAPI | > | disables interrupts --> | cpsw_poll() > | | handle Rx > Tx-interrupt ---> |-------------------------|------------- > | | ack Rx & Tx IRQ > | | enable interrupts > > When will the Tx interrupt get handled? Is it guaranteed that the DMA > logic will assert the Tx interrupt when the interrupts are enabled, even > though the interrupt was previously acked? It is not clear in the TRM. Need to check with the IP owner with this corner case. Simulating this will be difficult because even if driver misses the interrupt in next rx/tx interrupt it will be serviced. > > Another problem that I see is that the other interrupts (MISC) are not supposed > to be routed to the napi cpsw_poll() function at all. NAPI is for the tx/rx path > as far as I know. I agree that napi should not be used for MISC interrupts. Regards Mugunthan V N > > >> Regards >> Mugunthan V N > Regards > > -- Pantelis > -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/net/ethernet/ti/davinci_cpdma.c b/drivers/net/ethernet/ti/davinci_cpdma.c index f862918..afe14a6 100644 --- a/drivers/net/ethernet/ti/davinci_cpdma.c +++ b/drivers/net/ethernet/ti/davinci_cpdma.c @@ -487,6 +487,8 @@ int cpdma_ctlr_int_ctrl(struct cpdma_ctlr *ctlr, bool enable) void cpdma_ctlr_eoi(struct cpdma_ctlr *ctlr) { dma_reg_write(ctlr, CPDMA_MACEOIVECTOR, 0); + dma_reg_write(ctlr, CPDMA_MACEOIVECTOR, 1); + dma_reg_write(ctlr, CPDMA_MACEOIVECTOR, 2); } struct cpdma_chan *cpdma_chan_create(struct cpdma_ctlr *ctlr, int chan_num,
CPDMA interrupts are not properly acknowledged which leads to interrupt storm, only cpdma interrupt 0 is acknowledged in Davinci CPDMA driver. Changed cpdma_ctlr_eoi api to acknowledge 1 and 2 interrupts which are used for rx and tx respectively. Reported-by: Pantelis Antoniou <panto@antoniou-consulting.com> Signed-off-by: Mugunthan V N <mugunthanvnm@ti.com> --- drivers/net/ethernet/ti/davinci_cpdma.c | 2 ++ 1 file changed, 2 insertions(+)