Re: [RFC PATCH 13/17] net: ipa: Add support for IPA v2.x in the driver's QMI interface
From: Sireesh Kodali <hidden>
Date: 2021-10-18 18:33:08
Also in:
linux-arm-msm, lkml, phone-devel
On Thu Oct 14, 2021 at 4:00 AM IST, Alex Elder wrote:
On 9/19/21 10:08 PM, Sireesh Kodali wrote:quoted
On IPA v2.x, the modem doesn't send a DRIVER_INIT_COMPLETED, so we have to rely on the uc's IPA_UC_RESPONSE_INIT_COMPLETED to know when its ready. We add a function here that marks uc_ready = true. This function is called by ipa_uc.c when IPA_UC_RESPONSE_INIT_COMPLETED is handled.This should use the new ipa_mem_find() interface for getting the memory information for the ZIP region.
Got it, thanks
I don't know where the IPA_UC_RESPONSE_INIT_COMPLETED gets sent but I presume it ends up calling ipa_qmi_signal_uc_loaded().
IPA_UC_RESPONSE_INIT_COMPLETED is handled by the ipa_uc sub-driver. The handler calls ipa_qmi_signal_uc_loaded() once the response is received, at which point we know the uc has been inited.
I think actually the DRIVER_INIT_COMPLETE message from the modem is saying "I finished initializing the microcontroller." And I've wondered why there is a duplicate mechanism. Maybe there was a race or something.
This makes sense. Given that some modems rely on the IPA block for initialization, I wonder if Qualcomm decided it would be easier to allow the modem to complete the uc initialization and send the signal instead. Regards, Sireesh
-Alexquoted
Signed-off-by: Sireesh Kodali <redacted> Signed-off-by: Vladimir Lypak <redacted> --- drivers/net/ipa/ipa_qmi.c | 27 ++++++++++++++++++++++++++- drivers/net/ipa/ipa_qmi.h | 10 ++++++++++ 2 files changed, 36 insertions(+), 1 deletion(-)diff --git a/drivers/net/ipa/ipa_qmi.c b/drivers/net/ipa/ipa_qmi.c index 7e2fe701cc4d..876e2a004f70 100644 --- a/drivers/net/ipa/ipa_qmi.c +++ b/drivers/net/ipa/ipa_qmi.c@@ -68,6 +68,11 @@ * - The INDICATION_REGISTER request and INIT_COMPLETE indication are * optional for non-initial modem boots, and have no bearing on the * determination of when things are "ready" + * + * Note that on IPA v2.x, the modem doesn't send a DRIVER_INIT_COMPLETE + * request. Thus, we rely on the uc's IPA_UC_RESPONSE_INIT_COMPLETED to know + * when the uc is ready. The rest of the process is the same on IPA v2.x and + * later IPA versions */ #define IPA_HOST_SERVICE_SVC_ID 0x31@@ -345,7 +350,12 @@ init_modem_driver_req(struct ipa_qmi *ipa_qmi) req.hdr_proc_ctx_tbl_info.start + mem->size - 1; } - /* Nothing to report for the compression table (zip_tbl_info) */ + mem = &ipa->mem[IPA_MEM_ZIP]; + if (mem->size) { + req.zip_tbl_info_valid = 1; + req.zip_tbl_info.start = ipa->mem_offset + mem->offset; + req.zip_tbl_info.end = ipa->mem_offset + mem->size - 1; + } mem = ipa_mem_find(ipa, IPA_MEM_V4_ROUTE_HASHED); if (mem->size) {@@ -525,6 +535,21 @@ int ipa_qmi_setup(struct ipa *ipa) return ret; } +/* With IPA v2 modem is not required to send DRIVER_INIT_COMPLETE request to AP. + * We start operation as soon as IPA_UC_RESPONSE_INIT_COMPLETED irq is triggered. + */ +void ipa_qmi_signal_uc_loaded(struct ipa *ipa) +{ + struct ipa_qmi *ipa_qmi = &ipa->qmi; + + /* This is needed only on IPA 2.x */ + if (ipa->version > IPA_VERSION_2_6L) + return; + + ipa_qmi->uc_ready = true; + ipa_qmi_ready(ipa_qmi); +} + /* Tear down IPA QMI handles */ void ipa_qmi_teardown(struct ipa *ipa) {diff --git a/drivers/net/ipa/ipa_qmi.h b/drivers/net/ipa/ipa_qmi.h index 856ef629ccc8..4962d88b0d22 100644 --- a/drivers/net/ipa/ipa_qmi.h +++ b/drivers/net/ipa/ipa_qmi.h@@ -55,6 +55,16 @@ struct ipa_qmi { */ int ipa_qmi_setup(struct ipa *ipa); +/** + * ipa_qmi_signal_uc_loaded() - Signal that the UC has been loaded + * @ipa: IPA pointer + * + * This is called when the uc indicates that it is ready. This exists, because + * on IPA v2.x, the modem does not send a DRIVER_INIT_COMPLETED. Thus we have + * to rely on the uc's INIT_COMPLETED response to know if it was initialized + */ +void ipa_qmi_signal_uc_loaded(struct ipa *ipa); + /** * ipa_qmi_teardown() - Tear down IPA QMI handles * @ipa: IPA pointer