<html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:x="urn:schemas-microsoft-com:office:excel" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40"> <head> <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-7"> <meta name="Generator" content="Microsoft Word 15 (filtered medium)"> <style></style> </head> <body lang="EN-US" link="#0563C1" vlink="#954F72" style="word-wrap:break-word"> [AMD Official Use Only - Internal Distribution Only] <div class="WordSection1"> Hi Paul<o:p></o:p> <ol style="margin-top:0in" start="1" type="1"> <li class="MsoPlainText" style="mso-list:l1 level1 lfo2">The 50 ms is the whole full access time reduced, not one msleep(1), <o:p></o:p></li></ol> During amdgpu driver init, it will hit msleep(1) several times which increased the total time of full access. <o:p></o:p> <o:p> </o:p> I load amdgpu in the guest VM and collect VF’s full access time in the host, host dmesg was listed in the below:<o:p></o:p> In this time, the time reduced : 0.236847 s - 0.150411 s = 86.436 ms .<o:p></o:p> (The reason why it is 80+ms is that I add some code to program one register by psp. )<o:p></o:p> <table class="MsoNormalTable" border="1" cellspacing="0" cellpadding="0" width="562" style="width:421.5pt;margin-left:41.75pt;border-collapse:collapse;border:none"> <tbody> <tr style="height:14.5pt"> <td width="150" valign="top" style="width:112.2pt;border:solid windowtext 1.0pt;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> <o:p> </o:p> </td> <td width="130" nowrap="" valign="bottom" style="width:97.8pt;border:solid windowtext 1.0pt;border-left:none;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> VF Start Full access<o:p></o:p> </td> <td width="126" nowrap="" valign="bottom" style="width:94.5pt;border:solid windowtext 1.0pt;border-left:none;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> VF exit full access<o:p></o:p> </td> <td width="156" nowrap="" valign="bottom" style="width:117.0pt;border:solid windowtext 1.0pt;border-left:none;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> VF full access time<o:p></o:p> </td> </tr> <tr style="height:14.5pt"> <td width="150" valign="top" style="width:112.2pt;border:solid windowtext 1.0pt;border-top:none;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> msleep(1)<o:p></o:p> </td> <td width="130" nowrap="" valign="bottom" style="width:97.8pt;border-top:none;border-left:none;border-bottom:solid windowtext 1.0pt;border-right:solid windowtext 1.0pt;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> 295.9031 s<o:p></o:p> </td> <td width="126" nowrap="" valign="bottom" style="width:94.5pt;border-top:none;border-left:none;border-bottom:solid windowtext 1.0pt;border-right:solid windowtext 1.0pt;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> 296.0535 s<o:p></o:p> </td> <td width="156" nowrap="" valign="bottom" style="width:117.0pt;border-top:none;border-left:none;border-bottom:solid windowtext 1.0pt;border-right:solid windowtext 1.0pt;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> 0.150411 s<o:p></o:p> </td> </tr> <tr style="height:14.5pt"> <td width="150" valign="top" style="width:112.2pt;border:solid windowtext 1.0pt;border-top:none;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> usleep_range(10, 100)<o:p></o:p> </td> <td width="130" nowrap="" valign="bottom" style="width:97.8pt;border-top:none;border-left:none;border-bottom:solid windowtext 1.0pt;border-right:solid windowtext 1.0pt;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> 658.1791 s<o:p></o:p> </td> <td width="126" nowrap="" valign="bottom" style="width:94.5pt;border-top:none;border-left:none;border-bottom:solid windowtext 1.0pt;border-right:solid windowtext 1.0pt;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> 658.4159 s<o:p></o:p> </td> <td width="156" nowrap="" valign="bottom" style="width:117.0pt;border-top:none;border-left:none;border-bottom:solid windowtext 1.0pt;border-right:solid windowtext 1.0pt;padding:0in 5.4pt 0in 5.4pt;height:14.5pt"> 0.236847 s<o:p></o:p> </td> </tr> </tbody> </table> <o:p> </o:p> <ol style="margin-top:0in" start="2" type="1"> <li class="MsoPlainText" style="mso-list:l1 level1 lfo2">If I only change msleep(1) to usleep_range(10, 100), the polling time will reduced from 2 seconds to 0.2 seconds,<o:p></o:p></li></ol> So I change timeout from “timeout = 2000;” to “timeout = 20000;”<o:p></o:p> <o:p> </o:p> host dmesg with udelay_range(10, 100) in amdgpu:<o:p></o:p> [ 295.903102] gim info libgv: [4:0:0][amdgv_sched_enter_full_access:877] VF0 entered full access mode.<o:p></o:p> [ 295.906661] gim info libgv: [4:0:0][amdgv_ih_iv_ring_entry_process:254] PF_VF_MSGBUF_ACK received<o:p></o:p> [ 296.052903] gim info libgv: [4:0:0][amdgv_ih_iv_ring_entry_process:192] VF_PF_MSGBUF_VALID received<o:p></o:p> [ 296.052910] gim info libgv: [4:0:0][amdgv_ih_iv_ring_entry_process:205] Received Event: VF0, event = 0x2<o:p></o:p> [ 296.052914] gim info libgv: [4:0:0][amdgv_sched_event_queue_push_ex:193] queue event REL_GPU_INIT(0xef01) for VF0 of block(0xf0)<o:p></o:p> [ 296.052934] gim info libgv: [4:0:0][amdgv_sched_process_event:1582] process event REL_GPU_INIT (0xef01) for VF0 of block (0xf0)<o:p></o:p> [ 296.052944] gim info libgv: [4:0:0][navi12_gpuiov_set_mmsch_vfgate:904] mmsch mb ints disabled schedid = 4<o:p></o:p> [ 296.053334] gim info libgv: [4:0:0][navi12_psp_v11_set_mb_int:632] psp mailbox disabled for VF0<o:p></o:p> [ 296.053513] gim info libgv: [4:0:0][amdgv_sched_exit_full_access:976] VF0 exited full access.<o:p></o:p> <o:p> </o:p> Host demsg with msleep(1) in amdgpu:<o:p></o:p> [ 658.179053] gim info libgv: [4:0:0][amdgv_sched_enter_full_access:877] VF0 entered full access mode.<o:p></o:p> [ 658.182648] gim info libgv: [4:0:0][amdgv_ih_iv_ring_entry_process:254] PF_VF_MSGBUF_ACK received<o:p></o:p> [ 658.415227] gim info libgv: [4:0:0][amdgv_ih_iv_ring_entry_process:192] VF_PF_MSGBUF_VALID received<o:p></o:p> [ 658.415237] gim info libgv: [4:0:0][amdgv_ih_iv_ring_entry_process:205] Received Event: VF0, event = 0x2<o:p></o:p> [ 658.415241] gim info libgv: [4:0:0][amdgv_sched_event_queue_push_ex:193] queue event REL_GPU_INIT(0xef01) for VF0 of block(0xf0)<o:p></o:p> [ 658.415299] gim info libgv: [4:0:0][amdgv_sched_process_event:1582] process event REL_GPU_INIT (0xef01) for VF0 of block (0xf0)<o:p></o:p> [ 658.415311] gim info libgv: [4:0:0][navi12_gpuiov_set_mmsch_vfgate:904] mmsch mb ints disabled schedid = 4<o:p></o:p> [ 658.415719] gim info libgv: [4:0:0][navi12_psp_v11_set_mb_int:632] psp mailbox disabled for VF0<o:p></o:p> [ 658.415900] gim info libgv: [4:0:0][amdgv_sched_exit_full_access:976] VF0 exited full access.<o:p></o:p> <o:p> </o:p> <o:p> </o:p> <o:p> </o:p> ---------------------------------------------------------------------- <o:p></o:p> BW<o:p></o:p> Pengju Zhou<o:p></o:p> <o:p> </o:p> <o:p> </o:p> -----Original Message-----<o:p></o:p> From: Paul Menzel <<a href="mailto:pmenzel@molgen.mpg.de">pmenzel@molgen.mpg.de</a>> <o:p></o:p> Sent: Friday, December 25, 2020 6:44 AM<o:p></o:p> To: Zhou, Peng Ju <<a href="mailto:PengJu.Zhou@amd.com">PengJu.Zhou@amd.com</a>><o:p></o:p> Cc: <a href="mailto:amd-gfx@lists.freedesktop.org">amd-gfx@lists.freedesktop.org</a>; Deucher, Alexander <<a href="mailto:Alexander.Deucher@amd.com">Alexander.Deucher@amd.com</a>>; Koenig, Christian <<a href="mailto:Christian.Koenig@amd.com">Christian.Koenig@amd.com</a>><o:p></o:p> Subject: Re: [PATCH] drm/amdgpu: reduce the full access time by about 50ms<o:p></o:p> <o:p> </o:p> Dear Peng Ju,<o:p></o:p> <o:p> </o:p> <o:p> </o:p> Thank you for your patch.<o:p></o:p> <o:p> </o:p> Am 24.12.20 um 07:04 schrieb pengzhou:<o:p></o:p> <o:p> </o:p> Could you please configure your name in git:<o:p></o:p> <o:p> </o:p> git config --global user.name "Peng Zhou" # or Peng Ju Zhou<o:p></o:p> <o:p> </o:p> Also, please mention PSP in some way in the git commit message summary. <o:p></o:p> Maybe:<o:p></o:p> <o:p> </o:p> > drm/amdgpu: Reduce delay in PSP command submit by …<o:p></o:p> <o:p> </o:p> > The function msleep(1) can be delay to 10+ ms sometimes, which <o:p></o:p> > contributes a big delay during the full access time.<o:p></o:p> <o:p> </o:p> Do you have the Linux log messages with timestamps, where the delay can be seen?<o:p></o:p> <o:p> </o:p> > Changing msleep(1) to usleep_range(10, 100) and it can reduce about <o:p></o:p> > 50ms delay during full access time.<o:p></o:p> <o:p> </o:p> (Please wrap lines after 75 characters.)<o:p></o:p> <o:p> </o:p> `usleep_range(10, 100)` is 100 �s which is less then 1 ms (= 1.000 �s). <o:p></o:p> What datasheet specifies the needed delays?<o:p></o:p> <o:p> </o:p> > Signed-off-by: pengzhou <<a href="mailto:PengJu.Zhou@amd.com">PengJu.Zhou@amd.com</a>><o:p></o:p> > Change-Id: I151a07c55068d5c429553ef0e6668f024c0c0f3d<o:p></o:p> > ---<o:p></o:p> > drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 2 +-<o:p></o:p> > 1 file changed, 1 insertion(+), 1 deletion(-)<o:p></o:p> > <o:p></o:p> > diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c <o:p> </o:p> > b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c<o:p></o:p> > index 523d22db094b..ef69051681cf 100644<o:p></o:p> > --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c<o:p></o:p> > +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c<o:p></o:p> > @@ -282,7 +282,7 @@ psp_cmd_submit_buf(struct psp_context *psp,<o:p></o:p> > ras_intr = amdgpu_ras_intr_triggered();<o:p></o:p> > if (ras_intr)<o:p></o:p> > break;<o:p></o:p> > - msleep(1);<o:p></o:p> > + usleep_range(10, 100);<o:p></o:p> <o:p> </o:p> With `timeout = 2000`, this was a maximum of two seconds (or even 20 seconds judging from your commit message). With your change it seems the waiting time is reduced to 0.2 seconds.<o:p></o:p> <o:p> </o:p> I do not understand, how you reach 50 ms in the commit message title? <o:p></o:p> Only if the msleep would take 50 ms, which is unlikely.<o:p></o:p> <o:p> </o:p> > amdgpu_asic_invalidate_hdp(psp->adev, NULL);<o:p></o:p> > }<o:p></o:p> <o:p> </o:p> It’s great to see these kind of optimizations, as amdgpu takes 400 ms to load on my system.<o:p></o:p> <o:p> </o:p> In a followup the logging should be improved too. Maybe, print a warning, should it take longer than five milliseconds.<o:p></o:p> <o:p> </o:p> I tested that it still boots on my MSI B350M MORTAR (MS-7A37) with AMD Ryzen 3 2200G, but couldn’t determine if the patch improved the boot time in anyway due to absent logging.<o:p></o:p> <o:p> </o:p> <o:p> </o:p> Kind regards,<o:p></o:p> <o:p> </o:p> Paul<o:p></o:p> </div> </body> </html>