diff mbox series

[v4,1/2] EDAC/qcom: Get rid of hardcoded register offsets

Message ID 20221116143352.289303-2-manivannan.sadhasivam@linaro.org
State Superseded
Headers show
Series Fix crash when using Qcom LLCC/EDAC drivers | expand

Commit Message

Manivannan Sadhasivam Nov. 16, 2022, 2:33 p.m. UTC 
The LLCC EDAC register offsets varies between each SoC. Hardcoding the
register offsets won't work and will often result in crash due to
accessing the wrong locations.

Hence, get the register offsets from the LLCC driver matching the
individual SoCs.

Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
---
 drivers/edac/qcom_edac.c           | 116 ++++++++++++++---------------
 include/linux/soc/qcom/llcc-qcom.h |   6 --
 2 files changed, 58 insertions(+), 64 deletions(-)

Comments

Borislav Petkov Nov. 28, 2022, 12:16 p.m. UTC | #1 
On Wed, Nov 16, 2022 at 08:03:51PM +0530, Manivannan Sadhasivam wrote:
> The LLCC EDAC register offsets varies between each SoC. Hardcoding the
> register offsets won't work and will often result in crash due to
> accessing the wrong locations.
> 
> Hence, get the register offsets from the LLCC driver matching the
> individual SoCs.
> 
> Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
> ---
>  drivers/edac/qcom_edac.c           | 116 ++++++++++++++---------------
>  include/linux/soc/qcom/llcc-qcom.h |   6 --
>  2 files changed, 58 insertions(+), 64 deletions(-)

It looks to me like this patch needs to go to stable?

Also, the Fixes tag should probably be:

Fixes: 27450653f1db ("drivers: edac: Add EDAC driver support for QCOM SoCs")

Yes, no?

Also, please explain to me how you've tested this patch if it doesn't
even build?!

ERROR: modpost: "__devm_regmap_init_mmio_clk" [drivers/soc/qcom/llcc-qcom.ko] undefined!
make[1]: *** [scripts/Makefile.modpost:126: Module.symvers] Error 1
make: *** [Makefile:1944: modpost] Error 2

I guess because CONFIG_QCOM_LLCC is =y in your config while I have
CONFIG_QCOM_LLCC=m.

And I reported the same build error to you the last time. Did you not
see it?

For the next version, you'd need to fix all possible Kconfig build
errors before sending.
Manivannan Sadhasivam Nov. 28, 2022, 6:17 p.m. UTC | #2 
On Mon, Nov 28, 2022 at 01:16:53PM +0100, Borislav Petkov wrote:
> On Wed, Nov 16, 2022 at 08:03:51PM +0530, Manivannan Sadhasivam wrote:
> > The LLCC EDAC register offsets varies between each SoC. Hardcoding the
> > register offsets won't work and will often result in crash due to
> > accessing the wrong locations.
> > 
> > Hence, get the register offsets from the LLCC driver matching the
> > individual SoCs.
> > 
> > Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
> > ---
> >  drivers/edac/qcom_edac.c           | 116 ++++++++++++++---------------
> >  include/linux/soc/qcom/llcc-qcom.h |   6 --
> >  2 files changed, 58 insertions(+), 64 deletions(-)
> 
> It looks to me like this patch needs to go to stable?
> 

Well, yes but that would imply both LLCC and EDAC patches going together.
Splitting them will break the build, which is worse.

So I delibrately avoided CCing stable list. I'll ping them once both patches
are in mainline.

> Also, the Fixes tag should probably be:
> 
> Fixes: 27450653f1db ("drivers: edac: Add EDAC driver support for QCOM SoCs")
> 
> Yes, no?
> 

No. The actual breakage happened with, a6e9d7ef252c ("soc: qcom: llcc: Add
configuration data for SM8450 SoC"). I didn't add the fixes tag because, the
stable team might backport this patch automatically. And since the tag is not
present in the LLCC patch, it will break the build.

> Also, please explain to me how you've tested this patch if it doesn't
> even build?!
> 
> ERROR: modpost: "__devm_regmap_init_mmio_clk" [drivers/soc/qcom/llcc-qcom.ko] undefined!
> make[1]: *** [scripts/Makefile.modpost:126: Module.symvers] Error 1
> make: *** [Makefile:1944: modpost] Error 2
> 
> I guess because CONFIG_QCOM_LLCC is =y in your config while I have
> CONFIG_QCOM_LLCC=m.
> 
> And I reported the same build error to you the last time. Did you not
> see it?
> 
> For the next version, you'd need to fix all possible Kconfig build
> errors before sending.
> 

Sorry, it is because I only tried building for ARM64 architecture. The error
you are seeing is for x86-64 and I could now reproduce it as well.

Will submit the next iteration including a fix for that.

Thanks,
Mani

> -- 
> Regards/Gruss,
>     Boris.
> 
> https://people.kernel.org/tglx/notes-about-netiquette
Borislav Petkov Nov. 28, 2022, 6:35 p.m. UTC | #3 
On Mon, Nov 28, 2022 at 11:47:05PM +0530, Manivannan Sadhasivam wrote:
> Well, yes but that would imply both LLCC and EDAC patches going together.
> Splitting them will break the build, which is worse.

Sounds like you need to check out:

Documentation/process/stable-kernel-rules.rst

first.

Hint: there are provisions in there how to specify dependencies between
commits.

> Sorry, it is because I only tried building for ARM64 architecture. The
> error you are seeing is for x86-64 and I could now reproduce it as
> well.

Yes, because arch doesn't matter here - the .config does.

Therefore, as requested:

"For the next version, you'd need to fix all possible Kconfig build
errors before sending."

Thx.
Manivannan Sadhasivam Nov. 29, 2022, 5:53 a.m. UTC | #4 
On Mon, Nov 28, 2022 at 07:35:14PM +0100, Borislav Petkov wrote:
> On Mon, Nov 28, 2022 at 11:47:05PM +0530, Manivannan Sadhasivam wrote:
> > Well, yes but that would imply both LLCC and EDAC patches going together.
> > Splitting them will break the build, which is worse.
> 
> Sounds like you need to check out:
> 
> Documentation/process/stable-kernel-rules.rst
> 
> first.
> 
> Hint: there are provisions in there how to specify dependencies between
> commits.
> 

Hmm, I did miss reading this. Thanks for pointing out.

> > Sorry, it is because I only tried building for ARM64 architecture. The
> > error you are seeing is for x86-64 and I could now reproduce it as
> > well.
> 
> Yes, because arch doesn't matter here - the .config does.
> 

Well, arch does matter here. LLCC driver depends on ARCH_QCOM ||
COMPILE_TEST. In the case of ARCH_QCOM, the REGMAP_MMIO is implicitly
selected and you won't see the error with built-in or as a module.

Only if you unselect ARCH_QCOM and use x86_64 (or any other arch) with
COMPILE_TEST, then you'll see the error with both built-in and module.

> Therefore, as requested:
> 
> "For the next version, you'd need to fix all possible Kconfig build
> errors before sending."
> 

Sure.

Thanks,
Mani

> Thx.
> 
> -- 
> Regards/Gruss,
>     Boris.
> 
> https://people.kernel.org/tglx/notes-about-netiquette
diff mbox series

Patch

diff --git a/drivers/edac/qcom_edac.c b/drivers/edac/qcom_edac.c
index 97a27e42dd61..04df70b7fea3 100644
--- a/drivers/edac/qcom_edac.c
+++ b/drivers/edac/qcom_edac.c
@@ -21,30 +21,9 @@ 
 #define TRP_SYN_REG_CNT                 6
 #define DRP_SYN_REG_CNT                 8
 
-#define LLCC_COMMON_STATUS0             0x0003000c
 #define LLCC_LB_CNT_MASK                GENMASK(31, 28)
 #define LLCC_LB_CNT_SHIFT               28
 
-/* Single & double bit syndrome register offsets */
-#define TRP_ECC_SB_ERR_SYN0             0x0002304c
-#define TRP_ECC_DB_ERR_SYN0             0x00020370
-#define DRP_ECC_SB_ERR_SYN0             0x0004204c
-#define DRP_ECC_DB_ERR_SYN0             0x00042070
-
-/* Error register offsets */
-#define TRP_ECC_ERROR_STATUS1           0x00020348
-#define TRP_ECC_ERROR_STATUS0           0x00020344
-#define DRP_ECC_ERROR_STATUS1           0x00042048
-#define DRP_ECC_ERROR_STATUS0           0x00042044
-
-/* TRP, DRP interrupt register offsets */
-#define DRP_INTERRUPT_STATUS            0x00041000
-#define TRP_INTERRUPT_0_STATUS          0x00020480
-#define DRP_INTERRUPT_CLEAR             0x00041008
-#define DRP_ECC_ERROR_CNTR_CLEAR        0x00040004
-#define TRP_INTERRUPT_0_CLEAR           0x00020484
-#define TRP_ECC_ERROR_CNTR_CLEAR        0x00020440
-
 /* Mask and shift macros */
 #define ECC_DB_ERR_COUNT_MASK           GENMASK(4, 0)
 #define ECC_DB_ERR_WAYS_MASK            GENMASK(31, 16)
@@ -60,15 +39,6 @@ 
 #define DRP_TRP_INT_CLEAR               GENMASK(1, 0)
 #define DRP_TRP_CNT_CLEAR               GENMASK(1, 0)
 
-/* Config registers offsets*/
-#define DRP_ECC_ERROR_CFG               0x00040000
-
-/* Tag RAM, Data RAM interrupt register offsets */
-#define CMN_INTERRUPT_0_ENABLE          0x0003001c
-#define CMN_INTERRUPT_2_ENABLE          0x0003003c
-#define TRP_INTERRUPT_0_ENABLE          0x00020488
-#define DRP_INTERRUPT_ENABLE            0x0004100c
-
 #define SB_ERROR_THRESHOLD              0x1
 #define SB_ERROR_THRESHOLD_SHIFT        24
 #define SB_DB_TRP_INTERRUPT_ENABLE      0x3
@@ -86,9 +56,6 @@  enum {
 static const struct llcc_edac_reg_data edac_reg_data[] = {
 	[LLCC_DRAM_CE] = {
 		.name = "DRAM Single-bit",
-		.synd_reg = DRP_ECC_SB_ERR_SYN0,
-		.count_status_reg = DRP_ECC_ERROR_STATUS1,
-		.ways_status_reg = DRP_ECC_ERROR_STATUS0,
 		.reg_cnt = DRP_SYN_REG_CNT,
 		.count_mask = ECC_SB_ERR_COUNT_MASK,
 		.ways_mask = ECC_SB_ERR_WAYS_MASK,
@@ -96,9 +63,6 @@  static const struct llcc_edac_reg_data edac_reg_data[] = {
 	},
 	[LLCC_DRAM_UE] = {
 		.name = "DRAM Double-bit",
-		.synd_reg = DRP_ECC_DB_ERR_SYN0,
-		.count_status_reg = DRP_ECC_ERROR_STATUS1,
-		.ways_status_reg = DRP_ECC_ERROR_STATUS0,
 		.reg_cnt = DRP_SYN_REG_CNT,
 		.count_mask = ECC_DB_ERR_COUNT_MASK,
 		.ways_mask = ECC_DB_ERR_WAYS_MASK,
@@ -106,9 +70,6 @@  static const struct llcc_edac_reg_data edac_reg_data[] = {
 	},
 	[LLCC_TRAM_CE] = {
 		.name = "TRAM Single-bit",
-		.synd_reg = TRP_ECC_SB_ERR_SYN0,
-		.count_status_reg = TRP_ECC_ERROR_STATUS1,
-		.ways_status_reg = TRP_ECC_ERROR_STATUS0,
 		.reg_cnt = TRP_SYN_REG_CNT,
 		.count_mask = ECC_SB_ERR_COUNT_MASK,
 		.ways_mask = ECC_SB_ERR_WAYS_MASK,
@@ -116,9 +77,6 @@  static const struct llcc_edac_reg_data edac_reg_data[] = {
 	},
 	[LLCC_TRAM_UE] = {
 		.name = "TRAM Double-bit",
-		.synd_reg = TRP_ECC_DB_ERR_SYN0,
-		.count_status_reg = TRP_ECC_ERROR_STATUS1,
-		.ways_status_reg = TRP_ECC_ERROR_STATUS0,
 		.reg_cnt = TRP_SYN_REG_CNT,
 		.count_mask = ECC_DB_ERR_COUNT_MASK,
 		.ways_mask = ECC_DB_ERR_WAYS_MASK,
@@ -126,7 +84,7 @@  static const struct llcc_edac_reg_data edac_reg_data[] = {
 	},
 };
 
-static int qcom_llcc_core_setup(struct regmap *llcc_bcast_regmap)
+static int qcom_llcc_core_setup(struct llcc_drv_data *drv, struct regmap *llcc_bcast_regmap)
 {
 	u32 sb_err_threshold;
 	int ret;
@@ -135,31 +93,31 @@  static int qcom_llcc_core_setup(struct regmap *llcc_bcast_regmap)
 	 * Configure interrupt enable registers such that Tag, Data RAM related
 	 * interrupts are propagated to interrupt controller for servicing
 	 */
-	ret = regmap_update_bits(llcc_bcast_regmap, CMN_INTERRUPT_2_ENABLE,
+	ret = regmap_update_bits(llcc_bcast_regmap, drv->edac_reg_offset->cmn_interrupt_2_enable,
 				 TRP0_INTERRUPT_ENABLE,
 				 TRP0_INTERRUPT_ENABLE);
 	if (ret)
 		return ret;
 
-	ret = regmap_update_bits(llcc_bcast_regmap, TRP_INTERRUPT_0_ENABLE,
+	ret = regmap_update_bits(llcc_bcast_regmap, drv->edac_reg_offset->trp_interrupt_0_enable,
 				 SB_DB_TRP_INTERRUPT_ENABLE,
 				 SB_DB_TRP_INTERRUPT_ENABLE);
 	if (ret)
 		return ret;
 
 	sb_err_threshold = (SB_ERROR_THRESHOLD << SB_ERROR_THRESHOLD_SHIFT);
-	ret = regmap_write(llcc_bcast_regmap, DRP_ECC_ERROR_CFG,
+	ret = regmap_write(llcc_bcast_regmap, drv->edac_reg_offset->drp_ecc_error_cfg,
 			   sb_err_threshold);
 	if (ret)
 		return ret;
 
-	ret = regmap_update_bits(llcc_bcast_regmap, CMN_INTERRUPT_2_ENABLE,
+	ret = regmap_update_bits(llcc_bcast_regmap, drv->edac_reg_offset->cmn_interrupt_2_enable,
 				 DRP0_INTERRUPT_ENABLE,
 				 DRP0_INTERRUPT_ENABLE);
 	if (ret)
 		return ret;
 
-	ret = regmap_write(llcc_bcast_regmap, DRP_INTERRUPT_ENABLE,
+	ret = regmap_write(llcc_bcast_regmap, drv->edac_reg_offset->drp_interrupt_enable,
 			   SB_DB_DRP_INTERRUPT_ENABLE);
 	return ret;
 }
@@ -173,24 +131,28 @@  qcom_llcc_clear_error_status(int err_type, struct llcc_drv_data *drv)
 	switch (err_type) {
 	case LLCC_DRAM_CE:
 	case LLCC_DRAM_UE:
-		ret = regmap_write(drv->bcast_regmap, DRP_INTERRUPT_CLEAR,
+		ret = regmap_write(drv->bcast_regmap,
+				   drv->edac_reg_offset->drp_interrupt_clear,
 				   DRP_TRP_INT_CLEAR);
 		if (ret)
 			return ret;
 
-		ret = regmap_write(drv->bcast_regmap, DRP_ECC_ERROR_CNTR_CLEAR,
+		ret = regmap_write(drv->bcast_regmap,
+				   drv->edac_reg_offset->drp_ecc_error_cntr_clear,
 				   DRP_TRP_CNT_CLEAR);
 		if (ret)
 			return ret;
 		break;
 	case LLCC_TRAM_CE:
 	case LLCC_TRAM_UE:
-		ret = regmap_write(drv->bcast_regmap, TRP_INTERRUPT_0_CLEAR,
+		ret = regmap_write(drv->bcast_regmap,
+				   drv->edac_reg_offset->trp_interrupt_0_clear,
 				   DRP_TRP_INT_CLEAR);
 		if (ret)
 			return ret;
 
-		ret = regmap_write(drv->bcast_regmap, TRP_ECC_ERROR_CNTR_CLEAR,
+		ret = regmap_write(drv->bcast_regmap,
+				   drv->edac_reg_offset->trp_ecc_error_cntr_clear,
 				   DRP_TRP_CNT_CLEAR);
 		if (ret)
 			return ret;
@@ -203,16 +165,54 @@  qcom_llcc_clear_error_status(int err_type, struct llcc_drv_data *drv)
 	return ret;
 }
 
+struct qcom_llcc_syn_regs {
+	u32 synd_reg;
+	u32 count_status_reg;
+	u32 ways_status_reg;
+};
+
+static void get_reg_offsets(struct llcc_drv_data *drv, int err_type,
+			    struct qcom_llcc_syn_regs *syn_regs)
+{
+	const struct llcc_edac_reg_offset *edac_reg_offset = drv->edac_reg_offset;
+
+	switch (err_type) {
+	case LLCC_DRAM_CE:
+		syn_regs->synd_reg = edac_reg_offset->drp_ecc_sb_err_syn0;
+		syn_regs->count_status_reg = edac_reg_offset->drp_ecc_error_status1;
+		syn_regs->ways_status_reg = edac_reg_offset->drp_ecc_error_status0;
+		break;
+	case LLCC_DRAM_UE:
+		syn_regs->synd_reg = edac_reg_offset->drp_ecc_db_err_syn0;
+		syn_regs->count_status_reg = edac_reg_offset->drp_ecc_error_status1;
+		syn_regs->ways_status_reg = edac_reg_offset->drp_ecc_error_status0;
+		break;
+	case LLCC_TRAM_CE:
+		syn_regs->synd_reg = edac_reg_offset->trp_ecc_sb_err_syn0;
+		syn_regs->count_status_reg = edac_reg_offset->trp_ecc_error_status1;
+		syn_regs->ways_status_reg = edac_reg_offset->trp_ecc_error_status0;
+		break;
+	case LLCC_TRAM_UE:
+		syn_regs->synd_reg = edac_reg_offset->trp_ecc_db_err_syn0;
+		syn_regs->count_status_reg = edac_reg_offset->trp_ecc_error_status1;
+		syn_regs->ways_status_reg = edac_reg_offset->trp_ecc_error_status0;
+		break;
+	}
+}
+
 /* Dump Syndrome registers data for Tag RAM, Data RAM bit errors*/
 static int
 dump_syn_reg_values(struct llcc_drv_data *drv, u32 bank, int err_type)
 {
 	struct llcc_edac_reg_data reg_data = edac_reg_data[err_type];
+	struct qcom_llcc_syn_regs regs = { };
 	int err_cnt, err_ways, ret, i;
 	u32 synd_reg, synd_val;
 
+	get_reg_offsets(drv, err_type, &regs);
+
 	for (i = 0; i < reg_data.reg_cnt; i++) {
-		synd_reg = reg_data.synd_reg + (i * 4);
+		synd_reg = regs.synd_reg + (i * 4);
 		ret = regmap_read(drv->regmap, drv->offsets[bank] + synd_reg,
 				  &synd_val);
 		if (ret)
@@ -223,7 +223,7 @@  dump_syn_reg_values(struct llcc_drv_data *drv, u32 bank, int err_type)
 	}
 
 	ret = regmap_read(drv->regmap,
-			  drv->offsets[bank] + reg_data.count_status_reg,
+			  drv->offsets[bank] + regs.count_status_reg,
 			  &err_cnt);
 	if (ret)
 		goto clear;
@@ -234,7 +234,7 @@  dump_syn_reg_values(struct llcc_drv_data *drv, u32 bank, int err_type)
 		    reg_data.name, err_cnt);
 
 	ret = regmap_read(drv->regmap,
-			  drv->offsets[bank] + reg_data.ways_status_reg,
+			  drv->offsets[bank] + regs.ways_status_reg,
 			  &err_ways);
 	if (ret)
 		goto clear;
@@ -297,7 +297,7 @@  llcc_ecc_irq_handler(int irq, void *edev_ctl)
 	/* Iterate over the banks and look for Tag RAM or Data RAM errors */
 	for (i = 0; i < drv->num_banks; i++) {
 		ret = regmap_read(drv->regmap,
-				  drv->offsets[i] + DRP_INTERRUPT_STATUS,
+				  drv->offsets[i] + drv->edac_reg_offset->drp_interrupt_status,
 				  &drp_error);
 
 		if (!ret && (drp_error & SB_ECC_ERROR)) {
@@ -313,7 +313,7 @@  llcc_ecc_irq_handler(int irq, void *edev_ctl)
 			irq_rc = IRQ_HANDLED;
 
 		ret = regmap_read(drv->regmap,
-				  drv->offsets[i] + TRP_INTERRUPT_0_STATUS,
+				  drv->offsets[i] + drv->edac_reg_offset->trp_interrupt_0_status,
 				  &trp_error);
 
 		if (!ret && (trp_error & SB_ECC_ERROR)) {
@@ -340,7 +340,7 @@  static int qcom_llcc_edac_probe(struct platform_device *pdev)
 	int ecc_irq;
 	int rc;
 
-	rc = qcom_llcc_core_setup(llcc_driv_data->bcast_regmap);
+	rc = qcom_llcc_core_setup(llcc_driv_data, llcc_driv_data->bcast_regmap);
 	if (rc)
 		return rc;
 
diff --git a/include/linux/soc/qcom/llcc-qcom.h b/include/linux/soc/qcom/llcc-qcom.h
index bc2fb8343a94..d5b2d58e8857 100644
--- a/include/linux/soc/qcom/llcc-qcom.h
+++ b/include/linux/soc/qcom/llcc-qcom.h
@@ -57,9 +57,6 @@  struct llcc_slice_desc {
 /**
  * struct llcc_edac_reg_data - llcc edac registers data for each error type
  * @name: Name of the error
- * @synd_reg: Syndrome register address
- * @count_status_reg: Status register address to read the error count
- * @ways_status_reg: Status register address to read the error ways
  * @reg_cnt: Number of registers
  * @count_mask: Mask value to get the error count
  * @ways_mask: Mask value to get the error ways
@@ -68,9 +65,6 @@  struct llcc_slice_desc {
  */
 struct llcc_edac_reg_data {
 	char *name;
-	u64 synd_reg;
-	u64 count_status_reg;
-	u64 ways_status_reg;
 	u32 reg_cnt;
 	u32 count_mask;
 	u32 ways_mask;