Skip to contentSkip to conversation

ipmi: ssif_bmc: Move smbus_cmd assignment after cleanup

PATCH In review Cosmo Chou 3 participants 6 messages Jan 1, 2025 – Jan 6, 2025 IPMI

lore
Share & tools
lore

Cosmo Chou

author
#0 lore raw

PATCH 1/1 ipmi: ssif_bmc: Move smbus_cmd assignment after cleanup

Move smbus_cmd assignment to the end of process_smbus_cmd() to ensure the new command is not lost when complete_response() is triggered.

Fixes dd2bc5cc9e25 ("ipmi: ssif_bmc: Add SSIF BMC driver")
---
 drivers/char/ipmi/ssif_bmc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/char/ipmi/ssif_bmc.c +1 -1
index a14fafc583d4..7a4f52987a7d 100644
@@ -579,7 +579,6 @@ static void process_request_part(struct ssif_bmc_ctx *ssif_bmc)
 static void process_smbus_cmd(struct ssif_bmc_ctx *ssif_bmc, u8 *val)
 {
 	/* SMBUS command can vary (single or multi-part) */
-	ssif_bmc->part_buf.smbus_cmd = *val;
 	ssif_bmc->msg_idx = 1;
 	memset(&ssif_bmc->part_buf.payload[0], 0, MAX_PAYLOAD_PER_TRANSACTION);
 
@@ -596,6 +595,7 @@ static void process_smbus_cmd(struct ssif_bmc_ctx *ssif_bmc, u8 *val)
 		if (ssif_bmc->aborting)
 			ssif_bmc->aborting = false;
 	}
+	ssif_bmc->part_buf.smbus_cmd = *val;
 }
 
 static void on_read_requested_event(struct ssif_bmc_ctx *ssif_bmc, u8 *val)
Signature · 1 line
2.34.1
maintainer
#1 lore raw
Quoted 2 lines · Cosmo Chou

Move smbus_cmd assignment to the end of process_smbus_cmd() to ensure the new command is not lost when complete_response() is triggered.

Ok, I see, patch is applied. Thank you.

Quan, I assume this is ok. Please double-check it for me.

-corey

Quoted 30 lines
Fixes dd2bc5cc9e25 ("ipmi: ssif_bmc: Add SSIF BMC driver")
---
 drivers/char/ipmi/ssif_bmc.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/char/ipmi/ssif_bmc.c +1 -1
index a14fafc583d4..7a4f52987a7d 100644
@@ -579,7 +579,6 @@ static void process_request_part(struct ssif_bmc_ctx *ssif_bmc)
 static void process_smbus_cmd(struct ssif_bmc_ctx *ssif_bmc, u8 *val)
 {
 	/* SMBUS command can vary (single or multi-part) */
-	ssif_bmc->part_buf.smbus_cmd = *val;
 	ssif_bmc->msg_idx = 1;
 	memset(&ssif_bmc->part_buf.payload[0], 0, MAX_PAYLOAD_PER_TRANSACTION);
 
@@ -596,6 +595,7 @@ static void process_smbus_cmd(struct ssif_bmc_ctx *ssif_bmc, u8 *val)
 		if (ssif_bmc->aborting)
 			ssif_bmc->aborting = false;
 	}
+	ssif_bmc->part_buf.smbus_cmd = *val;
 }
 
 static void on_read_requested_event(struct ssif_bmc_ctx *ssif_bmc, u8 *val)
Signature · 1 line
2.34.1
#2 lore raw
Quoted 8 lines · Corey Minyard
Quoted 2 lines · Cosmo Chou

Move smbus_cmd assignment to the end of process_smbus_cmd() to ensure the new command is not lost when complete_response() is triggered.

Ok, I see, patch is applied. Thank you.

Quan, I assume this is ok. Please double-check it for me.

Yes, Corey,

This patch looks good to me and I think this can be improved a bit further in my reply for Cosmo.

- Quan

#3 lore raw
In reply to Cosmo Chou
Quoted 3 lines · Cosmo Chou

Move smbus_cmd assignment to the end of process_smbus_cmd() to ensure the new command is not lost when complete_response() is triggered.

Thanks Cosmo for the catch.

And, IMHO, the root cause is the memset() on part buffer called in complete_response() is not quite correct. In the current implementation, the complete_response() should only be called when the READ is completed, ie: only on I2C_SLAVE_STOP of READ transaction, otherwise all the info of current on-going request will be mistakenly cleared as in this case.

This patch is good and I wonder if we can make this a bit further as below?

drivers/char/ipmi/ssif_bmc.c +2 -2
index a14fafc583d4..310f17dd9511 100644
@@ -292,7 +292,6 @@ static void complete_response(struct ssif_bmc_ctx *ssif_bmc)
        ssif_bmc->nbytes_processed = 0;
        ssif_bmc->remain_len = 0;
        ssif_bmc->busy = false;
-       memset(&ssif_bmc->part_buf, 0, sizeof(struct ssif_part_buffer));
        wake_up_all(&ssif_bmc->wait_queue);
 }
 
@@ -744,9 +743,11 @@ static void on_stop_event(struct ssif_bmc_ctx *ssif_bmc, u8 *val)
                        ssif_bmc->aborting = true;
                }
        } else if (ssif_bmc->state == SSIF_RES_SENDING) {
-               if (ssif_bmc->is_singlepart_read || ssif_bmc->block_num == 0xFF)
+               if (ssif_bmc->is_singlepart_read || ssif_bmc->block_num == 0xFF) {
+                       memset(&ssif_bmc->part_buf, 0, sizeof(struct ssif_part_buffer));
                        /* Invalidate response buffer to denote it is 

sent */

                        complete_response(ssif_bmc);
+               }
                ssif_bmc->state = SSIF_READY;
        }

Thanks and Happy New Year.
- Quan

maintainer
#4 lore raw
Quoted 9 lines · Quan Nguyen
Quoted 3 lines · Cosmo Chou

Move smbus_cmd assignment to the end of process_smbus_cmd() to ensure the new command is not lost when complete_response() is triggered.

Thanks Cosmo for the catch.

Quoted 7 lines

And, IMHO, the root cause is the memset() on part buffer called in complete_response() is not quite correct. In the current implementation, the complete_response() should only be called when the READ is completed, ie:
only on I2C_SLAVE_STOP of READ transaction, otherwise all the info of current on-going request will be mistakenly cleared as in this case.

This patch is good and I wonder if we can make this a bit further as below?

Yes, this is probably more future proof.

Can you send me a formal patch, with a "Found-by:" for Cosmo? I'll replace Cosmo's patch.

Thanks,

-corey

Quoted 34 lines
drivers/char/ipmi/ssif_bmc.c +2 -2
index a14fafc583d4..310f17dd9511 100644
@@ -292,7 +292,6 @@ static void complete_response(struct ssif_bmc_ctx
 *ssif_bmc)
        ssif_bmc->nbytes_processed = 0;
        ssif_bmc->remain_len = 0;
        ssif_bmc->busy = false;
-       memset(&ssif_bmc->part_buf, 0, sizeof(struct ssif_part_buffer));
        wake_up_all(&ssif_bmc->wait_queue);
 }
@@ -744,9 +743,11 @@ static void on_stop_event(struct ssif_bmc_ctx
 *ssif_bmc, u8 *val)
                        ssif_bmc->aborting = true;
                }
        } else if (ssif_bmc->state == SSIF_RES_SENDING) {
-               if (ssif_bmc->is_singlepart_read || ssif_bmc->block_num ==
 0xFF)
+               if (ssif_bmc->is_singlepart_read || ssif_bmc->block_num ==
 0xFF) {
+                       memset(&ssif_bmc->part_buf, 0, sizeof(struct
 ssif_part_buffer));
                        /* Invalidate response buffer to denote it is sent

*/

                        complete_response(ssif_bmc);
+               }
                ssif_bmc->state = SSIF_READY;
        }

Thanks and Happy New Year.
- Quan

#5 lore raw
Quoted 16 lines · Corey Minyard
Quoted 15 lines · Quan Nguyen
Quoted 3 lines · Cosmo Chou

Move smbus_cmd assignment to the end of process_smbus_cmd() to ensure the new command is not lost when complete_response() is triggered.

Thanks Cosmo for the catch.

And, IMHO, the root cause is the memset() on part buffer called in complete_response() is not quite correct. In the current implementation, the complete_response() should only be called when the READ is completed, ie:
only on I2C_SLAVE_STOP of READ transaction, otherwise all the info of current on-going request will be mistakenly cleared as in this case.

Quoted 7 lines
Quoted 1 line

This patch is good and I wonder if we can make this a bit further as below?

Yes, this is probably more future proof.

Can you send me a formal patch, with a "Found-by:" for Cosmo? I'll replace Cosmo's patch.

Thanks Corey,

Will send the formal patch later.

Thanks,
-- Quan