Re: [PATCH 07/11 v5] coresight-etm: add CoreSight ETM/PTM driver
From: Mathieu Poirier
Date: Thu Sep 04 2014 - 13:19:50 EST
Thanks for the review - pls see comments below.
Mathieu
On 3 September 2014 02:37, Linus Walleij <linus.walleij@xxxxxxxxxx> wrote:
> On Wed, Aug 27, 2014 at 7:17 PM, <mathieu.poirier@xxxxxxxxxx> wrote:
>
>> From: Pratik Patel <pratikp@xxxxxxxxxxxxxx>
>>
>> This driver manages CoreSight ETM (Embedded Trace Macrocell) that
>> supports processor tracing. Currently supported version are ARM
>> ETMv3.x and PTM1.x.
>>
>> Signed-off-by: Pratik Patel <pratikp@xxxxxxxxxxxxxx>
>> Signed-off-by: Panchaxari Prasannamurthy <panchaxari.prasannamurthy@xxxxxxxxxx>
>> Signed-off-by: Mathieu Poirier <mathieu.poirier@xxxxxxxxxx>
>
> (...)
>> +/* Accessors for CP14 registers */
>> +#define dbg_read(reg) RCP14_##reg()
>> +#define dbg_write(val, reg) WCP14_##reg(val)
>> +#define etm_read(reg) RCP14_##reg()
>> +#define etm_write(val, reg) WCP14_##reg(val)
>> +
>> +/* MRC14 and MCR14 */
>> +#define MRC14(op1, crn, crm, op2) \
>> +({ \
>> +u32 val; \
>> +asm volatile("mrc p14, "#op1", %0, "#crn", "#crm", "#op2 : "=r" (val)); \
>> +val; \
>> +})
>> +
>> +#define MCR14(val, op1, crn, crm, op2) \
>> +({ \
>> +asm volatile("mcr p14, "#op1", %0, "#crn", "#crm", "#op2 : : "r" (val));\
>> +})
>
> OK... some ARMish funny assembly...
>
> (...)
>
> But this:
>
>> +static unsigned int etm_read_reg(u32 reg)
>> +{
>> + switch (reg) {
>> + case ETMCR:
>> + return etm_read(ETMCR);
>> + case ETMCCR:
>> + return etm_read(ETMCCR);
> (...)
>
> And this:
>
>> +static void etm_write_reg(u32 val, u32 reg)
>> +{
>> + switch (reg) {
>> + case ETMCR:
>> + etm_write(val, ETMCR);
>> + return;
>> + case ETMTRIGGER:
>> + etm_write(val, ETMTRIGGER);
>> + return;
>
> And then this:
>
>> +unsigned int etm_readl_cp14(u32 off)
>> +{
>> + return etm_read_reg(off);
>> +}
>> +
>> +void etm_writel_cp14(u32 val, u32 off)
>> +{
>> + etm_write_reg(val, off);
>> +}
I agree, this layer is redundant and can be squashed.
>
> And then this:
>
> +static inline void etm_writel(struct etm_drvdata *drvdata,
> + u32 val, u32 off)
> +{
> + if (drvdata->use_cp14)
> + etm_writel_cp14(val, off);
> + else
> + writel_relaxed(val, drvdata->base + off);
> +}
> +
> +static inline unsigned int etm_readl(struct etm_drvdata *drvdata, u32 off)
> +{
> + u32 val;
> +
> + if (drvdata->use_cp14)
> + val = etm_readl_cp14(off);
> + else
> + val = readl_relaxed(drvdata->base + off);
> + return val;
> +}
>
> Just looks like an extra layer or two or three of indirection of
> absolutely no use or value. It seems to be done because someone
> doesn't know how to get parameters to assembly inlines and
> added all the switch statements for get #defined enumerators
> from variables.
>
> The code actually using these layered accessors look like this:
>
> etmcr = etm_readl(drvdata, ETMCR);
>
> I would rewrite the last function like this:
>
> static inline unsigned int etm_readl(struct etm_drvdata *drvdata, u32 off)
> {
> u32 val;
>
> if (drvdata->use_cp14)
> asm_volatile()...
> else
> val = readl_relaxed(drvdata->base + off);
> return val;
> }
That unfortunately won't work. "u32 off" is a numerical value and it
works well in the case were CP14 accesses aren't needed. On the flip
side that numerical value isn't sufficient to deduce all 3 arguments
(crn, crm, op2) required by instructions mcr/mrc for the right access
to happen - there is simply no correlation between the offset of an
APB bus memory mapped address and the corresponding CP14 access.
I'd concur with your approach if it was just a matter of transforming
the C variable into an in-lined assembly instruction parameter but it
isn't the case. As indicated above though the
etm_writel_cp14/etm_write_reg layer can go but the rest has to stay.
Get back to me if you do not agree with my assessment.
>
> The asm_volatile() and parametrizing it is the trick. Just figure it out ;)
> http://www.ethernut.de/en/documents/arm-inline-asm.html
Yes, I have the same link bookmarked.
>
>> diff --git a/drivers/coresight/coresight-etm.h b/drivers/coresight/coresight-etm.h
> (...)
>> +struct etm_drvdata {
>> + void __iomem *base;
>> + struct device *dev;
>> + struct coresight_device *csdev;
>> + struct clk *clk;
>> + spinlock_t spinlock;
>> + int cpu;
>> + int port_size;
>> + u8 arch;
>> + bool use_cp14;
>> + bool enable;
>> + bool sticky_enable;
>> + bool boot_enable;
>> + bool os_unlock;
>> + u8 nr_addr_cmp;
>> + u8 nr_cntr;
>> + u8 nr_ext_inp;
>> + u8 nr_ext_out;
>> + u8 nr_ctxid_cmp;
>> + u8 reset;
>> + u32 etmccr;
>> + u32 etmccer;
>> + u32 traceid;
>> + u32 mode;
>> + u32 ctrl;
>> + u32 trigger_event;
>> + u32 startstop_ctrl;
>> + u32 enable_event;
>> + u32 enable_ctrl1;
>> + u32 fifofull_level;
>> + u8 addr_idx;
>> + u32 addr_val[ETM_MAX_ADDR_CMP];
>> + u32 addr_acctype[ETM_MAX_ADDR_CMP];
>> + u32 addr_type[ETM_MAX_ADDR_CMP];
>> + u8 cntr_idx;
>> + u32 cntr_rld_val[ETM_MAX_CNTR];
>> + u32 cntr_event[ETM_MAX_CNTR];
>> + u32 cntr_rld_event[ETM_MAX_CNTR];
>> + u32 cntr_val[ETM_MAX_CNTR];
>> + u32 seq_12_event;
>> + u32 seq_21_event;
>> + u32 seq_23_event;
>> + u32 seq_31_event;
>> + u32 seq_32_event;
>> + u32 seq_13_event;
>> + u32 seq_curr_state;
>> + u8 ctxid_idx;
>> + u32 ctxid_val[ETM_MAX_CTXID_CMP];
>> + u32 ctxid_mask;
>> + u32 sync_freq;
>> + u32 timestamp_event;
>> +};
>
> Unless all of this is self-evident I would kerneldoc it. If I don't
> understand something of this, and if you don't, who's going to
> understand it in a few years...
Yes, that needs to be better documented. I'll do the same for all the
other driver specific structure.
>
> Apart from this it looks OK. Albeit big. But that may be unavoidable.
>
> Yours,
> Linus Walleij
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/