Re: [PATCH 3/4 v4] OMAP: McBSP: Introduce caching in register write operations
Janusz Krzysztofik napisał(a): Thursday 03 December 2009 21:03:02 Tony Lindgren napisał(a): Hi, * Janusz Krzysztofik jkrzy...@tis.icnet.pl [091203 04:29]: Reserve space inside omap_mcbsp structure for storing cached copies of McBSP register values. Modify the MCBSP_WRITE() macro to update the cache with every register write operation. Introduce a new macro that reads from the cache instead of hardware. Applies on top of patch 2 from this series: [PATCH v3 2/4] OMAP: McBSP: Prepare register read/write macros API for caching Tested on OMAP1510 based Amstrad Delta using linux-omap for-next, commit 4421752e331cfb1d942b47ffdb26e451a8da58a0. Compile-tested with omap_3430sdp_defconfig. Signed-off-by: Janusz Krzysztofik jkrzy...@tis.icnet.pl --- Thursday 03 December 2009 08:49:21 Jarkko Nikula wrote: On Tue, 1 Dec 2009 04:15:50 +0100 Janusz Krzysztofik jkrzy...@tis.icnet.pl wrote: #define MCBSP_WRITE(mcbsp, reg, val) \ - omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, val) + omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, \ + mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / OMAP_MCBSP_REG_DRR1] \ + = val) +#define MCBSP_READ_CACHE(mcbsp, reg) \ + (mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / OMAP_MCBSP_REG_DRR1]) These divisions by DDR1 are confusing. Use rather sizeof: reg_cache[OMAP_MCBSP_REG_##reg / sizeof(*mcbsp-reg_cache)] Done. Jarkko, many thanks for your cooperation :). Janusz arch/arm/plat-omap/include/plat/mcbsp.h |5 + arch/arm/plat-omap/mcbsp.c |7 ++- 2 files changed, 11 insertions(+), 1 deletion(-) --- git/arch/arm/plat-omap/include/plat/mcbsp.h.orig2009-11-27 11:53:45.0 +0100 +++ git/arch/arm/plat-omap/include/plat/mcbsp.h 2009-12-01 03:37:21.0 +0100 @@ -415,6 +415,11 @@ struct omap_mcbsp { u16 max_tx_thres; u16 max_rx_thres; #endif +#ifdef CONFIG_ARCH_OMAP1 + u16 reg_cache[OMAP_MCBSP_REG_XCERH / sizeof(u16) + 1]; +#else + u32 reg_cache[OMAP_MCBSP_REG_RCCR / sizeof(u32) + 1]; +#endif }; extern struct omap_mcbsp **mcbsp_ptr; extern int omap_mcbsp_count; How about rather set: static u16 omap1_reg_cache[]; and static u32 omap2_reg_cache[]; Hi Tony, Let me understand more precisely what you mean before I submit next version. Did you mean something like: static u16 omap1_reg_cache[3][OMAP_MCBSP_REG_XCERH / sizeof(u16) + 1]; static u32 omap2_reg_cache[5][OMAP_MCBSP_REG_RCCR / sizeof(u32) + 1]; both put in arch/arm/plat-omap/mcbsp.c without any ifdefs? Tony, I think I've already found one part of the answer myself: you probably mean putting each table definition in a corresponding arch/arm/mach-omap[12]/mcbsp.c source file, don't you? What remains unclear for me is if you really intend to reserve static cache space for all McBSP ports supported by a machine, even if not supported by a particular cpu or not used on a particular board. --- git/arch/arm/plat-omap/mcbsp.c.orig 2009-12-01 03:19:56.0 +0100 +++ git/arch/arm/plat-omap/mcbsp.c2009-12-03 11:37:21.0 +0100 @@ -49,7 +49,12 @@ int omap_mcbsp_read(void __iomem *io_bas #define MCBSP_READ(mcbsp, reg) \ omap_mcbsp_read(mcbsp-io_base, OMAP_MCBSP_REG_##reg) #define MCBSP_WRITE(mcbsp, reg, val) \ - omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, val) + omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, \ + mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / \ +sizeof(*mcbsp-reg_cache)] = val) +#define MCBSP_READ_CACHE(mcbsp, reg) \ + (mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / \ + sizeof(*mcbsp-reg_cache)]) #define omap_mcbsp_check_valid_id(id) (id omap_mcbsp_count) #define id_to_mcbsp_ptr(id)mcbsp_ptr[id]; Then rather than doing this cache manipulation in the macro, just do it in the omap_mcbsp_read/write functions based on the cpu_class_is_omap1() accessing omap1_reg_cache or omap2_reg_cache. omap_mcbsp_wrtie() case seems clear to me, but: Did you mean defining a new omap_mcbsp_read_cache() function, similiar to omap_mcbsp_read(), but reading from cache instead of hardware? Or do you prefere if I modify omap_mcbsp_read() to be able to read from both hardware and cache, dependig on an additional argument, for example? Thanks, Janusz This way you can get rid of the ifdef else. But shouldn't declarations for both tables be placed in arch/arm/plat-omap/include/plat/mcbsp.h withing an ifdef else block? If yes, one ifdef else block would be replaced with another. Or did you mean placing those declarations inside existing ifdef block with defines for register offsets? Thanks, Janusz Regards, Tony -- To unsubscribe from this list: send the line unsubscribe linux-omap in the body of a message to majord...@vger.kernel.org More majordomo info at
Re: [PATCH 3/4 v4] OMAP: McBSP: Introduce caching in register write operations
* Janusz Krzysztofik jkrzy...@tis.icnet.pl [091204 04:56]: Janusz Krzysztofik napisał(a): snip I think I've already found one part of the answer myself: you probably mean putting each table definition in a corresponding arch/arm/mach-omap[12]/mcbsp.c source file, don't you? Well I did not initially, but sounds like that's the cleanest way to go. What remains unclear for me is if you really intend to reserve static cache space for all McBSP ports supported by a machine, even if not supported by a particular cpu or not used on a particular board. No idea :) But please consider what happens if we want to compile in omap2 + omap3 into a single binary. Or what happens if some newer omaps add more McBSP registers. snip Did you mean defining a new omap_mcbsp_read_cache() function, similiar to omap_mcbsp_read(), but reading from cache instead of hardware? Yeah something like that. That is more future proof. Or do you prefere if I modify omap_mcbsp_read() to be able to read from both hardware and cache, dependig on an additional argument, for example? That actually sounds like the best way to solve it to me! But shouldn't declarations for both tables be placed in arch/arm/plat-omap/include/plat/mcbsp.h withing an ifdef else block? If yes, one ifdef else block would be replaced with another. Or did you mean placing those declarations inside existing ifdef block with defines for register offsets? Ifdefs are OK if they just optimize out code when some omap is not compiled in. But if you do ifdef else, then it means trouble sooner or later. In this case you should probably have the cache initialized separately in mach-omap1/mcbsp.c and mach-omap2/mcbsp.c. Regards, Tony -- To unsubscribe from this list: send the line unsubscribe linux-omap in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 3/4 v4] OMAP: McBSP: Introduce caching in register write operations
Hi, * Janusz Krzysztofik jkrzy...@tis.icnet.pl [091203 04:29]: Reserve space inside omap_mcbsp structure for storing cached copies of McBSP register values. Modify the MCBSP_WRITE() macro to update the cache with every register write operation. Introduce a new macro that reads from the cache instead of hardware. Applies on top of patch 2 from this series: [PATCH v3 2/4] OMAP: McBSP: Prepare register read/write macros API for caching Tested on OMAP1510 based Amstrad Delta using linux-omap for-next, commit 4421752e331cfb1d942b47ffdb26e451a8da58a0. Compile-tested with omap_3430sdp_defconfig. Signed-off-by: Janusz Krzysztofik jkrzy...@tis.icnet.pl --- Thursday 03 December 2009 08:49:21 Jarkko Nikula wrote: On Tue, 1 Dec 2009 04:15:50 +0100 Janusz Krzysztofik jkrzy...@tis.icnet.pl wrote: #define MCBSP_WRITE(mcbsp, reg, val) \ - omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, val) + omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, \ + mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / OMAP_MCBSP_REG_DRR1] \ + = val) +#define MCBSP_READ_CACHE(mcbsp, reg) \ + (mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / OMAP_MCBSP_REG_DRR1]) These divisions by DDR1 are confusing. Use rather sizeof: reg_cache[OMAP_MCBSP_REG_##reg / sizeof(*mcbsp-reg_cache)] Done. Jarkko, many thanks for your cooperation :). Janusz arch/arm/plat-omap/include/plat/mcbsp.h |5 + arch/arm/plat-omap/mcbsp.c |7 ++- 2 files changed, 11 insertions(+), 1 deletion(-) --- git/arch/arm/plat-omap/include/plat/mcbsp.h.orig 2009-11-27 11:53:45.0 +0100 +++ git/arch/arm/plat-omap/include/plat/mcbsp.h 2009-12-01 03:37:21.0 +0100 @@ -415,6 +415,11 @@ struct omap_mcbsp { u16 max_tx_thres; u16 max_rx_thres; #endif +#ifdef CONFIG_ARCH_OMAP1 + u16 reg_cache[OMAP_MCBSP_REG_XCERH / sizeof(u16) + 1]; +#else + u32 reg_cache[OMAP_MCBSP_REG_RCCR / sizeof(u32) + 1]; +#endif }; extern struct omap_mcbsp **mcbsp_ptr; extern int omap_mcbsp_count; How about rather set: static u16 omap1_reg_cache[]; and static u32 omap2_reg_cache[]; --- git/arch/arm/plat-omap/mcbsp.c.orig 2009-12-01 03:19:56.0 +0100 +++ git/arch/arm/plat-omap/mcbsp.c2009-12-03 11:37:21.0 +0100 @@ -49,7 +49,12 @@ int omap_mcbsp_read(void __iomem *io_bas #define MCBSP_READ(mcbsp, reg) \ omap_mcbsp_read(mcbsp-io_base, OMAP_MCBSP_REG_##reg) #define MCBSP_WRITE(mcbsp, reg, val) \ - omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, val) + omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, \ + mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / \ + sizeof(*mcbsp-reg_cache)] = val) +#define MCBSP_READ_CACHE(mcbsp, reg) \ + (mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / \ + sizeof(*mcbsp-reg_cache)]) #define omap_mcbsp_check_valid_id(id)(id omap_mcbsp_count) #define id_to_mcbsp_ptr(id) mcbsp_ptr[id]; Then rather than doing this cache manipulation in the macro, just do it in the omap_mcbsp_read/write functions based on the cpu_class_is_omap1() accessing omap1_reg_cache or omap2_reg_cache. This way you can get rid of the ifdef else. Regards, Tony -- To unsubscribe from this list: send the line unsubscribe linux-omap in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH 3/4 v4] OMAP: McBSP: Introduce caching in register write operations
Thursday 03 December 2009 21:03:02 Tony Lindgren napisał(a): Hi, * Janusz Krzysztofik jkrzy...@tis.icnet.pl [091203 04:29]: Reserve space inside omap_mcbsp structure for storing cached copies of McBSP register values. Modify the MCBSP_WRITE() macro to update the cache with every register write operation. Introduce a new macro that reads from the cache instead of hardware. Applies on top of patch 2 from this series: [PATCH v3 2/4] OMAP: McBSP: Prepare register read/write macros API for caching Tested on OMAP1510 based Amstrad Delta using linux-omap for-next, commit 4421752e331cfb1d942b47ffdb26e451a8da58a0. Compile-tested with omap_3430sdp_defconfig. Signed-off-by: Janusz Krzysztofik jkrzy...@tis.icnet.pl --- Thursday 03 December 2009 08:49:21 Jarkko Nikula wrote: On Tue, 1 Dec 2009 04:15:50 +0100 Janusz Krzysztofik jkrzy...@tis.icnet.pl wrote: #define MCBSP_WRITE(mcbsp, reg, val) \ - omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, val) + omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, \ + mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / OMAP_MCBSP_REG_DRR1] \ + = val) +#define MCBSP_READ_CACHE(mcbsp, reg) \ + (mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / OMAP_MCBSP_REG_DRR1]) These divisions by DDR1 are confusing. Use rather sizeof: reg_cache[OMAP_MCBSP_REG_##reg / sizeof(*mcbsp-reg_cache)] Done. Jarkko, many thanks for your cooperation :). Janusz arch/arm/plat-omap/include/plat/mcbsp.h |5 + arch/arm/plat-omap/mcbsp.c |7 ++- 2 files changed, 11 insertions(+), 1 deletion(-) --- git/arch/arm/plat-omap/include/plat/mcbsp.h.orig2009-11-27 11:53:45.0 +0100 +++ git/arch/arm/plat-omap/include/plat/mcbsp.h 2009-12-01 03:37:21.0 +0100 @@ -415,6 +415,11 @@ struct omap_mcbsp { u16 max_tx_thres; u16 max_rx_thres; #endif +#ifdef CONFIG_ARCH_OMAP1 + u16 reg_cache[OMAP_MCBSP_REG_XCERH / sizeof(u16) + 1]; +#else + u32 reg_cache[OMAP_MCBSP_REG_RCCR / sizeof(u32) + 1]; +#endif }; extern struct omap_mcbsp **mcbsp_ptr; extern int omap_mcbsp_count; How about rather set: static u16 omap1_reg_cache[]; and static u32 omap2_reg_cache[]; Hi Tony, Let me understand more precisely what you mean before I submit next version. Did you mean something like: static u16 omap1_reg_cache[3][OMAP_MCBSP_REG_XCERH / sizeof(u16) + 1]; static u32 omap2_reg_cache[5][OMAP_MCBSP_REG_RCCR / sizeof(u32) + 1]; both put in arch/arm/plat-omap/mcbsp.c without any ifdefs? --- git/arch/arm/plat-omap/mcbsp.c.orig 2009-12-01 03:19:56.0 +0100 +++ git/arch/arm/plat-omap/mcbsp.c2009-12-03 11:37:21.0 +0100 @@ -49,7 +49,12 @@ int omap_mcbsp_read(void __iomem *io_bas #define MCBSP_READ(mcbsp, reg) \ omap_mcbsp_read(mcbsp-io_base, OMAP_MCBSP_REG_##reg) #define MCBSP_WRITE(mcbsp, reg, val) \ - omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, val) + omap_mcbsp_write(mcbsp-io_base, OMAP_MCBSP_REG_##reg, \ + mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / \ +sizeof(*mcbsp-reg_cache)] = val) +#define MCBSP_READ_CACHE(mcbsp, reg) \ + (mcbsp-reg_cache[OMAP_MCBSP_REG_##reg / \ + sizeof(*mcbsp-reg_cache)]) #define omap_mcbsp_check_valid_id(id) (id omap_mcbsp_count) #define id_to_mcbsp_ptr(id)mcbsp_ptr[id]; Then rather than doing this cache manipulation in the macro, just do it in the omap_mcbsp_read/write functions based on the cpu_class_is_omap1() accessing omap1_reg_cache or omap2_reg_cache. omap_mcbsp_wrtie() case seems clear to me, but: Did you mean defining a new omap_mcbsp_read_cache() function, similiar to omap_mcbsp_read(), but reading from cache instead of hardware? Or do you prefere if I modify omap_mcbsp_read() to be able to read from both hardware and cache, dependig on an additional argument, for example? Thanks, Janusz This way you can get rid of the ifdef else. Regards, Tony -- To unsubscribe from this list: send the line unsubscribe linux-omap in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line unsubscribe linux-omap in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html