The ETM4X driver now assume that all ETE as CPU system instructions accessed device, in fact the ETE device on some machines also accessed via MMIO.
Signed-off-by: Ruidong Tian tianruidong@linux.alibaba.com --- drivers/hwtracing/coresight/coresight-etm4x-core.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 285539104bcc..ad298c9cc87e 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -1103,8 +1103,9 @@ static bool etm4_init_iomem_access(struct etmv4_drvdata *drvdata, * with MMIO. But we cannot touch the OSLK until we are * sure this is an ETM. So rely only on the TRCDEVARCH. */ - if ((devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETMv4x_ARCH) { - pr_warn_once("TRCDEVARCH doesn't match ETMv4 architecture\n"); + if ((devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETMv4x_ARCH && + (devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETE_ARCH) { + pr_warn_once("TRCDEVARCH doesn't match ETMv4/ETE architecture\n"); return false; }
On 18/10/2023 08:05, Ruidong Tian wrote:
The ETM4X driver now assume that all ETE as CPU system instructions accessed device, in fact the ETE device on some machines also accessed via MMIO.
Signed-off-by: Ruidong Tian tianruidong@linux.alibaba.com
Why are we going backwards to MMIO from system instructions ? Is it because of an "unfriendly" hypervisor preventing access ?
As such, without a sufficiently acceptable explanation, I am reluctant to make this change
Suzuki
drivers/hwtracing/coresight/coresight-etm4x-core.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 285539104bcc..ad298c9cc87e 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -1103,8 +1103,9 @@ static bool etm4_init_iomem_access(struct etmv4_drvdata *drvdata, * with MMIO. But we cannot touch the OSLK until we are * sure this is an ETM. So rely only on the TRCDEVARCH. */
- if ((devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETMv4x_ARCH) {
pr_warn_once("TRCDEVARCH doesn't match ETMv4 architecture\n");
- if ((devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETMv4x_ARCH &&
(devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETE_ARCH) {pr_warn_once("TRCDEVARCH doesn't match ETMv4/ETE architecture\n");
Hi Suzuki,
Now ETM4X driver use MMIO or system instruction depends on this check in function etm4_init_csdev_access:
if (drvdata->base) return etm4_init_iomem_access(drvdata, csa);
This check always true if firmware provides a address range in ACPI table of ETE, and as a result, the ETE device in this case cannot be successfully probed.
I think OS should be compatible with this situation, no matter firmware how to organize ETE information in ACPI table. How do you feel about it?
Thank you
Ruidong Tian 在 2023/10/18 16:28, Suzuki K Poulose 写道:
On 18/10/2023 08:05, Ruidong Tian wrote:
The ETM4X driver now assume that all ETE as CPU system instructions accessed device, in fact the ETE device on some machines also accessed via MMIO.
Signed-off-by: Ruidong Tian tianruidong@linux.alibaba.com
Why are we going backwards to MMIO from system instructions ? Is it because of an "unfriendly" hypervisor preventing access ?
As such, without a sufficiently acceptable explanation, I am reluctant to make this change
Suzuki
drivers/hwtracing/coresight/coresight-etm4x-core.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 285539104bcc..ad298c9cc87e 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -1103,8 +1103,9 @@ static bool etm4_init_iomem_access(struct etmv4_drvdata *drvdata, * with MMIO. But we cannot touch the OSLK until we are * sure this is an ETM. So rely only on the TRCDEVARCH. */ - if ((devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETMv4x_ARCH) { - pr_warn_once("TRCDEVARCH doesn't match ETMv4 architecture\n"); + if ((devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETMv4x_ARCH && + (devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETE_ARCH) { + pr_warn_once("TRCDEVARCH doesn't match ETMv4/ETE architecture\n"); return false; }
Hi
On 18/10/2023 10:30, Ruidong Tian wrote:
Hi Suzuki,
Now ETM4X driver use MMIO or system instruction depends on this check in function etm4_init_csdev_access:
if (drvdata->base) return etm4_init_iomem_access(drvdata, csa);
This check always true if firmware provides a address range in ACPI table of ETE, and as a result, the ETE device in this case cannot be successfully probed.
I think OS should be compatible with this situation, no matter firmware how to organize ETE information in ACPI table. How do you feel about it?
My question is not about "What the patch does". But, why can't we use system instructions on your system, when ETE was designed to be used with that in the first place and get rid of the MMIO.
Suzuki
Thank you
Ruidong Tian 在 2023/10/18 16:28, Suzuki K Poulose 写道:
On 18/10/2023 08:05, Ruidong Tian wrote:
The ETM4X driver now assume that all ETE as CPU system instructions accessed device, in fact the ETE device on some machines also accessed via MMIO.
Signed-off-by: Ruidong Tian tianruidong@linux.alibaba.com
Why are we going backwards to MMIO from system instructions ? Is it because of an "unfriendly" hypervisor preventing access ?
As such, without a sufficiently acceptable explanation, I am reluctant to make this change
Suzuki
drivers/hwtracing/coresight/coresight-etm4x-core.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 285539104bcc..ad298c9cc87e 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -1103,8 +1103,9 @@ static bool etm4_init_iomem_access(struct etmv4_drvdata *drvdata, * with MMIO. But we cannot touch the OSLK until we are * sure this is an ETM. So rely only on the TRCDEVARCH. */ - if ((devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETMv4x_ARCH) { - pr_warn_once("TRCDEVARCH doesn't match ETMv4 architecture\n"); + if ((devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETMv4x_ARCH && + (devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETE_ARCH) { + pr_warn_once("TRCDEVARCH doesn't match ETMv4/ETE architecture\n"); return false; }
Hi Suzuki
You are right, I review armv9 Spec again, and find that ETE only support system instructions access. This patch is meaningless and need to drop it.
Ruidong
在 2023/10/18 17:36, Suzuki K Poulose 写道:
Hi
On 18/10/2023 10:30, Ruidong Tian wrote:
Hi Suzuki,
Now ETM4X driver use MMIO or system instruction depends on this check in function etm4_init_csdev_access:
if (drvdata->base) return etm4_init_iomem_access(drvdata, csa);
This check always true if firmware provides a address range in ACPI table of ETE, and as a result, the ETE device in this case cannot be successfully probed.
I think OS should be compatible with this situation, no matter firmware how to organize ETE information in ACPI table. How do you feel about it?
My question is not about "What the patch does". But, why can't we use system instructions on your system, when ETE was designed to be used with that in the first place and get rid of the MMIO.
Suzuki
Thank you
Ruidong Tian 在 2023/10/18 16:28, Suzuki K Poulose 写道:
On 18/10/2023 08:05, Ruidong Tian wrote:
The ETM4X driver now assume that all ETE as CPU system instructions accessed device, in fact the ETE device on some machines also accessed via MMIO.
Signed-off-by: Ruidong Tian tianruidong@linux.alibaba.com
Why are we going backwards to MMIO from system instructions ? Is it because of an "unfriendly" hypervisor preventing access ?
As such, without a sufficiently acceptable explanation, I am reluctant to make this change
Suzuki
drivers/hwtracing/coresight/coresight-etm4x-core.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/hwtracing/coresight/coresight-etm4x-core.c b/drivers/hwtracing/coresight/coresight-etm4x-core.c index 285539104bcc..ad298c9cc87e 100644 --- a/drivers/hwtracing/coresight/coresight-etm4x-core.c +++ b/drivers/hwtracing/coresight/coresight-etm4x-core.c @@ -1103,8 +1103,9 @@ static bool etm4_init_iomem_access(struct etmv4_drvdata *drvdata, * with MMIO. But we cannot touch the OSLK until we are * sure this is an ETM. So rely only on the TRCDEVARCH. */ - if ((devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETMv4x_ARCH) { - pr_warn_once("TRCDEVARCH doesn't match ETMv4 architecture\n"); + if ((devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETMv4x_ARCH && + (devarch & ETM_DEVARCH_ID_MASK) != ETM_DEVARCH_ETE_ARCH) { + pr_warn_once("TRCDEVARCH doesn't match ETMv4/ETE architecture\n"); return false; }
-
Ruidong Tian -
Suzuki K Poulose