This reverts commit fd48124e09825797bdc8ff0120f2401030c618ee.
The cleanup in this commit removes the qsel to addr mappings in
rtw_get_ff_hwaddr. The underlying assumption is that rtw_write_port
uses its addr parameter only for the high_queue check.
This is obviously incorrect as rtw_write_port calls
ffaddr2pipehdl(pdvobj, addr);
where addr is mapped to a usb bulk endpoint.
Unfortunately, testing did not show any problems. The Edimax V2 on which I
tested has two bulk out endpoints. I guess that with the incorrect patch,
addr could only be 0 (no high queue) or 6 (high queue), both of which were
mapped to the first bulk out endpoint. Data transfers did still work...
Signed-off-by: Martin Kaiser <[email protected]>
---
Hello Greg,
sorry for introducing a regression in commit fd48124e0982 ("staging:
r8188eu: simplify rtw_get_ff_hwaddr").
Could you take this revert before the 6.3 pull request?
Thanks,
Martin
drivers/staging/r8188eu/core/rtw_xmit.c | 32 +++++++++++++++++++++-
drivers/staging/r8188eu/include/rtw_xmit.h | 7 +++++
2 files changed, 38 insertions(+), 1 deletion(-)
diff --git a/drivers/staging/r8188eu/core/rtw_xmit.c b/drivers/staging/r8188eu/core/rtw_xmit.c
index d0761ac2e695..df88b3e29e77 100644
--- a/drivers/staging/r8188eu/core/rtw_xmit.c
+++ b/drivers/staging/r8188eu/core/rtw_xmit.c
@@ -1568,9 +1568,39 @@ static int rtw_br_client_tx(struct adapter *padapter, struct sk_buff **pskb)
u32 rtw_get_ff_hwaddr(struct xmit_frame *pxmitframe)
{
+ u32 addr;
struct pkt_attrib *pattrib = &pxmitframe->attrib;
- return pattrib->qsel == 0x11 ? HIGH_QUEUE_INX : 0;
+ switch (pattrib->qsel) {
+ case 0:
+ case 3:
+ addr = BE_QUEUE_INX;
+ break;
+ case 1:
+ case 2:
+ addr = BK_QUEUE_INX;
+ break;
+ case 4:
+ case 5:
+ addr = VI_QUEUE_INX;
+ break;
+ case 6:
+ case 7:
+ addr = VO_QUEUE_INX;
+ break;
+ case 0x10:
+ addr = BCN_QUEUE_INX;
+ break;
+ case 0x11:/* BC/MC in PS (HIQ) */
+ addr = HIGH_QUEUE_INX;
+ break;
+ case 0x12:
+ default:
+ addr = MGT_QUEUE_INX;
+ break;
+ }
+
+ return addr;
}
/*
diff --git a/drivers/staging/r8188eu/include/rtw_xmit.h b/drivers/staging/r8188eu/include/rtw_xmit.h
index e4e5af198eee..feeac85aedb0 100644
--- a/drivers/staging/r8188eu/include/rtw_xmit.h
+++ b/drivers/staging/r8188eu/include/rtw_xmit.h
@@ -26,7 +26,14 @@
#define XMIT_BE_QUEUE (2)
#define XMIT_BK_QUEUE (3)
+#define VO_QUEUE_INX 0
+#define VI_QUEUE_INX 1
+#define BE_QUEUE_INX 2
+#define BK_QUEUE_INX 3
+#define BCN_QUEUE_INX 4
+#define MGT_QUEUE_INX 5
#define HIGH_QUEUE_INX 6
+#define TXCMD_QUEUE_INX 7
#define HW_QUEUE_ENTRY 8
--
2.30.2
On Sat, Feb 11, 2023 at 07:32:05PM +0100, Martin Kaiser wrote:
> This reverts commit fd48124e09825797bdc8ff0120f2401030c618ee.
>
> The cleanup in this commit removes the qsel to addr mappings in
> rtw_get_ff_hwaddr. The underlying assumption is that rtw_write_port
> uses its addr parameter only for the high_queue check.
>
> This is obviously incorrect as rtw_write_port calls
> ffaddr2pipehdl(pdvobj, addr);
> where addr is mapped to a usb bulk endpoint.
>
> Unfortunately, testing did not show any problems. The Edimax V2 on which I
> tested has two bulk out endpoints. I guess that with the incorrect patch,
> addr could only be 0 (no high queue) or 6 (high queue), both of which were
> mapped to the first bulk out endpoint. Data transfers did still work...
>
> Signed-off-by: Martin Kaiser <[email protected]>
> ---
>
> Hello Greg,
>
> sorry for introducing a regression in commit fd48124e0982 ("staging:
> r8188eu: simplify rtw_get_ff_hwaddr").
>
> Could you take this revert before the 6.3 pull request?
>
> Thanks,
> Martin
>
I feel like the ancient `git revert` script is not at all in line with
current standards and sets people up for failure. This one at least
has a commit message. But
1) The subject doesn't have a correct patch prefix.
2) "commit fd48124e09825797bdc8ff0120f2401030c618ee" is not human
readable or how we describe commits these days with a 12 char hash.
3) There is no fixes tag.
regards,
dan carpenter
This reverts commit fd48124e0982 ("staging: r8188eu: simplify
rtw_get_ff_hwaddr").
The cleanup in this commit removes the qsel to addr mappings in
rtw_get_ff_hwaddr. The underlying assumption is that rtw_write_port
uses its addr parameter only for the high_queue check.
This is obviously incorrect as rtw_write_port calls
ffaddr2pipehdl(pdvobj, addr);
where addr is mapped to a usb bulk endpoint.
Unfortunately, testing did not show any problems. The Edimax V2 on which I
tested has two bulk out endpoints. I guess that with the incorrect patch,
addr could only be 0 (no high queue) or 6 (high queue), both of which were
mapped to the first bulk out endpoint. Data transfers did still work...
Fixes: fd48124e0982 ("staging: r8188eu: simplify rtw_get_ff_hwaddr")
Signed-off-by: Martin Kaiser <[email protected]>
---
v2:
- use correct patch prefix
- add a Fixes tag
- use 12-digit hash to refer to the reverted commit
drivers/staging/r8188eu/core/rtw_xmit.c | 32 +++++++++++++++++++++-
drivers/staging/r8188eu/include/rtw_xmit.h | 7 +++++
2 files changed, 38 insertions(+), 1 deletion(-)
diff --git a/drivers/staging/r8188eu/core/rtw_xmit.c b/drivers/staging/r8188eu/core/rtw_xmit.c
index d0761ac2e695..df88b3e29e77 100644
--- a/drivers/staging/r8188eu/core/rtw_xmit.c
+++ b/drivers/staging/r8188eu/core/rtw_xmit.c
@@ -1568,9 +1568,39 @@ static int rtw_br_client_tx(struct adapter *padapter, struct sk_buff **pskb)
u32 rtw_get_ff_hwaddr(struct xmit_frame *pxmitframe)
{
+ u32 addr;
struct pkt_attrib *pattrib = &pxmitframe->attrib;
- return pattrib->qsel == 0x11 ? HIGH_QUEUE_INX : 0;
+ switch (pattrib->qsel) {
+ case 0:
+ case 3:
+ addr = BE_QUEUE_INX;
+ break;
+ case 1:
+ case 2:
+ addr = BK_QUEUE_INX;
+ break;
+ case 4:
+ case 5:
+ addr = VI_QUEUE_INX;
+ break;
+ case 6:
+ case 7:
+ addr = VO_QUEUE_INX;
+ break;
+ case 0x10:
+ addr = BCN_QUEUE_INX;
+ break;
+ case 0x11:/* BC/MC in PS (HIQ) */
+ addr = HIGH_QUEUE_INX;
+ break;
+ case 0x12:
+ default:
+ addr = MGT_QUEUE_INX;
+ break;
+ }
+
+ return addr;
}
/*
diff --git a/drivers/staging/r8188eu/include/rtw_xmit.h b/drivers/staging/r8188eu/include/rtw_xmit.h
index e4e5af198eee..feeac85aedb0 100644
--- a/drivers/staging/r8188eu/include/rtw_xmit.h
+++ b/drivers/staging/r8188eu/include/rtw_xmit.h
@@ -26,7 +26,14 @@
#define XMIT_BE_QUEUE (2)
#define XMIT_BK_QUEUE (3)
+#define VO_QUEUE_INX 0
+#define VI_QUEUE_INX 1
+#define BE_QUEUE_INX 2
+#define BK_QUEUE_INX 3
+#define BCN_QUEUE_INX 4
+#define MGT_QUEUE_INX 5
#define HIGH_QUEUE_INX 6
+#define TXCMD_QUEUE_INX 7
#define HW_QUEUE_ENTRY 8
--
2.30.2
Thus wrote Dan Carpenter ([email protected]):
> On Sat, Feb 11, 2023 at 07:32:05PM +0100, Martin Kaiser wrote:
> > This reverts commit fd48124e09825797bdc8ff0120f2401030c618ee.
> > The cleanup in this commit removes the qsel to addr mappings in
> > rtw_get_ff_hwaddr. The underlying assumption is that rtw_write_port
> > uses its addr parameter only for the high_queue check.
> > This is obviously incorrect as rtw_write_port calls
> > ffaddr2pipehdl(pdvobj, addr);
> > where addr is mapped to a usb bulk endpoint.
> > Unfortunately, testing did not show any problems. The Edimax V2 on which I
> > tested has two bulk out endpoints. I guess that with the incorrect patch,
> > addr could only be 0 (no high queue) or 6 (high queue), both of which were
> > mapped to the first bulk out endpoint. Data transfers did still work...
> > Signed-off-by: Martin Kaiser <[email protected]>
> > ---
> > Hello Greg,
> > sorry for introducing a regression in commit fd48124e0982 ("staging:
> > r8188eu: simplify rtw_get_ff_hwaddr").
> > Could you take this revert before the 6.3 pull request?
> > Thanks,
> > Martin
> I feel like the ancient `git revert` script is not at all in line with
> current standards and sets people up for failure. This one at least
> has a commit message. But
> 1) The subject doesn't have a correct patch prefix.
> 2) "commit fd48124e09825797bdc8ff0120f2401030c618ee" is not human
> readable or how we describe commits these days with a 12 char hash.
> 3) There is no fixes tag.
Thanks, I fixed these points in v2.
Best regards,
Martin
On Mon, Feb 13, 2023 at 08:54:08PM +0100, Martin Kaiser wrote:
> This reverts commit fd48124e0982 ("staging: r8188eu: simplify
> rtw_get_ff_hwaddr").
>
> The cleanup in this commit removes the qsel to addr mappings in
> rtw_get_ff_hwaddr. The underlying assumption is that rtw_write_port
> uses its addr parameter only for the high_queue check.
>
> This is obviously incorrect as rtw_write_port calls
> ffaddr2pipehdl(pdvobj, addr);
> where addr is mapped to a usb bulk endpoint.
>
> Unfortunately, testing did not show any problems. The Edimax V2 on which I
> tested has two bulk out endpoints. I guess that with the incorrect patch,
> addr could only be 0 (no high queue) or 6 (high queue), both of which were
> mapped to the first bulk out endpoint. Data transfers did still work...
>
> Fixes: fd48124e0982 ("staging: r8188eu: simplify rtw_get_ff_hwaddr")
> Signed-off-by: Martin Kaiser <[email protected]>
> ---
> v2:
> - use correct patch prefix
> - add a Fixes tag
> - use 12-digit hash to refer to the reverted commit
Thanks, Martin!
regards,
dan carpenter