After recent reports ([1], [2]) of older platforms (particularly 8150 and
7180) breaking after DDR sleep stats introduction, revert the following:
Commit 73380e2573c3 ("soc: qcom: stats: fix 64-bit division")
Commit e84e61bdb97c ("soc: qcom: stats: Add DDR sleep stats")
The feature itself is rather useful for debugging DRAM power management,
however it looks like the shared RPMh stats data structures differ on
previous SoCs.
Revert its addition for now to un-break booting on these earlier SoCs,
while I try to come up with a better way to enable it conditionally.
[1] https://lore.kernel.org/linux-arm-msm/[email protected]/
[2] https://lore.kernel.org/linux-arm-msm/CAD=FV=XX4wLg1NNVL15RK4D4tLvuSzZyUv=k_tS4bSb3=7QJzQ@mail.gmail.com/
Reported-by: Dmitry Baryshkov <[email protected]>
Reported-by: Doug Anderson <[email protected]>
Signed-off-by: Konrad Dybcio <[email protected]>
---
drivers/soc/qcom/qcom_stats.c | 187 +-----------------------------------------
1 file changed, 1 insertion(+), 186 deletions(-)
diff --git a/drivers/soc/qcom/qcom_stats.c b/drivers/soc/qcom/qcom_stats.c
index 5ec8a754b22b..0216fc24f2ca 100644
--- a/drivers/soc/qcom/qcom_stats.c
+++ b/drivers/soc/qcom/qcom_stats.c
@@ -3,7 +3,6 @@
* Copyright (c) 2011-2021, The Linux Foundation. All rights reserved.
*/
-#include <linux/bitfield.h>
#include <linux/debugfs.h>
#include <linux/device.h>
#include <linux/io.h>
@@ -12,7 +11,6 @@
#include <linux/platform_device.h>
#include <linux/seq_file.h>
-#include <linux/soc/qcom/qcom_aoss.h>
#include <linux/soc/qcom/smem.h>
#include <clocksource/arm_arch_timer.h>
@@ -24,20 +22,8 @@
#define LAST_ENTERED_AT_OFFSET 0x8
#define LAST_EXITED_AT_OFFSET 0x10
#define ACCUMULATED_OFFSET 0x18
-#define DDR_DYNAMIC_OFFSET 0x1c
- #define DDR_OFFSET_MASK GENMASK(9, 0)
#define CLIENT_VOTES_OFFSET 0x20
-#define ARCH_TIMER_FREQ 19200000
-#define DDR_MAGIC_KEY1 0xA1157A75 /* leetspeak "ALLSTATS" */
-#define DDR_MAX_NUM_ENTRIES 20
-
-#define DDR_VOTE_DRV_MAX 18
-#define DDR_VOTE_DRV_ABSENT 0xdeaddead
-#define DDR_VOTE_DRV_INVALID 0xffffdead
-#define DDR_VOTE_X GENMASK(27, 14)
-#define DDR_VOTE_Y GENMASK(13, 0)
-
struct subsystem_data {
const char *name;
u32 smem_item;
@@ -62,7 +48,6 @@ struct stats_config {
bool appended_stats_avail;
bool dynamic_offset;
bool subsystem_stats_in_smem;
- bool ddr_stats;
};
struct stats_data {
@@ -83,25 +68,6 @@ struct appended_stats {
u32 reserved[3];
};
-struct ddr_stats_entry {
- u32 name;
- u32 count;
- u64 dur;
-} __packed;
-
-struct ddr_stats {
- u32 key;
- u32 entry_count;
-#define MAX_DDR_STAT_ENTRIES 20
- struct ddr_stats_entry entry[MAX_DDR_STAT_ENTRIES];
-} __packed;
-
-struct ddr_stats_data {
- struct device *dev;
- void __iomem *base;
- struct qmp *qmp;
-};
-
static void qcom_print_stats(struct seq_file *s, const struct sleep_stats *stat)
{
u64 accumulated = stat->accumulated;
@@ -152,108 +118,6 @@ static int qcom_soc_sleep_stats_show(struct seq_file *s, void *unused)
return 0;
}
-#define DDR_NAME_TYPE GENMASK(15, 8)
- #define DDR_NAME_TYPE_LPM 0
- #define DDR_NAME_TYPE_FREQ 1
-
-#define DDR_NAME_LPM_NAME GENMASK(7, 0)
-
-#define DDR_NAME_FREQ_MHZ GENMASK(31, 16)
-#define DDR_NAME_FREQ_CP_IDX GENMASK(4, 0)
-static void qcom_ddr_stats_print(struct seq_file *s, struct ddr_stats_entry *entry)
-{
- u32 cp_idx, name;
- u8 type;
-
- type = FIELD_GET(DDR_NAME_TYPE, entry->name);
-
- switch (type) {
- case DDR_NAME_TYPE_LPM:
- name = FIELD_GET(DDR_NAME_LPM_NAME, entry->name);
-
- seq_printf(s, "LPM | Type 0x%2x\tcount: %u\ttime: %llums\n",
- name, entry->count, entry->dur);
- break;
- case DDR_NAME_TYPE_FREQ:
- cp_idx = FIELD_GET(DDR_NAME_FREQ_CP_IDX, entry->name);
- name = FIELD_GET(DDR_NAME_FREQ_MHZ, entry->name);
-
- /* Neither 0Mhz nor 0 votes is very interesting */
- if (!name || !entry->count)
- return;
-
- seq_printf(s, "Freq | %dMHz (idx %u)\tcount: %u\ttime: %llums\n",
- name, cp_idx, entry->count, entry->dur);
- break;
- default:
- seq_printf(s, "Unknown data chunk (type = 0x%x count = 0x%x dur = 0x%llx)\n",
- type, entry->count, entry->dur);
- }
-}
-
-static int qcom_ddr_stats_show(struct seq_file *s, void *unused)
-{
- struct ddr_stats_data *ddrd = s->private;
- struct ddr_stats ddr;
- struct ddr_stats_entry *entry = ddr.entry;
- u32 entry_count, stats_size;
- u32 votes[DDR_VOTE_DRV_MAX];
- int i, ret;
-
- /* Request a stats sync, it may take some time to update though.. */
- ret = qmp_send(ddrd->qmp, "{class: ddr, action: freqsync}");
- if (ret) {
- dev_err(ddrd->dev, "failed to send QMP message\n");
- return ret;
- }
-
- entry_count = readl(ddrd->base + offsetof(struct ddr_stats, entry_count));
- if (entry_count > DDR_MAX_NUM_ENTRIES)
- return -EINVAL;
-
- /* We're not guaranteed to have DDR_MAX_NUM_ENTRIES */
- stats_size = sizeof(ddr);
- stats_size -= DDR_MAX_NUM_ENTRIES * sizeof(*entry);
- stats_size += entry_count * sizeof(*entry);
-
- /* Copy and process the stats */
- memcpy_fromio(&ddr, ddrd->base, stats_size);
-
- for (i = 0; i < ddr.entry_count; i++) {
- /* Convert the period to ms */
- entry[i].dur = div_u64(entry[i].dur, ARCH_TIMER_FREQ / MSEC_PER_SEC);
- }
-
- for (i = 0; i < ddr.entry_count; i++)
- qcom_ddr_stats_print(s, &entry[i]);
-
- /* Ask AOSS to dump DDR votes */
- ret = qmp_send(ddrd->qmp, "{class: ddr, res: drvs_ddr_votes}");
- if (ret) {
- dev_err(ddrd->dev, "failed to send QMP message\n");
- return ret;
- }
-
- /* Subsystem votes */
- memcpy_fromio(votes, ddrd->base + stats_size, sizeof(u32) * DDR_VOTE_DRV_MAX);
-
- for (i = 0; i < DDR_VOTE_DRV_MAX; i++) {
- u32 ab, ib;
-
- if (votes[i] == DDR_VOTE_DRV_ABSENT || votes[i] == DDR_VOTE_DRV_INVALID)
- ab = ib = votes[i];
- else {
- ab = FIELD_GET(DDR_VOTE_X, votes[i]);
- ib = FIELD_GET(DDR_VOTE_Y, votes[i]);
- }
-
- seq_printf(s, "Vote | AB = %5u\tIB = %5u\n", ab, ib);
- }
-
- return 0;
-}
-
-DEFINE_SHOW_ATTRIBUTE(qcom_ddr_stats);
DEFINE_SHOW_ATTRIBUTE(qcom_soc_sleep_stats);
DEFINE_SHOW_ATTRIBUTE(qcom_subsystem_sleep_stats);
@@ -316,56 +180,13 @@ static void qcom_create_subsystem_stat_files(struct dentry *root,
&qcom_subsystem_sleep_stats_fops);
}
-static int qcom_create_ddr_stats_files(struct device *dev,
- struct dentry *root,
- void __iomem *reg,
- const struct stats_config *config)
-{
- struct ddr_stats_data *ddrd;
- u32 key, stats_offset;
- struct dentry *dent;
-
- /* Nothing to do */
- if (!config->ddr_stats)
- return 0;
-
- ddrd = devm_kzalloc(dev, sizeof(*ddrd), GFP_KERNEL);
- if (!ddrd)
- return dev_err_probe(dev, -ENOMEM, "Couldn't allocate DDR stats data\n");
-
- ddrd->dev = dev;
-
- /* Get the offset of DDR stats */
- stats_offset = readl(reg + DDR_DYNAMIC_OFFSET) & DDR_OFFSET_MASK;
- ddrd->base = reg + stats_offset;
-
- /* Check if DDR stats are present */
- key = readl(ddrd->base);
- if (key != DDR_MAGIC_KEY1)
- return 0;
-
- dent = debugfs_create_file("ddr_sleep_stats", 0400, root, ddrd, &qcom_ddr_stats_fops);
- if (IS_ERR(dent))
- return PTR_ERR(dent);
-
- /* QMP is only necessary for DDR votes */
- ddrd->qmp = qmp_get(dev);
- if (IS_ERR(ddrd->qmp)) {
- dev_err(dev, "Couldn't get QMP mailbox: %ld. DDR votes won't be available.\n",
- PTR_ERR(ddrd->qmp));
- debugfs_remove(dent);
- }
-
- return 0;
-}
-
static int qcom_stats_probe(struct platform_device *pdev)
{
void __iomem *reg;
struct dentry *root;
const struct stats_config *config;
struct stats_data *d;
- int i, ret;
+ int i;
config = device_get_match_data(&pdev->dev);
if (!config)
@@ -387,11 +208,6 @@ static int qcom_stats_probe(struct platform_device *pdev)
qcom_create_subsystem_stat_files(root, config);
qcom_create_soc_sleep_stat_files(root, reg, d, config);
- ret = qcom_create_ddr_stats_files(&pdev->dev, root, reg, config);
- if (ret) {
- debugfs_remove_recursive(root);
- return ret;
- };
platform_set_drvdata(pdev, root);
@@ -438,7 +254,6 @@ static const struct stats_config rpmh_data = {
.appended_stats_avail = false,
.dynamic_offset = false,
.subsystem_stats_in_smem = true,
- .ddr_stats = true,
};
static const struct of_device_id qcom_stats_table[] = {
---
base-commit: 11651f8cb2e88372d4ed523d909514dc9a613ea3
change-id: 20231214-topic-undo_ddr_stats-e3ef145692f6
Best regards,
--
Konrad Dybcio <[email protected]>
Hi,
On Thu, Dec 14, 2023 at 4:25 AM Konrad Dybcio <[email protected]> wrote:
>
> After recent reports ([1], [2]) of older platforms (particularly 8150 and
> 7180) breaking after DDR sleep stats introduction, revert the following:
>
> Commit 73380e2573c3 ("soc: qcom: stats: fix 64-bit division")
> Commit e84e61bdb97c ("soc: qcom: stats: Add DDR sleep stats")
>
> The feature itself is rather useful for debugging DRAM power management,
> however it looks like the shared RPMh stats data structures differ on
> previous SoCs.
>
> Revert its addition for now to un-break booting on these earlier SoCs,
> while I try to come up with a better way to enable it conditionally.
>
> [1] https://lore.kernel.org/linux-arm-msm/[email protected]/
> [2] https://lore.kernel.org/linux-arm-msm/CAD=FV=XX4wLg1NNVL15RK4D4tLvuSzZyUv=k_tS4bSb3=7QJzQ@mail.gmail.com/
>
> Reported-by: Dmitry Baryshkov <[email protected]>
> Reported-by: Doug Anderson <[email protected]>
> Signed-off-by: Konrad Dybcio <[email protected]>
> ---
> drivers/soc/qcom/qcom_stats.c | 187 +-----------------------------------------
> 1 file changed, 1 insertion(+), 186 deletions(-)
Tested-by: Douglas Anderson <[email protected]>
On 14.12.2023 13:25, Konrad Dybcio wrote:
> After recent reports ([1], [2]) of older platforms (particularly 8150 and
> 7180) breaking after DDR sleep stats introduction, revert the following:
>
> Commit 73380e2573c3 ("soc: qcom: stats: fix 64-bit division")
> Commit e84e61bdb97c ("soc: qcom: stats: Add DDR sleep stats")
>
> The feature itself is rather useful for debugging DRAM power management,
> however it looks like the shared RPMh stats data structures differ on
> previous SoCs.
>
> Revert its addition for now to un-break booting on these earlier SoCs,
> while I try to come up with a better way to enable it conditionally.
>
> [1] https://lore.kernel.org/linux-arm-msm/[email protected]/
> [2] https://lore.kernel.org/linux-arm-msm/CAD=FV=XX4wLg1NNVL15RK4D4tLvuSzZyUv=k_tS4bSb3=7QJzQ@mail.gmail.com/
>
> Reported-by: Dmitry Baryshkov <[email protected]>
> Reported-by: Doug Anderson <[email protected]>
> Signed-off-by: Konrad Dybcio <[email protected]>
> ---
Arnd, since Bjorn seems to be MIA, could you please pick this directly
so that it gets into the next RC? Un-breaking booting on some platforms
would be very welcome :D
Konrad
On Sat, Dec 16, 2023 at 01:05:53AM +0100, Konrad Dybcio wrote:
> On 14.12.2023 13:25, Konrad Dybcio wrote:
> > After recent reports ([1], [2]) of older platforms (particularly 8150 and
> > 7180) breaking after DDR sleep stats introduction, revert the following:
> >
> > Commit 73380e2573c3 ("soc: qcom: stats: fix 64-bit division")
> > Commit e84e61bdb97c ("soc: qcom: stats: Add DDR sleep stats")
> >
> > The feature itself is rather useful for debugging DRAM power management,
> > however it looks like the shared RPMh stats data structures differ on
> > previous SoCs.
> >
> > Revert its addition for now to un-break booting on these earlier SoCs,
> > while I try to come up with a better way to enable it conditionally.
> >
> > [1] https://lore.kernel.org/linux-arm-msm/[email protected]/
> > [2] https://lore.kernel.org/linux-arm-msm/CAD=FV=XX4wLg1NNVL15RK4D4tLvuSzZyUv=k_tS4bSb3=7QJzQ@mail.gmail.com/
> >
> > Reported-by: Dmitry Baryshkov <[email protected]>
> > Reported-by: Doug Anderson <[email protected]>
> > Signed-off-by: Konrad Dybcio <[email protected]>
> > ---
> Arnd, since Bjorn seems to be MIA, could you please pick this directly
> so that it gets into the next RC? Un-breaking booting on some platforms
> would be very welcome :D
>
I'm confused, the two offending commits are staged for v6.8. Which -rc
do you want this applied in?! And you posted this patch yesterday...
Regards,
Bjorn
On 16.12.2023 05:35, Bjorn Andersson wrote:
> On Sat, Dec 16, 2023 at 01:05:53AM +0100, Konrad Dybcio wrote:
>> On 14.12.2023 13:25, Konrad Dybcio wrote:
>>> After recent reports ([1], [2]) of older platforms (particularly 8150 and
>>> 7180) breaking after DDR sleep stats introduction, revert the following:
>>>
>>> Commit 73380e2573c3 ("soc: qcom: stats: fix 64-bit division")
>>> Commit e84e61bdb97c ("soc: qcom: stats: Add DDR sleep stats")
>>>
>>> The feature itself is rather useful for debugging DRAM power management,
>>> however it looks like the shared RPMh stats data structures differ on
>>> previous SoCs.
>>>
>>> Revert its addition for now to un-break booting on these earlier SoCs,
>>> while I try to come up with a better way to enable it conditionally.
>>>
>>> [1] https://lore.kernel.org/linux-arm-msm/[email protected]/
>>> [2] https://lore.kernel.org/linux-arm-msm/CAD=FV=XX4wLg1NNVL15RK4D4tLvuSzZyUv=k_tS4bSb3=7QJzQ@mail.gmail.com/
>>>
>>> Reported-by: Dmitry Baryshkov <[email protected]>
>>> Reported-by: Doug Anderson <[email protected]>
>>> Signed-off-by: Konrad Dybcio <[email protected]>
>>> ---
>> Arnd, since Bjorn seems to be MIA, could you please pick this directly
>> so that it gets into the next RC? Un-breaking booting on some platforms
>> would be very welcome :D
>>
>
> I'm confused, the two offending commits are staged for v6.8. Which -rc
> do you want this applied in?! And you posted this patch yesterday...
Uh right, sorry, I should think twice before sending such emails late
into the night..
Konrad
On Thu, 14 Dec 2023 13:25:15 +0100, Konrad Dybcio wrote:
> After recent reports ([1], [2]) of older platforms (particularly 8150 and
> 7180) breaking after DDR sleep stats introduction, revert the following:
>
> Commit 73380e2573c3 ("soc: qcom: stats: fix 64-bit division")
> Commit e84e61bdb97c ("soc: qcom: stats: Add DDR sleep stats")
>
> The feature itself is rather useful for debugging DRAM power management,
> however it looks like the shared RPMh stats data structures differ on
> previous SoCs.
>
> [...]
Applied, thanks!
[1/1] Revert "soc: qcom: stats: Add DDR sleep stats"
commit: a7dc6343519752eb6d86bfa78378a8af5da1f475
Best regards,
--
Bjorn Andersson <[email protected]>