On 12/30/22 13:57, Adrian Hunter wrote:
On 30/12/22 08:42, Marek Vasut wrote:
On 12/29/22 13:51, Adrian Hunter wrote:
On 26/10/22 12:20, Marek Vasut wrote:
On 10/26/22 08:07, Adrian Hunter wrote:
On 25/10/22 22:15, Marek Vasut wrote:
On Xilinx ZynqMP, the reg_capabilities (SDIO) Register
https://www.xilinx.com/htmldocs/registers/ug1087/sdio___reg_capabilities.html#
Absolute Address 0x00FF160040 (SD0)
Reset Value 0x280737EC6481
really reads 0x200737EC6481 . The interesting part is the
top 32 bits, which are SDHCI_CAPABILITIES_1 = 0x2007. The
missing 0x800 is SDHCI_RETUNING_TIMER_COUNT_MASK=0, which
makes the SDHCI core disable retuning timer.
Fix this up here by explicitly setting tuning_count to 8
as it should be, otherwise an eMMC might fail in various
thermal conditions
Note that the diff is best shown with -w option, this makes it
visible what happened with !sdhci_arasan->has_cqe conditional,
which is placed between sdhci_setup_host() and __sdhci_add_host()
calls. Since sdhci_add_host() is also a sequence of these two
calls and host->tuning_count must be overriden before calling
overriden -> overridden
Fixed
__sdhci_add_host(), call the two calls separately and do all
the adjustments between them in either case.
Signed-off-by: Marek Vasut <marex@xxxxxxx>
---
Cc: Michal Simek <michal.simek@xxxxxxxxxx>
Cc: Adrian Hunter <adrian.hunter@xxxxxxxxx>
Cc: Ulf Hansson <ulf.hansson@xxxxxxxxxx>
Cc: linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
To: linux-mmc@xxxxxxxxxxxxxxx
---
drivers/mmc/host/sdhci-of-arasan.c | 57 ++++++++++++++++++++----------
1 file changed, 38 insertions(+), 19 deletions(-)
diff --git a/drivers/mmc/host/sdhci-of-arasan.c b/drivers/mmc/host/sdhci-of-arasan.c
index 3997cad1f793d..465498f2a7c0f 100644
--- a/drivers/mmc/host/sdhci-of-arasan.c
+++ b/drivers/mmc/host/sdhci-of-arasan.c
@@ -1521,37 +1521,56 @@ static int sdhci_arasan_register_sdclk(struct sdhci_arasan_data *sdhci_arasan,
return 0;
}
-static int sdhci_arasan_add_host(struct sdhci_arasan_data *sdhci_arasan)
+static int sdhci_arasan_add_host(struct sdhci_arasan_data *sdhci_arasan,
+ struct device *dev)
{
struct sdhci_host *host = sdhci_arasan->host;
struct cqhci_host *cq_host;
bool dma64;
int ret;
- if (!sdhci_arasan->has_cqe)
- return sdhci_add_host(host);
-
ret = sdhci_setup_host(host);
if (ret)
return ret;
- cq_host = devm_kzalloc(host->mmc->parent,
- sizeof(*cq_host), GFP_KERNEL);
- if (!cq_host) {
- ret = -ENOMEM;
- goto cleanup;
- }
+ /*
+ * On Xilinx ZynqMP, the reg_capabilities (SDIO) Register
+ *
+ * https://www.xilinx.com/htmldocs/registers/ug1087/sdio___reg_capabilities.html#
+ * Absolute Address 0x00FF160040 (SD0)
+ * Reset Value 0x280737EC6481
+ *
+ * really reads 0x200737EC6481 . The interesting part is the
+ * top 32 bits, which are SDHCI_CAPABILITIES_1 = 0x2007. The
+ * missing 0x800 is SDHCI_RETUNING_TIMER_COUNT_MASK=0, which
+ * makes the SDHCI core disable retuning timer.
Are you aware that caps can be changed in DT via "sdhci-caps" and
"sdhci-caps-mask" ?
No, I wasn't aware of those.
Is that the preferred approach to this fix, over handling it in the driver ?
I guess ideally. Mainline does not really need the driver
fix because it seems it can be done by DT. Older kernels
are a separate issue really.
I think the driver-side fix would be preferable, because it also fixes systems which use legacy DTs without the sdhci-caps properties, which would be all ZynqMP systems thus far.
You could backport support of the properties "sdhci-caps"
and "sdhci-caps-mask".
This won't help. Vivado (the xilinx FPGA design tool) is capable of generating DTs, so you can end up with a combination of new Linux kernel and old generated DT, which is still missing the sdhci-caps/sdhci-caps-mask .
That is a bit sad. You might want to push for changing that situation.
Send an updated patch then.
Xilinx Device Tree Generator, which is the tool for DT generation, was never
designed to be directly used without any change. It was designed to help you to
describe the system as much as possible. It means you get the base and you need
to change things which are not properly described. That's why just do it.
Thanks,
Michal
