TI wlcore firmware log fixes
Subsystems (3)
cover TI wlcore firmware log fixes
Hi,
(I'm assuming this is the correct way to submit for wireless-drivers-next.)
The following series fixes a number of issues with the firmware logging on TI wireless devices, noticed while looking at AP mode issues on the WL18xx family of devices.
Patch 1 tidies up the use of "fw_log.actual_buff_size" removing multiple unnecessary endian conversions by caching the CPU endian value in a local variable "actual_size".
Patch 2 makes the buffer calculations more obvious and adds comments to describe what is going on. In particular, the addition of "addr_ptr" eliminates several "addr + internal_fw_addrbase" calculations in the code.
Patch 3 fixes an error in the calculation of "clean_addr" when it hits the end of the buffer, which was causing the kernel console to spew "errors" when firmware tracing is in effect.
Patch 4 removes the error message fixed in patch 3, which can still occur when we race with the firmware reading the log structure. The write pointer is where the firmware is writing its next message to, whereas "clean_ptr" is the point that we've read the firmware log to. It is fine that these may not match.
As this is a debug facility, I don't see any urgent need to push these patches into -rc nor stable kernels.
drivers/net/wireless/ti/wlcore/event.c | 67 +++++++++++++++++++--------------- 1 file changed, 38 insertions(+), 29 deletions(-)
Signature · 2 lines
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/ FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!
PATCH 1/4 net: wlcore: tidy up use of fw_log.actual_buff_size
Tidy up the use of fw_log.actual_buff_size - rather than reading it multiple times and applying the endian conversion, read it once into actual_len and use that instead.
--- drivers/net/wireless/ti/wlcore/event.c | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-)
drivers/net/wireless/ti/wlcore/event.c +10 -10
@@ -40,7 +40,7 @@ int wlcore_event_fw_logger(struct wl1271 *wl) buffer =kzalloc (WL18XX_LOGGER_SDIO_BUFF_MAX ,GFP_KERNEL );if (!buffer) {wl1271_error ("Fail to allocate fw logger memory" ); - fw_log.actual_buff_size =cpu_to_le32 (0 ); + actual_len =0 ; goto out; }
@@ -49,30 +49,30 @@ int wlcore_event_fw_logger(struct wl1271 *wl)if (ret <0 ) {wl1271_error ("Fail to read logger buffer, error_id = %d" , ret); - fw_log.actual_buff_size =cpu_to_le32 (0 ); + actual_len =0 ; goto free_out; }memcpy (&fw_log, buffer,sizeof (fw_log)); -if (le32_to_cpu (fw_log.actual_buff_size) ==0 ) + actual_len =le32_to_cpu (fw_log.actual_buff_size); +if (actual_len ==0 ) goto free_out; - actual_len =le32_to_cpu (fw_log.actual_buff_size); start_loc = (le32_to_cpu (fw_log.buff_read_ptr) - internal_fw_addrbase) - addr; end_buff_addr +=le32_to_cpu (fw_log.max_buff_size); available_len = end_buff_addr - (le32_to_cpu (fw_log.buff_read_ptr) - internal_fw_addrbase); - actual_len =min (actual_len, available_len); - len = actual_len; +/* Copy initial part from end of ring buffer */ + len =min (actual_len, available_len);wl12xx_copy_fwlog (wl, &buffer[start_loc], len); - clear_addr = addr + start_loc +le32_to_cpu (fw_log.actual_buff_size) + - internal_fw_addrbase; + clear_addr = addr + start_loc + actual_len + internal_fw_addrbase; - len =le32_to_cpu (fw_log.actual_buff_size) - len; +/* Copy any remaining part from beginning of ring buffer */ + len = actual_len - len;if (len) { wl12xx_copy_fwlog(wl, &buffer[WL18XX_LOGGER_BUFF_OFFSET],
@@ -93,7 +93,7 @@ int wlcore_event_fw_logger(struct wl1271 *wl) free_out:kfree (buffer); out: -return le32_to_cpu (fw_log.actual_buff_size); +return actual_len; }EXPORT_SYMBOL_GPL (wlcore_event_fw_logger);
Signature · 1 line
2.20.1
Re: [wireless-drivers-next,1/4] wlcore: tidy up use of fw_log.actual_buff_size
Quoted 5 lines · Russell King
Tidy up the use of fw_log.actual_buff_size - rather than reading it multiple times and applying the endian conversion, read it once into
actual_lenand use that instead.Signed-off-by Russell King <[email protected]>
4 patches applied to wireless-drivers-next.git, thanks.
913112398d5e wlcore: tidy up use of fw_log.actual_buff_size 98e94771cadc wlcore: make some of the fwlog calculations more obvious 87ab9cbaee7c wlcore: fix bug reading fwlog01de6fe49ca4 wlcore: fix read pointer update
Signature · 3 lines
https://patchwork.kernel.org/project/linux-wireless/patch/[email protected]/ https://wireless.wiki.kernel.org/en/developers/documentation/submittingpatches
PATCH 2/4 net: wlcore: make some of the fwlog calculations more obvious
In reply to Russell KingMake some of the fwlog calculations more obvious by calculating bits that get used and documenting what they are. Validate the read pointer while we're at it to ensure we do not overflow the data block we have allocated and read.
--- drivers/net/wireless/ti/wlcore/event.c | 43 +++++++++++++++++--------- 1 file changed, 28 insertions(+), 15 deletions(-)
drivers/net/wireless/ti/wlcore/event.c +28 -15
@@ -29,11 +29,13 @@ int wlcore_event_fw_logger(struct wl1271 *wl)u8 *buffer;u32 internal_fw_addrbase =WL18XX_DATA_RAM_BASE_ADDRESS ;u32 addr =WL18XX_LOGGER_SDIO_BUFF_ADDR ; -u32 end_buff_addr =WL18XX_LOGGER_SDIO_BUFF_ADDR + -WL18XX_LOGGER_BUFF_OFFSET ; +u32 addr_ptr; +u32 buff_start_ptr; +u32 buff_read_ptr; +u32 buff_end_ptr;u32 available_len;u32 actual_len; -u32 clear_addr; +u32 clear_ptr;size_t len;u32 start_loc;
@@ -59,17 +61,29 @@ int wlcore_event_fw_logger(struct wl1271 *wl)if (actual_len ==0 ) goto free_out; - start_loc = (le32_to_cpu (fw_log.buff_read_ptr) - - internal_fw_addrbase) - addr; - end_buff_addr +=le32_to_cpu (fw_log.max_buff_size); - available_len = end_buff_addr - - (le32_to_cpu (fw_log.buff_read_ptr) - - internal_fw_addrbase); +/* Calculate the internal pointer to the fwlog structure */ + addr_ptr = internal_fw_addrbase + addr; -/* Copy initial part from end of ring buffer */ +/* Calculate the internal pointers to the start and end of log buffer */ + buff_start_ptr = addr_ptr +WL18XX_LOGGER_BUFF_OFFSET ; + buff_end_ptr = buff_start_ptr +le32_to_cpu (fw_log.max_buff_size); + +/* Read the read pointer and validate it */ + buff_read_ptr =le32_to_cpu (fw_log.buff_read_ptr); +if (buff_read_ptr < buff_start_ptr || + buff_read_ptr >= buff_end_ptr) { +wl1271_error ("buffer read pointer out of bounds: %x not in (%x-%x)\n" , + buff_read_ptr, buff_start_ptr, buff_end_ptr); + goto free_out; + } + + start_loc = buff_read_ptr - addr_ptr; + available_len = buff_end_ptr - buff_read_ptr; + +/* Copy initial part up to the end of ring buffer */ len =min (actual_len, available_len);wl12xx_copy_fwlog (wl, &buffer[start_loc], len); - clear_addr = addr + start_loc + actual_len + internal_fw_addrbase; + clear_ptr = addr_ptr + start_loc + actual_len;/* Copy any remaining part from beginning of ring buffer */ len = actual_len - len;
@@ -77,14 +91,13 @@ int wlcore_event_fw_logger(struct wl1271 *wl)wl12xx_copy_fwlog (wl, &buffer[WL18XX_LOGGER_BUFF_OFFSET ], len); - clear_addr = addr +WL18XX_LOGGER_BUFF_OFFSET + len + - internal_fw_addrbase; + clear_ptr = addr_ptr +WL18XX_LOGGER_BUFF_OFFSET + len; }/* double check that clear address and write pointer are the same */ -if (clear_addr !=le32_to_cpu (fw_log.buff_write_ptr)) { +if (clear_ptr !=le32_to_cpu (fw_log.buff_write_ptr)) {wl1271_error ("Calculate of clear addr Clear = %x, write = %x" , - clear_addr,le32_to_cpu (fw_log.buff_write_ptr)); + clear_ptr,le32_to_cpu (fw_log.buff_write_ptr)); }/* indicate FW about Clear buffer */
Signature · 1 line
2.20.1
PATCH 3/4 net: wlcore: fix bug reading fwlog
In reply to Russell KingWith logging enabled, it has been observed that the driver spews messages such as:
wlcore: ERROR Calculate of clear addr Clear = 204025b0, write = 204015b0
The problem occurs because 204025b0 is the end of the buffer, and 204015b0 is the beginning, and the calculation for "clear"ing the buffer does not take into account that if we read to the very end of the ring buffer, we are actually at the beginning of the buffer.
Fix this.
--- drivers/net/wireless/ti/wlcore/event.c | 2 ++ 1 file changed, 2 insertions(+)
drivers/net/wireless/ti/wlcore/event.c +2 -0
@@ -84,6 +84,8 @@ int wlcore_event_fw_logger(struct wl1271 *wl) len =min (actual_len, available_len);wl12xx_copy_fwlog (wl, &buffer[start_loc], len); clear_ptr = addr_ptr + start_loc + actual_len; +if (clear_ptr == buff_end_ptr) + clear_ptr = buff_start_ptr;/* Copy any remaining part from beginning of ring buffer */ len = actual_len - len;
Signature · 1 line
2.20.1
PATCH 4/4 net: wlcore: fix read pointer update
In reply to Russell KingWhen reading the fw_log structure from the device's memory, we could race with the firmware updating the actual_buff_size and buff_write_ptr members of this structure. This would lead to bytes being dropped from the log.
Fix this by writing back the actual - now fixed - clear_ptr which reflects where we read up to in the buffer.
This also means that we must not check that the clear_ptr matches the current write pointer, so remove that check.
--- drivers/net/wireless/ti/wlcore/event.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)
drivers/net/wireless/ti/wlcore/event.c +2 -8
@@ -96,15 +96,9 @@ int wlcore_event_fw_logger(struct wl1271 *wl) clear_ptr = addr_ptr +WL18XX_LOGGER_BUFF_OFFSET + len; } -/* double check that clear address and write pointer are the same */ -if (clear_ptr !=le32_to_cpu (fw_log.buff_write_ptr)) { -wl1271_error ("Calculate of clear addr Clear = %x, write = %x" , - clear_ptr,le32_to_cpu (fw_log.buff_write_ptr)); - } - -/* indicate FW about Clear buffer */ +/* Update the read pointer */ ret =wlcore_write32 (wl, addr +WL18XX_LOGGER_READ_POINT_OFFSET , - fw_log.buff_write_ptr); + clear_ptr); free_out:kfree (buffer); out:
Signature · 1 line
2.20.1