Skip to contentSkip to conversation

tidspbridge: don't flood the mailbox

PATCH In review Felipe Contreras 6 participants 21 messages Mar 18, 2009 – Mar 20, 2009

lore
Share & tools
lore
#0 lore raw

cover tidspbridge: don't flood the mailbox

This patch series is meant to be applied on top of the previous one (A) and fixes a very nasty problem that causes interrupts to be constantly dropped since the mailbox is full most of the time. The details are explained in patch 0001.

Flooding the mailbox causes reliability issues (messages lost) and performance issues (busy looping waiting for a slot in the mailbox).

Felipe Contreras (3):
tidspbridge: don't flood the mailbox on MemUnMap
tidspbridge: cleanup and remove HW_MMU_TLBFlushAll
tidspbridge: decreate timeout to a saner value

 drivers/dsp/bridge/hw/MMUAccInt.h   |    3 --
 drivers/dsp/bridge/hw/MMURegAcM.h   |   14 ----------
 drivers/dsp/bridge/hw/hw_mmu.c      |    9 ------
 drivers/dsp/bridge/hw/hw_mmu.h      |    2 -
 drivers/dsp/bridge/wmd/tiomap3430.c |   50 +++++++++++++++++++++-------------
 drivers/dsp/bridge/wmd/tiomap_sm.c  |    2 +-
 6 files changed, 32 insertions(+), 48 deletions(-)
#1 lore raw

PATCH 1/3 tidspbridge: don't flood the mailbox on MemUnMap

From: Felipe Contreras <[email protected]>

Impact: improves performance and reliability

Check the hardware state of the DSP before sending the command to wake it up thus avoiding to flood the mailbox unnecessarily.

By doing that the mailbox is free most of the time which dramatically decreases CPU usage since the check for empty slots is done in a busy loop.

Also, a free mailbox means less timeouts, so now most messages are posted which improves reliability.

MemMap is doing the flush properly, so this patch also refactors the code into a common flush_all function.

The problem can be triggered by doing many memmap/unmaps, just like TI's OpenMAX IL implementation.

---
 drivers/dsp/bridge/wmd/tiomap3430.c |   42 +++++++++++++++++++---------------
 1 files changed, 23 insertions(+), 19 deletions(-)
drivers/dsp/bridge/wmd/tiomap3430.c +23 -19
index c0f4ffc..b538ef7 100644
@@ -235,6 +235,25 @@ static struct WMD_DRV_INTERFACE drvInterfaceFxns = {
 	WMD_MSG_SetQueueId,
 };
 
+static inline void flush_all(struct WMD_DEV_CONTEXT *pDevContext)
+{
+	struct CFG_HOSTRES resources;
+	u32 temp = 0;
+
+	CFG_GetHostResources((struct CFG_DEVNODE *)DRV_GetFirstDevExtension(), &resources);
+	HW_PWRST_IVA2RegGet(resources.dwPrmBase, &temp);
+
+	if ((temp & HW_PWR_STATE_ON) == HW_PWR_STATE_OFF) {
+		/* IVA domain is not in ON state*/
+		DBG_Trace(DBG_LEVEL7, "temp value is 0x%x\n", temp);
+		CLK_Enable(SERVICESCLK_iva2_ck);
+		WakeDSP(pDevContext, NULL);
+		HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+		CLK_Disable(SERVICESCLK_iva2_ck);
+	} else
+		HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+}
+
 /*
  *  ======== WMD_DRV_Entry ========
  *  purpose:
@@ -1283,17 +1302,14 @@ static DSP_STATUS WMD_BRD_MemMap(struct WMD_DEV_CONTEXT *hDevContext,
 	struct WMD_DEV_CONTEXT *pDevContext = hDevContext;
 	struct HW_MMUMapAttrs_t hwAttrs;
 	u32 numOfActualTabEntries = 0;
-	u32 temp = 0;
-	struct CFG_HOSTRES resources;
 	u32 *pPhysAddrPageTbl = NULL;
 	struct vm_area_struct *vma;
 	struct mm_struct *mm = current->mm;
+	u32 temp = 0;
 
 	DBG_Trace(DBG_ENTER, "> WMD_BRD_MemMap hDevContext %x, pa %x, va %x, "
 		 "size %x, ulMapAttr %x\n", hDevContext, ulMpuAddr, ulVirtAddr,
 		 ulNumBytes, ulMapAttr);
-	status = CFG_GetHostResources(
-		 (struct CFG_DEVNODE *)DRV_GetFirstDevExtension(), &resources);
 	if (ulNumBytes == 0)
 		return DSP_EINVALIDARG;
 
@@ -1421,16 +1437,7 @@ func_cont:
 	 * This is called from here instead from PteUpdate to avoid unnecessary
 	 * repetition while mapping non-contiguous physical regions of a virtual
 	 * region */
-	HW_PWRST_IVA2RegGet(resources.dwPrmBase, &temp);
-	if ((temp & HW_PWR_STATE_ON) == HW_PWR_STATE_OFF) {
-		/* IVA domain is not in ON state*/
-		DBG_Trace(DBG_LEVEL7, "temp value is 0x%x\n", temp);
-		CLK_Enable(SERVICESCLK_iva2_ck);
-		WakeDSP(pDevContext, NULL);
-		HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
-		CLK_Disable(SERVICESCLK_iva2_ck);
-	} else
-		HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+	flush_all(pDevContext);
 	DBG_Trace(DBG_ENTER, "< WMD_BRD_MemMap status %x\n", status);
 	return status;
 }
@@ -1571,8 +1578,7 @@ static DSP_STATUS WMD_BRD_MemUnMap(struct WMD_DEV_CONTEXT *hDevContext,
 	 /* It is better to flush the TLB here, so that any stale old entries
 	 * get flushed */
 EXIT_LOOP:
-	CHNLSM_InterruptDSP2(pDevContext, MBX_PM_DSPWAKEUP);
-	HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+	flush_all(pDevContext);
 	DBG_Trace(DBG_LEVEL1, "WMD_BRD_MemUnMap vaCurr %x, pteAddrL1 %x "
 		  "pteAddrL2 %x\n", vaCurr, pteAddrL1, pteAddrL2);
 	DBG_Trace(DBG_ENTER, "< WMD_BRD_MemUnMap status %x remBytes %x, "
@@ -2055,9 +2061,7 @@ func_cont:
 	 * This is called from here instead from PteUpdate to avoid unnecessary
 	 * repetition while mapping non-contiguous physical regions of a virtual
 	 * region */
-	/* Waking up DSP before calling TLB Flush */
-	CHNLSM_InterruptDSP2(pDevContext, MBX_PM_DSPWAKEUP);
-	HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+	flush_all(pDevContext);
 	DBG_Trace(DBG_LEVEL7, "< WMD_BRD_MemMap  at end status %x\n", status);
 	return status;
 }
Signature · 1 line
1.6.2.1.287.g9a8be
#2 lore raw

PATCH 2/3 tidspbridge: cleanup and remove HW_MMU_TLBFlushAll

From: Felipe Contreras <[email protected]>

It doesn't make sense to have layers and layers of constants and defines to turn one bit on.

---
 drivers/dsp/bridge/hw/MMUAccInt.h   |    3 ---
 drivers/dsp/bridge/hw/MMURegAcM.h   |   14 --------------
 drivers/dsp/bridge/hw/hw_mmu.c      |    9 ---------
 drivers/dsp/bridge/hw/hw_mmu.h      |    2 --
 drivers/dsp/bridge/wmd/tiomap3430.c |   12 ++++++++++--
 5 files changed, 10 insertions(+), 30 deletions(-)
drivers/dsp/bridge/hw/MMUAccInt.h +0 -3
index 78e1d15..6ca1573 100644
@@ -41,7 +41,6 @@
 #define EASIL1_MMUMMU_LD_TLBWriteRegister32   (MMU_BASE_EASIL1 + 214)
 #define EASIL1_MMUMMU_CAMWriteRegister32   (MMU_BASE_EASIL1 + 226)
 #define EASIL1_MMUMMU_RAMWriteRegister32 (MMU_BASE_EASIL1 + 268)
-#define EASIL1_MMUMMU_GFLUSHGlobalFlushWrite32 (MMU_BASE_EASIL1 + 317)
 #define EASIL1_MMUMMU_FLUSH_ENTRYWriteRegister32  (MMU_BASE_EASIL1 + 322)
 
 /* Register offset address definitions */
@@ -73,7 +72,5 @@
 #define MMU_MMU_LOCK_BaseValue_OFFSET   10
 #define MMU_MMU_LOCK_CurrentVictim_MASK   0x3f0
 #define MMU_MMU_LOCK_CurrentVictim_OFFSET    4
-#define MMU_MMU_GFLUSH_GlobalFlush_MASK 0x1
-#define MMU_MMU_GFLUSH_GlobalFlush_OFFSET   0
 
 #endif /* _MMU_ACC_INT_H */
drivers/dsp/bridge/hw/MMURegAcM.h +0 -14
index a130b1a..e46fdcb 100644
@@ -239,20 +239,6 @@
 }
 
 
-#define MMUMMU_GFLUSHGlobalFlushWrite32(baseAddress, value)\
-{\
-    const u32 offset = MMU_MMU_GFLUSH_OFFSET;\
-    register u32 data = RD_MEM_32_VOLATILE((baseAddress)+offset);\
-    register u32 newValue = (value);\
-    _DEBUG_LEVEL_1_EASI(EASIL1_MMUMMU_GFLUSHGlobalFlushWrite32);\
-    data &= ~(MMU_MMU_GFLUSH_GlobalFlush_MASK);\
-    newValue <<= MMU_MMU_GFLUSH_GlobalFlush_OFFSET;\
-    newValue &= MMU_MMU_GFLUSH_GlobalFlush_MASK;\
-    newValue |= data;\
-    WR_MEM_32_VOLATILE(baseAddress+offset, newValue);\
-}
-
-
 #define MMUMMU_FLUSH_ENTRYWriteRegister32(baseAddress, value)\
 {\
     const u32 offset = MMU_MMU_FLUSH_ENTRY_OFFSET;\
drivers/dsp/bridge/hw/hw_mmu.c +0 -9
index da7e092..ab65de0 100644
@@ -212,15 +212,6 @@ HW_STATUS HW_MMU_VictimNumSet(const u32 baseAddress,
     return status;
 }
 
-HW_STATUS HW_MMU_TLBFlushAll(const u32 baseAddress)
-{
-    HW_STATUS status = RET_OK;
-
-    MMUMMU_GFLUSHGlobalFlushWrite32(baseAddress, HW_SET);
-
-    return status;
-}
-
 HW_STATUS HW_MMU_EventAck(const u32 baseAddress, u32 irqMask)
 {
     HW_STATUS status = RET_OK;
drivers/dsp/bridge/hw/hw_mmu.h +0 -2
index 924f32b..4783276 100644
@@ -91,8 +91,6 @@ extern HW_STATUS HW_MMU_TLBFlush(const u32 baseAddress,
 				    u32 virtualAddr,
 				    u32 pageSize);
 
-extern HW_STATUS HW_MMU_TLBFlushAll(const u32 baseAddress);
-
 extern HW_STATUS HW_MMU_TLBAdd(const u32     baseAddress,
 				  u32	   physicalAddr,
 				  u32	   virtualAddr,
drivers/dsp/bridge/wmd/tiomap3430.c +10 -2
index b538ef7..30e0cb3 100644
@@ -88,6 +88,9 @@
 #define MMU_LARGE_PAGE_MASK      0xFFFF0000
 #define MMU_SMALL_PAGE_MASK      0xFFFFF000
 #define PAGES_II_LVL_TABLE   512
+
+#define MMU_GFLUSH 0x60
+
 /* Forward Declarations: */
 static DSP_STATUS WMD_BRD_Monitor(struct WMD_DEV_CONTEXT *pDevContext);
 static DSP_STATUS WMD_BRD_Read(struct WMD_DEV_CONTEXT *pDevContext,
@@ -235,6 +238,11 @@ static struct WMD_DRV_INTERFACE drvInterfaceFxns = {
 	WMD_MSG_SetQueueId,
 };
 
+static inline void tlb_flush_all(const u32 base)
+{
+    __raw_writeb(__raw_readb(base + MMU_GFLUSH) | 1, base + MMU_GFLUSH);
+}
+
 static inline void flush_all(struct WMD_DEV_CONTEXT *pDevContext)
 {
 	struct CFG_HOSTRES resources;
@@ -248,10 +256,10 @@ static inline void flush_all(struct WMD_DEV_CONTEXT *pDevContext)
 		DBG_Trace(DBG_LEVEL7, "temp value is 0x%x\n", temp);
 		CLK_Enable(SERVICESCLK_iva2_ck);
 		WakeDSP(pDevContext, NULL);
-		HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+		tlb_flush_all(pDevContext->dwDSPMmuBase);
 		CLK_Disable(SERVICESCLK_iva2_ck);
 	} else
-		HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+		tlb_flush_all(pDevContext->dwDSPMmuBase);
 }
 
 /*
Signature · 1 line
1.6.2.1.287.g9a8be
#3 lore raw

PATCH 3/3 tidspbridge: decreate timeout to a saner value

From: Felipe Contreras <[email protected]>

---
 drivers/dsp/bridge/wmd/tiomap_sm.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/dsp/bridge/wmd/tiomap_sm.c +1 -1
index 1b31162..535dc13 100644
@@ -144,7 +144,7 @@ DSP_STATUS CHNLSM_InterruptDSP2(struct WMD_DEV_CONTEXT *pDevContext,
 
 		pDevContext->dwBrdState = BRD_RUNNING;
 	}
-	timeout = jiffies + msecs_to_jiffies(10);
+	timeout = jiffies + msecs_to_jiffies(1);
 	while (fifo_full((void __iomem *) resources.dwMboxBase, 0)) {
 		if (time_after(jiffies, timeout)) {
 			printk(KERN_ERR "dspbridge: timed out waiting for mailbox\n");
Signature · 1 line
1.6.2.1.287.g9a8be
#4 lore raw

RE: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Hi Felipe,

I am seeing with this patch because of the timeout:

DSP device detected !!
DSPProcessor_Attach succeeded.
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
...

Did you see any issue when you change to 1ms? Maybe we can use a bigger timeout.

Please let me know your comments.

Regards,
Fernando.

-----Original Message-----
From: Felipe Contreras [mailto:[email protected]]
Sent: Tuesday, March 17, 2009 7:27 PM
To: [email protected]
Cc: Kanigeri, Hari; Hiroshi DOYU; Ameya Palande; Guzman Lugo, Fernando; Felipe Contreras
Subject: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

From: Felipe Contreras <[email protected]>

---
 drivers/dsp/bridge/wmd/tiomap_sm.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
drivers/dsp/bridge/wmd/tiomap_sm.c +1 -1
index 1b31162..535dc13 100644
@@ -144,7 +144,7 @@ DSP_STATUS CHNLSM_InterruptDSP2(struct WMD_DEV_CONTEXT *pDevContext,
 
 		pDevContext->dwBrdState = BRD_RUNNING;
 	}
-	timeout = jiffies + msecs_to_jiffies(10);
+	timeout = jiffies + msecs_to_jiffies(1);
 	while (fifo_full((void __iomem *) resources.dwMboxBase, 0)) {
 		if (time_after(jiffies, timeout)) {
 			printk(KERN_ERR "dspbridge: timed out waiting for mailbox\n");
Signature · 1 line
1.6.2.1.287.g9a8be
#5 lore raw

Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Quoted 21 lines · Fernando

Hi Felipe,

       I am seeing with this patch because of the timeout:

DSP device detected !! DSPProcessor_Attach succeeded.
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
...

Quoted 1 line

Did you see any issue when you change to 1ms? Maybe we can use a bigger timeout.

Did you apply patch #1 of the B series?

I didn't see any timeout on my tests.

Signature · 5 lines
Felipe Contreras
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#6 lore raw

RE: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Yes, I applied this; in fact I have applied all the patches. If I increase the timeout there are no problems. The test I run creates 4 process and each one does several a lot of calls to InputChnl and OutputChnl, so I think this test is using the mailbox a lot and would be better a bigger timeout. What do you think?

Regards,
Fernando.
-----Original Message-----
From: Felipe Contreras [mailto:[email protected]]
Sent: Thursday, March 19, 2009 3:04 PM
To: Guzman Lugo, Fernando
Cc: [email protected]; Kanigeri, Hari; Hiroshi DOYU; Ameya Palande; Felipe Contreras
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Quoted 21 lines · Fernando

Hi Felipe,

       I am seeing with this patch because of the timeout:

DSP device detected !! DSPProcessor_Attach succeeded.
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
dspbridge: timed out waiting for mailbox
...

Quoted 1 line

Did you see any issue when you change to 1ms? Maybe we can use a bigger timeout.

Did you apply patch #1 of the B series?

I didn't see any timeout on my tests.

Signature · 6 lines
Felipe Contreras

--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#7 lore raw

Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Quoted 2 lines · Fernando

Yes, I applied this; in fact I have applied all the patches. If I increase the timeout there are no problems. The test I run creates 4 process and each one does several a lot of calls to InputChnl and OutputChnl, so I think this test is using the mailbox a lot and would be better a bigger timeout. What do you think?

How fast are these messages sent? Can you track down which functions are calling CHNLSM_InterruptDSP2 and making these timeouts happen.

I think it's safe to leave the timeout at 10, but that means it's possible the code will be busy-looping up to 10 ms which will increase the CPU load. Somebody from Nokia (Siarhei?) suggested to idle-wait for the mbox empty irq, I think that's the best way to implement this, but at least for the use cases I'm interested in (video encoding/decoding) timeouts don't seem to be an issue anymore.

Signature · 1 line
Felipe Contreras
#8 lore raw

RE: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

This is a stress test, it creates 4 processes and each process will do 1000 transfers using streams, so the trace is:

STRM_Issue -> WMD_CHNL_AddIOReq -> IO_Schedule

IO_Schedule schedules a call to IO_DPC using task let.

IO_DPC -> IO_DispatchChnl -> InputChnl -> CHNLSM_InterruptDSP2

Also IO_DispatchChnl -> OutputChnl -> CHNLSM_InterruptDSP2.

As we can call a lot CHNLSM_InterruptDSP2 in this test, there is a problem with the timeout. However running other tests, videos and mp3 there no problems. I think we should change to 10ms, only to make sure there is no problem when CHNLSM_InterruptDSP2 is called a lot.

Let me know if you are agreed. Or have some comments about it.

Regards,
Fernando.

-----Original Message-----
From: Felipe Contreras [mailto:[email protected]]
Sent: Thursday, March 19, 2009 3:35 PM
To: Guzman Lugo, Fernando
Cc: [email protected]; Kanigeri, Hari; Hiroshi DOYU; Ameya Palande; Felipe Contreras
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Quoted 2 lines · Fernando

Yes, I applied this; in fact I have applied all the patches. If I increase the timeout there are no problems. The test I run creates 4 process and each one does several a lot of calls to InputChnl and OutputChnl, so I think this test is using the mailbox a lot and would be better a bigger timeout. What do you think?

How fast are these messages sent? Can you track down which functions are calling CHNLSM_InterruptDSP2 and making these timeouts happen.

I think it's safe to leave the timeout at 10, but that means it's possible the code will be busy-looping up to 10 ms which will increase the CPU load. Somebody from Nokia (Siarhei?) suggested to idle-wait for the mbox empty irq, I think that's the best way to implement this, but at least for the use cases I'm interested in (video encoding/decoding) timeouts don't seem to be an issue anymore.

Signature · 1 line
Felipe Contreras
#9 lore raw

Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Quoted 9 lines · Fernando

This is a stress test, it creates 4 processes and each process will do 1000 transfers using streams, so the trace is:

STRM_Issue ->  WMD_CHNL_AddIOReq  -> IO_Schedule

IO_Schedule schedules a call to IO_DPC using task let.

Quoted 8 lines

IO_DPC -> IO_DispatchChnl -> InputChnl -> CHNLSM_InterruptDSP2

Also      IO_DispatchChnl -> OutputChnl -> CHNLSM_InterruptDSP2.

As we can call a lot CHNLSM_InterruptDSP2 in this test, there is a problem with the timeout. However running other tests, videos and mp3 there no problems. I think we should change to 10ms, only to make sure there is no problem when CHNLSM_InterruptDSP2 is called a lot.

Let me know if you are agreed. Or have some comments about it.

Well again, the best way to implement the wait for a slot in the mailbox is with interrupts, not busy-looping. If we busy-loop too much that would increase the CPU usage, and that would be bad.

That's why I want to use the 1ms timeout; to find issues that cause increase in CPU load.

But for now I think 10ms is the safest, as it's the current value. If somebody wants to pin-point issues, then the timeout should be decreased.

Signature · 5 lines
Felipe Contreras
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#10 lore raw

Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

From: ext Felipe Contreras <[email protected]>
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value
Date: Fri, 20 Mar 2009 01:06:16 +0100

Quoted 15 lines · Fernando
Quoted 14 lines

This is a stress test, it creates 4 processes and each process will do 1000 transfers using streams, so the trace is:

STRM_Issue ->  WMD_CHNL_AddIOReq  -> IO_Schedule

IO_Schedule schedules a call to IO_DPC using task let.

IO_DPC -> IO_DispatchChnl -> InputChnl -> CHNLSM_InterruptDSP2

Also      IO_DispatchChnl -> OutputChnl -> CHNLSM_InterruptDSP2.

Quoted 7 lines
Quoted 3 lines

As we can call a lot CHNLSM_InterruptDSP2 in this test, there is a problem with the timeout. However running other tests, videos and mp3 there no problems. I think we should change to 10ms, only to make sure there is no problem when CHNLSM_InterruptDSP2 is called a lot.

Let me know if you are agreed. Or have some comments about it.

Well again, the best way to implement the wait for a slot in the mailbox is with interrupts, not busy-looping. If we busy-loop too much that would increase the CPU usage, and that would be bad.

I think that s/w queuing of messages would be more efficient to allow multiple senders to continue thier work anyway, especially in the case of having these streamings.

Quoted 10 lines

That's why I want to use the 1ms timeout; to find issues that cause increase in CPU load.

But for now I think 10ms is the safest, as it's the current value. If somebody wants to pin-point issues, then the timeout should be decreased.

Signature · 1 line
Felipe Contreras

--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#11 lore raw

Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

From: Hiroshi DOYU <[email protected]>
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value
Date: Fri, 20 Mar 2009 06:54:04 +0200 (EET)

Quoted 23 lines

From: ext Felipe Contreras <[email protected]>
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value
Date: Fri, 20 Mar 2009 01:06:16 +0100

Quoted 19 lines · Fernando
Quoted 17 lines

This is a stress test, it creates 4 processes and each process will do 1000 transfers using streams, so the trace is:

STRM_Issue ->  WMD_CHNL_AddIOReq  -> IO_Schedule

IO_Schedule schedules a call to IO_DPC using task let.

IO_DPC -> IO_DispatchChnl -> InputChnl -> CHNLSM_InterruptDSP2

Also      IO_DispatchChnl -> OutputChnl -> CHNLSM_InterruptDSP2.

As we can call a lot CHNLSM_InterruptDSP2 in this test, there is a problem with the timeout. However running other tests, videos and mp3 there no problems. I think we should change to 10ms, only to make sure there is no problem when CHNLSM_InterruptDSP2 is called a lot.

Let me know if you are agreed. Or have some comments about it.

Quoted 7 lines
Quoted 3 lines

Well again, the best way to implement the wait for a slot in the mailbox is with interrupts, not busy-looping. If we busy-loop too much that would increase the CPU usage, and that would be bad.

I think that s/w queuing of messages would be more efficient to allow multiple senders to continue thier work anyway, especially in the case of having these streamings.

Migrating to the existing mailbox driver("plat-omap/mailbox.c") is also another option to support multiple OMAP architectures at the same time. That mailbox driver has already had a message queuing feature inside and it's buildable on OMAP3.
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#12 lore raw

Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Quoted 31 lines · Hiroshi DOYU

From: Hiroshi DOYU <[email protected]>
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value
Date: Fri, 20 Mar 2009 06:54:04 +0200 (EET)

Quoted 27 lines

From: ext Felipe Contreras <[email protected]>
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value
Date: Fri, 20 Mar 2009 01:06:16 +0100

Quoted 22 lines · Fernando
Quoted 9 lines

This is a stress test, it creates 4 processes and each process will do 1000 transfers using streams, so the trace is:

STRM_Issue ->  WMD_CHNL_AddIOReq  -> IO_Schedule

IO_Schedule schedules a call to IO_DPC using task let.

Quoted 8 lines

IO_DPC -> IO_DispatchChnl -> InputChnl -> CHNLSM_InterruptDSP2

Also      IO_DispatchChnl -> OutputChnl -> CHNLSM_InterruptDSP2.

As we can call a lot CHNLSM_InterruptDSP2 in this test, there is a problem with the timeout. However running other tests, videos and mp3 there no problems. I think we should change to 10ms, only to make sure there is no problem when CHNLSM_InterruptDSP2 is called a lot.

Let me know if you are agreed. Or have some comments about it.

Well again, the best way to implement the wait for a slot in the mailbox is with interrupts, not busy-looping. If we busy-loop too much that would increase the CPU usage, and that would be bad.

Quoted 8 lines
Quoted 3 lines

I think that s/w queuing of messages would be more efficient to allow multiple senders to continue thier work anyway, especially in the case of having these streamings.

Migrating to the existing mailbox driver("plat-omap/mailbox.c") is also another option to support multiple OMAP architectures at the same time. That mailbox driver has already had a message queuing feature inside and it's buildable on OMAP3.

Yeap, I have it in my to-do list, so if somebody doesn't beat me to it I play with that.

Is there a way to play with the existing mailbox driver independently of dsp-bridge?

Signature · 5 lines
Felipe Contreras
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#13 lore raw

RE: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Felipe,

Quoted 58 lines

-----Original Message-----
From: Felipe Contreras [mailto:[email protected]]
Sent: Friday, March 20, 2009 3:48 PM
To: Hiroshi DOYU
Cc: Gupta, Ramesh; Kanigeri, Hari; Guzman Lugo, Fernando; [email protected]; [email protected]
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Quoted 7 lines · Hiroshi DOYU

From: Hiroshi DOYU <[email protected]>
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value
Date: Fri, 20 Mar 2009 06:54:04 +0200 (EET)

Quoted 2 lines

From: ext Felipe Contreras <[email protected]>
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout

to a saner

Quoted 4 lines
Quoted 4 lines

value
Date: Fri, 20 Mar 2009 01:06:16 +0100

Quoted 1 line

On Fri, Mar 20, 2009 at 2:00 AM, Guzman Lugo, Fernando

Quoted 4 lines · <[email protected]> wrote:
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines

This is a stress test, it creates 4 processes and each

process will do 1000 transfers using streams, so the trace is:

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

STRM_Issue ->  WMD_CHNL_AddIOReq  -> IO_Schedule

IO_Schedule schedules a call to IO_DPC using task let.

IO_DPC -> IO_DispatchChnl -> InputChnl -> CHNLSM_InterruptDSP2

Also      IO_DispatchChnl -> OutputChnl ->

CHNLSM_InterruptDSP2.

Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines

As we can call a lot CHNLSM_InterruptDSP2 in this

test, there is a problem with the timeout. However running other tests, videos and mp3 there no problems. I think we should change to 10ms, only to make sure there is no problem when CHNLSM_InterruptDSP2 is called a lot.

Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 2 lines

Let me know if you are agreed. Or have some comments about it.

Well again, the best way to implement the wait for a slot in the mailbox is with interrupts, not busy-looping. If we

busy-loop too

Quoted 3 lines
Quoted 3 lines
Quoted 1 line

much that would increase the CPU usage, and that would be bad.

I think that s/w queuing of messages would be more

efficient to allow

Quoted 3 lines
Quoted 2 lines

multiple senders to continue thier work anyway, especially in the case of having these streamings.

Quoted 8 lines
Quoted 2 lines

Migrating to the existing mailbox driver("plat-omap/mailbox.c") is also another option to support multiple OMAP architectures

at the same

Quoted 2 lines

time. That mailbox driver has already had a message queuing feature inside and it's buildable on OMAP3.

Yeap, I have it in my to-do list, so if somebody doesn't beat me to it I play with that.

We already started looking into this :), made some progress, will provide more updates in a day or two.

Regards
Ramesh Gupta G
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#14 lore raw

Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Quoted 18 lines · Ramesh

Felipe,

Quoted 15 lines

-----Original Message-----
From: Felipe Contreras [mailto:[email protected]]
Sent: Friday, March 20, 2009 3:48 PM
To: Hiroshi DOYU
Cc: Gupta, Ramesh; Kanigeri, Hari; Guzman Lugo, Fernando; [email protected]; [email protected]
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Quoted 3 lines · Hiroshi DOYU

From: Hiroshi DOYU <[email protected]>
Migrating to the existing mailbox driver("plat-omap/mailbox.c") is also another option to support multiple OMAP architectures

at the same

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

time. That mailbox driver has already had a message queuing feature inside and it's buildable on OMAP3.

Yeap, I have it in my to-do list, so if somebody doesn't beat me to it I play with that.

We already started looking into this :), made some progress, will provide more updates in a day or two.

Sweet! :)

Signature · 1 line
Felipe Contreras
#15 lore raw

Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Quoted 22 lines · Felipe Contreras
Quoted 21 lines · Ramesh

Felipe,

Quoted 18 lines

-----Original Message-----
From: Felipe Contreras [mailto:[email protected]]
Sent: Friday, March 20, 2009 3:48 PM
To: Hiroshi DOYU
Cc: Gupta, Ramesh; Kanigeri, Hari; Guzman Lugo, Fernando; [email protected]; [email protected]
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

Quoted 3 lines · Hiroshi DOYU

From: Hiroshi DOYU <[email protected]>
Migrating to the existing mailbox driver("plat-omap/mailbox.c") is also another option to support multiple OMAP architectures

at the same

Quoted 2 lines

time. That mailbox driver has already had a message queuing feature inside and it's buildable on OMAP3.

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

Yeap, I have it in my to-do list, so if somebody doesn't beat me to it I play with that.

We already started looking into this :), made some progress, will provide more updates in a day or two.

Sweet! :)

Good to hear. For the common features, we need to have code that can be shared among all the DSP implementations.

Tony

#16 lore raw

Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value

In reply to Hiroshi DOYU
Quoted 23 lines · Hiroshi DOYU

From: ext Felipe Contreras <[email protected]>
Subject: Re: [PATCH B 3/3] tidspbridge: decreate timeout to a saner value
Date: Fri, 20 Mar 2009 01:06:16 +0100

Quoted 19 lines · Fernando
Quoted 17 lines

This is a stress test, it creates 4 processes and each process will do 1000 transfers using streams, so the trace is:

STRM_Issue ->  WMD_CHNL_AddIOReq  -> IO_Schedule

IO_Schedule schedules a call to IO_DPC using task let.

IO_DPC -> IO_DispatchChnl -> InputChnl -> CHNLSM_InterruptDSP2

Also      IO_DispatchChnl -> OutputChnl -> CHNLSM_InterruptDSP2.

As we can call a lot CHNLSM_InterruptDSP2 in this test, there is a problem with the timeout. However running other tests, videos and mp3 there no problems. I think we should change to 10ms, only to make sure there is no problem when CHNLSM_InterruptDSP2 is called a lot.

Let me know if you are agreed. Or have some comments about it.

Quoted 7 lines
Quoted 3 lines

Well again, the best way to implement the wait for a slot in the mailbox is with interrupts, not busy-looping. If we busy-loop too much that would increase the CPU usage, and that would be bad.

I think that s/w queuing of messages would be more efficient to allow multiple senders to continue thier work anyway, especially in the case of having these streamings.

Indeed. But what would happen if the application is sending messages way too fast for the DSP to handle? For example, some encoding algorithm might be too heavy, and if we are in a live situation, like video call, then it's ok to drop messages, but user-space needs to be notified of these and adjust the quality-of-service. But of course some other messages, like control messages (start, stop, etc.) should not be dropped, ever, so they must be queued.

Signature · 5 lines
Felipe Contreras
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#17 lore raw

Re: [PATCH B 2/3] tidspbridge: cleanup and remove HW_MMU_TLBFlushAll

In reply to Felipe Contreras
Quoted 7 lines · Felipe Contreras

From: Felipe Contreras <[email protected]>

It doesn't make sense to have layers and layers of constants and defines to turn one bit on.

---

I thought a bit more about this:

drivers/dsp/bridge/wmd/tiomap3430.c static struct WMD_DRV_INTERFACE drvInterfaceFxns = { · line 245

Quoted 5 lines
drivers/dsp/bridge/wmd/tiomap3430.c +5 -0
+static inline void tlb_flush_all(const u32 base)
+{
+    __raw_writeb(__raw_readb(base + MMU_GFLUSH) | 1, base + MMU_GFLUSH);
+}
+

Originally it was the equivalent of __raw_writel(__raw_readl but since we are dealing only with one bit I chose 'b' instead of 'l'. Is that correct?

Signature · 5 lines
Felipe Contreras
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#18 lore raw

RE: [PATCH B 1/3] tidspbridge: don't flood the mailbox on MemUnMap

In reply to Felipe Contreras

Felipe,

Quoted 3 lines

Check the hardware state of the DSP before sending the command to wake it up thus avoiding to flood the mailbox unnecessarily.

-- So this check was missing in Unmap part. Good catch.

drivers/dsp/bridge/wmd/tiomap3430.c static DSP_STATUS WMD_BRD_MemMap(struct WMD_DEV_CONTEXT *hDevContext, · line 1308

Quoted 7 lines
drivers/dsp/bridge/wmd/tiomap3430.c +1 -2
-	u32 temp = 0;
-	struct CFG_HOSTRES resources;
 	u32 *pPhysAddrPageTbl = NULL;
 	struct vm_area_struct *vma;
 	struct mm_struct *mm = current->mm;
+	u32 temp = 0;

-- Is temp variable needed ?

Thank you,
Best regards,
Hari

Quoted 138 lines

-----Original Message-----
From: Felipe Contreras [mailto:[email protected]]
Sent: Tuesday, March 17, 2009 8:27 PM
To: [email protected]
Cc: Kanigeri, Hari; Hiroshi DOYU; Ameya Palande; Guzman Lugo, Fernando; Felipe Contreras
Subject: [PATCH B 1/3] tidspbridge: don't flood the mailbox on MemUnMap

From: Felipe Contreras <[email protected]>

Impact: improves performance and reliability

Check the hardware state of the DSP before sending the command to wake it up thus avoiding to flood the mailbox unnecessarily.

By doing that the mailbox is free most of the time which dramatically decreases CPU usage since the check for empty slots is done in a busy loop.

Also, a free mailbox means less timeouts, so now most messages are posted which improves reliability.

MemMap is doing the flush properly, so this patch also refactors the code into a common flush_all function.

The problem can be triggered by doing many memmap/unmaps, just like TI's OpenMAX IL implementation.

---
 drivers/dsp/bridge/wmd/tiomap3430.c |   42 +++++++++++++++++++-----------
----
 1 files changed, 23 insertions(+), 19 deletions(-)
drivers/dsp/bridge/wmd/tiomap3430.c +19 -0
index c0f4ffc..b538ef7 100644
@@ -235,6 +235,25 @@ static struct WMD_DRV_INTERFACE drvInterfaceFxns = {
 	WMD_MSG_SetQueueId,
 };
 
+static inline void flush_all(struct WMD_DEV_CONTEXT *pDevContext)
+{
+	struct CFG_HOSTRES resources;
+	u32 temp = 0;
+
+	CFG_GetHostResources((struct CFG_DEVNODE
 *)DRV_GetFirstDevExtension(), &resources);
+	HW_PWRST_IVA2RegGet(resources.dwPrmBase, &temp);
+
+	if ((temp & HW_PWR_STATE_ON) == HW_PWR_STATE_OFF) {
+		/* IVA domain is not in ON state*/
+		DBG_Trace(DBG_LEVEL7, "temp value is 0x%x\n", temp);
+		CLK_Enable(SERVICESCLK_iva2_ck);
+		WakeDSP(pDevContext, NULL);
+		HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+		CLK_Disable(SERVICESCLK_iva2_ck);
+	} else
+		HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+}
+
 /*
  *  ======== WMD_DRV_Entry ========

* purpose:

drivers/dsp/bridge/wmd/tiomap3430.c +1 -2
@@ -1283,17 +1302,14 @@ static DSP_STATUS WMD_BRD_MemMap(struct
 WMD_DEV_CONTEXT *hDevContext,
 	struct WMD_DEV_CONTEXT *pDevContext = hDevContext;
 	struct HW_MMUMapAttrs_t hwAttrs;
 	u32 numOfActualTabEntries = 0;
-	u32 temp = 0;
-	struct CFG_HOSTRES resources;
 	u32 *pPhysAddrPageTbl = NULL;
 	struct vm_area_struct *vma;
 	struct mm_struct *mm = current->mm;
+	u32 temp = 0;
 
 	DBG_Trace(DBG_ENTER, "> WMD_BRD_MemMap hDevContext %x, pa %x, va %x,

"
"size %x, ulMapAttr %x\n", hDevContext, ulMpuAddr,
ulVirtAddr,

drivers/dsp/bridge/wmd/tiomap3430.c +0 -2
 		 ulNumBytes, ulMapAttr);
-	status = CFG_GetHostResources(
-		 (struct CFG_DEVNODE *)DRV_GetFirstDevExtension(),

&resources);

drivers/dsp/bridge/wmd/tiomap3430.c +0 -0
 	if (ulNumBytes == 0)
 		return DSP_EINVALIDARG;
@@ -1421,16 +1437,7 @@ func_cont:
 	 * This is called from here instead from PteUpdate to avoid

unnecessary
* repetition while mapping non-contiguous physical regions of a
virtual

drivers/dsp/bridge/wmd/tiomap3430.c +1 -10
 	 * region */
-	HW_PWRST_IVA2RegGet(resources.dwPrmBase, &temp);
-	if ((temp & HW_PWR_STATE_ON) == HW_PWR_STATE_OFF) {
-		/* IVA domain is not in ON state*/
-		DBG_Trace(DBG_LEVEL7, "temp value is 0x%x\n", temp);
-		CLK_Enable(SERVICESCLK_iva2_ck);
-		WakeDSP(pDevContext, NULL);
-		HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
-		CLK_Disable(SERVICESCLK_iva2_ck);
-	} else
-		HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+	flush_all(pDevContext);
 	DBG_Trace(DBG_ENTER, "< WMD_BRD_MemMap status %x\n", status);
 	return status;
 }
@@ -1571,8 +1578,7 @@ static DSP_STATUS WMD_BRD_MemUnMap(struct
 WMD_DEV_CONTEXT *hDevContext,
 	 /* It is better to flush the TLB here, so that any stale old

entries

drivers/dsp/bridge/wmd/tiomap3430.c +1 -2
 	 * get flushed */
 EXIT_LOOP:
-	CHNLSM_InterruptDSP2(pDevContext, MBX_PM_DSPWAKEUP);
-	HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+	flush_all(pDevContext);
 	DBG_Trace(DBG_LEVEL1, "WMD_BRD_MemUnMap vaCurr %x, pteAddrL1 %x "
 		  "pteAddrL2 %x\n", vaCurr, pteAddrL1, pteAddrL2);
 	DBG_Trace(DBG_ENTER, "< WMD_BRD_MemUnMap status %x remBytes %x, "
@@ -2055,9 +2061,7 @@ func_cont:
 	 * This is called from here instead from PteUpdate to avoid

unnecessary
* repetition while mapping non-contiguous physical regions of a
virtual

drivers/dsp/bridge/wmd/tiomap3430.c +1 -3
 	 * region */
-	/* Waking up DSP before calling TLB Flush */
-	CHNLSM_InterruptDSP2(pDevContext, MBX_PM_DSPWAKEUP);
-	HW_MMU_TLBFlushAll(pDevContext->dwDSPMmuBase);
+	flush_all(pDevContext);
 	DBG_Trace(DBG_LEVEL7, "< WMD_BRD_MemMap  at end status %x\n",

status);
return status;
}
--
1.6.2.1.287.g9a8be

#19 lore raw

Re: [PATCH B 1/3] tidspbridge: don't flood the mailbox on MemUnMap

Quoted 16 lines · Hari

Felipe,

Quoted 3 lines

Check the hardware state of the DSP before sending the command to wake it up thus avoiding to flood the mailbox unnecessarily.

-- So this check was missing in Unmap part. Good catch.

Quoted 7 lines

-     u32 temp = 0;
-     struct CFG_HOSTRES resources;
      u32 *pPhysAddrPageTbl = NULL;
      struct vm_area_struct *vma;
      struct mm_struct *mm = current->mm;
+     u32 temp = 0;

Quoted 1 line

-- Is temp variable needed ?

You tell me :)

I just moved code around. I don't truly understand what it's doing. Maybe the read is needed for some kind of sync?

I've removed the tlb flush all completely and I didn't notice any problems, is it really needed?

Signature · 5 lines
Felipe Contreras
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#20 lore raw

RE: [PATCH B 1/3] tidspbridge: don't flood the mailbox on MemUnMap

Quoted 8 lines

You tell me :)

I just moved code around. I don't truly understand what it's doing. Maybe the read is needed for some kind of sync?

I've removed the tlb flush all completely and I didn't notice any problems, is it really needed?

-- Ignore my previous comment. I think temp is used for other purpose in the same function. I thought temp was used by only the part of the code section that you stripped out.

Thank you,
Best regards,
Hari

Quoted 31 lines

-----Original Message-----
From: Felipe Contreras [mailto:[email protected]]
Sent: Wednesday, March 18, 2009 8:30 AM
To: Kanigeri, Hari
Cc: [email protected]; Hiroshi DOYU; Ameya Palande; Guzman Lugo, Fernando; Felipe Contreras
Subject: Re: [PATCH B 1/3] tidspbridge: don't flood the mailbox on MemUnMap

Quoted 16 lines · Hari

Felipe,

Quoted 3 lines

Check the hardware state of the DSP before sending the command to wake it up thus avoiding to flood the mailbox unnecessarily.

-- So this check was missing in Unmap part. Good catch.

Quoted 7 lines

-     u32 temp = 0;
-     struct CFG_HOSTRES resources;
      u32 *pPhysAddrPageTbl = NULL;
      struct vm_area_struct *vma;
      struct mm_struct *mm = current->mm;
+     u32 temp = 0;

Quoted 1 line

-- Is temp variable needed ?

You tell me :)

Quoted 8 lines

I just moved code around. I don't truly understand what it's doing. Maybe the read is needed for some kind of sync?

I've removed the tlb flush all completely and I didn't notice any problems, is it really needed?

--
Felipe Contreras

--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html