Staging: sxg: Removed unnecessary checks while taking Transmit Locks
Fix the locking issue of locks in transmit code path. There was an unnecessary check for interrupt context in transmit code path. Removed that. Signed-off-by: LinSysSoft Sahara Team <saharaproj@linsyssoft.com> Signed-off-by: Mithlesh Thukral <mithlesh@linsyssoft.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
This commit is contained in:
Родитель
7c66b14b6b
Коммит
c5e5cf5a18
|
@ -119,7 +119,7 @@ static int sxg_poll(struct napi_struct *napi, int budget);
|
|||
static int sxg_process_isr(struct adapter_t *adapter, u32 MessageId);
|
||||
static u32 sxg_process_event_queue(struct adapter_t *adapter, u32 RssId,
|
||||
int *sxg_napi_continue, int *work_done, int budget);
|
||||
static void sxg_complete_slow_send(struct adapter_t *adapter, int irq_context);
|
||||
static void sxg_complete_slow_send(struct adapter_t *adapter);
|
||||
static struct sk_buff *sxg_slow_receive(struct adapter_t *adapter,
|
||||
struct sxg_event *Event);
|
||||
static void sxg_process_rcv_error(struct adapter_t *adapter, u32 ErrorStatus);
|
||||
|
@ -1274,7 +1274,7 @@ static int sxg_process_isr(struct adapter_t *adapter, u32 MessageId)
|
|||
}
|
||||
/* Slowpath send completions */
|
||||
if (Isr & SXG_ISR_SPSEND) {
|
||||
sxg_complete_slow_send(adapter, 1);
|
||||
sxg_complete_slow_send(adapter);
|
||||
}
|
||||
/* Dump */
|
||||
if (Isr & SXG_ISR_UPC) {
|
||||
|
@ -1477,11 +1477,10 @@ static u32 sxg_process_event_queue(struct adapter_t *adapter, u32 RssId,
|
|||
*
|
||||
* Arguments -
|
||||
* adapter - A pointer to our adapter structure
|
||||
* irq_context - An integer to denote if we are in interrupt context
|
||||
* Return
|
||||
* None
|
||||
*/
|
||||
static void sxg_complete_slow_send(struct adapter_t *adapter, int irq_context)
|
||||
static void sxg_complete_slow_send(struct adapter_t *adapter)
|
||||
{
|
||||
struct sxg_xmt_ring *XmtRing = &adapter->XmtRings[0];
|
||||
struct sxg_ring_info *XmtRingInfo = &adapter->XmtRingZeroInfo;
|
||||
|
@ -1496,12 +1495,7 @@ static void sxg_complete_slow_send(struct adapter_t *adapter, int irq_context)
|
|||
* This means two different processors can both be running/
|
||||
* through this loop. Be *very* careful.
|
||||
*/
|
||||
if(irq_context) {
|
||||
if(!spin_trylock(&adapter->XmtZeroLock))
|
||||
goto lock_busy;
|
||||
}
|
||||
else
|
||||
spin_lock_irqsave(&adapter->XmtZeroLock, flags);
|
||||
spin_lock_irqsave(&adapter->XmtZeroLock, flags);
|
||||
|
||||
SXG_TRACE(TRACE_SXG, SxgTraceBuffer, TRACE_NOISY, "CmpSnds",
|
||||
adapter, XmtRingInfo->Head, XmtRingInfo->Tail, 0);
|
||||
|
@ -1545,36 +1539,23 @@ static void sxg_complete_slow_send(struct adapter_t *adapter, int irq_context)
|
|||
* chimney send, which results in a double trip
|
||||
* in SxgTcpOuput
|
||||
*/
|
||||
if(irq_context)
|
||||
spin_unlock(&adapter->XmtZeroLock);
|
||||
else
|
||||
spin_unlock_irqrestore(
|
||||
&adapter->XmtZeroLock, flags);
|
||||
spin_unlock_irqrestore(
|
||||
&adapter->XmtZeroLock, flags);
|
||||
|
||||
SxgSgl->DumbPacket = NULL;
|
||||
SXG_COMPLETE_DUMB_SEND(adapter, skb,
|
||||
FirstSgeAddress,
|
||||
FirstSgeLength);
|
||||
SXG_FREE_SGL_BUFFER(adapter, SxgSgl, NULL,
|
||||
irq_context);
|
||||
SXG_FREE_SGL_BUFFER(adapter, SxgSgl, NULL);
|
||||
/* and reacquire.. */
|
||||
if(irq_context) {
|
||||
if(!spin_trylock(&adapter->XmtZeroLock))
|
||||
goto lock_busy;
|
||||
}
|
||||
else
|
||||
spin_lock_irqsave(&adapter->XmtZeroLock, flags);
|
||||
spin_lock_irqsave(&adapter->XmtZeroLock, flags);
|
||||
}
|
||||
break;
|
||||
default:
|
||||
ASSERT(0);
|
||||
}
|
||||
}
|
||||
if(irq_context)
|
||||
spin_unlock(&adapter->XmtZeroLock);
|
||||
else
|
||||
spin_unlock_irqrestore(&adapter->XmtZeroLock, flags);
|
||||
lock_busy:
|
||||
spin_unlock_irqrestore(&adapter->XmtZeroLock, flags);
|
||||
SXG_TRACE(TRACE_SXG, SxgTraceBuffer, TRACE_NOISY, "CmpSnd",
|
||||
adapter, XmtRingInfo->Head, XmtRingInfo->Tail, 0);
|
||||
}
|
||||
|
@ -2468,7 +2449,7 @@ static int sxg_dumb_sgl(struct sxg_x64_sgl *pSgl,
|
|||
*/
|
||||
|
||||
spin_unlock_irqrestore(&adapter->XmtZeroLock, flags);
|
||||
sxg_complete_slow_send(adapter, 0);
|
||||
sxg_complete_slow_send(adapter);
|
||||
spin_lock_irqsave(&adapter->XmtZeroLock, flags);
|
||||
SXG_GET_CMD(XmtRing, XmtRingInfo, XmtCmd, SxgSgl);
|
||||
if (XmtCmd == NULL) {
|
||||
|
@ -3781,22 +3762,16 @@ static void sxg_allocate_sgl_buffer_complete(struct adapter_t *adapter,
|
|||
unsigned long sgl_flags;
|
||||
SXG_TRACE(TRACE_SXG, SxgTraceBuffer, TRACE_NOISY, "AlSglCmp",
|
||||
adapter, SxgSgl, Length, 0);
|
||||
if(!in_irq())
|
||||
spin_lock_irqsave(&adapter->SglQLock, sgl_flags);
|
||||
else
|
||||
spin_lock(&adapter->SglQLock);
|
||||
spin_lock_irqsave(&adapter->SglQLock, sgl_flags);
|
||||
adapter->AllSglBufferCount++;
|
||||
/* PhysicalAddress; */
|
||||
SxgSgl->PhysicalAddress = PhysicalAddress;
|
||||
/* Initialize backpointer once */
|
||||
SxgSgl->adapter = adapter;
|
||||
InsertTailList(&adapter->AllSglBuffers, &SxgSgl->AllList);
|
||||
if(!in_irq())
|
||||
spin_unlock_irqrestore(&adapter->SglQLock, sgl_flags);
|
||||
else
|
||||
spin_unlock(&adapter->SglQLock);
|
||||
spin_unlock_irqrestore(&adapter->SglQLock, sgl_flags);
|
||||
SxgSgl->State = SXG_BUFFER_BUSY;
|
||||
SXG_FREE_SGL_BUFFER(adapter, SxgSgl, NULL, in_irq());
|
||||
SXG_FREE_SGL_BUFFER(adapter, SxgSgl, NULL);
|
||||
SXG_TRACE(TRACE_SXG, SxgTraceBuffer, TRACE_NOISY, "XAlSgl",
|
||||
adapter, SxgSgl, Length, 0);
|
||||
}
|
||||
|
|
|
@ -244,20 +244,14 @@ struct sxg_stats {
|
|||
}
|
||||
|
||||
/* SGL macros */
|
||||
#define SXG_FREE_SGL_BUFFER(_pAdapt, _Sgl, _NB, _irq) { \
|
||||
if(!_irq) \
|
||||
spin_lock_irqsave(&(_pAdapt)->SglQLock, sgl_flags); \
|
||||
else \
|
||||
spin_lock(&(_pAdapt)->SglQLock); \
|
||||
#define SXG_FREE_SGL_BUFFER(_pAdapt, _Sgl, _NB) { \
|
||||
spin_lock_irqsave(&(_pAdapt)->SglQLock, sgl_flags); \
|
||||
(_pAdapt)->FreeSglBufferCount++; \
|
||||
ASSERT((_pAdapt)->AllSglBufferCount >= (_pAdapt)->FreeSglBufferCount); \
|
||||
ASSERT(!((_Sgl)->State & SXG_BUFFER_FREE)); \
|
||||
(_Sgl)->State = SXG_BUFFER_FREE; \
|
||||
InsertTailList(&(_pAdapt)->FreeSglBuffers, &(_Sgl)->FreeList); \
|
||||
if(!_irq) \
|
||||
spin_unlock_irqrestore(&(_pAdapt)->SglQLock, sgl_flags); \
|
||||
else \
|
||||
spin_unlock(&(_pAdapt)->SglQLock); \
|
||||
spin_unlock_irqrestore(&(_pAdapt)->SglQLock, sgl_flags); \
|
||||
}
|
||||
|
||||
/*
|
||||
|
@ -280,7 +274,7 @@ struct sxg_stats {
|
|||
if(!_irq) \
|
||||
spin_lock_irqsave(&(_pAdapt)->SglQLock, sgl_flags); \
|
||||
else \
|
||||
spin_lock(&(_pAdapt)->SglQLock); \
|
||||
spin_lock_irqsave(&(_pAdapt)->SglQLock, sgl_flags); \
|
||||
if((_pAdapt)->FreeSglBufferCount) { \
|
||||
ASSERT(!(IsListEmpty(&(_pAdapt)->FreeSglBuffers))); \
|
||||
_ple = RemoveHeadList(&(_pAdapt)->FreeSglBuffers); \
|
||||
|
@ -294,7 +288,7 @@ struct sxg_stats {
|
|||
if(!_irq) \
|
||||
spin_unlock_irqrestore(&(_pAdapt)->SglQLock, sgl_flags);\
|
||||
else \
|
||||
spin_unlock(&(_pAdapt)->SglQLock); \
|
||||
spin_unlock_irqrestore(&(_pAdapt)->SglQLock, sgl_flags);\
|
||||
}
|
||||
|
||||
/*
|
||||
|
|
Загрузка…
Ссылка в новой задаче