OMAP2/3/4: DMA: reset controller during init
Subsystems (2)
PATCH 1/1 OMAP2/3/4: DMA: reset controller during init
If we are softbooting another kernel using kexec, DMA controller state is not known when we are performing omap_init_dma(). It is possible that some DMA channels are already active. For example after kexec we get:
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
To prevent any weird things happening, we perform soft reset for the controller and disable all per channel interrupts.
--- arch/arm/plat-omap/dma.c | 32 ++++++++++++++++++++++++++++++++ arch/arm/plat-omap/include/plat/dma.h | 4 ++++ 2 files changed, 36 insertions(+), 0 deletions(-)
arch/arm/plat-omap/dma.c +32 -0
@@ -2024,6 +2024,31 @@ void omap_dma_global_context_restore(void)/*----------------------------------------------------------------------------*/ +/** +* omap_dma_reset() - perform software reset for the DMA controller +*/ +static void omap_dma_reset (void ) +{ +u32 v; + +if (cpu_class_is_omap1 ()) +return ; + + v =dma_read (OCP_SYSCONFIG ); + v |=0x2 ;/* software reset */ +dma_write (v,OCP_SYSCONFIG ); + +/* wait until reset is complete */ +while ((dma_read (SYSSTATUS ) &0x1 ) ==0 ) +cpu_relax (); + +/* disable per channel interrupts */ +dma_write (0 ,IRQENABLE_L0 ); +dma_write (0 ,IRQENABLE_L1 ); +dma_write (0 ,IRQENABLE_L2 ); +dma_write (0 ,IRQENABLE_L3 ); +} +static int __init omap_init_dma (void ) {unsigned long base;
@@ -2069,6 +2094,13 @@ static int __init omap_init_dma(void) } } +/* +* Make sure that DMA is in reset state before doing anything else. If +* we are coming from softboot (e.g kexec) DMA is probably configured so +* we explicitly reset it here to get into known state. +*/ +omap_dma_reset (); +if (cpu_is_omap15xx ()) {printk (KERN_INFO "DMA support for OMAP15xx initialized\n" ); dma_chan_count =9 ;
arch/arm/plat-omap/include/plat/dma.h +4 -0
@@ -134,6 +134,10 @@#define OMAP1_DMA_REVISION 0#define OMAP1_DMA_IRQSTATUS_L0 0#define OMAP1_DMA_IRQENABLE_L0 0 +#define OMAP1_DMA_IRQENABLE_L1 0 +#define OMAP1_DMA_IRQENABLE_L2 0 +#define OMAP1_DMA_IRQENABLE_L3 0 +#define OMAP1_DMA_SYSSTATUS 0#define OMAP1_DMA_OCP_SYSCONFIG 0#define OMAP_DMA4_HW_ID 0#define OMAP_DMA4_CAPS_0_L 0
Signature · 1 line
1.5.6.5
Quoted 10 lines · Mika Westerberg
If we are softbooting another kernel using kexec, DMA controller state is not known when we are performing
omap_init_dma(). It is possible that some DMA channels are already active. For example after kexec we get:<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
Quoted 4 lines
To prevent any weird things happening, we perform soft reset for the controller and disable all per channel interrupts.
Signed-off-by Mika Westerberg <[email protected]>
This is a good fix, but we get reset of DMA (and all other blocks) for free when switching to hwmod. Here's a good reason to convert DMA to hwmod.
Kevin
Quoted 12 lines · Kevin Hilman
Quoted 10 lines · Mika Westerberg
If we are softbooting another kernel using kexec, DMA controller state is not known when we are performing
omap_init_dma(). It is possible that some DMA channels are already active. For example after kexec we get:<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
Quoted 8 lines
Quoted 4 lines
To prevent any weird things happening, we perform soft reset for the controller and disable all per channel interrupts.
Signed-off-by Mika Westerberg <[email protected]>This is a good fix, but we get reset of DMA (and all other blocks) for free when switching to hwmod. Here's a good reason to convert DMA to hwmod.
Hmm, do we have existing DMA hwmod patches somewhere that work on all omaps?
Tony
Quoted 16 lines · Tony Lindgren
Quoted 15 lines · Kevin Hilman
Quoted 13 lines · Mika Westerberg
If we are softbooting another kernel using kexec, DMA controller state is not known when we are performing
omap_init_dma(). It is possible that some DMA channels are already active. For example after kexec we get:<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5To prevent any weird things happening, we perform soft reset for the controller and disable all per channel interrupts.
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
Signed-off-by Mika Westerberg <[email protected]>This is a good fix, but we get reset of DMA (and all other blocks) for free when switching to hwmod. Here's a good reason to convert DMA to hwmod.
Hmm, do we have existing DMA hwmod patches somewhere that work on all omaps?
No, that's my point.
Rather than implement reset here (then remove it with DMA hwmods) I'd rather just see a DMA hwmods added, and then we get reset for free.
Kevin
Quoted 24 lines · Kevin Hilman
Quoted 22 lines · Tony Lindgren
Quoted 20 lines · Kevin Hilman
Quoted 10 lines · Mika Westerberg
If we are softbooting another kernel using kexec, DMA controller state is not known when we are performing
omap_init_dma(). It is possible that some DMA channels are already active. For example after kexec we get:<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5Quoted 4 lines
To prevent any weird things happening, we perform soft reset for the controller and disable all per channel interrupts.
Signed-off-by Mika Westerberg <[email protected]>This is a good fix, but we get reset of DMA (and all other blocks) for free when switching to hwmod. Here's a good reason to convert DMA to hwmod.
Quoted 6 lines
Quoted 1 line
Hmm, do we have existing DMA hwmod patches somewhere that work on all omaps?
No, that's my point.
Rather than implement reset here (then remove it with DMA hwmods) I'd rather just see a DMA hwmods added, and then we get reset for free.
Sure. But until we have something available we should get the fix part in.
AFAIK just disabling the channel interrupt in omap_clear_dma() should do the trick, see my other reply.
Regards,
Tony
Quoted 25 lines · Tony Lindgren
Quoted 24 lines · Kevin Hilman
Quoted 22 lines · Tony Lindgren
Quoted 20 lines · Kevin Hilman
Quoted 10 lines · Mika Westerberg
If we are softbooting another kernel using kexec, DMA controller state is not known when we are performing
omap_init_dma(). It is possible that some DMA channels are already active. For example after kexec we get:<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5Quoted 4 lines
To prevent any weird things happening, we perform soft reset for the controller and disable all per channel interrupts.
Signed-off-by Mika Westerberg <[email protected]>This is a good fix, but we get reset of DMA (and all other blocks) for free when switching to hwmod. Here's a good reason to convert DMA to hwmod.
Quoted 8 lines
Quoted 6 lines
Quoted 1 line
Hmm, do we have existing DMA hwmod patches somewhere that work on all omaps?
No, that's my point.
Rather than implement reset here (then remove it with DMA hwmods) I'd rather just see a DMA hwmods added, and then we get reset for free.
Sure. But until we have something available we should get the fix part in.
which will reduce the likelyhood of anyone doing the DMA hwmods until some sucker like me does it since nobody else will. :(
I'm just looking for more ways to motivate others to help in hwmod conversion.
Kevin
Quoted 31 lines · Kevin Hilman
Quoted 29 lines · Tony Lindgren
Quoted 28 lines · Kevin Hilman
Quoted 16 lines · Tony Lindgren
Quoted 15 lines · Kevin Hilman
Quoted 13 lines · Mika Westerberg
If we are softbooting another kernel using kexec, DMA controller state is not known when we are performing
omap_init_dma(). It is possible that some DMA channels are already active. For example after kexec we get:<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5To prevent any weird things happening, we perform soft reset for the controller and disable all per channel interrupts.
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
Signed-off-by Mika Westerberg <[email protected]>This is a good fix, but we get reset of DMA (and all other blocks) for free when switching to hwmod. Here's a good reason to convert DMA to hwmod.
Hmm, do we have existing DMA hwmod patches somewhere that work on all omaps?
No, that's my point.
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
Rather than implement reset here (then remove it with DMA hwmods) I'd rather just see a DMA hwmods added, and then we get reset for free.
Sure. But until we have something available we should get the fix part in.
which will reduce the likelyhood of anyone doing the DMA hwmods until some sucker like me does it since nobody else will. :(
Yeah I feel your pain.
Anyways, doing that hwmod change is too late for this coming merge window if we don't have the patches ready. So we still need to fix the problem.
Quoted 2 lines
I'm just looking for more ways to motivate others to help in hwmod conversion.
Yes, and in addition to that at least plat-omap/dma.c, plat-omap/gpio.c could use some serious clean-up.. And i2c-omap.c too.
Tony
Quoted 34 lines · Mika Westerberg
If we are softbooting another kernel using kexec, DMA controller state is not known when we are performing
omap_init_dma(). It is possible that some DMA channels are already active. For example after kexec we get:<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5To prevent any weird things happening, we perform soft reset for the controller and disable all per channel interrupts.
Signed-off-by Mika Westerberg <[email protected]>--- arch/arm/plat-omap/dma.c | 32 ++++++++++++++++++++++++++++++++ arch/arm/plat-omap/include/plat/dma.h | 4 ++++ 2 files changed, 36 insertions(+), 0 deletions(-)arch/arm/plat-omap/dma.c +7
-0@@ -2024,6 +2024,31 @@ void omap_dma_global_context_restore(void)/*----------------------------------------------------------------------------*/ +/** +* omap_dma_reset() - perform software reset for the DMA controller +*/ +static void omap_dma_reset (void ) +{ +u32 v; +
arch/arm/plat-omap/dma.c void omap_dma_global_context_restore(void) · line 2043
Quoted 10 lines
arch/arm/plat-omap/dma.c +10
-0+if (cpu_class_is_omap1 ()) +return ; + + v =dma_read (OCP_SYSCONFIG ); + v |=0x2 ;/* software reset */ +dma_write (v,OCP_SYSCONFIG ); + +/* wait until reset is complete */ +while ((dma_read (SYSSTATUS ) &0x1 ) ==0 ) +cpu_relax ();
This reset part seems to be mach-omap2 specific.
arch/arm/plat-omap/dma.c void omap_dma_global_context_restore(void) · line 2050
Quoted 6 lines
arch/arm/plat-omap/dma.c +6
-0+/* disable per channel interrupts */ +dma_write (0 ,IRQENABLE_L0 ); +dma_write (0 ,IRQENABLE_L1 ); +dma_write (0 ,IRQENABLE_L2 ); +dma_write (0 ,IRQENABLE_L3 ); +}
For a minimal fix, how about just disable the interrupt in omap_clear_dma()? We are already calling that from omap_init_dma().
Regards,
Tony
Quoted 1 line · ext Tony Lindgren
* Mika Westerberg <[email protected]> [100503 05:52]:
(...)
Quoted 20 lines
Quoted 7 lines
(quoted diff) +7
-0+/** + * omap_dma_reset() - perform software reset for the DMA controller + */ +static void omap_dma_reset(void) +{ + u32 v; +Review
Quoted 10 lines
(quoted diff) +10
-0+ if (cpu_class_is_omap1()) + return; + + v = dma_read(OCP_SYSCONFIG); + v |= 0x2; /* software reset */ + dma_write(v, OCP_SYSCONFIG); + + /* wait until reset is complete */ + while ((dma_read(SYSSTATUS) & 0x1) == 0) + cpu_relax();This reset part seems to be mach-omap2 specific.
Quoted 9 lines
Review
Quoted 6 lines
(quoted diff) +6
-0+ /* disable per channel interrupts */ + dma_write(0, IRQENABLE_L0); + dma_write(0, IRQENABLE_L1); + dma_write(0, IRQENABLE_L2); + dma_write(0, IRQENABLE_L3); +}For a minimal fix, how about just disable the interrupt in
omap_clear_dma()? We are already calling that fromomap_init_dma().
If we do it in omap_clear_dma() then omap_dma_global_context_restore() will return with interrupts masked which probably breaks things.
How about something like following patch?
(checking cpu_class_is_omap2() below in omap_init_dma() is probably redundant
but seems to be consitent with other code).
Thanks,
MW
Quoted 1 line
From
74389ff1de06e5adb84b978bf1d169c6d802db1bMon Sep 17 00:00:00 2001
From: Mika Westerberg <[email protected]>
Date: Tue, 4 May 2010 09:32:29 +0300
Subject: [PATCH] OMAP2/3/4: DMA: disable channel interrupts in omap_init_dma()
If we are softbooting another kernel using kexec, DMA controller state is not known when we are performing omap_init_dma(). It is possible that some DMA channels are already active. For example after kexec we get:
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
To prevent any weird things happening, we disable all channel interrupts during init.
--- arch/arm/plat-omap/dma.c | 27 +++++++++++++++++++-------- 1 files changed, 19 insertions(+), 8 deletions(-)
arch/arm/plat-omap/dma.c +19 -8
@@ -709,6 +709,21 @@ static inline void omap2_enable_irq_lch(int lch)spin_unlock_irqrestore (&dma_chan_lock, flags); } +static inline void omap2_disable_irq_lch (int lch) +{ +u32 val; +unsigned long flags; + +if (!cpu_class_is_omap2 ()) +return ; + +spin_lock_irqsave (&dma_chan_lock, flags); + val =dma_read (IRQENABLE_L0 ); + val &= ~(1 << lch); +dma_write (val,IRQENABLE_L0 ); +spin_unlock_irqrestore (&dma_chan_lock, flags); +} +int omap_request_dma (int dev_id,const char *dev_name,void (*callback)(int lch,u16 ch_status,void *data),void *data,int *dma_ch_out)
@@ -807,14 +822,7 @@ void omap_free_dma(int lch) }if (cpu_class_is_omap2 ()) { -u32 val; - -spin_lock_irqsave (&dma_chan_lock, flags); -/* Disable interrupts */ - val =dma_read (IRQENABLE_L0 ); - val &= ~(1 << lch); -dma_write (val,IRQENABLE_L0 ); -spin_unlock_irqrestore (&dma_chan_lock, flags); +omap2_disable_irq_lch (lch);/* Clear the CSR register and IRQ status register */ dma_write (OMAP2_DMA_CSR_CLEAR_MASK ,CSR (lch));
@@ -2107,6 +2115,9 @@ static int __init omap_init_dma(void)for (ch =0 ; ch < dma_chan_count; ch++) {omap_clear_dma (ch); +if (cpu_class_is_omap2 ()) +omap2_disable_irq_lch (ch); + dma_chan[ch].dev_id =-1 ; dma_chan[ch].next_lch =-1 ;
Signature · 1 line
1.5.6.5
Quoted 33 lines · Mika Westerberg
Quoted 1 line · ext Tony Lindgren
* Mika Westerberg <[email protected]> [100503 05:52]:
(...)
Quoted 20 lines
Quoted 7 lines
(quoted diff) +7
-0+/** + * omap_dma_reset() - perform software reset for the DMA controller + */ +static void omap_dma_reset(void) +{ + u32 v; +Review
Quoted 10 lines
(quoted diff) +10
-0+ if (cpu_class_is_omap1()) + return; + + v = dma_read(OCP_SYSCONFIG); + v |= 0x2; /* software reset */ + dma_write(v, OCP_SYSCONFIG); + + /* wait until reset is complete */ + while ((dma_read(SYSSTATUS) & 0x1) == 0) + cpu_relax();This reset part seems to be mach-omap2 specific.
Quoted 9 lines
Review
Quoted 6 lines
(quoted diff) +6
-0+ /* disable per channel interrupts */ + dma_write(0, IRQENABLE_L0); + dma_write(0, IRQENABLE_L1); + dma_write(0, IRQENABLE_L2); + dma_write(0, IRQENABLE_L3); +}For a minimal fix, how about just disable the interrupt in
omap_clear_dma()? We are already calling that fromomap_init_dma().
Quoted 4 lines
If we do it in
omap_clear_dma()thenomap_dma_global_context_restore()will return with interrupts masked which probably breaks things.How about something like following patch?
Tony, Kevin
Any comments on this patch?
Thanks,
MW
Quoted 84 lines
(checking
cpu_class_is_omap2()below inomap_init_dma()is probably redundant
but seems to be consitent with other code).Thanks,
MWQuoted 1 line
From
74389ff1de06e5adb84b978bf1d169c6d802db1bMon Sep 17 00:00:00 2001From: Mika Westerberg <[email protected]>
Date: Tue, 4 May 2010 09:32:29 +0300
Subject: [PATCH] OMAP2/3/4: DMA: disable channel interrupts inomap_init_dma()If we are softbooting another kernel using kexec, DMA controller state is not known when we are performing
omap_init_dma(). It is possible that some DMA channels are already active. For example after kexec we get:<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5
<4>IRQ 0020 for non-allocated DMAchannel 5To prevent any weird things happening, we disable all channel interrupts during init.
Signed-off-by Mika Westerberg <[email protected]>--- arch/arm/plat-omap/dma.c | 27 +++++++++++++++++++-------- 1 files changed, 19 insertions(+), 8 deletions(-)arch/arm/plat-omap/dma.c +19
-8@@ -709,6 +709,21 @@ static inline void omap2_enable_irq_lch(int lch)spin_unlock_irqrestore (&dma_chan_lock, flags); } +static inline void omap2_disable_irq_lch (int lch) +{ +u32 val; +unsigned long flags; + +if (!cpu_class_is_omap2 ()) +return ; + +spin_lock_irqsave (&dma_chan_lock, flags); + val =dma_read (IRQENABLE_L0 ); + val &= ~(1 << lch); +dma_write (val,IRQENABLE_L0 ); +spin_unlock_irqrestore (&dma_chan_lock, flags); +} +int omap_request_dma (int dev_id,const char *dev_name,void (*callback)(int lch,u16 ch_status,void *data),void *data,int *dma_ch_out)@@ -807,14 +822,7 @@ void omap_free_dma(int lch) }if (cpu_class_is_omap2 ()) { -u32 val; - -spin_lock_irqsave (&dma_chan_lock, flags); -/* Disable interrupts */ - val =dma_read (IRQENABLE_L0 ); - val &= ~(1 << lch); -dma_write (val,IRQENABLE_L0 ); -spin_unlock_irqrestore (&dma_chan_lock, flags); +omap2_disable_irq_lch (lch);/* Clear the CSR register and IRQ status register */ dma_write (OMAP2_DMA_CSR_CLEAR_MASK ,CSR (lch));@@ -2107,6 +2115,9 @@ static int __init omap_init_dma(void)for (ch =0 ; ch < dma_chan_count; ch++) {omap_clear_dma (ch); +if (cpu_class_is_omap2 ()) +omap2_disable_irq_lch (ch); + dma_chan[ch].dev_id =-1 ; dma_chan[ch].next_lch =-1 ;Signature · 1 line
1.5.6.5
Kevin Hilman
Acked-byQuoted 34 lines · Mika Westerberg
Quoted 33 lines · Mika Westerberg
Quoted 1 line · ext Tony Lindgren
* Mika Westerberg <[email protected]> [100503 05:52]:
(...)
Quoted 29 lines
Quoted 7 lines
(quoted diff) +7
-0+/** + * omap_dma_reset() - perform software reset for the DMA controller + */ +static void omap_dma_reset(void) +{ + u32 v; +Review
Quoted 10 lines
(quoted diff) +10
-0+ if (cpu_class_is_omap1()) + return; + + v = dma_read(OCP_SYSCONFIG); + v |= 0x2; /* software reset */ + dma_write(v, OCP_SYSCONFIG); + + /* wait until reset is complete */ + while ((dma_read(SYSSTATUS) & 0x1) == 0) + cpu_relax();This reset part seems to be mach-omap2 specific.
Review
Quoted 6 lines
(quoted diff) +6
-0+ /* disable per channel interrupts */ + dma_write(0, IRQENABLE_L0); + dma_write(0, IRQENABLE_L1); + dma_write(0, IRQENABLE_L2); + dma_write(0, IRQENABLE_L3); +}For a minimal fix, how about just disable the interrupt in
omap_clear_dma()? We are already calling that fromomap_init_dma().
Quoted 8 lines
Quoted 4 lines
If we do it in
omap_clear_dma()thenomap_dma_global_context_restore()will return with interrupts masked which probably breaks things.How about something like following patch?
Tony, Kevin
Any comments on this patch?
Looks good to me.
PATCH 1/1 OMAP2/3/4: DMA: reset controller during init
In reply to Mika WesterbergThis patch has been applied to the linux-omap
by youw fwiendly patch wobot.
Branch in linux-omap: for-next
Initial commit ID (Likely to change): 738e377ff6c45d27695b39a67e3be876d4b5dae5
PatchWorks
http://patchwork.kernel.org/patch/96684/
Git (Likely to change, and takes a while to get mirrored)
http://git.kernel.org/?p=linux/kernel/git/tmlind/linux-omap-2.6.git;a=commit;h=738e377ff6c45d27695b39a67e3be876d4b5dae5